Aller au contenu
Corporate & Investment Banking Mon parcours English

Accueil / Capital Markets & Securities Services / Capital Markets / Module 1

1Equity Capital Markets (ECM)

From Origination to Aftermarket · The Primary Equity Issuance Process

Comment une entreprise vend une part d’elle-même au public, et comment le prix se fabrique.

Lecture
3 h 20
Sections
64
Exercices
25
Questions de rappel
24
Questions de QCM
20
Simulateurs
6

Le texte de ce module est encore en anglais : la version française arrive. Les outils interactifs, eux, sont déjà en français.

Avec qui ce métier travaille
l’émetteur et les investisseurs passent le KYC avant toutClient Onboarding and KYC in Global Markets : l’émetteur et les investisseurs passent le KYC avant tout5KYCles ordres des investisseursSales & Distribution : les ordres des investisseurs6Salesrecherche connectée, derrière le murResearch & Investment Advisory : recherche connectée, derrière le mur7Researchstabilisation et premières négociationsEquity Trading (Cash & Algo) : stabilisation et premières négociations9Equity Tradingengagement de prise fermeRisk & P&L : engagement de prise ferme26Risk & P&LISIN de l’action, LEI des partiesMarket Data & Reference Data : ISIN de l’action, LEI des parties28Market & Ref Dataemprunt de titres pour le greenshoeRepo and Securities Financing : emprunt de titres pour le greenshoe14Repolivraison contre paiement des actions allouéesSettlement — DvP, Fails, and the Move to T+1 : livraison contre paiement des actions allouées20Settlementregistre et dépositaire centralCustody & Asset Servicing : registre et dépositaire central21Custodyallocation justifiée, stabilisation déclaréeRegulatory Reporting : allocation justifiée, stabilisation déclarée29Reg Reportingle produit de l’émission et le bilan engagéTreasury & Funding : le produit de l’émission et le bilan engagé15Treasuryallocations à enregistrerTrade Capture & Booking Models : allocations à enregistrer16Bookingcollatéral du prêt de titresCollateral Management : collatéral du prêt de titres19Collateral1

Ce métier dans l’atelier

This module teaches ECM from zero, but goes deep enough for a professional

It assumes no prior knowledge of investment banking and builds every idea from its foundations. Yet it goes deep enough to be useful to someone who already trades or structures deals for a living.

Wherever a concept is technical, we stop and explain it in plain language before using it again. Wherever a term is jargon, we define it once, clearly, and then rely on it.

Two threads run through everything: the business and the technology

The business thread covers what equity issuance is, why companies and investors do it, how a deal is priced and allocated, and where deals go wrong.

The technology thread covers the systems that hold a deal together, how a single investor order travels front to back, and why data quality at one step decides whether the next five steps work.

These two threads are not separate. In modern capital markets, the process and the platform are the same thing seen from two angles.

How to read this module

Read it as a narrative, not a reference card

Each part assumes you have read the ones before it. If you are short on time, read the section titles and the "In a nutshell" boxes first, then come back for the detail. The titles are written to carry the argument on their own.

Each coloured box plays a fixed role

Knowing what each box signals lets you skim or dive as you choose.

Key insight distils the one idea from a section worth remembering even if you forget everything around it. Going deeper adds nuance for readers who already have the fundamentals. IT & data lens connects the business step to the systems and data beneath it. Worked example grounds an abstract mechanic in concrete numbers, usually through a running case. Watch out flags a common misunderstanding or a place where deals fail. In a nutshell closes a section with a short recap. Common misconception names a plausible wrong answer and dismantles it. Practice sets out calculations and judgement calls to attempt yourself. Explain it back asks you to teach the idea to someone outside finance, which is the sharpest test of whether you have it.

The module is built to be worked, not only read

Reading this document end to end will leave you able to follow a conversation about equity issuance. That is not the same as being able to hold one. The difference is made by four kinds of box that ask something of you rather than telling you something, and skipping them will cost you most of the value here.

Practice boxes contain the calculations and judgement calls the desk actually makes: a theoretical ex-rights price, an allocation across four competing investors, a greenshoe settled in a falling market. Attempt them before looking anything up. Getting one wrong and then finding out why is worth more than reading the correct method three times.

Explain it back boxes ask you to teach the idea to a specific person who does not work in finance. This is the sharpest available test of whether you understand something or merely recognise it, because jargon becomes unavailable at exactly the moment you were about to reach for it. If you cannot get through the explanation without the words the module gave you, you have not finished learning the section.

Common misconception boxes name the plausible wrong answer directly. Most of these are beliefs held confidently by people who work in the industry, which is why they are worth stating rather than leaving to be inferred.

Check your understanding boxes close each part with recall questions. The answers are deliberately not printed beside them; they are collected in the Answer Key near the end of the document. Retrieving an answer from memory is what fixes it. Reading one you have just been shown mostly produces the feeling of knowing, which fades within a day.

A suggested path through the document

The module runs to about 160 pages and is not designed to be read in one sitting. Four sessions works well for most readers.

The first session covers Part 1 and the opening of Part 2, up to the end of preparation. The second finishes Part 2, which is the heart of the module and the one everything else depends on. The third covers Parts 3 and 4, the actors and the systems, which will read very differently depending on whether you arrived here from a business or a technology background. The fourth covers Part 5, the summary and the cheat sheet, and is the right moment to return to the practice boxes from the earlier parts, because spacing the retrieval out is what makes it stick.

Readers who already work in capital markets can treat Part 1 as revision and start at Part 2, with one exception: the sections on equity-linked products and on listing venue are frequently new even to experienced practitioners.

What you will master

By the end of this module you will be able to do five things. Each maps to one part of what follows.

You will understand the ECM process end to end

You will walk through every step from the moment a bank wins a mandate to the moment a newly listed stock trades quietly on its own. You will understand how an IPO differs from a follow-on and from an overnight placement, not just in name but in mechanics.

You will be able to map every actor

You will know who does what inside the bank and outside it, how these parties interact through the life of a deal, and where information barriers between teams turn into real engineering constraints.

You will understand where the price actually comes from

You will know how a book is built, how demand at different prices becomes a single number, which valuation methods set the starting range, and why allocating shares is as much a political decision as a quantitative one.

You will be able to trace data front to back

You will follow a deal record and an investor order through the CRM, the bookbuilding platform, the allocation engine, settlement, and regulatory reporting. You will understand why an error early in that chain is so expensive to fix later.

You will recognise the risk and transformation landscape

You will know the regulations that shape every step, the failure modes that have killed real deals, and the places where better technology genuinely changes the economics of the business.

1Context and the Big Picture

Before any process makes sense, you need the map. This part defines what ECM is, places it inside the bank, and lays out the products, players, and market forces that everything else builds on.

Equity is just selling a slice of the company

A company that needs money has two honest choices: borrow or sell ownership

A company needs capital to grow, and it can raise it from outside in two ways. It can borrow, which means taking on debt it must repay with interest. Or it can sell a piece of itself, which means giving new owners a share of everything the company will ever earn.

That second path is equity. A share of stock is nothing more mysterious than a legal claim on a slice of a company: a slice of its profits, a slice of its assets if it is ever wound up, and usually a vote on how it is run.

ECM is the bank function that turns that idea into a public transaction

Equity Capital Markets, almost always shortened to ECM, sells that equity to public investors. It runs the process when a company sells shares for the first time, when an already-listed company sells more, and when a large holder offloads a block overnight.

The primary-versus-secondary distinction is the one to fix first

A surprising amount of confusion comes from blurring two markets that sound similar.

In the primary market, the company sells newly created shares and receives the cash. The company's bank account grows. This is where ECM mostly lives.

In the secondary market, investors trade existing shares among themselves, and not a cent reaches the company. Your everyday image of the stock market, prices flickering on a screen, is the secondary market. ECM is in the business of creating and placing new shares, not the minute-by-minute trading that follows.

ECM sits at the awkward junction between advisory and trading

A bank is many businesses under one roof

An investment bank is not one thing. It is a collection of businesses, and ECM sits at a specific, slightly awkward junction between two of them.

On one side is advisory, also called coverage. These bankers build long relationships with a company's chief executive and chief financial officer, advise on mergers, and originate the idea that the company should raise capital. Their world is relationships, discretion, and advice.

On the other side is the markets business, the trading floor. These people deal with investors all day, know what the buy-side wants to own this week, and live by execution. Their world is speed, distribution, and price.

ECM is the hinge that carries a deal from one world to the other

An equity raise starts as an advisory idea and ends as a market execution. Someone has to carry it across that gap without dropping it, translating a private strategic decision into a public transaction that hundreds of investors can price and buy.

That someone is ECM. It has to speak the language of the boardroom and the language of the trading desk in the same afternoon, and its systems have to connect a relationship-driven origination world to a fast, data-heavy execution world. This dual identity explains much of why its technology is hard.

Six products cover the entire field

Under the ECM umbrella sit a small number of distinct products. They share DNA but differ enormously in complexity, speed, and risk. Learn these five and you can classify almost any equity deal you will ever see.

The IPO is the first-time listing, and the most complex product there is

An Initial Public Offering is a company selling shares to the public for the very first time and listing on an exchange. The company has never been public, so everything is built from scratch: the disclosure document, the governance, the investor relationships, the first-ever price. An IPO typically takes four to six months and involves a small army of lawyers, auditors, and bankers.

The follow-on is a top-up by a company that is already listed

A Follow-On Offering, also called a Secondary Public Offering or SPO, is an already-listed company selling additional shares. Because the company is already public, much of the hard scaffolding exists: a track record, a market price, existing disclosure. A follow-on is therefore faster, often weeks rather than months.

Watch the word "secondary," which is used in two senses. A secondary public offering can still raise new money for the company; the word describes the timing, that it comes after the IPO. That is different from a secondary sale of shares, where existing shares change hands and the company gets nothing.

The rights issue protects existing owners from being diluted

A Rights Issue offers new shares first to existing shareholders, in proportion to what they already own, usually at a discount. Own one percent of the company and you get the right to buy one percent of the new shares before anyone else.

This protects existing owners from being diluted against their will. Rights issues are common in Europe and often used when a company needs a large amount of capital relative to its size, for example to repair a damaged balance sheet.

The ABB is the overnight sale, priced before the market reopens

An Accelerated Bookbuild, or ABB, is the sports car of ECM. A block of shares is sold to institutional investors overnight, sometimes in hours, with almost no marketing. No roadshow, no new prospectus; it launches after the market closes and prices before it opens.

ABBs are used when speed and discretion matter more than the last cent of price, for example when a private equity fund wants to sell down a stake cleanly without spooking the market over several days.

The block trade moves existing shares and raises nothing for the company

A Block Trade is a large parcel of existing shares sold from one holder to investors. No new shares are created, and the company itself raises nothing. It is a pure secondary sale. The bank often buys the block onto its own books first, at a negotiated price, then distributes it, taking the risk it cannot resell at a profit.

Two axes organize these six products more usefully than any list. The first is whether new shares are created, which decides whether the company raises money at all: IPOs, follow-ons, and rights issues do; block trades do not; ABBs can go either way. The second is speed, which is really a proxy for how much disclosure and marketing the deal needs. Every difference in process downstream flows from where a deal sits on these two axes.

The word "ECM deal" tells you almost nothing about the technology required, and this is the single most important lesson for anyone building or buying ECM systems. The systems needed to run an IPO are roughly ten times heavier than those needed for an ABB: full prospectus workflow, multi-party document control, roadshow logistics, a structured multi-day bookbuild, and a heavy compliance overlay. Before scoping any ECM work, the first question is never "how big is the deal" but "which product is it," because the product sets the entire shape of the platform.

Rights issues deserve their own mechanics

The core idea: offer new shares to existing owners first, at a discount

A rights issue raises equity from the people who already own the company. Each shareholder receives the right to buy new shares in proportion to their current holding, at a price set below the market. The discount is not a gift; it is what makes the offer attractive enough to be taken up without a marketing campaign.

The design solves a fairness problem. If a company simply sold new shares to outsiders, existing owners would see their stake diluted. By offering the new shares pro-rata first, a rights issue lets current owners keep their percentage of the company if they choose to pay.

The rights themselves are tradable, which changes everything

A shareholder who does not want to put in more money is not trapped. In most European rights issues the rights trade separately on the market for a couple of weeks, in what is called nil-paid form. A holder can sell the right to someone else, who then exercises it and pays for the shares.

This is why a rights issue has two prices in play at once: the price of the share and the price of the right. They are linked by simple arithmetic, and that arithmetic gives us the single most useful number in the whole product, the theoretical ex-rights price.

TERP: the price the share should settle at after the raise

The theoretical ex-rights price, or TERP, is the weighted-average price of the old shares and the new discounted shares combined. It is what the share is worth once the new, cheaper shares dilute the old ones. Every party watches it, because the value of a right is essentially the gap between the market price and TERP.

Deep discounts remove the need for underwriting, at a cost

A company that discounts steeply enough almost guarantees take-up. If the subscription price sits far below the market, the rights are so valuable that virtually every holder either exercises or sells them, and the raise succeeds even in a shaky market. This is the deeply-discounted rights issue, common when a company must raise capital urgently, for example to repair its balance sheet.

The cost is optics and dilution. A deep discount means issuing more shares to raise the same amount, so ownership shifts more, and a very deep discount can signal distress.

The rump: what happens to shares nobody took up

Some rights always go unexercised. Holders forget, or cannot pay, or the rights were never sold. The shares those rights would have bought are gathered into a block called the rump, and the banks place the rump with institutional investors after the offer closes. Any proceeds above the subscription price are typically returned to the shareholders who did not take up their rights, which is why even a lapsed holder usually gets something back.

A rights issue is a heavier data exercise than it looks, because entitlements must be tracked per shareholder and then reconciled. The registrar and the depository maintain who was entitled to how many rights, who exercised, who sold, and who lapsed, across a shifting register while the nil-paid rights trade. The system has to handle fractional entitlements, the separate trading line for the rights, and the final rump placement and its proceeds distribution. Corporate-actions platforms, not bookbuilding platforms, carry most of this load, which is why rights issues lean on a different part of the technology estate than an IPO.

Equity-linked completes the set: a convertible is a bond that can turn into shares

The instrument is a loan with an escape hatch into equity

A convertible bond is an ordinary corporate bond carrying one extra clause: the holder may exchange it for a fixed number of shares instead of taking the cash back at maturity. The company borrows at a coupon well below what it would pay on straight debt. The investor accepts that thin coupon because the right to convert may end up worth far more than the interest given up. Each side trades certainty for optionality.

The number of shares a bond converts into is fixed when the deal prices, and is quoted as a conversion price. A bond with a face value of 1,000 that converts into 40 shares has a conversion price of 25. If the share trades at 20 on the day of issue, that conversion price sits 25% above the market, and that gap is the conversion premium. It is the most negotiated number in the deal, because it decides how far the share has to run before the investor's option is worth anything.

Two floors hold the value up, and that is the entire appeal

A convertible is always worth at least the higher of two separate things. The first is its bond floor: what the paper would fetch as plain debt, discounting the coupons and the principal at the issuer's ordinary borrowing rate. The second is its parity: the market value, right now, of the shares it converts into. When the equity collapses, the bond floor catches the instrument. When the equity runs, parity carries it.

In practice the instrument trades above both lines rather than on them. The gap is the value of the embedded option, the right and never the obligation to convert, and it is widest when the share sits near the conversion price, because that is where the outcome is most genuinely uncertain. Far below, conversion is fanciful and the thing is simply a bond. Far above, conversion is certain and the thing is equity in a slow disguise.

Figure 1 A convertible is floored by its value as debt and lifted by its value as equity. The shaded wedge is the option the issuer sells and the investor buys.

The issuer is selling volatility in order to buy cheap funding

An issuer reaches for a convertible when straight equity looks too expensive and straight debt looks too dear. The coupon saving is immediate and visible, often one to two percentage points below what the same company pays on its ordinary bonds. Dilution is deferred, and when it eventually arrives it arrives at a price above where the shares trade today rather than at the discount a follow-on would demand. For a management team convinced its stock is undervalued, that combination is close to irresistible.

The cost is that the company has written an option on its own shares, usually cheaply. If the stock triples, holders convert at a price that now looks absurdly generous and existing shareholders carry the dilution. Treasurers who dislike that outcome buy it back through a call spread overlay, a derivative purchased alongside the bond that raises the effective conversion price. It is an elegant fix and it is not a free one.

The buyers split into two camps that behave nothing alike

Outright funds buy convertibles the way they buy any credit, holding the paper for its coupon and its equity optionality and behaving much like the long-only investors met elsewhere in this module. They are patient, relationship-driven, and on most deals they are a minority of the book.

The other camp is convertible arbitrage. These hedge funds buy the bond and immediately short the underlying share in the ratio the embedded option implies, a quantity called the delta. They are expressing no view whatever on the company. They are buying cheap volatility, hedging away everything else, and trading the hedge as the share moves. The consequence catches first-time issuers by surprise: launching an equity-linked deal creates real, mechanical selling in the underlying stock on day one, because much of the book is putting its short on. That is not the market rejecting the deal. It is the deal working exactly as designed.

Term What it means Why it is negotiated hard
Conversion premium How far above today's share price the conversion price sits Sets how much upside the issuer keeps; commonly 25-40% for a solid credit
Coupon Annual interest paid until conversion or maturity Traded directly against the premium: a richer premium demands a richer coupon
Bond floor Value as straight debt, given the issuer's credit spread Defines the investor's downside, so a weak credit must pay more elsewhere
Implied volatility The volatility the embedded option is priced at The real subject of the negotiation; arbitrage buyers price the deal off this number
Call protection Years before the issuer may force conversion Stops the issuer cutting the investor's option short
Delta Shares to short per bond to be equity-neutral Determines how much hedging flow hits the stock on day one

Equity-linked is the one ECM product whose pricing engine does not live in the equity stack at all. Valuing a convertible needs a credit spread, a volatility surface, a dividend forecast and an option model, which is derivatives infrastructure rather than bookbuilding infrastructure. Banks price these deals on the equity-derivatives platform and then push the result into the same syndicate, allocation and settlement chain as everything else.

That seam has operational consequences. The instrument settles as a bond, with an ISIN of its own and a coupon calendar behind it, while the arbitrage hedge settles as equity. One deal therefore lands in two different downstream systems on two different timetables, and the reconciliation between them is a standing source of breaks.

The numbers that frame the business

A few reference figures make the field concrete

Global IPO volume in 2024 was roughly 123 billion dollars, across about 1,215 listings. That is a moderate, recovering year, not a boom. These 2024 figures remain the last full-year reference as of mid-2026; treat every such number here as illustrative of scale rather than current market data.

The bank's fee, the gross spread, typically runs three to seven percent of the money raised. The higher end applies to smaller, riskier, or more complex deals; the lower end to large, straightforward ones. That fee is the economic engine of the whole business.

IPO, follow-on, and ABB are three genuinely different processes

Line them up and the differences become vivid

An IPO, a follow-on, and an accelerated bookbuild are not three sizes of the same thing. Almost every later part of the module is an elaboration of one of the rows below, so it is worth studying slowly.

Dimension IPO Follow-On (FO / SPO) Accelerated Bookbuild (ABB)
Timeline 4-6 months 2-4 weeks 24-48 hours
Prospectus Full document, 200-400 pages, regulator-approved Supplement only, if already shelf-registered None; existing disclosure is sufficient
Roadshow 2-3 weeks, 80+ investor meetings Limited, 1-3 days None
Bookbuilding 5-10 days, highly structured 1-3 days Overnight, evening to morning
Price discovery Range set through full valuation analysis Discount to the market price, ~3-5% Discount to the closing price, ~3-7%
Typical gross spread 3-7% 2-3% 1-2%
Greenshoe Yes, usually 15% Sometimes Rare
Regulatory complexity Very high Medium Low
Key IT systems Full stack: CRM, bookbuilding, allocation, settlement Bookbuilding plus settlement Minimal: order collection plus settlement

The timeline row explains all the others

An IPO takes months because the company is being introduced to the public for the first time, and the law requires a thorough, approved disclosure document before a single share is sold. A follow-on takes weeks because the company is already known and paperwork can be reused. An ABB takes hours because it leans entirely on disclosure that already exists and skips marketing.

The fee falls as the process speeds up, and that is no accident

Price tracks difficulty in this business. An IPO carries months of work and real reputational risk, so it earns three to seven percent. An ABB is a single night of risk on an already-public company, so it earns one to two percent.

League tables keep score and quietly distort behaviour

A league table ranks banks by how much ECM business they did

League tables are compiled and published by data vendors, principally Dealogic, Refinitiv (now part of the London Stock Exchange Group), and Bloomberg. On the surface they look like harmless scorekeeping. In reality they shape the whole competitive game.

They matter for four connected reasons

They drive mandate wins. A chief financial officer choosing banks looks at the table first, taking a high ranking as proof of capability and market access.

They drive banker pay. Bonuses are tied, formally or informally, to league table position, so bankers care about them personally and intensely.

They can distort pricing. Because the ranking rewards volume, a bank may price a deal cheaply, or take on a marginal deal, simply to earn "credit" for the size and climb the table.

They compound into franchise value. A bank consistently in the top three can charge higher fees and attract better talent, which helps it stay in the top three. Success feeds on itself.

An illustrative ranking shows the shape of the market

Rank Bank ECM volume (illustrative)
1 Goldman Sachs $28.5B
2 JP Morgan $26.8B
3 Morgan Stanley $24.2B
4 BofA Securities $18.9B
5 Citi $15.3B
6 Barclays $12.1B
7 UBS $10.8B

Working out which bank deserves "credit" for a deal is a genuinely hard data problem. A deal may have three lead banks and four junior ones, and splitting the credit is contentious enough that banks employ people to make sure their deals are counted the most favourable way. This feeds internal dashboards and competitive-intelligence tools that pull from vendor feeds such as the Dealogic and Refinitiv APIs. The deal-attribution logic matters far more to banks than its size would suggest.

A traditional IPO is the default, but not the only route

Three alternatives come and go with fashion and market conditions

The SPAC is a blank-cheque shortcut. A Special Purpose Acquisition Company is a listed shell with no operations that raises money on a promise to find a private company to merge with. When it does, that company becomes public without a conventional IPO. SPACs boomed in 2020-21 and collapsed almost as fast as investors soured on their incentives and results.

The direct listing skips the underwriter. A company lists its existing shares without issuing new ones and without banks building a book. It simply starts trading and the market finds the price. Because no new shares are sold, no fresh capital is raised; the point is to let holders sell and to set a public price. Spotify in 2018 and Coinbase in 2021 are the landmark cases, and both were famous enough not to need the marketing machine of a normal IPO.

The dual-track keeps two options open. A company prepares an IPO and a private sale at the same time and decides late, based on which delivers more. It is expensive to run both, but each path keeps the other honest on price.

Alternative What it is Trade-off
SPAC Blank-cheque company merges with a target to take it public Fast and flexible, but reputationally damaged; peaked in 2021, now marginal
Direct listing Existing shares listed with no new issuance and no underwriter Sets a public price and lets holders sell, but raises no new capital
Dual-track IPO and M&A sale prepared in parallel, winner chosen late Maximises leverage on price, but doubles cost and complexity

Where a company lists is a strategic choice, and it sets the rules for everything after

The venue decides the rulebook, the investor base, and the index

A listing venue is not a neutral piece of plumbing; it is a package. Choosing an exchange chooses a disclosure regime, a corporate-governance code, a settlement infrastructure, a peer group the company will be compared against forever, and, most consequentially, an index family. Two companies with identical financials can be worth measurably different amounts depending on where their shares are admitted, and every advisor in the room knows it.

The pull towards the deepest pool is strong. A US listing offers the largest concentration of specialist capital, analysts who follow the sector closely, and valuation multiples that in many sectors simply run higher. The push back is equally real: US disclosure and litigation exposure are heavier, the internal-controls regime is demanding, and a European mid-cap that lists in New York often finds itself the smallest company its new analysts cover, and therefore the least noticed.

Free float is the number that quietly gates admission

Every venue sets a minimum proportion of shares that must sit in public hands, and that proportion, the free float, is one of the few genuinely hard constraints in deal design. Shares held by founders, sponsors and strategic holders do not count. A company that wants to sell only a sliver of itself may discover it is not eligible to list at all, or that it qualifies for a junior segment with a different index treatment.

Free float then goes on mattering long after admission. It drives index weighting, since major index families weight by free-float-adjusted market capitalization rather than by total capitalization. It drives liquidity, and thin liquidity widens spreads, deters institutional buyers and depresses the multiple. A deal structured to minimize dilution can therefore end up destroying more value through illiquidity than it saved through scarcity.

Index inclusion is a demand event the deal team plans for deliberately

Trillions of dollars track indices mechanically, and a fund that tracks an index must hold what the index holds, at the weight the index specifies, regardless of what it thinks of the company. When a newly listed share enters a major index, those funds become forced buyers on a known date. That is the single most predictable source of demand a new listing will ever see.

Index families run fast-entry rules that let a large IPO join at the next scheduled review, or sooner, rather than waiting the customary period. Qualifying turns on size, free float, and the venue and share class involved. Deal teams model this explicitly: the expected index demand shapes how much stock the syndicate is willing to place with flippers, how the lock-up is timed, and how aggressively stabilization defends the price in the weeks before a review date.

Share structure can forfeit the prize entirely. Dual-class shares, which give founders enhanced voting rights, have historically been restricted or excluded by several index families. A founder who insists on control may therefore be trading away a permanent, mechanical bid for the stock.

Consideration What the company is really choosing Where it bites later
Disclosure regime How much it must publish, how often, in what format Ongoing reporting cost, and the internal systems that produce it
Governance code Board composition, committee structure, independence tests Board recruitment, sometimes years before the deal
Free float minimum How much of itself it must genuinely sell Deal size, dilution, and index weighting
Index eligibility Whether passive money can ever own it A large, permanent, price-insensitive block of demand
Peer group and analyst coverage Who it is compared against The multiple, and how well the equity story travels
Settlement infrastructure Which CSD and currency the shares live in Investor accessibility and post-trade cost

The market breathes, and the cycle sets the tempo

ECM is far more cyclical than most of a bank

It swings with the mood of the market, and the swing is essential to understanding why the business, and investment in its technology, expands and contracts.

The recent cycle tells the story cleanly

2021 was a record year. Rates were near zero, tech valuations were stretched, and the SPAC craze was in full flow. Global issuance reached roughly 460 billion dollars on EY's count, close to four times the level the market would fall to two years later.

2022 and 2023 were a sharp correction. Central banks raised rates hard to fight inflation. Higher rates hurt exactly the high-growth companies most likely to go public, investors rotated into safer bonds, and the IPO window slammed shut. Many companies simply waited.

2024 was a gradual, selective recovery, favouring quality over quantity and rebuilding a pipeline for the years ahead.

Figure 2 The ECM cycle: a record window in 2021, a sharp closure through 2022 and 2023, and a selective recovery. Approximate figures, shown to convey the shape of the cycle.

The cycle matters because it sets the rhythm of investment

ECM activity is a leading indicator of where capital markets are heading. When the window is open, banks race to build capacity and platforms. When it closes, the priority swings to cutting cost and consolidating systems. Anyone working in or around this business needs to read the cycle, because it decides whether this quarter's conversation is about growth or efficiency.

Work these through before reading on. Full solutions are in the Answer Key.

1.1 TERP

A company's shares trade at 48. It announces a 1-for-3 rights issue at a subscription price of 32. Calculate the theoretical ex-rights price, and then calculate the value of one right.

For every 3 existing shares at 48, one new share is issued at 32. Four shares now exist where three did before.

TERP = (3 × 48 + 1 × 32) / 4 = (144 + 32) / 4 = 44.

The right to subscribe one new share is worth TERP minus the subscription price, so 44 − 32 = 12. Since three existing shares carry the entitlement to one new share, the value attaching to each existing share held is 12 / 3 = 4.

The two figures answer different questions and are routinely confused. Twelve is what the right is worth to whoever ends up exercising it; four is what each share a holder already owns is carrying.

1.2 Doing nothing

A shareholder in that company owns 3,000 shares and takes no action at all, letting the rights lapse without selling them. Calculate the value they held before the announcement and the value of their holding at TERP afterwards. State in one sentence where the difference went.

Before the announcement: 3,000 × 48 = 144,000.

After, at TERP, having taken no action: 3,000 × 44 = 132,000. The loss is 12,000.

That value did not evaporate. The shareholder was entitled to subscribe 1,000 new shares at 32 when they were worth 44, a gain of 12 each, or exactly 12,000. By lapsing, they handed it to whoever bought the shares instead.

This is the whole point of the rights structure, and the reason the discount matters far less than it appears to. A shareholder who takes up their rights is close to indifferent to how deep the discount is. A shareholder who does nothing is punished in direct proportion to it. In most markets a rump placement recovers some of this for lapsed holders, which softens the outcome without changing the principle.

1.3 Product selection

For each situation, name the product you would recommend and give the single decisive reason. A listed industrial group needs 400 million within 48 hours to fund an acquisition it must announce on Monday. A private equity fund wants to exit 12% of a company it took public two years ago. A loss-making biotech with a volatile share price needs 250 million and believes its stock is badly undervalued. A family-controlled manufacturer needs 900 million, which is more than half its current market capitalisation.

400 million in 48 hours: an accelerated bookbuild. Nothing else is fast enough. The company is already listed, so existing disclosure carries the deal and no prospectus is required, which is precisely what removes the months from the timetable.

A sponsor exiting 12%: a block trade. The shares already exist and belong to the sponsor, so this is a secondary sale. No new shares are created and the company receives nothing, which is the defining feature.

A loss-making biotech that thinks it is undervalued: a convertible. Two reasons converge. The company refuses to issue equity at what it considers a depressed price, and a volatile share makes the embedded option genuinely valuable, so investors pay well for it. Dilution is deferred and arrives at a premium.

900 million against a market capitalisation of under 1.8 billion: a rights issue. At this size relative to the company, existing holders must be offered the shares first, and the family can preserve control by taking up its entitlement rather than being diluted by a placement.

1.4 Reading a comparison

Using the IPO, follow-on and ABB comparison table, explain why the gross spread on an ABB is roughly a quarter of the spread on an IPO. Your answer should refer to at least three specific rows of that table.

The spread compensates work, risk and duration, and the ABB carries far less of all three.

Timeline: 24 to 48 hours against four to six months. The bank commits a full deal team for a fraction of a percent of the time. Prospectus: none against a regulator-approved document of 200 to 400 pages, removing the largest single block of legal and accounting cost. Roadshow: none against two to three weeks and 80 or more meetings, removing most of the distribution effort. Regulatory complexity: low against very high. Key IT systems: order collection and settlement against the full front-to-back stack.

The honest summary is that an IPO fee buys months of process risk and reputational exposure, while an ABB fee buys one night of price risk on a company the market already knows how to value.

  1. In which market, primary or secondary, does the company actually receive cash?

  2. Why do the systems for an IPO cost roughly ten times more than those for an ABB?

  3. Name one way league tables distort behaviour.

  4. What distinguishes a direct listing from an IPO?

1

  1. TechCo's shares change hands on the exchange the day after listing. How much of that money reaches TechCo?

  2. Why is the gross spread on an accelerated bookbuild (ABB) roughly a quarter of an IPO's?

  3. TechCo considers a direct listing instead of an IPO. What would it actually get?

  4. A bank accepts a marginal deal mainly to climb a ranking. Which mechanism is at work?

2The End-to-End Flow

This is the heart of the module. We walk an equity deal from the first client conversation to a stable listing, stopping at each stage to explain what happens, why, who is involved, and what the technology must do. We use an IPO as the reference case because it is the most complete; the shorter products are simplifications of it.

Hold the whole arc in your head before the detail

A full IPO runs four to six months through seven stages

The seven stages are origination and mandate, preparation and due diligence, marketing and roadshow, bookbuilding, pricing and allocation, settlement and listing, and finally stabilisation and aftermarket.

Effort is not spread evenly across them. The first two stages are long and preparatory. The middle three are short, intense, and where the deal is actually made or broken. The last two clean up and support the new stock.

Stage Typical duration What it produces Who leads
Origination and mandate Ongoing Engagement letter, mandate Coverage banking, ECM origination
Preparation and due diligence 8-12 weeks Prospectus, valuation range Legal, compliance, auditors
Marketing and roadshow 2-3 weeks Investor feedback, indications of demand Sales, syndicate, investor relations
Bookbuilding 5-10 days Order book, demand curve Syndicate desk, sales
Pricing and allocation 1-2 days Final price, allocation table Syndicate, ECM, issuer
Settlement and listing T+1 to T+2 Admitted securities, cash delivered CSD, exchange, operations
Stabilisation and aftermarket ~30 days Orderly trading, research coverage Trading desk, research

The pressure spikes hard at bookbuilding and pricing

Picture the intensity of the deal as a curve over time. It rises gently through preparation, climbs through the roadshow, and spikes at bookbuilding and pricing, where the price is set, the shares are allocated, and a mistake cannot be walked back. Then it falls away through settlement.

If you remember one image from this part, make it that spike. The deal is won or lost in the few days around pricing, and everything before it exists to make those few days go well.

Figure 3 Deal intensity across the seven stages: effort and risk peak at bookbuilding and pricing.

Step 1 — Winning the mandate is a competition, not a given

A deal begins in a relationship, long before any shares are sold

Origination is how a bank identifies that a company needs equity, convinces it to raise, and wins the job of running the raise.

It usually starts with a coverage banker, whose role is to maintain a deep relationship with a set of companies in a sector. This banker notices a trigger: the company wants to fund growth, pay down debt, finance an acquisition, or let early investors cash out. Any of these can create a need for equity, and the banker's job is to see it, sometimes before the company has fully articulated it.

The pitch book is the bank making its case in writing

Once there is an idea, ECM origination builds a pitch book that lays out the market backdrop, comparable companies, a preliminary view of valuation, and a proposed structure. It is the bank saying: here is what your company is worth, here is how we would sell it, here is why we are the right bank.

Banks compete in a "beauty contest," and one to three win

The company invites several banks to present, and they compete for the mandate. Each pitches to the chief financial officer and the board, competing on credibility, valuation view, distribution reach, and, inevitably, fees.

The company then selects one, two, or three banks to lead. These lead banks are called bookrunners, and being chosen as one is the prize the whole origination effort was chasing.

Many IPOs are only half a process: the dual track

A private-equity owner selling a business rarely commits to a listing at all. It runs a dual track: preparing an IPO and a trade sale in parallel, right up to the point where one of them has to win. The IPO team builds the prospectus and the equity story while the M&A team runs an auction of strategic and financial buyers. Both processes are real, both are expensive, and only one will complete.

The logic is competitive tension. An auction with a credible listing behind it forces bidders to price against a live public-market alternative rather than against each other alone. The seller keeps optionality until the last possible moment, which is worth a great deal when equity markets can shut in a fortnight.

For the bank, a dual track is a demanding thing to sit inside. The same client is running two mutually exclusive transactions through two different divisions, each with its own diligence stream, its own advisers and its own confidentiality perimeter. Information that is routine on one track can be highly sensitive on the other, and the teams are often walled from each other. The ECM team may spend four months preparing a deal that is cancelled three days before launch because a trade buyer paid more. The fee for that work is typically small or nil, which is precisely why banks fight to be on both tracks rather than one.

The engagement letter fixes the fee, split three ways

The terms are documented in an engagement letter. The fee is the gross spread, split by convention into roughly twenty percent for managing the process, twenty percent for underwriting the risk, and sixty percent as a selling concession for actually placing the shares. There may also be a discretionary incentive fee if the company is pleased with the outcome.

The four terms you will meet constantly
Term Plain meaning
Bookrunner A lead bank that controls the order book, pricing, and allocation. The most powerful role on the deal.
Gross spread The total fee, as a percentage of proceeds, split into management, underwriting, and selling portions.
Greenshoe An over-allotment option, usually up to 15%, that lets banks sell extra shares and later decide whether to keep them.
Syndicate The full group of banks on the deal, sharing distribution and, on underwritten deals, risk.

The technology of origination is relationship technology. Deal pipelines live in a CRM, most often Salesforce for financial services, sometimes Dealogic, sometimes proprietary. Pitch books draw on data tools such as FactSet, Capital IQ, and PitchBook for the comparable-company data. One quiet pain point sits here: fee tracking. Who is owed what, across a syndicate and across management, underwriting, and selling splits, is very often still managed in spreadsheets, with all the version-control and error risk that implies.

The syndicate is a hierarchy, and the fee split reveals who actually runs the deal

One deal, many banks, and a strict order of precedence

A large IPO is never run by one bank alone, and the banks on the cover are not equals. They form a syndicate, and the syndicate is a hierarchy in which position determines both how much work a bank does and how much it is paid. Reading the cover of a prospectus from left to right is reading a ranking.

At the top sits the global coordinator, sometimes two. This bank owns the process: it runs the timetable, controls the order book, drives the pricing recommendation and takes the blame if the deal breaks. Below it, joint bookrunners also see the book and contribute meaningfully to distribution, but they do not own the machinery. Further down, co-lead managers and co-managers bring specific investor relationships, sector credibility or geographic reach, receive far less economics, and in many cases never see the full book at all.

That last point is the one outsiders miss. Being named on a deal and having access to its order book are different things, and the difference is deliberate. The fewer institutions that can see live demand, the less the information leaks.

The fee is one number to the issuer and three payments to the banks

The issuer negotiates a single headline percentage, the gross spread, and experiences it as one cost. Inside the syndicate that number immediately splits into three distinct pots, each paying for something different.

The management fee compensates the work of running the process: the structuring, the documentation, the coordination of lawyers and accountants over months. The underwriting fee pays for balance-sheet risk, and belongs to whoever actually stands behind the shares. The selling concession, by far the largest slice, is paid for placing stock with real buyers, and is allocated according to who brought the demand. A bank that generates orders earns; a bank that lends only its name does not.

Figure 4 The gross spread splits three ways by function, then across the syndicate by seniority. Illustrative percentages.

Discretion at the end keeps everyone honest

Issuers increasingly hold back part of the fee and award it after the deal is done. This incentive fee, or discretionary pool, is typically a slice of the total that the company allocates at its own judgement once the outcome is visible. It rewards the banks whose advice proved right, whose research travelled, and whose sales desks delivered genuine long-only demand rather than easy hedge-fund orders.

The mechanism exists to solve a real problem. Fixed economics agreed at the mandate stage pay a bank identically whether it worked the deal hard or coasted on the strength of others. A discretionary tail restores the link between effort and reward at exactly the moment when effort is hardest to observe.

Fee component What it actually pays for Who captures it
Management fee Running the process end to end: structuring, documents, timetable Weighted heavily to the global coordinator
Underwriting fee Bearing the commitment risk on unsold shares Split in proportion to underwriting commitments
Selling concession Placing stock with genuine buyers Allocated by demand actually generated
Incentive / discretionary fee Judged quality of advice and distribution Awarded by the issuer after pricing
Expenses Legal, accounting, printing, roadshow logistics Reimbursed by the issuer, not a margin

League-table credit is a data problem before it is a marketing one. When six banks each claim to have led the same transaction, the data vendors that compile the tables apportion credit by a published methodology, and banks maintain dedicated teams to submit, contest and reconcile those submissions. The bank's own deal database is the source of truth for that process, which is why the CRM record from Step 1 has consequences that outlast the deal by years. A mandate recorded with the wrong role or the wrong economics quietly costs ranking.

How the bank actually takes risk

"Underwriting" means the bank stands behind the deal, but not always the same way

The word underwriting hides a spectrum, and where a deal sits on that spectrum decides who loses money if it goes wrong. Two arrangements sit at the ends.

Firm commitment: the bank guarantees the money

In a firm-commitment, or bought, deal, the bank agrees to buy the shares itself and then resell them. The company is guaranteed its proceeds the moment the deal is struck. If the bank cannot resell at the expected price, that loss is the bank's, not the company's. The bank is taking real principal risk onto its own balance sheet.

Best efforts: the bank promises to try, not to deliver

In a best-efforts deal, the bank agrees only to use its best efforts to place the shares. If demand falls short, the deal is scaled back or pulled, and the shortfall is the company's problem, not the bank's. The bank earns a smaller fee because it is taking less risk.

Most large IPOs are underwritten on a firm-commitment basis, but the commitment usually only becomes firm at the very end, at pricing, once the book already shows the demand is there. The bank is rarely as exposed as the label suggests, because it does not commit until it can see the buyers.

The bought deal and the ABB: risk taken deliberately, overnight

An accelerated bookbuild or a block trade often is a genuine bought deal. The bank buys a large block from a seller at an agreed price, usually a small discount to the market, and then has to resell it, sometimes within hours. Here the principal risk is real and immediate. If the market drops before the bank has placed the block, the loss lands on the bank.

This is why banks compete for these trades on price but price them carefully. Winning an ABB by offering the seller too generous a level, then failing to place the block, is one of the fastest ways for a markets desk to lose money.

Sub-underwriting spreads the risk, especially in rights issues

A single bank rarely wants to carry the whole backstop on a large rights issue, so it lays the risk off. It recruits sub-underwriters, usually large institutional investors, who agree, for a fee, to buy any shares left unsold. The bank has effectively bought insurance against the rump, paying part of its fee to the institutions that provide the backstop.

Arrangement Who bears the risk Typical use
Firm commitment / bought deal The bank buys and resells; loss is the bank's Most IPOs (committed at pricing), ABBs, block trades
Best efforts The company; deal scaled or pulled if short Smaller or riskier deals, some follow-ons
Sub-underwritten backstop Spread to institutional sub-underwriters for a fee Rights issues, large capital raises

Before a bank commits, the deal runs an internal gauntlet

A bank does not simply say yes to a mandate. Before it puts its name and its balance sheet behind a deal, the transaction has to clear a series of internal gates. This governance is invisible from outside, but it shapes which deals happen and how fast.

The bank is underwriting its own reputation, so it vets the deal first

Every deal a bank leads carries its name into the market and its balance sheet into the underwriting. If the company turns out to be fraudulent, or the prospectus misleads, or the stock collapses, the bank shares the blame, the liability, and the reputational damage. So the bank scrutinises a deal before agreeing to run it, sometimes as hard as the investors will later.

Two committees gate the decision

A new-business or commitment committee decides whether to pursue the deal at all. Senior bankers and control functions weigh the client, the sector, the reputational fit, and the odds of success. A deal that looks risky to the franchise can be declined here, however large the fee.

An underwriting or commitment committee decides the terms of the risk. When the bank is about to commit its balance sheet, at pricing on a bought deal, or when agreeing a backstop, a separate committee signs off on the price, the size, and the exposure. This is where the firm formally accepts the principal risk discussed earlier.

Onboarding and KYC gate the client

Before any of this, the client has to be onboarded. Know-your-customer and anti-money-laundering checks establish who the company and its owners really are, and screen them against sanctions and financial-crime risk. A company that cannot clear onboarding never reaches a committee, and slow onboarding is a common, unglamorous cause of deal delay.

Gate Question it answers Who owns it
KYC / onboarding Who is this client, and are they acceptable? Compliance, operations
New-business committee Should we pursue this deal at all? Banking seniors, control functions
Underwriting committee On what terms do we commit our balance sheet? Risk, capital markets, finance

Deal governance is a workflow-and-audit problem, which is why it shows up in the technology estate. Each gate is an approval step that must be requested, evidenced, decided, and recorded, and regulators expect the record to be reconstructable. Onboarding runs on KYC and screening platforms; committee approvals run on workflow tools with a documented decision trail. The same immutability principle from the data chain applies here: a sign-off that could be edited after the fact is worthless as evidence that the bank did its checks.

Step 2 — Preparation turns the company into an approved, disclosable case

The longest phase assembles everything the market will need

With the mandate won, the deal enters its longest phase. For eight to twelve weeks, a large team assembles everything the market needs to evaluate the company, under legal conditions that make accuracy a matter of liability, not just professionalism.

Three streams of work run in parallel, and it helps to see them as three separate crafts happening at once.

The legal and regulatory stream produces the prospectus

Lawyers draft the prospectus, prepare the listing application, and manage filings with the relevant regulators. Alongside it, they conduct legal due diligence, combing through contracts, litigation, and corporate structure to make sure nothing material is hidden.

The financial and valuation stream builds the numbers and the narrative

This stream runs financial due diligence and audit, develops the equity story, and builds the valuation. The equity story is the narrative about why the company is worth owning. It is not marketing fluff; it is the frame through which every number will be read, and getting it right is a genuine skill.

The marketing and distribution stream sets up the intense weeks to come

This stream identifies which investors to target, prepares the marketing materials, plans the roadshow logistics, and sets the media strategy. It starts later and runs shorter, but it prepares the sprint.

The prospectus is the document everything hinges on

A prospectus is a legal document, often 200 to 400 pages, that describes the company completely and honestly to investors. It covers the business, strategy, financials, risks, management, ownership, and the terms of the offer. A regulator must approve it before shares can be sold.

Understand why it matters so much. The prospectus is the legal basis on which investors decide to buy. If it misstates something material or omits it, investors who lose money can sue, and the banks share in that liability. It is a marketing tool and a legal shield at once, and those purposes pull in opposite directions, which is part of why it is so laborious to draft.

The prospectus is a document-control nightmare and therefore a technology problem. Ten or more parties edit it at once: the company, two or three banks, two sets of lawyers, the auditors. Everyone must work from the current version, changes must be tracked, and the filed version must be provably correct. This is why specialised platforms exist for exactly this job, such as Workiva and DFIN's ActiveDisclosure, and why virtual data rooms like Intralinks and Datasite hold the underlying material under tight access control. Version control across many parties is one of the defining pain points of the preparation phase.

Preparation workstream Core deliverables Typical duration
Legal and regulatory Prospectus, legal due diligence, filings, listing application 8-12 weeks
Financial and valuation Financial due diligence and audit, equity story, valuation 6-10 weeks
Marketing and distribution Investor targeting, materials, roadshow logistics, media strategy 4-6 weeks

Becoming listable: the company itself has to change, not just its paperwork

A private company is not a public company with a prospectus attached

Preparation is usually described as producing documents, which understates it considerably. A public company is a different kind of organisation from a private one, and much of the work in this phase is rebuilding the company so that it can survive the obligations it is about to accept. The prospectus is the visible output. The rewiring underneath is the harder part, and it is the part that most often sets the timetable.

The gap is easiest to see in the reporting calendar. A private company may close its books over several weeks and answer to a handful of shareholders who will wait. A listed company must publish results within a fixed window, to a fixed standard, knowing that a delay is itself a disclosable event that moves the share price. Closing the books in five days rather than five weeks is not a documentation exercise. It is a finance-function rebuild with a systems programme inside it.

The board is rebuilt before the deal, not after

Governance codes require a board that looks nothing like a founder-led private company's. A meaningful proportion of directors must be genuinely independent. An audit committee must exist, must be chaired by someone with real financial expertise, and must be capable of challenging management rather than agreeing with it. Remuneration and nomination committees follow.

Recruiting those people takes six to twelve months, and it cannot be rushed at the end, because investors read the board as a signal of how seriously the company takes its new obligations. Companies that arrive at a roadshow with a thin board are asked about it in every meeting.

Reporting, controls and the audit trail all move up a level

Accounting standards frequently have to change, because the framework a private company reports under is often not the one its listing venue accepts. Converting the historical accounts, restating comparatives and having them re-audited is months of work performed under deadline. Where it goes wrong, it goes wrong loudly: a restatement discovered late in the process can delay a deal by a quarter.

Internal controls over financial reporting come next. The company must be able to demonstrate, and its auditors must be able to attest, that the numbers it publishes are produced by a controlled process rather than by capable people working heroically in spreadsheets. This is where the readiness work turns unambiguously into a technology programme: consolidation systems, access controls, segregation of duties, evidenced approvals, and a data lineage clean enough to survive an audit.

Sustainability and non-financial reporting has become the newest addition to that list, and the most underestimated. The data required sits outside finance, in operations and human resources and the supply chain, in systems that were never built to be audited. Companies routinely find that their environmental data cannot be assured to the standard their disclosure now demands.

Readiness workstream What it really requires Typical lead time
Board and committees Independent directors, audit committee with financial expertise 6-12 months
Accounting conversion Restated and re-audited historical accounts 4-9 months
Financial close Closing the books in days, not weeks 6-18 months
Internal controls Documented, tested, evidenced control environment 9-18 months
Investor relations IR function, disclosure policy, results calendar 3-6 months
Non-financial reporting Auditable sustainability and governance data 9-24 months

This section is where the technology thread runs directly through the business thread rather than alongside it. Every readiness requirement is a systems requirement in disguise: a fast close needs an integrated consolidation platform, evidenced controls need workflow and immutable approval logs, disclosure obligations need a document and filing pipeline, and assured non-financial reporting needs data collection from systems that were never designed to produce audited output. A bank assessing whether a client is genuinely ready is, in substance, running a technology due diligence.

Interlude — The price range comes from three lenses on value

A range is not guessed; it is triangulated

Before the deal goes to market it needs a price range, an indicative band such as 19 to 23 euros per share. That range comes from valuation, and valuation in practice means looking at the company through three lenses and triangulating. No single method is trusted alone, because each is strong where the others are weak.

Lens one — DCF values the company on its own fundamentals

Discounted cash flow, or DCF, values a company as the sum of all the cash it will generate in future, discounted to today. You project free cash flows for several years, estimate a terminal value beyond that, and discount everything back using a rate that reflects the risk, the weighted average cost of capital, or WACC.

Its strength is rigour; its weakness is fragility. DCF values the company on fundamentals rather than on today's mood. But the answer is exquisitely sensitive to assumptions, and a small change in growth, discount rate, or terminal value swings it widely. For young, fast-growing companies, where the cash is mostly distant, DCF becomes especially fragile.

Lens two — Comps price the company against its listed peers

Trading comparables, or "comps," value a company using the market multiples of similar listed companies. Take a set of peers, calculate multiples such as enterprise value to revenue or to EBITDA, and apply them to your company. If peers trade at ten times revenue and your company has 300 million in revenue, comps suggest about three billion.

Comps usually anchor the range, because they are grounded in real, current prices and are intuitive to the investors who will buy. Their weakness is peer selection and market mood: the wrong peers or a euphoric market distort the answer.

Lens three — Precedents show what buyers have actually paid

Precedent transactions value a company using the multiples paid in recent deals in the same sector. They capture what buyers have really been willing to pay, including any premium for control. The weakness is that transaction data can be stale, and every deal has unique circumstances.

The three lenses combine into one range

Banks run all three and combine them with judgement. Comps anchor the range because they are the most market-grounded. DCF tests whether fundamentals support the upside. Precedents suggest the ceiling, since acquisition multiples usually include a control premium a public offering will not. Where the three overlap, you have your range.

Method What it captures Strength Weakness
Discounted cash flow Intrinsic value from projected cash Forward-looking, independent of sentiment Highly assumption-sensitive; fragile for high-growth firms
Trading comparables What the market pays for peers now Market-based, intuitive, real-time Depends on peer choice; distorted by market mood
Precedent transactions What buyers have paid in deals Reflects real willingness to pay, captures premiums Data can be stale; every deal is idiosyncratic
Figure 5 Triangulating a price range from three valuation lenses.

For all its intellectual weight, valuation almost always lives in Excel, and that is not a criticism; spreadsheets are flexible in ways that matter for bespoke analysis. The data comes from Bloomberg, FactSet, and Capital IQ. The realistic automation opportunity is not to replace the model but to auto-populate the tedious parts, above all the comparable-company tables that must be refreshed constantly as peer prices move.

Step 2¼ — The analysts speak before management does

Research comes first, and it is written from the far side of the wall

Long before the chief executive meets a single investor, the banks' research analysts do. The sequence is deliberate and it is regulated. The company presents its business to the analysts of the syndicate banks at an analyst presentation, a full day of detail with no marketing gloss on it. Those analysts then go away and write independent research on the company: their own forecasts, their own valuation, their own view.

Because these analysts work for banks running the deal, their research is called connected research, and it is fenced about with rules. The analysts sit on the public side of the information barrier and must be formally brought over, or wall-crossed, to receive the deal information at all. Their published views must be their own, must not be shown to the deal team for approval, and must not be edited by the issuer. A bank that lets its syndicate desk influence a connected analyst's number has committed one of the more serious offences available to it.

The regulatory pressure has run one way for a decade. Rules separating research from investment banking have hardened, connected research has grown more constrained, and some jurisdictions have moved to open the analyst presentation to unconnected analysts as well, so that the company's story is not framed exclusively by the banks it is paying.

Pre-deal investor education builds the price expectation before any price exists

Armed with published research, the analysts go to see investors. This is pre-deal investor education, universally abbreviated to PDIE, and it typically runs one to two weeks. Management is nowhere near it. The analyst walks the investor through the sector, the company, the model and a valuation, and then does the thing the exercise actually exists for: listens.

The feedback is collected systematically. Which parts of the equity story landed and which were met with scepticism. Where each investor's own valuation lands, and on what multiple. Whether they would participate at all, and at what sort of level. Those responses flow back to the syndicate desk and are aggregated into a picture of where real demand might sit.

The price range is built from this feedback, not from the valuation work alone. The three lenses of the previous section produce a defensible theoretical range. PDIE tells the bank what the market will actually pay. Where the two disagree, PDIE wins, because a range that ignores investor feedback produces an undersubscribed book and a failed deal.

The quiet period follows, and it is a hard stop

Once the deal is formally announced, connected analysts stop publishing and stop speaking about the company. This blackout or quiet period runs through the offering and for a defined window after listing, to prevent the bank's research from being used as a marketing instrument at the moment it would be most effective. Coverage resumes only after that window closes, which is why a newly listed company can go weeks with no analyst commentary at all.

Stage Who is in the room What it produces Typical duration
Analyst presentation Company, connected analysts Analysts' independent models 1 day
Research publication Analysts alone Connected research with valuation ranges 1-2 weeks later
PDIE Analysts and investors Feedback on story and valuation 1-2 weeks
Range setting Syndicate desk, issuer The published indicative price range 1-2 days
Blackout Nobody publishes Silence through the offering Announcement to post-listing

PDIE is the point where the information barrier becomes an access-control problem with a clock on it. Analysts must be wall-crossed to work on the deal and walled off again afterwards, each crossing logged with a timestamp, an approver and a scope. Their entitlements to deal systems have to be granted for a defined window and revoked cleanly at blackout, and the surveillance systems must be able to prove, months later, exactly who could see what and when. Manual wall-crossing registers kept in spreadsheets are still surprisingly common, and they are exactly the control a regulator asks to see first.

Step 2½ — Marketing turns a prepared deal into real demand

Between a finished prospectus and an open order book sits the marketing phase, where the bank tests and builds appetite. It is short, but it is where a deal quietly succeeds or fails before bookbuilding even starts.

Pilot fishing tests the water before anyone is committed

Well before the formal launch, the banks sound out a handful of trusted investors in confidence. This is pilot fishing, or pre-sounding. The banks describe the deal, float a valuation, and listen. Are these investors interested? At what price does their enthusiasm fade?

Because this shares deal information before the deal is public, the investors are wall-crossed, brought formally over the Chinese Wall and made insiders, with the crossing logged by compliance. Pilot fishing is how banks avoid the humiliation of launching a deal the market does not want. In the TechCo case, pilot fishing revealed strong appetite but real resistance above nine times revenue, which shaped the range before a single formal order arrived.

Cornerstone and anchor investors give a deal its backbone

A cornerstone investor commits to a fixed allocation before the book even opens, and agrees to be named. This is powerful. A respected fund publicly taking, say, a hundred million euros of the deal tells the rest of the market that a sophisticated buyer has done the work and is convinced. Cornerstones are especially common in Asian and European IPOs.

An anchor investor is a softer version: a large, high-quality order placed early in the bookbuild that anchors the book and pulls others in, though usually without the public naming or the lock-up that a cornerstone accepts. Both exist to solve the same problem, the fear of being first, by making sure someone credible has already committed.

The roadshow is the management sales tour

For two to three weeks, the company's management presents to investors city by city. London, Frankfurt, Amsterdam, New York, Boston: an IPO roadshow can run to eighty or more meetings. Management tells the equity story, investors probe it, and the banks gather the feedback that will shape the final range. The roadshow is exhausting and it is decisive, because institutional investors are buying the management team as much as the numbers.

Deals are marketed across jurisdictions under different rulebooks

A global offering is not sold the same way everywhere, because securities law differs by country. Two provisions of United States law shape how almost every international equity deal is structured, and they are worth knowing by name.

Rule 144A opens the US institutional market without a full US listing

Rule 144A lets shares be sold privately to large, sophisticated US institutions, called QIBs, without the full registration a public US offering demands. A QIB, a qualified institutional buyer, is an institution big enough that regulators assume it can look after itself. Through a 144A tranche, a European IPO can tap deep US institutional money without becoming a fully SEC-registered US listing.

Regulation S covers the offering outside the United States

Regulation S is the counterpart that governs offers made outside the US. In practice a typical cross-border deal is marketed as a "Reg S / 144A" offering: Reg S to investors in Europe, Asia, and elsewhere, and 144A to US QIBs in parallel. The structure lets one deal reach global demand while staying inside each jurisdiction's rules.

Testing-the-waters lets some issuers gauge demand early

In the US, reforms let certain issuers hold exploratory conversations with institutional investors before formally filing. Testing-the-waters is the American cousin of pilot fishing, a way to gauge appetite early without committing to a public process that might then have to be abandoned in view of the whole market.

Marketing tool What it does Why it matters
Pilot fishing / pre-sounding Confidential sounding of a few wall-crossed investors Catches a weak deal before it launches publicly
Cornerstone investor Named, locked-up commitment before the book opens Signals quality; de-risks the launch
Anchor investor Large early order that anchors the book Pulls in followers without public naming
Roadshow Two to three weeks of management meetings Sells the equity story; sets the final range
Reg S / 144A structure Parallel offering under EU and US rules Reaches global demand within each jurisdiction

Step 3 — Bookbuilding is where the price is actually discovered

This is the centre of gravity of the whole deal

Bookbuilding is the process of collecting investor demand at different prices and using it to discover the single price at which the deal should be done. Understand bookbuilding and you understand ECM; almost everything else supports this step.

The logic, built from nothing

The company and its banks have a range, but a range is not a price. To find the price, they need to know how many shares investors will buy at each level. So they ask.

Over several days, the banks' sales teams contact institutional investors and invite orders. Each order specifies how many shares the investor wants and how price-sensitive that demand is. As orders arrive, the syndicate desk aggregates them into the book, which shows demand by price, investor type, and geography, updated continuously.

The book produces a demand curve, and guidance steers it

From the book emerges a demand curve, the familiar downward slope: more demand at lower prices, less at higher ones. If demand is strong across the range, the banks issue price guidance to nudge expectations upward, narrowing the range toward its top and testing whether the strength holds.

When the book closes, the banks calculate the oversubscription ratio, how many times over the deal could have been sold. A healthy IPO is often three to ten times covered. Under about 1.5 times, the deal is in trouble.

Figure 6 The order book read as a demand curve. Every candidate price implies a different level of cover, and the syndicate chooses a point on this line rather than a number in isolation. Illustrative.

Read the curve from right to left and the pricing decision becomes obvious. At the top of the range, only the most convinced buyers remain and the book barely covers. Move left and demand accumulates quickly, because each step down brings in the investors whose limits sat just below. The syndicate is not choosing a price so much as choosing where on this curve it wants to stand, and every position on it trades proceeds against safety.

The curve also explains why the shape of the book matters more than its height. Two deals can both be three times covered and be in completely different states. One is covered by orders that hold firm all the way to the top of the range, meaning the curve is flat and the price could go higher. The other is covered only because demand piles up at the bottom, meaning the curve is steep and any push upward will collapse it. The multiple is identical; the deals are not.

Order types reveal conviction, not just size

The strike order signals the strongest conviction

A strike order, or market order, says "I will buy at whatever the final price turns out to be." It carries no price limit, and banks value it highly, because a buyer willing to pay any price within reason truly wants the stock.

The limit order is rational but more cautious

A limit order says "I will buy, but only at or below a specific price." It is price-sensitive by design, a perfectly reasonable order that nonetheless signals more caution than a strike.

The step order maps out one investor's own demand curve

A step order specifies different quantities at different prices, for example more shares if the price is lower. It is a graduated bid.

Order type What it says Signal to the bank
Strike (market) order Buy at any final price Strongest conviction; most valued
Limit order Buy only at or below a set price Price-sensitive; more cautious
Step order Different quantities at different prices Graduated demand; nuanced

Bookbuilding runs on specialised platforms. Ipreo, now part of S&P Global, is the market standard; Dealogic is widely used; and the top banks often run proprietary systems. The platform aggregates orders in real time from many sales people talking to many investors, deduplicates them, and presents the book by price, investor type, and geography. Data quality here is mission-critical in the most literal sense. A single mistyped order, wrong quantity or wrong type, flows straight into the allocation decision and can cause a real, expensive error.

Step 4 — Pricing and allocation is the most consequential night

Two decisions remain, and both are settled the night before trading

With the book closed, the banks must set the final price and decide who gets the shares. These are the most consequential and most delicate choices of the entire deal.

Pricing balances three forces at once

Setting the final price is a balancing act.

The demand curve sets the floor. At what price does demand still comfortably cover the deal, with a safe cushion?

Investor quality firms or softens the price. A book of long-only holders, who tend to hold, supports a firmer price than one full of hedge funds who may sell on day one.

The aftermarket target caps the ambition. The banks deliberately leave a little on the table, pricing so the stock rises perhaps ten to fifteen percent on day one. Price too high and there is no pop, the stock may break its issue price, and everyone remembers a failure. Price too low and the company left money behind. The sweet spot is a modest, healthy gain.

The deliberate first-day pop, often called underpricing, has been studied for decades. To the issuer it is a cost: the shares could have been sold higher. To the banks it rewards the investors who took early risk and keeps them coming back. The honest reading is that a modest pop is the price of a durable, repeatable market. A market where every IPO priced to perfection and half fell on day one would frighten investors away from IPOs altogether. The pop is a subsidy that keeps the primary market functioning, paid by issuers, collected by investors, brokered by banks.

Underpricing is structural, not a mistake that keeps happening

New issues are priced below where they trade almost everywhere, in almost every decade, and that persistence is the clue. If underpricing were simply banks getting the number wrong, the errors would cancel out over time. They do not. Something in the structure of the transaction produces it, and three explanations do most of the work.

Uninformed buyers demand a discount to show up at all

Some investors know more about a given company than others, and the informed ones bid only for deals they judge cheap. An uninformed investor therefore faces an unpleasant asymmetry: when a deal is genuinely attractive, informed money competes and the uninformed investor is scaled back to almost nothing; when a deal is poor, informed money stays away and the uninformed investor gets everything asked for. Filling every order in full is a bad sign, and being cut hard is a good one.

An investor who understands this will not participate at fair value, because the average deal they actually receive is worse than the average deal offered. The discount exists to make participation worthwhile despite that adverse selection. This is the winner's curse applied to new issues, and it explains why a market that needs breadth of demand cannot price at fair value.

The discount is what the bank pays investors for the truth

Bookbuilding asks investors to reveal what they really think a company is worth, at a moment when saying "I would pay more" is directly against their own interest. Nothing compels honesty. An investor who understates their enthusiasm gets a cheaper price.

The discount is how the bank buys candour. Investors who bid aggressively and early, and whose orders help the syndicate move the price upward, are rewarded with generous allocations in an underpriced deal. Those who hang back and bid at the bottom are punished with poor allocations. Underpricing is the currency in which the bank pays for the information that made accurate pricing possible in the first place, which is why the range so often moves up during a strong bookbuild rather than being set correctly at the start.

A quiet first day is worth paying for

A stock that closes below its offer price on day one damages people who cannot easily be repaired. The investors who bought are immediately underwater and remember it at the follow-on. The employees who watched the listing conclude the company is worth less than they were told. The company's next visit to the equity market is measurably harder.

Against that, the cost of a modest discount is a one-off transfer to the initial buyers. Issuers who dislike the arithmetic are entitled to push, and some do. Most conclude that buying a stable aftermarket and a satisfied shareholder base is cheap at the price.

The three explanations are complements rather than rivals, and they point at different remedies. If underpricing is mostly adverse selection, better disclosure narrows it. If it is mostly payment for information revealed, better price discovery mechanics narrow it. If it is mostly insurance against a broken listing, only a calmer market narrows it.

This is also the honest answer to the question of whether digital bookbuilding platforms will eliminate the discount. They attack the second explanation and to some extent the first, by making demand visible earlier and to more people. They do nothing whatever about the third. Expect the discount to compress, not to disappear.

Allocation is where four interests collide

Allocation decides who receives shares, and how many, from a book that asked for several times what is available. It is inherently political, because the parties want different things and cannot all be satisfied.

The issuer wants stable, long-term shareholders who will not flip on day one. The banks want to reward their best clients, because those relationships drive future fees. Investors want the largest allocation at the lowest price. And regulators want the whole thing demonstrably fair, a concern embodied in MiFID II.

The criteria that decide who gets what
Allocation criterion Typical weight Why it matters
Investor quality (long-only vs hedge fund) High Long-only holders are stable and less likely to flip
Order timing (early vs late) Medium Early orders show conviction and build momentum
Price sensitivity (strike vs limit) High A strike order signals the strongest, least price-sensitive demand
Relationship value to the bank Medium Rewarding top clients drives future business, within fairness rules
Geographic diversification Low to medium The issuer may want a broad, international base

Retail tranches and the clawback that balances them

Many offerings split the deal between institutions and the public. Managing that split, and rebalancing it when demand is lopsided, has its own mechanics worth knowing.

A deal is often carved into an institutional and a retail tranche

When an offering includes the public, it is usually divided into two pools. The institutional tranche goes to the funds and investors we have met throughout; the retail tranche is reserved for individual investors, often through their banks or brokers. The split is set in advance, for example ninety percent institutional and ten percent retail.

The retail tranche serves a purpose beyond raising money. It builds a public constituency, spreads the shares widely, and, for a well-known consumer brand, turns customers into shareholders. It also adds distribution complexity and, in some markets, real political sensitivity about fairness of access.

The clawback rebalances the tranches when demand is lopsided

Demand rarely arrives in the exact proportions the tranches were set at. Sometimes retail investors pile in far beyond their reserved slice; sometimes institutions do. A fixed split would then leave one group heavily oversubscribed while the other has room to spare.

The clawback is the mechanism that shifts shares between the tranches to follow the demand. If the retail tranche is many times oversubscribed while the institutional book is merely covered, a clawback moves shares from institutions to retail, enlarging the public allocation. It can run the other way too. The triggers and step-sizes are set out in advance in the offer terms, so the rebalancing is rules-based rather than improvised.

Clawbacks are a data and reconciliation exercise under time pressure. The two tranches are often collected through different channels, institutional orders through the bookbuilding platform, retail through banks and brokers feeding a separate aggregation system. To fire a clawback correctly, both books must be captured, reconciled, and combined in near real time on pricing night, then both re-scaled consistently. A mismatch between the two systems is exactly the kind of error that cascades into wrong allocations and confirmations.

Steps 5 and 6 — Settlement, listing, and stabilisation make the deal real

The price is set, but the deal is not done

The shares must be created and delivered, the cash must move, the stock must start trading, and for a month the banks must support it. These steps are less dramatic than pricing, but they are where a deal becomes real and where operational risk concentrates.

Settlement moves shares against cash, simultaneously

Several things now happen in a tight sequence. New shares are created and registered in a central securities depository such as Euroclear or DTCC, and assigned an ISIN, the unique code for the security. Investors are told what they have been allocated, through sales contacts and formal SWIFT messages (the securities messaging standard now migrating from legacy MT to ISO 20022). They then pay, moving cash through large-value systems such as TARGET2 or Fedwire. Settlement now completes fast: the United States moved to a T+1 cycle, one business day after the trade, in May 2024, and Europe and the UK are set to follow in 2027, which compresses the window in which anything can go wrong.

The mechanism that ties shares and cash together is delivery versus payment, or DVP. Shares and cash change hands simultaneously, so neither side is exposed to the other defaulting. It sounds obvious, but building systems that guarantee this simultaneity across institutions is serious infrastructure. Finally, the stock is admitted to trading, and the first trade sets the opening price, the number the whole market watches.

Stabilisation supports the stock for about thirty days

For roughly thirty days the lead bank acts as stabilisation manager, with three tools.

Buying shares to support the price

The bank can buy shares in the open market at or below the issue price to support the stock. This is one of the very few situations where deliberately supporting a price is legal rather than manipulation, precisely because it is disclosed and tightly bounded.

The greenshoe, an elegant two-in-one instrument

The banks deliberately sell up to fifteen percent more shares than the base deal, going temporarily short. If the stock trades up, they exercise the greenshoe, buying those extra shares from the company at the issue price to cover the short, and the extra capital goes to the company. If the stock trades down, they buy shares back in the open market to cover, and that buying supports the price. The greenshoe is thus both a way to upsize a success and a built-in shock absorber.

Figure 7 One instrument, three outcomes. The greenshoe resolves whichever of them the market delivers.

None of this works without a stock borrow

Stabilisation and the greenshoe both rest on a mechanism the previous pages quietly assumed: the bank can sell shares it does not own. Selling 115 million shares when 100 million exist leaves the syndicate short 15 million, and a short position has to be covered on settlement day whatever the market does. The greenshoe covers it only if the shoe is exercised. In the case where the stock falls and the shoe lapses, the shares must come from somewhere in the days before the bank has bought them back.

They come from a stock borrow. A large existing holder — on an IPO, typically the selling shareholder or a founder — lends shares to the syndicate under a securities lending agreement, against collateral and for a fee. The bank delivers the borrowed shares to settle its over-allotment on time, then returns them later, either from market purchases made during stabilisation or from the shares delivered when the shoe is exercised.

The borrow is what converts a timing mismatch into a manageable position. Without it, the over-allotment could not settle, because delivery obligations do not wait for the stabilisation period to conclude.

Borrow scarcity is a real constraint, not a formality

A newly listed company has almost no lendable stock. Everything is locked up, held by the sponsor, or sitting with investors who have just received it and have no lending programme in place. The borrow therefore has to be arranged in advance, with a named lender, as part of the deal structure. Where it cannot be arranged in size, the over-allotment is reduced accordingly, which is one of the less visible reasons a greenshoe is sometimes smaller than the customary 15%.

The same constraint governs the equity-linked market from Part 1 in a more acute form. Convertible arbitrage funds cannot buy the bond unless they can short the underlying share, and they cannot short it unless the borrow exists. An issuer whose stock is hard to borrow will find its convertible priced materially worse, or find the deal simply cannot be done at the size intended. The availability of borrow is, in that market, a component of the price.

Securities lending introduces a second, parallel settlement chain running alongside the deal, with its own instructions, its own collateral movements marked to market daily, and its own recall risk if the lender wants the shares back. Operations therefore reconciles two positions rather than one: the deal position and the borrow position, which must net to the intended exposure at every point. Firms that treat lending as a back-office utility discover during a stabilisation period that it is a front-office dependency with a settlement deadline attached.

Research coverage, after a deliberate delay

Once a cooling-off period passes, usually twenty-five to forty days after listing, the banks' analysts initiate coverage. The delay preserves at least the appearance of independence between the bankers who sold the deal and the analysts who now rate it.

Post-pricing step When What happens
Pricing Day 0, evening Final price fixed, allocations decided
Trading begins Day 1 Stock admitted; the opening print sets the tone
Settlement (DVP) Day 1-2 Shares and cash exchange simultaneously via the depository
Stabilisation Day 1-30 Lead bank supports the stock; greenshoe managed
Research coverage Day 25-40 Analysts initiate after the cooling-off period

Life after listing: the deal is a beginning, not an end

An IPO makes a company public, but the equity story continues for years. The same ECM toolkit reappears through a listed company's life, and one date in particular, the lock-up expiry, deserves attention because it can move the stock sharply.

The lock-up ends, and a wave of shares can hit the market

Recall that insiders, founders, early investors, and employees, agree not to sell for a period after listing, typically 90 to 180 days. That lock-up protects the new stock from a flood of supply while the market is still finding its feet.

When the lock-up expires, that protection lifts all at once, and the market knows the date in advance. Suddenly a large block of shares becomes eligible to sell, and the fear of that supply, the overhang, often pushes the stock down in the days around expiry, even before anyone actually sells. A well-advised company manages the expiry carefully, sometimes staggering lock-ups or arranging an orderly sell-down rather than letting insiders hit the market at random.

Selling down a stake is its own mini-deal

A large holder who wants to exit after the lock-up does not simply dump shares into the market. Doing so would crash the price. Instead they run a controlled sell-down, usually an accelerated bookbuild or a block trade, the very products from Part 1. A private equity fund exiting its position over two or three post-lock-up placements is a common pattern, and each placement is a compressed version of the whole ECM process.

The full public-equity lifecycle, in one view

Seen from a distance, a company's equity life is a sequence of ECM events, not a single IPO. It lists, its lock-up expires, early investors sell down through ABBs, it returns for a follow-on when it needs growth capital, and it may run a rights issue if it needs to repair its balance sheet. Each event is a different product, but each rests on the same machinery of origination, bookbuilding, pricing, allocation, and settlement.

Post-IPO event Typical product What drives it
Lock-up expiry (No deal, but supply overhang) Calendar; insiders become free to sell
Insider / PE sell-down ABB or block trade Early investors exit in an orderly way
Growth capital raise Follow-on (FO / SPO) Company needs new equity to expand
Balance-sheet repair Rights issue Company needs large capital, treats owners fairly

These are the calculations a syndicate desk does under time pressure. Full solutions are in the Answer Key.

2.1 Reading the book

A deal is offering 40 million shares in a range of 19 to 23. At 23 the book holds orders for 52 million shares; at 21.50 it holds 124 million; at 19 it holds 310 million. Calculate the cover at each of the three prices. The syndicate recommends 21.50 rather than 23. Give the two strongest arguments for that recommendation, and the strongest argument against it.

Cover at each price: 52 / 40 = 1.3× at 23; 124 / 40 = 3.1× at 21.50; 310 / 40 = 7.75× at 19.

The two strongest arguments for 21.50. First, 1.3× is not a functioning book. There is no room to allocate selectively, a single large order withdrawing overnight leaves the deal uncovered, and every investor is filled in full, which is the condition described in the underpricing section as the signal of a bad deal. Second, at 3.1× the syndicate can build the shareholder register it wants, reward the investors who moved the price up, and leave real buyers unsatisfied, which is what supports the stock in the days after listing.

The strongest argument against. The company raises 40 million × 1.50 = 60 million less. If the demand at 23 is genuine, of good quality and likely to hold, that is a permanent transfer from the issuer to the initial buyers, and no aftermarket argument makes it costless. The issuer is entitled to weigh 60 million today against a smoother listing.

2.2 Allocation under scarcity

The same deal prices at 21.50 with 40 million shares available and 124 million demanded. Four investors are in the book: a sovereign wealth fund that placed a strike order for 30 million and has held every previous deal it bought; a long-only fund that bid 25 million at a limit of 22; a hedge fund that bid 40 million at strike and flipped 90% of its last three allocations within a week; and a second long-only fund that bid 12 million at strike but only entered the book on the final morning. Allocate the 40 million shares between them and justify each line in one sentence.

One defensible allocation of the 40 million shares:

Sovereign wealth fund — 17 million (57% of its order). A price-insensitive strike order from a holder with a perfect retention record is the backbone of the register, and this is the order that most deserves to be filled.

Long-only fund bidding 25 million at a limit of 22 — 13 million (52%). The limit is satisfied at 21.50, so the order is live and the buyer is exactly the type the issuer wants. Being slightly price-sensitive, it is filled a shade behind the strike order.

Hedge fund bidding 40 million at strike — 4 million (10%). The largest order in the book receives the smallest fill. A 90% flip rate on the last three deals means this order is not demand for the shares but demand for the first-day move. Some allocation is right, since aftermarket liquidity has value, but not much.

Long-only fund bidding 12 million at strike on the final morning — 6 million (50%). Good quality, but it arrived after the price had been formed by others, and contributed nothing to price discovery. It is filled respectably and not generously.

The general principle: allocation rewards conviction, timing and retention, not order size. The largest order in this book gets the smallest allocation, and that is the system functioning correctly rather than failing.

2.3 The greenshoe

A deal sells 100 million shares at 20, with a 15% greenshoe. The syndicate sells 115 million shares to investors on pricing night. The stock opens at 18.50 and drifts to 18 over the following fortnight, at which point the bank has bought 15 million shares in the market at an average of 18.20. State whether the shoe is exercised, calculate the profit or loss on the stabilisation activity, and say who receives it.

The syndicate sold 115 million shares while only 100 million existed, leaving it short 15 million. The stock fell, so the bank covered that short by buying in the market rather than by calling on the issuer. The shoe is not exercised. It lapses.

Stabilisation result: sold 15 million at 20 = 300 million; bought 15 million at an average of 18.20 = 273 million. The gain is 27 million.

Who receives it is set in the underwriting agreement. Market practice is that it accrues to the syndicate, split among the underwriters, as compensation for taking on the stabilisation obligation, though some agreements share it with the issuer.

The uncomfortable feature is worth stating plainly. The mechanism generates a profit for the banks precisely when the deal has gone badly for the issuer and its new investors, and generates nothing when the deal goes well. Nobody designed that as a reward for failure — it falls out of the structure of a short position that can be covered either at the offer price or in a falling market — but it is a fair question for an issuer to raise when negotiating who keeps it.

2.4 The clawback

An offer is split 80% institutional and 20% retail, over a total of 50 million shares. The retail tranche is 9 times subscribed and the institutional tranche is 2.5 times subscribed. A clawback provision moves up to 10 percentage points of the deal to retail if retail subscription exceeds 5 times. Calculate the final tranche sizes in shares and the resulting cover on each tranche.

Initial split of 50 million shares: institutional 80% = 40 million; retail 20% = 10 million.

Demand: retail at 9× its tranche = 90 million shares; institutional at 2.5× its tranche = 100 million shares.

Retail subscription exceeds the 5× trigger, so the clawback fires in full and moves 10 percentage points across.

Final tranches: retail 30% = 15 million shares; institutional 70% = 35 million shares.

Resulting cover: retail 90 / 15 = 6.0×; institutional 100 / 35 = 2.86×.

Note what the mechanism does and does not achieve. Retail remains substantially more oversubscribed than institutional even after the clawback, because a capped shift cannot equalise a 9× tranche against a 2.5× one. The clawback narrows the unfairness rather than removing it, which is the design intent: retail investors get a larger share of a hot deal without the institutional book being hollowed out to the point where the register loses its long-term holders.

2.5 Fee arithmetic

A 600 million IPO carries a 3.5% gross spread, split 20% management, 20% underwriting, 60% selling concession, with a further 0.75% of deal size held back as a discretionary incentive fee. Calculate the total fee pool in euros. Then calculate what a joint bookrunner receives if it holds 25% of the underwriting commitment, generated 15% of the placed demand, and is awarded 20% of the discretionary pool.

Gross spread: 600 million × 3.5% = 21.0 million. Discretionary pool: 600 million × 0.75% = 4.5 million. Total fee pool: 25.5 million.

Decomposition of the gross spread: management 20% = 4.2 million; underwriting 20% = 4.2 million; selling concession 60% = 12.6 million.

The joint bookrunner receives 25% of the underwriting fee (0.25 × 4.2 = 1.05 million), 15% of the selling concession (0.15 × 12.6 = 1.89 million), and 20% of the discretionary pool (0.20 × 4.5 = 0.9 million). Total: 3.84 million.

No management fee is attributed, since none was specified for this bank; that pot is weighted heavily to the global coordinator, which is exactly the point. This bank underwrote a quarter of the risk and earned a quarter of the payment for it, but generated only 15% of the demand and was paid accordingly. The selling concession is where the difference between a bank that works a deal and a bank that lends its balance sheet becomes visible in cash.

Explain bookbuilding to someone who has sold a house. They already understand the core of it, and the analogy is closer than it looks: a guide price, viewings, offers that arrive over days, an agent who knows which buyers are serious, and a seller who can see the whole picture while each buyer sees only their own bid.

Then push the analogy until it breaks, because that is where the learning is. Two things happen in a bookbuild that never happen in a house sale. The seller can choose a buyer who bid less, and routinely does. And the seller deliberately accepts slightly less than the highest achievable price. Explain both, and explain why they are connected.

      1. A share trades at 20 before a 1-for-4 rights issue priced at 15. What is the TERP?

      2. Why is the quality of demand more important than the oversubscription multiple?

      3. In a firm-commitment deal, who bears the loss if the shares cannot be resold at the expected price?

      4. What are the two things the greenshoe achieves?

      2

      1. A share trades at 20 before a 1-for-4 rights issue priced at 15. What is the TERP?

      2. Two books are each covered five times. Which one should reassure the syndicate more?

      3. TechCo's syndicate sells 23 million shares for 20 million on offer. The share falls below the offer price. What happens?

      4. In a firm-commitment deal, the shares cannot be resold at the expected price. Who bears the loss?

      3Roles, Actors, and Interactions

      A deal is a coordination problem before it is anything else. This part maps the cast: who sits around the deal, who does what inside the bank, and why the walls the bank builds between its own teams become one of the hardest engineering constraints in the whole business.

      Four camps surround a deal, and the bank sits in the middle

      Strip a deal down and you find four participants

      Understanding how they interact is the fastest way to understand why the bank's systems look the way they do.

      The issuer is the company raising capital, represented by its chief financial officer and board. It wants capital on good terms and a shareholder base it can live with.

      The bank, specifically its ECM function, is the intermediary that makes the deal happen.

      The investors, the buy-side, are the institutions that buy the shares: pension funds, asset managers, hedge funds, sovereign wealth funds, and, in smaller amounts, retail.

      The market infrastructure is the plumbing that makes ownership real: the exchange that lists and trades the shares, and the central securities depository that records ownership and settles trades.

      The bank is one party to almost every interaction

      Watch how the four interact through the deal, and one fact jumps out. The issuer pitches to the bank, and the bank shepherds the prospectus to approval. The bank runs the roadshow to investors, and investors submit orders back to the bank. The bank recommends a price to the issuer, and confirms allocations back to investors. The bank sends settlement instructions to the infrastructure, and the infrastructure delivers shares against payment.

      In nearly every one of these, the bank is one of the two parties. The issuer and the investors barely talk to each other directly at all. The bank sits between them, translating, and also between both of them and the infrastructure.

      Interaction From To Stage
      Mandate and pitch Issuer Bank Origination
      Prospectus approval Bank Regulator / issuer Preparation
      Roadshow and marketing Bank Investors Marketing
      Orders submitted Investors Bank Bookbuilding
      Pricing recommendation Bank Issuer Pricing
      Allocation confirmation Bank Investors Allocation
      Settlement instructions Bank Infrastructure Settlement
      Delivery versus payment Infrastructure Bank / investors Settlement
      Figure 8 The bank intermediates almost every interaction in a deal.

      The issuer is not one actor either, and its people want different things

      Six roles inside the company, with genuinely divergent incentives

      "The issuer" is shorthand for a group of people who do not agree. Treating the client as a single voice is the most common mistake a junior banker makes, and it produces advice that satisfies nobody in the room.

      The chief financial officer owns the transaction and is usually the bank's day-to-day counterparty. They are optimising for a successful listing at a defensible price and, more quietly, for not being blamed. That second objective explains more behaviour than the first: a CFO will often prefer a slightly lower price with a certain outcome to a higher price with execution risk attached.

      The chief executive owns the equity story and will meet the investors. They care about valuation as a public verdict on the company they built, and they are the person most likely to push for the top of the range. They are also the person least available for the process work, which is why timetables slip around their diary.

      The treasurer cares about the proceeds and what they fund, and is frequently the only person in the room who thinks about the settlement mechanics of receiving several hundred million euros on a specific morning.

      The general counsel owns the prospectus and the liability that comes with it. Their instinct is to disclose more, because every omission is a potential claim, and this puts them in structural tension with the marketing instinct to tell a clean story.

      The head of investor relations is often hired for the deal and will live with its consequences longest. They are the one person whose interests align almost perfectly with building a durable register rather than maximising the price.

      The board and its audit committee approve the transaction and carry the governance responsibility. The audit committee in particular is the body that must be satisfied the numbers are right, and a newly recruited independent chair who is not yet comfortable is entirely capable of delaying a deal.

      Selling shareholders sit alongside the company and want something else entirely

      When existing holders are selling into the deal, a second client appears with an opposed interest. The company wants a high price and a stable register. A selling shareholder wants a high price and liquidity, and cares about the register only to the extent it affects their remaining stake.

      A private equity sponsor is the sharpest version of this. It is exiting to a timetable set by its own fund life, it is highly price-sensitive because the exit multiple determines its returns, and it is running the dual track that may cancel the listing entirely. It is also, usually, the most experienced party in the room, having done this many times more often than the management team it is selling.

      Founders behave differently again. They are frequently selling for the first time, are emotionally attached to the valuation as a judgement on their life's work, and may be constrained by lock-ups that a sponsor negotiated away. Where founders retain enhanced voting rights, their preferences carry weight disproportionate to their economic stake, and as Part 1 noted, that structure can cost the company index eligibility.

      Actor Optimising for Where it collides with the bank's advice
      CFO Certain execution, defensible price Prefers safety when the bank sees room to push
      CEO Valuation as public verdict Pushes the range beyond what the book supports
      Treasurer Proceeds and their timing Indifferent to register quality
      General counsel Complete disclosure Resists the clean narrative marketing wants
      Investor relations A durable shareholder base Aligned with the bank, rarely senior enough to decide
      Board / audit committee Governance and defensible numbers Can delay for reasons the market never learns
      PE sponsor Exit price and certainty Runs a dual track that may cancel the deal
      Founders Valuation and control Trades index eligibility for voting rights

      The buy-side is not one investor but several

      The whole module leans on the words "long-only" and "hedge fund," so it is worth stopping to map the buy-side properly. Who actually buys an equity deal, and why does the bank care so much which of them it is?

      Why the type of investor matters as much as the size of the order

      An allocation is not just a quantity; it is a bet on behaviour. The issuer and the banks are trying to assemble a shareholder base that will hold the stock and support it, not one that will sell into the first rally. So they care intensely about which kind of investor is behind each order, because each kind behaves differently.

      Two traits sort the buy-side more usefully than any label: how long they intend to hold, and how sensitive they are to price. A holder who will own the stock for years and is not chasing a quick gain is worth more to a new listing than a trader who will flip it on day one.

      The main investor types, and how each behaves in a deal

      Long-only funds are the backbone of a good book. Mutual funds, pension funds, and insurance funds buy to hold, often for years, and rarely use leverage. They are price-sensitive but patient, and issuers prize them because they stabilise the register. When a bank says a book is "high quality," it usually means long-only demand is heavy.

      Pension and sovereign wealth funds are the most patient money of all. They invest on very long horizons and can take large positions, which makes them ideal cornerstones and anchors. Their presence signals confidence to everyone else.

      Hedge funds bring liquidity but also churn. They are more likely to flip an allocation for a quick day-one gain, and their demand is often price-sensitive and opportunistic. They are not unwelcome; they add depth and liquidity to a book. But a deal dominated by hedge-fund demand is a warning sign for aftermarket stability.

      Retail investors, where a deal includes them, add breadth and a public constituency. Individuals buy smaller amounts but in large numbers, and a retail tranche can build goodwill and a loyal shareholder base, though it brings distribution complexity and, in some markets, political sensitivity about who gets access.

      Investor type Holding horizon Price sensitivity Why the issuer wants (or fears) them
      Long-only (mutual, insurance) Years Sensitive but patient The stable backbone of a good book
      Pension / sovereign wealth Very long Low; strategic Ideal cornerstones; signal confidence
      Hedge funds Days to months High; opportunistic Add liquidity, but prone to flipping
      Retail Mixed Low individually Breadth and goodwill; distribution complexity

      ESG reaches the deal as an eligibility test, not a preference

      For a large part of the book, sustainability is a hard constraint

      The buy-side taxonomy above sorts investors by horizon and price sensitivity. A second axis now cuts across it. A substantial share of European institutional money operates under mandates that exclude certain activities outright, or that require the fund to evidence sustainability characteristics in its holdings. For those investors the question is not whether they like the company's environmental record. It is whether they are permitted to own it at all.

      This changes what a "no" means during marketing. An investor who declines on valuation can be brought back with a lower price. An investor whose mandate excludes the sector cannot be brought back at any price, and a syndicate that fails to distinguish the two will misread its own demand curve badly.

      Fund classification decides who may buy before anyone reads the prospectus

      European funds are classified by the sustainability characteristics they claim, and those claims constrain what the fund can hold. A fund marketed on sustainability characteristics that buys a company failing its own stated screens has a compliance problem, not a performance problem, and the penalty for misclassification is reputational and regulatory rather than financial.

      The practical effect on a deal is that a portion of the potential book is filtered before the roadshow. Data vendors score the issuer, sometimes from public information the company has never reviewed, and those scores feed the eligibility screens of funds the syndicate would otherwise have targeted. A company can discover during PDIE that it has been screened out by investors it never met, on the basis of a rating it did not know existed.

      The disclosure burden lands back on the issuer as a data problem

      Part 2 covered non-financial reporting as a readiness workstream. The reason it matters commercially appears here: the data is not produced for regulators, it is produced for the screens that decide who can buy the stock. Issuers who treat sustainability disclosure as a compliance exercise consistently under-invest in the data quality, and then find that the ratings derived from it are worse than their actual performance warrants.

      The correction is slow. Ratings are refreshed on the vendor's cycle, not the company's, so a data problem identified during PDIE cannot be fixed before pricing.

      ESG dimension How it reaches the deal What the syndicate must do differently
      Exclusion screens Sector or activity ruled out by mandate Remove from the target list; never a pricing conversation
      Fund classification Fund's own sustainability claims constrain holdings Confirm eligibility before spending roadshow time
      Vendor ratings Third-party score feeds eligibility screens Check the issuer's public score early, expect no quick fix
      Governance factors Dual-class shares, board independence Overlaps directly with index eligibility from Part 1
      Engagement investors Will buy, then push for change Treat as long-term holders with conditions attached

      Market infrastructure is a chain of specialists, and each one owns a different truth

      Four institutions stand between an allocation and a shareholder

      "Infrastructure" is the camp readers skim, and it is the camp that decides whether the deal actually completes. Between the moment the syndicate agrees who gets what and the moment a name appears on a share register, four distinct institutions each perform one irreducible function. They are frequently confused with one another, so it is worth separating them cleanly.

      The exchange admits the security and hosts the trading

      The exchange runs two businesses that outsiders merge into one. The listing function admits the security: it reviews eligibility against its own rulebook, checks free float and governance, approves the admission document alongside the regulator, and then polices the company's continuing obligations for the rest of its listed life. The trading function operates the order book on which the shares change hands, runs the opening auction on listing morning, and can halt trading when disclosure obligations are breached.

      For the deal, the listing function matters far more than the trading one. It is a gate with a queue and a timetable, and its approval is one of the few dependencies the bank cannot accelerate with effort.

      The central securities depository holds the definitive record

      The CSD is where the shares legally exist. Securities today are dematerialised, meaning there is no certificate anywhere, only an entry in a book. The CSD keeps that book, and settlement is a change to it: an entry moves from the seller's account to the buyer's while cash moves the other way.

      The consequence is worth stating precisely. When operations sends a settlement instruction, it is not moving anything. It is asking the institution that maintains the definitive record to amend it. Every other system in the bank holds a copy, a projection or an expectation of what the CSD says. Only the CSD is the fact.

      The central counterparty absorbs the risk between trade and settlement

      Once shares trade on the exchange, a CCP typically steps into the middle of every bargain by novation, becoming buyer to every seller and seller to every buyer. If one side defaults before settlement, the CCP absorbs it, funded by the margin it collects from members.

      A CCP usually plays no part in the primary transaction itself, because the deal is settled bilaterally between the syndicate and each allocated investor. It matters intensely from the opening auction onwards, which is precisely why stabilisation trades and the aftermarket behave differently from the placement that preceded them.

      The registrar and the paying agent maintain the relationship afterwards

      The registrar, or transfer agent, maintains the register of members: the legally significant list of who owns the company. Where investors hold through the custody chain, the register may show a nominee rather than the ultimate holder, which is why identifying a company's real shareholders is a research exercise rather than a database query.

      The paying agent handles the flows that follow: dividends, corporate action elections, and, on the deal itself, the receipt of subscription monies. On a rights issue the paying agent is central rather than peripheral, because it is the entity that collects the cash and reconciles it against the acceptances.

      The custody chain is the reason nothing is as direct as it looks

      An investor rarely faces the CSD. It instructs its custodian, which may instruct a global custodian, which instructs a sub-custodian in the market concerned, which faces the CSD. A single allocation can therefore travel through four institutions before it settles, and each link is an opportunity for an instruction to mismatch.

      This chain explains several things that otherwise look like inefficiency. It explains why settlement instructions carry so much identifying data, since each link must know precisely whose account to touch. It explains why a wrong standing settlement instruction fails so expensively, as seen in the lineage exercise. And it explains why shortening the settlement cycle is genuinely hard: compressing T+2 to T+1 removes an entire overnight batch window from a chain of four parties operating in different time zones.

      Institution The one thing it owns What breaks if it is wrong
      Exchange (listing) Admission to trading and continuing obligations The deal cannot list on the planned date
      Exchange (trading) The order book and the opening auction No orderly price formation on day one
      CSD The definitive record of ownership Settlement fails; nobody legally owns the shares
      CCP Counterparty risk between trade and settlement A member default propagates into the market
      Registrar / transfer agent The register of members The company cannot identify or pay its owners
      Paying agent Cash flows to and from holders Subscription monies or dividends go astray
      Custodian chain The investor's route to the CSD Instructions mismatch and trades fail

      Every institution above is a system integration with its own protocol, its own cut-off times and its own reference data. The bank does not have one settlement interface; it has one per market, and often one per CSD within a market. This is the single largest source of the "why can't we just" frustration that technologists new to capital markets bring with them. The answer is almost always that the counterparty on the other side of the interface is a monopoly infrastructure with a published specification and no interest in accommodating one bank's preferences.

      Chinese Walls turn the org chart into an access-control problem

      The problem: one bank knows things it must not share with itself

      In the course of a deal, the banking team learns material non-public information, usually shortened to MNPI. They know a company is about to raise capital, or be acquired, before the world does. That information is valuable and, if misused, illegal to trade on.

      Now consider that the same bank also trades shares all day, publishes research, and talks to investors constantly. If the banking team's secret knowledge leaked to any of those, the bank could trade ahead of the market, tip favoured clients, or publish conveniently timed research. That is insider dealing, and it is exactly what the walls exist to prevent.

      The barriers separate four functions

      The bank walls off banking, which knows deal details, from three others. Research must stay independent. Trading must not front-run. Sales must be restricted until it is deliberately brought into the loop.

      Wall-crossing is the deliberate act of bringing someone over

      When a salesperson needs briefing on an upcoming deal so they can sound out investors, compliance formally crosses them over the wall. It records who was crossed, when, by whom, and under what approval, and from that moment the salesperson is restricted too.

      Why this belongs in a technology module

      Every part of a Chinese Wall has to be enforced by systems, not just by trust. You cannot police information flow across a global bank of thousands by asking people to be careful. It has to be built into the plumbing.

      Chinese Wall enforcement is one of the largest single areas of compliance technology spend in a corporate and investment bank, and it decomposes into several systems. Restricted-list management maintains an automated list of securities under restriction and blocks trading and research on those names. Communications surveillance uses natural-language processing to scan emails, Bloomberg chats, and Teams messages for signs of information leaking across a wall. Access controls govern who can open which repository, deal room, or drive, based on their wall status. A wall-crossing log keeps a compliance-auditable record of every crossing and its approval chain. And personal-account-dealing systems monitor employees' own trades, with pre-clearance workflows. Vendors such as NICE Actimize, Behavox, and StarCompliance specialise here, and banks layer proprietary systems on top.

      Six internal teams run the deal, and the wall runs through them

      The "bank" is a specific set of teams, not a monolith

      Each has its own role, its own place relative to the Chinese Wall, and its own technology footprint. Attaching each step of the process to the team that runs it makes every earlier part of the module click into place.

      ECM origination wins the deal

      Senior bankers within investment banking who build client relationships, pitch for mandates, and win the deal. They live on the private side of the wall, deep in MNPI. Their technology is CRM, deal-pipeline tracking, and pitch automation.

      ECM syndicate runs the execution

      The desk within capital markets that runs the bookbuild, coordinates the other banks, and proposes the price and allocation. It is the operational engine of execution. Its technology is the bookbuilding platform and the allocation engine.

      Equity sales distributes the deal

      The people within global markets who distribute the deal to investors, collect the orders, and relay market colour. Their technology is investor CRM and order management.

      Equity research brackets the deal, from the far side of the wall

      Analysts in their own department who produce independent analysis, educate investors before a deal within strict limits, and cover the stock afterward. Because independence is the point, research lives on the public side of the wall. Its technology is research-publishing platforms and the Chinese Wall controls that keep it separate.

      Legal and compliance govern every step

      Control functions that review the prospectus, manage filings, and police the fairness of allocation. Their technology is compliance systems and regulatory portals.

      Operations closes the deal out

      The back office that handles the unglamorous but essential mechanics: share creation, settlement, DVP, and reconciliation. Its technology is settlement systems and connectivity to the depository.

      Team Division Core role Primary technology
      ECM origination Investment banking Pitch clients, win mandates, sector expertise CRM, deal pipeline, pitch automation
      ECM syndicate Capital markets Run bookbuilding, coordinate syndicate, propose price and allocation Bookbuilding platform, allocation engine
      Equity sales Global markets Distribute to investors, collect orders, market colour Investor CRM, order management
      Equity research Research Independent analysis, pre-deal education, post-deal coverage Research platforms, Chinese Wall controls
      Legal and compliance Control functions Prospectus review, filings, allocation fairness Compliance systems, regulatory portals
      Operations Operations Share creation, settlement, DVP, reconciliation Settlement systems, CSD connectivity

      The Chinese Wall runs straight through the table

      Origination and syndicate are private-side, inside the deal. Research must stay public-side, independent. Sales moves across the wall deal by deal, crossed when needed and restricted afterward. The org chart is not just a management diagram; it is a map of who may know what, and that map is exactly what the access-control systems have to encode.

      Full solutions are in the Answer Key.

      3.1 Build the book you want

      You are the syndicate lead on a 500 million IPO for a stable, dividend-paying utility. Specify the investor mix you would target as percentages of the placed deal across long-only funds, pension and sovereign wealth funds, hedge funds and retail. Then do the same for a high-growth software company with no profits, and explain each difference you introduced.

      Stable, dividend-paying utility. Long-only 45%, pension and sovereign wealth 30%, hedge funds 10%, retail 15%. The buyer of a utility is buying a yield stream, so the register should be dominated by patient income investors whose holding horizon matches the asset. Pension and sovereign money is unusually well suited and should be courted for cornerstone positions. Hedge fund participation is kept low because there is little volatility to trade and less need for the liquidity they provide. Retail participation is meaningful, since utilities are widely understood and often carry a domestic shareholder-base expectation.

      High-growth software company with no profits. Long-only 40%, pension and sovereign wealth 15%, hedge funds 30%, retail 15%. Three changes matter. Specialist growth-focused long-only funds replace generalist income funds, since valuation here rests on forecasts rather than yield. Pension and sovereign allocation falls sharply because the risk profile does not fit most mandates. Hedge fund allocation rises substantially, and deliberately: an unprofitable, high-multiple stock will be volatile, and a register with no fast money in it produces a share nobody can trade in size, which widens spreads and deters the very institutions the company needs later.

      3.2 A wall-crossing decision

      Your syndicate desk wants to sound out a large fund about a deal that has not been announced. That fund's portfolio manager currently holds a position in the company's listed peer. Set out what has to happen before the conversation, what the fund is prevented from doing afterwards, what record the bank must keep, and which system holds each part of that record.

      Before the conversation. Compliance must approve the crossing. The fund must agree to be wall-crossed, usually by explicit consent on a recorded line. The scope of what will be disclosed is defined in advance and not exceeded. The company is added to the bank's restricted list if it is not already there.

      Afterwards. The fund is in possession of inside information and may not trade in the company's securities until the information is public or the crossing is cleansed. The position in the listed peer is the complication: trading it may be permissible, but if the inside information is price-relevant to the peer, that trade is also restricted, and this is exactly the judgement the compliance approval exists to make.

      The record. Who was crossed, at which firm, by which banker, at what time, what was disclosed, on whose approval, and when the information was cleansed.

      Where each part lives. Compliance approval and the crossing log sit in the wall-crossing or control-room system. The restricted-list entry sits in the restricted-list system, which feeds trading controls automatically. The recorded consent sits in the voice-recording archive. Entitlements to deal systems are granted in the identity platform for a defined window. Trade surveillance monitors the fund's activity from that point forward. The audit trail is append-only, because a regulator will ask for it months later and a mutable log proves nothing.

      3.3 Trace a conflict

      For each situation, name the interest in conflict, the party disadvantaged, and the specific control that addresses it. The bank's research analyst wants to publish a note on a company its ECM desk is pitching. The syndicate desk wants to allocate generously to a hedge fund that is also a large trading client of the same bank. The bank's asset-management arm wants to buy into a deal the bank is underwriting.

      Research wanting to publish on a pitch target. The conflict is between the bank's interest in winning a mandate and the analyst's duty of independence. The disadvantaged party is the investing public relying on that research. The control is the information barrier plus research independence rules: the ECM desk may not tell research what it is pitching, may not see the note before publication, and may not influence the rating.

      Allocating generously to a hedge fund that is a large trading client. The conflict is between the bank's own commercial relationship and the issuer's interest in a durable register. The disadvantaged party is the issuer. The control is a documented allocation policy with criteria agreed with the issuer in advance, an immutable record of the allocation rationale, and issuer sign-off on the final allocation.

      The bank's asset-management arm buying into a deal the bank underwrites. The conflict is between the underwriter's interest in placing stock and the asset manager's duty to its own clients. The disadvantaged parties are the funds' end investors. The control is the information barrier between the two entities, the asset manager bidding at arm's length with no preferential information, and disclosure of the affiliate participation.

      3.5 Follow one allocation to a shareholder

      An allocation of 500,000 shares is agreed on pricing night for a fund in Stockholm. Name every institution the shares or the instruction pass through before that fund's name could appear on the register, state what each one does, and identify the two points where a mismatch would cause the trade to fail.

      The chain runs: syndicate allocation, confirmed to the fund; the bank's operations instructing its settlement agent; the agent instructing the CSD to deliver against payment; on the other side, the fund instructing its custodian, which instructs a global custodian, which instructs a sub-custodian in the market, which faces the same CSD; the CSD matching the two instructions and moving the entry; and finally the registrar, which records the holding, usually in the name of a nominee in the custody chain rather than the fund itself.

      The two failure points are both matching problems. The first is at the CSD, where the delivering and receiving instructions must agree on quantity, security, settlement date and cash amount; any disagreement leaves the trade unmatched and unsettled. The second is the standing settlement instruction used to build the delivery: if it names the wrong account or the wrong sub-custodian, the instruction is internally valid, matches nothing, and fails.

      The instructive part is the last step. The fund may never appear on the register at all, because the nominee does. This is why a company cannot simply query a database to learn who owns it, and why identifying the real shareholder base is a paid service rather than a lookup.

      3.4 Map the wall

      Using the six-team table, draw the information barrier as a single line through the organisation. For each team, state which side it sits on and name one piece of information it must never receive from the other side.

      The line runs between the private side and the public side. Private side: ECM origination, ECM syndicate, legal and compliance in their deal capacity. Public side: equity research, equity sales, and the trading desks. Operations sits behind the deal but works on executed transactions rather than live confidential ones.

      One item each side must never receive from the other: research must never receive the deal timetable or the pitch, because knowing a mandate is being sought compromises the independence of the note. Sales must never receive the live book totals before they are released as formal guidance, because that is tradeable information about supply and demand. Origination must never receive research's unpublished rating or forecast, because it would be tempted to market the deal on it.

      The instructive part of drawing the line is that syndicate sits on the private side while sales sits on the public side, yet the two must talk constantly during a bookbuild. That conversation is the most carefully controlled channel in the bank, and it is managed by controlling what may be said rather than by preventing the conversation.

      Explain to a colleague why a single bank is allowed to advise a company on its listing, publish research about that company, sell its shares to investors, and trade those shares for its own account, all at once, without that being a scandal.

      The explanation has to do two things at the same time: make the conflicts vivid rather than minimising them, and then show why the controls make the arrangement workable. If your explanation makes the bank sound blameless, you have not made the conflicts vivid enough. If it makes the arrangement sound indefensible, you have not explained the controls.

          1. Why are the bank's systems the integration hub of a deal?

          2. What is wall-crossing?

          3. Which team runs the bookbuild, and which team must be walled off from it?

          4. Carrying forward from Part 2 — why does an investor's order type change how the syndicate reads the order?

          5. Carrying forward from Part 1 — which of the six products would you expect to attract the highest proportion of hedge-fund demand, and why?

          3

          1. Which team runs the bookbuild, and which must never see it?

          2. What is wall-crossing?

          3. An investor places a strike order. What does it signal to the syndicate?

          4. Why are the bank's systems the integration hub of a deal?

          4Systems, Data, and IT

          Now we turn the deal over and look at its underside. Every step is carried by a system, and every decision leaves a data object behind that the next step depends on. This part builds the technology picture in three layers, the functional architecture, the real vendors, and the data lineage that stitches them together, then closes with the regulations that shape every system.

          The functional architecture is a chain of six blocks

          Ask the clean question first: what functions does any deal need?

          Ignore brand names for a moment. What functions does a bank need to run an equity deal, whatever software it buys? The answer is a chain of six functional blocks aligned to the stages of the process, plus capabilities that cut across all of them. Every real system is a branded implementation of one of these blocks.

          The chain, read front to back

          At origination, a deal pipeline and CRM tracks clients, manages the pipeline, and supports pitching. In preparation, document management handles prospectus drafting, version control, and the data room. For marketing, an investor database holds profiles, mandates, and targeting. During bookbuilding, a bookbuilding platform collects orders, aggregates demand, and provides analytics. At allocation, an allocation engine applies the rules, runs fairness checks, and confirms who gets what. At settlement, a settlement and operations capability connects to the depository, manages DVP, and reconciles.

          Functional block Stage served What it does
          Deal pipeline and CRM Origination Client tracking, pipeline management, pitch support
          Document management Preparation Prospectus drafting, version control, data room
          Investor database Marketing Investor profiles, mandates, targeting
          Bookbuilding platform Bookbuilding Order collection, demand aggregation, analytics
          Allocation engine Allocation Rules, fairness checks, compliance, confirmation
          Settlement and operations Settlement CSD connectivity, DVP, cash management, reconciliation

          Four capabilities cut across every stage

          Some capabilities belong to no single stage because they touch all of them.

          Compliance and Chinese Wall controls run restricted lists, wall-crossing logs, surveillance, and allocation-fairness checks across the whole deal. Regulatory reporting handles prospectus filings, stabilisation disclosure, and transaction reporting. Reference data creates the ISIN and sets up issuer and instrument static data so every other system can recognise the new security. Communication and workflow carries syndicate messaging, roadshow logistics, and deal-status tracking.

          Figure 9 The functional architecture: six front-to-back blocks wrapped in cross-cutting capabilities.

          Real vendors fill each block, and where banks build tells you where the edge is

          Each block maps to a few dominant vendors

          Knowing the names, and knowing where proprietary builds win, is part of being literate in this business.

          CRM and deal pipeline is dominated by Salesforce for financial services in new builds, with Dealogic and proprietary systems also present. Document management leans on Workiva and DFIN's ActiveDisclosure for drafting and disclosure, with Intralinks and Datasite the standard data rooms; DFIN is strongest where US SEC filings are involved. Bookbuilding is led by Ipreo, now part of S&P Global, close to a market standard, alongside Dealogic, while the very top banks such as Goldman Sachs and JP Morgan run proprietary systems. The investor database is served by Ipreo's corporate products, FactSet, and Bloomberg, though proprietary investor data is itself an edge, so banks guard their own. The allocation engine is typically proprietary, tightly integrated with bookbuilding, because allocation rules are bank-specific and sensitive. Compliance and conflicts is led by NICE Actimize, StarCompliance, and Behavox.

          Functional block Leading vendors Notes
          Deal pipeline and CRM Salesforce Financial Services, Dealogic, proprietary Salesforce dominant in new builds
          Document management Workiva, DFIN ActiveDisclosure, Intralinks, Datasite DFIN strong for US SEC; Datasite and Intralinks for the data room
          Bookbuilding platform Ipreo (S&P Global), Dealogic, proprietary Ipreo is the market standard; top banks build their own
          Investor database Ipreo BD Corporate, FactSet, Bloomberg Proprietary investor data is a competitive edge
          Allocation engine Typically proprietary, integrated with bookbuilding Custom-built because allocation rules are bank-specific
          Compliance and conflicts NICE Actimize, StarCompliance, Behavox Chinese Wall, personal-account dealing, surveillance

          There is a pattern in where banks buy versus build, and it predicts where transformation money flows. Banks buy for the blocks where the function is common across the industry and there is no edge in doing it differently: document management, the data room, communications surveillance. They build for the blocks where their own logic is an advantage or is too sensitive to outsource: the allocation engine above all, and, at the very top, the bookbuilding platform itself. When you see a proprietary system, ask what edge it protects; when you see a vendor system, ask why the bank decided the function was a commodity. The buy-versus-build line is a map of where the bank thinks its real advantages lie.

          Data lineage stitches the blocks into one machine

          This is the most important idea in the technology picture

          Data lineage means the path a piece of data travels through the systems, and how it transforms at each step. In ECM, understanding lineage is the difference between seeing a collection of systems and seeing a single machine.

          Follow one investor order on its journey

          It begins upstream, before the order exists, as a deal record in the CRM: the client, the structure, the fee terms. From that record the investor database produces a target list. When a targeted investor bids, an order is born in the bookbuilding platform, carrying identity, quantity, price, and type. The platform aggregates it into a book, expressed as a demand curve. From the book, the allocation engine produces an allocation table. That flows into settlement as a confirmation, sent as a structured SWIFT message, then as a settlement instruction moving shares against cash under DVP. Finally the same facts resurface in regulatory reporting.

          See the whole chain as one line

          CRM deal record → investor target list → order → aggregated book and demand curve → allocation table → SWIFT confirmation → DVP settlement instruction → reconciliation → regulatory report.

          Figure 10 Front-to-back data lineage: every downstream object depends on an upstream decision.
          Stage System Data object created
          Origination CRM Deal record: client, structure, fee terms
          Marketing Investor database Investor target list
          Bookbuilding Bookbuilding platform Order, then aggregated book and demand curve
          Allocation Allocation engine Allocation table: who gets what
          Confirmation Settlement / messaging SWIFT confirmation (ISO 20022; legacy MT515-518)
          Settlement CSD connectivity DVP settlement instruction; cash and shares reconciliation
          Reporting Regulatory reporting MiFID II report, allocation rationale, MAR stabilisation disclosure

          Every downstream object depends on an upstream decision

          This is a chain of dependency, and chains of dependency have a brutal property: an error at the top cascades all the way down. The allocation table exists because of the book; the book exists because of the orders; the orders were placed against a target list; the target list came from the deal record.

          The cascade is also the business case for straight-through processing, or STP, the aspiration to pass data from one system to the next automatically, without re-keying or manual reconciliation. Every manual hand-off in the lineage chain is a place where a human can introduce an error or a delay. STP replaces those hand-offs with validated, automated flows, so a clean order at the top produces a clean settlement at the bottom with no one retyping anything in between. It is called the holy grail of post-trade precisely because achieving it end to end, across systems from different vendors and different eras, is genuinely hard.

          Anatomy of an order, and the book it joins

          The lineage chain talks about an "order" and a "book" as if they were simple. At the data level they are not, and looking inside them shows exactly where errors and duplicates creep in.

          An order is a small record with surprisingly heavy consequences

          An order object is a handful of fields, but every field feeds a downstream decision. Get one wrong and the cascade from Part 4 begins.

          The core fields are the investor identity, the quantity, the price or price type, and the order type. Around them sit fields that shape allocation: the sales contact who took the order, the timestamp, the geography, the investor classification, and any conditions. On a cross-border deal, the order also carries which tranche and jurisdiction it belongs to, Reg S or 144A.

          Field What it holds Why it matters downstream
          Investor identity (with LEI) Who is buying Deduplication, allocation, regulatory reporting
          Quantity How many shares Book size, oversubscription, allocation
          Price / price type Limit price, or strike Demand curve, price sensitivity signal
          Order type Strike, limit, step Conviction signal, allocation weighting
          Timestamp When it arrived Early-order credit, audit trail
          Sales contact & channel Who took it, through which desk Deduplication, credit attribution
          Investor classification Long-only, hedge fund, retail Book quality, allocation policy
          Tranche / jurisdiction Reg S or 144A, retail or institutional Clawback, compliance, reporting

          The book is an aggregation, and deduplication is where it gets hard

          The book is not just a pile of orders; it is orders aggregated and cleaned. The same investor often appears more than once, having spoken to two different salespeople at the bank, or placed orders through several of the syndicate banks. If those are counted as separate demand, the book looks bigger than it is, and the deal can be mispriced on phantom appetite.

          Deduplication is therefore a mission-critical, and genuinely difficult, piece of logic. The system must recognise that "BlackRock via the London desk" and "BlackRock via New York" are the same buyer, net their orders sensibly, and do it in real time as the book moves. The LEI helps, but fund structures, sub-accounts, and naming inconsistencies make this one of the least glamorous and most important problems in the whole platform.

          Under the hood: identifiers, integration, and immutable logs

          The lineage chain explains what flows between systems. This section goes one level deeper, into the plumbing that makes the flow reliable: how things are named, how systems talk, and why some records can never be changed.

          Nothing flows cleanly until everything is named the same way

          The quiet precondition for straight-through processing is shared identity. If two systems call the same investor by two different names, or the same security by two different codes, the data cannot be matched automatically and a human has to intervene. Reference data, the master records for securities, issuers, and counterparties, is the golden source that every other system defers to.

          Three identifiers do most of the work, and knowing what each names prevents a lot of confusion.

          ISIN names the security

          The International Securities Identification Number identifies a specific instrument, for example TechCo's newly listed ordinary share. It is created as part of setting up the new security, and every order, allocation, and settlement instruction carries it.

          LEI names the legal entity

          The Legal Entity Identifier identifies the parties, the issuer and each investing institution, as legal entities. Regulators mandated the LEI after the financial crisis precisely so that "who is on the other side of this trade" could be answered unambiguously across the whole market. Transaction reports under MiFID II require it.

          FIGI names the instrument in a vendor-neutral way

          The Financial Instrument Global Identifier is an open, vendor-neutral code that many data platforms use to map an instrument across systems that each have their own internal ticker. Where ISIN is the regulatory standard, FIGI often does the mapping work inside the technology estate.

          Identifier Names Where it bites
          ISIN The security (this share) Every order, allocation, settlement instruction
          LEI The legal entity (issuer, investor) Regulatory reporting, counterparty identification
          FIGI The instrument, vendor-neutral Cross-system mapping inside the estate

          How the systems actually talk: batch, API, and events

          The lineage chain is only as fast as the connections between its links, and there are three ways to make those connections.

          Batch is the legacy default

          Batch integration moves data in scheduled bundles, often overnight. System A writes a file, system B reads it hours later. It is simple and robust, and it is everywhere in banks because it is decades old. Its weakness is latency: nothing is current until the next batch runs, so a book that moves through the day is stale in the systems that only see it at night.

          APIs make data available on demand

          An application programming interface lets one system ask another for data in real time and get an immediate answer. Replacing overnight files with APIs is much of what "modernisation" means in practice. It turns a nightly snapshot into a live query.

          Event streaming pushes changes the instant they happen

          Event streaming, using technologies such as Kafka, flips the model: instead of asking, systems subscribe and are pushed each change as it occurs. When an order lands, an event fires, and every downstream system that cares, risk, compliance, the book view, updates at once. For a live bookbuild, where the picture changes second by second, streaming is what makes a genuinely real-time platform possible.

          The migration from batch to API to event streaming is the single biggest structural change under modern ECM platforms, and it is not just a speed upgrade. Batch systems force reconciliation, because two systems that sync overnight will drift apart during the day and must be checked against each other. Event-driven systems that share one stream of truth reduce the need for reconciliation in the first place. Removing reconciliation, not just speeding it up, is where the real cost and risk come out.

          Some records must be impossible to change

          Regulators do not only want the data; they want proof the data was not altered afterwards. An allocation rationale is worthless as evidence if someone could quietly rewrite it once the regulator starts asking questions. So the systems that hold compliance-critical records, wall-crossing logs, allocation decisions, surveillance alerts, are built to be append-only and immutable: you can add to them, but you cannot edit or delete what is already there.

          This is why audit and lineage tooling matters as much as the business systems. Platforms that track data lineage and enforce governance, such as Collibra, let a bank show not just what a number is but where it came from and who touched it. When MiFID II asks why a particular investor received a particular allocation, the answer has to be reconstructable from an unbroken, tamper-evident record.

          The allocation engine is a rules engine with a conscience

          The allocation engine deserves a closer look, because it is where discretion and automation meet. It takes the closed book and applies the allocation criteria, investor quality, timing, price sensitivity, relationship, geography, as weighted rules to propose a split. But it is not fully automatic. The syndicate overrides it, and every override has to be captured with a reason.

          The engineering challenge is to encode judgement without pretending to remove it. The engine must run the rules, flag where a human overrode them, deduplicate the same investor appearing through several sales channels, check the result against the fairness constraints, and record the whole chain immutably. That combination, a rules engine plus a mandatory, tamper-evident audit of every human deviation, is why banks build the allocation engine themselves rather than buying it.

          The messages that move a deal: FIX and SWIFT

          Systems do not pass data by magic; they exchange standardised messages. Two messaging worlds carry an equity deal, one for the front office and one for the back, and knowing where each rules removes a lot of mystery from the lineage chain.

          FIX carries the front-office conversation

          The Financial Information eXchange protocol, universally called FIX, is the language of electronic trading. It is how order and execution information moves between investors, brokers, and exchanges in real time. When an order is routed electronically, or an execution is confirmed back, it usually travels as a FIX message.

          FIX matters here because it is fast, standardised, and session-based, built for the low-latency front office where things happen in milliseconds. In the primary-market context, FIX and FIX-like electronic channels increasingly carry order flow into the bookbuilding process that once arrived only by phone and email, which is part of what makes digital bookbuilding possible.

          SWIFT carries the post-trade instructions

          Once a deal is priced and allocated, the messages shift to the SWIFT network, the backbone of secure inter-bank messaging. SWIFT is where confirmations, settlement instructions, and status messages travel between the bank, the investors' custodians, and the central securities depository.

          The traditional securities messages are the MT5xx family: an allocation or trade confirmation as an MT515, settlement instructions in the MT54x range, and status or settlement confirmations as MT548. These are the structured messages that turn an allocation decision into an actual movement of shares against cash.

          Figure 11 Two messaging worlds, one deal. FIX carries the conversation at the front; SWIFT carries the instructions that move securities and cash.

          ISO 20022 is replacing MT, but not everywhere at the same speed

          The MT messages are being replaced by a richer standard, ISO 20022, often called MX. It carries far more structured data than the terse legacy MT format, which makes automation, screening and regulatory reporting cleaner. What is easy to get wrong, and frequently is, is that this is not one migration running on one timetable.

          The payments world has already crossed over. SWIFT ended the coexistence period for cross-border payments and reporting on 22 November 2025, retiring the legacy MT 1xx, 2xx and 9xx families in favour of their ISO 20022 equivalents. For those message types the transition is complete rather than in progress.

          The securities world has not. The MT 54x settlement messages described above are ISO 15022 messages carried over SWIFT, and they were explicitly outside the scope of that cutover. Securities messaging is migrating to ISO 20022 on a longer and looser timetable, pulled along by market infrastructures that are already native to the new standard rather than by a single industry deadline. In Europe the central settlement platform operates in ISO 20022, so banks translate at the boundary; elsewhere the older syntax remains entirely normal.

          The practical consequence for anyone assessing a post-trade estate is that translation is not a temporary bridge to be decommissioned next year. Firms will be maintaining mappings between ISO 15022 and ISO 20022 for securities for years after their payments equivalents were switched off, and a programme plan that assumes one cutover date for "the ISO migration" is planning for something that is not happening.

          Layer Standard Carries Status
          Front office FIX Orders, executions, real-time flow Stable, no migration
          Post-trade securities ISO 15022, SWIFT MT 54x Confirmations, settlement instructions Migrating to ISO 20022, no single deadline
          Post-trade securities (target) ISO 20022 / MX (sese, semt) Richer structured instructions and statements Native on major infrastructures
          Cross-border payments ISO 20022 / MX Payment instructions and reporting Migration complete, MT retired November 2025

          Deal data is a security problem before it is an architecture problem

          The bank is holding the most tradeable secret in the market

          For the months before a deal is announced, a bank holds information that is worth money to anyone who obtains it. That a company is about to list, at what valuation, with which cornerstone investors, and above all at what price the book is being struck: each of these is material, non-public and directly actionable. This is MNPI, and protecting it is not one control among many. It is the reason much of the architecture in this part exists at all.

          The threat model is unusual because the most dangerous actor is often an authorised one. External attackers matter, but the person who can do the most damage is someone with a legitimate login who looks at a deal they have no business seeing, or forwards a file to a personal address on their last week of employment. Perimeter security answers the first problem and does nothing for the second.

          Figure 12 Defence in depth around deal information. Each layer answers a different failure, and no single layer is sufficient.

          Four layers, each answering a different failure

          Data-layer controls assume the attacker already has the file. Encryption at rest and in transit is the baseline. Virtual data rooms add watermarking that stamps each viewer's identity onto every page, dynamically, so a leaked document identifies its source. Download restrictions, expiring links and view-only rendering all reduce the number of uncontrolled copies in the world, which is the quantity that actually determines leak risk.

          Identity-layer controls assume the attacker is an insider. Entitlements are granted per deal, not per role, so working on one transaction gives no visibility of another. Wall-crossing produces a timestamped record of who was brought over, by whom, and for what scope. The joiner-mover-leaver process matters more here than almost anywhere in the bank, because an analyst who moves from research to syndicate must lose one set of access at the moment they gain another, and stale entitlements accumulated across a career are the single most common finding in an access review.

          Monitoring-layer controls assume the first two layers have already failed. Electronic communications surveillance reads email and chat for the patterns that precede a leak. Data-loss prevention watches for deal documents moving towards personal mail, removable media or unsanctioned cloud storage. Trade surveillance runs the other direction, looking for unusual trading in a name shortly before an announcement, including in the personal accounts of staff, which is why personal-account dealing must be pre-cleared and is monitored automatically.

          Human-layer controls assume all technology fails eventually. Need-to-know discipline keeps deal teams small. Code names replace company names in calendars, file paths and chat. Physical measures still matter: locked deal rooms, clean-desk enforcement, screen filters, and the unglamorous rule that nobody discusses live transactions in a lift or on a train.

          The attacks that actually succeed are boring

          The successful compromises in this space are rarely sophisticated. Targeted phishing of a deal team during a live transaction remains the most reliable route in, because the attacker knows exactly what the target is working on and can craft a message that fits the day's work perfectly. Compromise of a third party is next: the printer, the translation agency, the small law firm and the investor-relations consultancy all hold the same documents as the bank while rarely holding the same defences.

          Settlement fraud deserves separate mention because it converts directly into cash. An attacker who can alter payment instructions during the days around closing can redirect very large sums, and the window is narrow enough that the fraud is often discovered only after funds have moved. This is why standing settlement instructions are held as controlled reference data with dual authorisation on change, and why a call-back on a changed instruction is a control rather than a courtesy.

          Control The failure it is built for Where it commonly breaks
          Per-deal entitlements An insider browsing deals they are not on Access granted for cover and never revoked
          Wall-crossing log Proving who knew what and when Kept manually, incomplete under time pressure
          VDR watermarking Identifying the source of a leaked document Bypassed by screenshots and photographs
          E-comms surveillance Deal information leaving through chat or email New channels adopted faster than they are monitored
          Personal-account dealing rules Staff trading on what they know Accounts held by family members
          Third-party risk assessment Suppliers holding deal data with weaker controls Assessed at onboarding, never reassessed
          Dual authorisation on settlement instructions Redirected payments at closing Urgency used as the reason to skip the second check

          Resilience: the systems cannot fail on listing day

          Most software can afford a bad ten minutes. A bookbuilding platform on pricing night, or an exchange on listing morning, cannot. Deal-critical systems live under a resilience standard closer to air-traffic control than to ordinary business software.

          Some moments have no second chance

          Pricing happens once, on a single evening, and the listing opens once, at a fixed time the next morning. If the bookbuilding platform freezes while the syndicate is finalising the book, or the allocation engine stalls at the moment allocations must go out, there is no rerun. The window is fixed by the market, and a failure inside it can break the deal or cost real money.

          This is why the systems that carry the critical steps are held to a much higher availability standard than the systems around them. It is not enough for them to be correct; they must be correct while never going down at the one moment that matters.

          Resilience is built from redundancy, failover, and rehearsal

          Redundancy means no single point of failure. Critical systems run in more than one place, so that if a server, a data centre, or a network link dies, another is already carrying the load. Nothing important depends on one machine.

          Failover means the switch is automatic and fast. When a component fails, traffic moves to the standby without a human deciding to flip a switch, because on pricing night there is no time to page an engineer and wait.

          Disaster recovery and business continuity mean the whole site can be lost and the deal still completes. Banks maintain recovery targets, how quickly systems must be back and how little data may be lost, and they rehearse them. A recovery plan that has never been tested is a hope, not a plan.

          Concept What it guarantees Why it matters on listing day
          Redundancy No single point of failure A dead server does not stop pricing
          Automatic failover Fast switch to standby No time to intervene manually at the peak
          Disaster recovery (DR) Recover from losing a whole site The deal completes even after a major outage
          Business continuity (BCP) People and process keep running Operations continue if a location goes dark

          Nobody deploys on pricing night

          The cheapest resilience control in the industry is refusing to change anything. Banks run change freezes around the moments that cannot be repeated: the pricing night, the listing morning, the settlement date, and the reporting deadlines that follow. During a freeze no release reaches production, no configuration is altered, and no infrastructure work is scheduled, unless an emergency process is invoked with senior authorisation and a written justification.

          The reason is arithmetic rather than caution. Most production incidents are caused by changes. Removing changes during the window when an incident would be unrecoverable removes most of the risk at almost no cost. The discipline extends to the vendors in the chain, which is why a bookbuilding platform provider will publish its own freeze calendar and why banks compare that calendar against their deal pipeline.

          The failure mode of a freeze is the pile-up behind it. Teams that cannot deploy for three weeks deploy everything at once when the freeze lifts, which concentrates risk into a single large release rather than eliminating it. Firms that handle this well shorten freezes by making individual releases smaller and reversible, so that fewer moments need protecting in the first place.

          You cannot recover from what you cannot see

          Resilience assumes that failures are noticed, and that assumption is routinely wrong. Systems fail silently far more often than they fail loudly. A feed stops updating and shows the last good value. A message queue backs up while every component reports itself healthy. A batch job completes successfully having processed nothing.

          Observability answers this by requiring systems to expose what they are actually doing rather than whether they are running. The distinction is between monitoring the machine and monitoring the work: not "is the bookbuilding platform up?" but "are orders arriving at the rate we would expect for a live deal, and is the book total consistent with the sum of its orders?" Business-level checks catch the failures that infrastructure checks miss entirely.

          Service levels then convert this into an accountable commitment. A service level agreement fixes what availability the business is entitled to and what happens when it is not delivered. Two numbers do most of the work: the recovery time objective, how long a system may take to come back, and the recovery point objective, how much data may be lost in the process. For a system holding a live order book, a recovery point objective measured in minutes is unacceptable, because minutes of lost orders during a bookbuild is a set of investor relationships damaged and a price struck on incomplete demand.

          Those two numbers are the honest test of whether resilience has been designed or merely asserted. Ask what they are for the allocation engine on pricing night. If nobody can answer, the resilience plan is a document rather than a capability.

          Tax and currency are settlement problems long before they are advisory ones

          Transaction taxes change the economics of where a deal is done

          Several European markets levy a tax on securities transactions, and the detail varies enough that it materially affects structuring. The rates differ, the taxable events differ, and so do the exemptions — new issues are frequently exempt while secondary transfers are not, which is precisely the distinction between the primary and secondary markets that opened Part 1.

          Two features make this an engineering problem rather than a footnote. The first is that liability often attaches to the instrument's country of issue rather than the place of trading, so a bank in one jurisdiction can owe tax in another on a trade between two parties in a third. The second is that the bank is usually the collecting agent. It calculates, withholds and remits on behalf of clients, which means the calculation must be correct at the moment of settlement and evidenced afterwards.

          The consequence is a tax engine sitting in the settlement path, driven by reference data — instrument domicile, investor status, exemption certificates — that has to be right. It is the same golden-source problem in a form where errors have a statutory penalty attached.

          Withholding on dividends creates an obligation that outlives the deal

          When the company pays a dividend, tax is withheld at source, and the rate depends on where the investor is resident and which treaty applies. Investors entitled to a lower rate must either claim relief at source, which requires documentation to be in place before the payment date, or reclaim afterwards through a process that is slow and administratively expensive.

          This lands on the deal indirectly but consistently. The custody chain from earlier in this part means the paying agent frequently cannot see the ultimate beneficial owner, only a nominee, so establishing entitlement requires information to travel back up a chain built to pass securities down it. Cross-border withholding relief remains one of the least automated processes in the whole post-trade estate, and European reform efforts to standardise it are ongoing.

          Alongside it sits the automatic exchange of tax information between jurisdictions, which obliges the firm to classify account holders, determine their tax residence and report holdings to authorities annually. That classification is reference data about the investor, captured at onboarding, and it decays as clients restructure.

          A cross-border deal is denominated in one currency and paid for in several

          Where the offer is priced in one currency and investors settle in another, foreign exchange enters the settlement path. The book may be struck in euros while a US investor funds in dollars and an Asian investor in yen, and someone must convert, at a rate fixed at a point in time, with the difference between the assumed and achieved rate landing on somebody's profit and loss.

          Depositary receipts are the structural answer to part of this. They let a company's shares trade in a market and currency other than its home one, with a depositary bank holding the underlying shares and issuing receipts against them. The receipt trades and settles locally in local currency, while the underlying stays where it was issued. The arrangement widens the investor base considerably and introduces a ratio, a depositary, a second settlement venue and a fee structure that all have to be modelled.

          Dual listings do the same thing more expensively, by admitting the same shares to two exchanges with two settlement infrastructures. The company then lives with two rulebooks, two continuing-obligation regimes and the reconciliation between them.

          Item Where it bites The system that must be right
          Transaction taxes Calculated and withheld at settlement Tax engine plus instrument domicile reference data
          Dividend withholding Every payment for the life of the holding Investor tax status, treaty rates, documentation store
          Tax information reporting Annual obligation on the firm Onboarding classification, kept current
          FX at settlement Cross-border subscriptions Rate capture, timing, and who bears the slippage
          Depositary receipts A second venue and currency Depositary interface, ratio handling, dual settlement
          Dual listing Two rulebooks, two infrastructures Duplicated reporting and continuing-obligations tooling

          Why the stack looks the way it does, and why it is so hard to change

          Nobody designed this; it accumulated

          The architecture described so far reads as though someone drew it. No bank's actual estate looks like that. What exists is the residue of thirty years of decisions, each individually reasonable: a settlement platform written when the bank was smaller, a bookbuilding system bought during a boom because building it would have taken too long, an allocation engine written in-house because no vendor understood the desk, and three CRM systems because the bank made two acquisitions.

          The cost of that history is rarely a system that does not work. It is the integration surface between systems that were never meant to meet. Each pairing needs a mapping, each mapping needs an owner, and each owner leaves eventually. Technical debt in capital markets is mostly this: undocumented translation logic sitting between systems, understood by fewer people every year.

          The consequence is that change is priced in coupling, not in code. Replacing the allocation engine is a contained piece of work. Replacing the reference-data store that eleven systems read from is not, and the difference has nothing to do with the difficulty of the software. The most expensive systems to change are usually the least sophisticated ones, because simplicity is what allowed everything else to depend on them.

          Buy or build is a question about competitive advantage

          Banks buy where the function is identical across the industry and build where it is not. Settlement messaging, document management, market data and reference data are bought, because there is no version of DVP instruction generation that constitutes a competitive edge, and a vendor amortising its development across forty clients will always be cheaper.

          Allocation logic, pricing analytics and the investor-relationship graph are built, because those encode judgement the bank considers proprietary. An allocation engine embodies a specific view of which investors deserve stock and why. Handing that to a vendor means handing it to competitors.

          The mapping is useful as a diagnostic. Show someone a bank's buy-versus-build split and they can tell you where that bank believes its edge lies, and whether the belief is still true. A bank that has built its own document management is not being clever; it is carrying cost for no return.

          Decision driver Points towards buying Points towards building
          Differentiation The function is identical at every bank The logic encodes proprietary judgement
          Regulatory change Vendor absorbs it across all clients Bank absorbs it alone
          Integration depth Sits at the edge of the estate Touches many internal systems
          Talent No scarce skill required Requires people already on the desk
          Total cost Licence plus configuration Build cost plus maintenance for its whole life

          Reference data is the golden source problem, and it is never quite solved

          Almost every failure traced in this part comes back to two systems disagreeing about a fact. Which legal entity placed the order. Which fund it belongs to. What the instrument is called. Whether the client is one client or three. The discipline that answers this is master data management, and its object is a golden source: one system designated as authoritative for each type of reference data, with everything else consuming rather than maintaining its own copy.

          The theory is uncontroversial and the practice is difficult, because designating a golden source means telling teams that their local copy is now wrong. Sales desks maintain client records that are richer than the official ones. Operations keeps a spreadsheet of settlement instructions because the official store is slow to update. Each of these is a rational local decision that degrades the whole.

          Data quality has to be measured to be managed, and the measures that work are boring: completeness, how many records carry every mandatory field; timeliness, how long a change takes to propagate; consistency, how often two systems disagree about the same entity. Firms that publish these numbers to the desks that create the data improve them. Firms that publish them only to the data team do not.

          The operating model decides whether any of this improves

          The persistent divide in banks is between technology delivered as projects and technology owned as products. A project has a budget, a scope and an end date, after which the team disperses and the system enters maintenance. This explains a great deal of what is wrong with the estate: every system was delivered by a team that no longer exists, and nobody has owned the mapping between two of them since 2016.

          The product model puts a standing team on a system for its whole life, with a named owner accountable for its roadmap and its defects, and a business counterpart who prioritises alongside them. The advantage is not delivery speed. It is that somebody is still there in year four, when the undocumented translation logic needs to change.

          The consulting-shaped mistake is to treat this as a reorganisation. Renaming project managers as product owners changes nothing unless the funding model changes with it, because a bank that still funds work in annual project bundles will keep producing orphaned systems whatever the job titles say.

          Six regulations shape every system in the chain

          Technology here is not free to do whatever is convenient

          A dense web of regulation governs every step, and each regulation lands as a concrete requirement on the systems. You do not need to be a lawyer, but you cannot design or assess ECM technology without knowing the six regimes that matter most and what each demands of the machine.

          Prospectus Regulation — the document and its approval

          EU 2017/1129 governs the content, format, and approval of the prospectus, and lets an approved prospectus be "passported" across the EU without country-by-country re-approval. It drives document-management platforms, filing systems, and multi-jurisdiction coordination.

          Market Abuse Regulation — the teeth behind the Chinese Wall

          MAR, EU 596/2014, governs insider dealing and manipulation. It requires insider lists, stabilisation disclosure, and prevention of the misuse of inside information. It drives communications surveillance, restricted lists, insider-list tools, and stabilisation reporting. MAR is what turns Chinese Walls from a policy into monitored controls.

          MiFID II and MiFIR — fairness, research, and reporting

          EU 2014/65 is sprawling, but three parts touch ECM directly. It imposes allocation-fairness requirements through the rules known as RTS 29, which is why allocation decisions now carry a documented rationale. It forced the unbundling of research from trading commissions, so research must be paid for explicitly. And it requires detailed transaction reporting. It drives allocation audit trails, research-billing systems, and trade reporting.

          Listing rules — admission and ongoing obligations

          Each exchange's listing rules govern admission requirements, ongoing disclosure, and governance. They are exchange-specific rather than a single regime, and they drive listing-application systems and disclosure monitoring.

          GDPR — the investor is a data subject

          GDPR, EU 2016/679, governs personal data, which in ECM means investor data above all: consent for marketing, retention limits, and protection. It drives data governance in the CRM and the investor database, and consent-management workflows.

          Operational resilience regulation — resilience as a supervised obligation

          Everything in the previous section used to be a matter of internal standards. In the European Union it is now law. The Digital Operational Resilience Act, universally shortened to DORA, is R8egulation (EU) 2022/2554. It has applied since 17 January 2025, with no transition period, and converts operational resilience from good practice into a supervised requirement with a defined shape.

          Four obligations do most of the work. Firms must maintain an ICT risk management framework owned by the management body rather than delegated to technology. They must report major incidents to their regulator within prescribed windows, which forces incident classification to become rigorous rather than a matter of judgement after the fact. They must test resilience on a schedule, including threat-led penetration testing for the most significant firms, and testing means rehearsal rather than documentation review. And they must manage third-party ICT risk through a register of every provider supporting an important function, with contractual rights of audit, exit plans, and concentration analysis.

          That last obligation is the one that reaches furthest into the architecture in this part. A bank running its bookbuilding on a vendor platform and its infrastructure in a public cloud must be able to name what would happen if either became unavailable, and to demonstrate a credible exit. The uncomfortable answer for much of the industry is that the market has consolidated onto a handful of providers, so a single vendor outage would degrade the same function at many banks simultaneously. Concentration risk is precisely what the regulation exists to surface.

          The United Kingdom reached similar ground by a different route, through the operational resilience regime that requires firms to define their important business services, set impact tolerances for each, and prove they can stay within them under severe but plausible scenarios. The vocabulary differs; the demand is the same. Name what must not break, say how long it may be broken for, and show your evidence.

          Regulation Scope Main IT impact
          DORA / operational resilience ICT risk, incident reporting, resilience testing, third-party risk Named important services, impact tolerances, vendor register, tested exit plans
          Prospectus Regulation (EU 2017/1129) Prospectus content, format, approval, EU passporting Document management, filing systems, multi-jurisdiction coordination
          Market Abuse Regulation (EU 596/2014) Insider dealing, manipulation, stabilisation disclosure, insider lists Surveillance, restricted lists, insider-list and stabilisation tools
          MiFID II / MiFIR (EU 2014/65) Allocation fairness, research unbundling, transaction reporting Allocation audit trail, research billing, trade reporting
          Listing rules (exchange-specific) Admission, ongoing obligations, governance Listing-application systems, disclosure monitoring
          GDPR (EU 2016/679) Investor data protection, consent, retention limits CRM and investor-database governance, consent management

          Full solutions are in the Answer Key.

          4.1 Trace the cascade

          An investor's order is captured with the correct fund name but the legal entity identifier of that fund's sister vehicle. Nobody notices until settlement. Walk the error forward through every system in the lineage chain, stating at each step what happens, whether the system could plausibly have detected the problem, and what the eventual consequence is. Then state where you would place a single control to catch it, and why there rather than anywhere else.

          Bookbuilding platform. The order is accepted. The fund name is right, so nothing looks wrong to the syndicate desk. Detectable in principle, by validating the identifier against the name in reference data, but only if that check exists.

          Allocation engine. Shares are allocated to the wrong legal entity. Undetectable here: the engine has no independent view of which entity is correct and is faithfully processing what it was given.

          Confirmation. The allocation is confirmed to the investor. This is the last cheap moment to catch it, and it is caught only if someone on the investor's side reads the entity on the confirmation rather than the fund name and the quantity. Frequently nobody does.

          Settlement instruction. The instruction is generated against the wrong entity, so it carries the wrong standing settlement instruction and the wrong custodian account. The trade fails to settle.

          Consequence. A failed settlement on listing day, remediated manually under time pressure, with a buy-in risk, a cash cost, an unhappy investor, and a transaction report filed with an incorrect counterparty that must now be corrected and, depending on the regime, self-reported.

          Where to place one control: at capture, in the bookbuilding platform, validating the identifier against the entity name in the golden source before the order is accepted. Everywhere downstream the error is already embedded in objects other systems have acted on, and each step multiplies the cost of unwinding it. This is the cascade principle in its most concrete form: the only cheap place to fix a data error is the place it was created.

          4.2 Set the objectives

          For each of the bookbuilding platform, the allocation engine and the regulatory reporting pipeline, propose a recovery time objective and a recovery point objective, and justify each in one sentence with reference to what is lost while the system is down. State which of the three you would spend the most on protecting, and why.

          Bookbuilding platform: RTO 15 minutes, RPO near zero. Orders arrive continuously during a live bookbuild and each lost order is an investor relationship damaged and a price struck on incomplete demand. Losing even a few minutes of orders is not recoverable by asking investors to resubmit, because they will not all do so and the ones who do may change their terms.

          Allocation engine: RTO 1 hour, RPO near zero. It is used intensively but briefly. An hour of unavailability during the day is absorbed; an hour on pricing night is a crisis, which is the argument for treating its recovery time as a function of the calendar rather than a constant.

          Regulatory reporting pipeline: RTO 24 hours, RPO zero. Reporting deadlines are measured in days, so a day of downtime is survivable. Losing a single reportable transaction is not, because completeness is the obligation being supervised.

          Spend the most on the bookbuilding platform. It is the only one of the three whose failure destroys information that cannot be reconstructed from anywhere else. Allocation logic can be re-run and reports can be re-generated, both from data that still exists. An order that never arrived is simply gone.

          4.3 Buy or build

          A bank is replacing four systems: its investor targeting analytics, its prospectus document workflow, its allocation engine, and its settlement instruction generator. Decide buy or build for each, and give the decisive criterion in each case. One of these decisions should be genuinely close; identify which and explain the tension.

          Prospectus document workflow: buy. Versioning, redlining and controlled distribution are identical at every bank and confer no advantage whatever.

          Settlement instruction generator: buy. Message formats are industry standards under active regulatory change, and a vendor absorbs the ISO 20022 migration across its whole client base rather than each bank funding it alone.

          Allocation engine: build. It encodes the bank's proprietary judgement about which investors deserve stock and why. That logic is the product of the relationships the bank has spent decades building, and it is not something to hand to a vendor who also serves competitors.

          Investor targeting analytics is the close one. The analytical techniques are generic and vendors do them well, which argues for buying. But the data the analytics run on — who held what, who flipped, who told the truth during PDIE — is among the bank's most genuinely proprietary assets, which argues for building. The resolution most banks reach is a hybrid: buy the analytical engine, keep the data and the relationship graph in-house, and accept the integration cost. The tension is real because the value lives in the data while the difficulty lives in the algorithms, and the buy-versus-build question assumes those travel together.

          4.4 Find the golden source failure

          Three systems disagree about a client. The CRM holds one record for a global asset manager. The bookbuilding platform holds four, one per regional desk that has ever traded with the bank. Settlement holds nine, one per fund. Identify which of these is wrong, or argue that none of them is. Then state what a golden source would have to look like to serve all three, and what each system would have to give up.

          None of the three is wrong. Each is correct for its own purpose, and that is exactly why this problem is hard.

          The CRM holds one record because relationship coverage is organised around the institution: one client, one relationship, one revenue picture. The bookbuilding platform holds four because orders arrive from regional desks that make their own decisions, and collapsing them would destroy the information that four separate teams wanted stock. Settlement holds nine because shares deliver to funds, and a fund is the legal entity that actually owns the securities.

          A golden source therefore cannot be a flat list. It has to be a hierarchy: the institution at the top, decision-making desks beneath it, and legal entities at the leaves, each with its own identifier, and each level linked to the one above. The LEI names the leaves and nothing names the top, which is why the parent level is almost always the bank's own construction and the hardest part to maintain.

          What each system gives up: the CRM stops treating its single record as complete and accepts that revenue must be aggregated up from the leaves. The bookbuilding platform stops creating client records locally and consumes the hierarchy, which means a new fund cannot be onboarded by a salesperson at midnight. Settlement gives up least, but must accept the shared identifier rather than its own. The bookbuilding platform's concession is the one that meets real resistance, because it removes flexibility precisely when the desk feels it needs it most.

          4.6 Where does the tax go wrong?

          A US fund subscribes to a euro-denominated European IPO and later receives its first dividend. Identify three separate points at which a tax or currency error could occur, name the reference data each depends on, and say which of the three would be caught automatically and which would surface only when the investor complains.

          At subscription, the transaction tax. Whether a levy applies depends on the instrument's country of issue and whether a new issue is exempt in that jurisdiction. The reference data is instrument domicile and issue type. This is normally caught automatically, because the calculation sits in the settlement path and a failure to remit is visible to the tax authority.

          At subscription, the currency. The fund funds in dollars against a euro-denominated deal, so a rate is struck at some point by someone. The reference data is the agreed conversion arrangement and the timing convention. This surfaces immediately in a profit-and-loss break, so it is caught, though often after the fact and by the wrong team.

          At the dividend, the withholding rate. The rate depends on the holder's tax residence and the applicable treaty, and relief at source requires documentation lodged before the payment date. The reference data is the investor's tax status captured at onboarding, which decays silently as clients restructure. This is the one that surfaces only when the investor complains, because over-withholding produces a correct-looking payment that is simply smaller than it should have been. Nothing fails, no exception is raised, and the error is discovered months later when someone attempts a reclaim.

          4.5 Two messages, one deal

          An order is placed by FIX and settles through SWIFT. Name the specific message that carries each step from order to settlement confirmation, and state what would go wrong if the two chains disagreed about the quantity.

          Order to allocation, over FIX. The investor's order arrives as a new order single, tag 35=D, carrying quantity, limit and the identity of the ordering entity. The allocation is returned as an execution report, tag 35=8.

          Allocation to settlement, over SWIFT. The bank instructs its custodian to deliver against payment, MT 543, while the investor's custodian instructs receipt against payment, MT 541. Confirmations return as MT 545 and MT 547. Under the ISO 20022 migration these become structured XML equivalents in the sese message family.

          If the two chains disagreed about quantity, the settlement instruction would not match its counterparty instruction at the CSD and the trade would fail. The failure is at least visible, which is the saving grace: matching at the depository is a genuine reconciliation point that catches quantity mismatches before securities move. Quantity is checked. The entity identifier of exercise 4.1 is not checked the same way, which is precisely why that error runs all the way to failure while this one does not.

          Explain the difference between FIX and SWIFT to someone who understands email but nothing about markets, using no financial jargon at all.

          The useful analogy is about urgency and reversibility rather than about technology. One system carries conversation, where a message is cheap, fast and easily superseded by the next one. The other carries instruction, where a message moves property and money and cannot simply be corrected by sending another. Build your explanation on that distinction, and the reason the industry runs two entirely separate messaging worlds becomes obvious rather than historical accident.

              1. Why is shared reference data a precondition for straight-through processing?

              2. What does an LEI identify, and how does that differ from an ISIN?

              3. Why does event streaming reduce the need for reconciliation?

              4. Carrying forward from Part 3 — why is the wall-crossing log a systems requirement rather than a compliance formality?

              5. Carrying forward from Part 2 — which data object created during bookbuilding causes the most downstream damage if it is wrong, and where does the damage surface?

              4

              1. What is the difference between an LEI and an ISIN?

              2. Which object created during bookbuilding does the most damage downstream if it is wrong?

              3. Which pair of messaging standards matches each world correctly?

              4. On listing day, what recovery objectives apply to the live book?

              5Risks, Pain Points, and Transformation

              Deals fail, and they fail in patterns. This part studies where equity deals go wrong through three real cases, then turns to the other side: the places where better technology genuinely changes the economics of the business rather than just tidying it up.

              A MECE map of where ECM risk lives

              Before the war stories, a clean taxonomy

              The three cases below are memorable, but memory is not a framework. To reason about risk you need a map that covers the whole territory without overlaps. ECM risk sorts cleanly into six types, from the market outward to the back office.

              Market risk is the risk that conditions move against the deal: rates rise, sentiment sours, the window shuts. It is the largest risk and the one nobody controls, and it is the 2022-23 drought in a phrase.

              Execution risk is the risk that the deal is run badly: mispriced, poorly allocated, launched into the wrong week. Facebook's pricing sits here. Unlike market risk, this one is squarely the banks' to manage.

              Reputational risk is the risk that the deal damages the franchise even if it clears. A deal that breaks its issue price, or an allocation that looks like favouritism, costs the bank future mandates. Reputation is slow to build and fast to lose.

              Liability and litigation risk is the risk that the disclosure was wrong and someone sues. It traces straight back to the prospectus and to the quality of due diligence, and it is why the prospectus is a legal shield, not just a brochure. WeWork lived here before the deal even priced.

              Conflict-of-interest risk is the risk that the bank's many roles collide: banking against research, one client against another, the bank's book against its clients'. It is what Chinese Walls exist to contain.

              Operational and settlement risk is the risk that something breaks in the plumbing: a failed settlement, a data error that cascades, a system outage on listing day. The NASDAQ glitch during Facebook's debut is the classic example.

              Risk type What it is Who owns it Main defence
              Market Conditions move against the deal Nobody; systemic Flexible cost base, diversified revenue, timing
              Execution Deal priced, allocated, or timed badly The banks Skilled syndicate, disciplined pricing
              Reputational Deal clears but damages the franchise The banks Fair allocation, a healthy aftermarket
              Liability / litigation Disclosure was wrong; investors sue Banks and issuer jointly Due diligence, prospectus quality
              Conflict of interest The bank's roles collide The banks Chinese Walls, surveillance, documented allocation
              Operational / settlement The plumbing breaks Operations and IT STP, immutable audit, resilient systems
              Figure 13 The six risk types placed by how controllable they are and how fast they bite.

              How to tell a good deal from a lucky one

              A deal that pops is not automatically a deal done well

              The market applauds a big first-day jump, but a banker reads the same event more carefully. Judging a deal properly means looking at several indicators together, because any one of them can mislead.

              First-day performance is the headline: where the stock opens and closes against the issue price. The target is a modest pop, roughly ten to fifteen percent. Too little suggests the deal was overpriced and nearly broke; too much suggests it was underpriced and the company left money behind.

              Aftermarket performance is the deeper test: how the stock trades over the following weeks and months, ideally measured against its sector index so that a rising or falling market is stripped out. A deal that pops on day one and sinks in month two was not a success; it was a well-timed exit for flippers.

              Book quality looks past the oversubscription headline to what the demand was made of: how much came from long-only holders versus hedge funds, how concentrated it was, how price-sensitive. A 5x book of conviction money is worth more than a 10x book of momentum.

              The flip rate measures how many shares allocated to investors were sold in the first days. A low flip rate means the shares went to holders, as intended; a high flip rate means the allocation missed its target, whatever the day-one price did.

              Deal completion is the simplest and most brutal indicator across a franchise: what fraction of launched deals actually price and list, rather than being pulled. A bank that prices everything it launches is either very good or too cautious; a bank that pulls often is misreading the market.

              Indicator What it measures A good result
              First-day performance Open and close vs issue price A modest pop, ~10-15%
              Aftermarket performance Trading over weeks, vs the sector index Holds or outperforms after the pop
              Book quality Composition of demand, not just its size Weighted to long-only conviction
              Flip rate Share of allocation sold in the first days Low; shares stayed with holders
              Deal completion Launched deals that actually price High, without reckless pricing

              Three real failures cover the three ways a deal dies

              Nothing teaches a market faster than watching it break

              Three episodes, all real, cover the three main ways an equity deal can go wrong: a bad company, a bad price, and a bad market. Each maps onto a step you already understand from Part 2.

              WeWork, 2019 — killed by its own disclosure

              What happened

              WeWork, a co-working office company, tried to go public in 2019 at a target valuation near 47 billion dollars, and the attempt collapsed before it reached the market. When the company filed its prospectus, the S-1, investors and journalists read it closely and did not like what they found: serious governance problems, losses of roughly 1.9 billion dollars, and a tangle of related-party transactions between the company and its own chief executive.

              Root cause

              The disclosure itself was the trigger. The prospectus is meant to be a marketing document and a legal shield at once, and here the act of disclosing the facts honestly destroyed the deal, because the facts were bad and the market had not seen them clearly before. The valuation was cut from 47 billion toward 15, and then the IPO was pulled. The chief executive departed. The company later reached the public market through a SPAC in 2021 at around 9 billion, and eventually filed for bankruptcy.

              The lesson

              Due diligence and prospectus quality are not box-ticking. A weak business exposed by honest disclosure, or a strong business undone by careless disclosure, can kill a deal outright. This lands directly on Step 2.

              Facebook, 2012 — priced with no margin for error

              What happened

              Facebook went public in 2012 in a 16-billion-dollar offering, priced at 38 dollars, the top of its range. The stock opened at 42, then fell below the issue price within hours, not helped by technical failures on NASDAQ that delayed and confused trading. Within months it had fallen to 18 dollars.

              Root cause

              The deal was priced to extract the maximum, leaving no cushion. At the very top of the range there was no room for the modest first-day pop that stabilises a new stock. When it broke below the issue price, the lead banks struggled to support it, and the episode was compounded by allegations that some analysts had shared cautious views selectively with large investors. Lawsuits and a regulatory investigation followed, and the reputational damage lingered for years.

              The lesson

              Pricing at the absolute top of the range removes the shock absorber. The company captures a few extra cents per share and risks a broken deal, months of bad headlines, and lasting reputational cost. Aftermarket performance is how the market remembers both the company and its banks. This lands on Step 4.

              The 2022-23 drought — a window nobody could open

              What happened

              Through 2022 and 2023, dozens of planned IPOs were postponed or abandoned. Central banks raised rates sharply, in the US from near zero toward 5.5 percent, to fight inflation. Higher rates crushed the valuations of exactly the high-growth companies most likely to list, investors rotated into now-attractive bonds, and volatility made risk appetite evaporate. Global IPO proceeds fell around 65 percent from the 2021 peak.

              Root cause

              The cause was outside anyone's control. No amount of good execution can open a window the macro environment has closed. Banks lost hundreds of millions in ECM revenue, teams were cut, and the pipeline only rebuilt late in 2024.

              The lesson

              Market risk is systemic and uncontrollable at the level of a single deal. The only defences are structural: flexible cost bases that can shrink when the window closes, and diversified revenue that does not depend on any one market being open. It is the clearest illustration of the cyclicality from Part 1, seen from the inside.

              Case Failure type Step it maps to Lesson
              WeWork 2019 Company and disclosure Preparation and due diligence Prospectus quality can make or break a deal
              Facebook 2012 Pricing Pricing and allocation Pricing to the top leaves no room for error
              2022-23 drought Market The market cycle Macro risk is uncontrollable; only structure defends against it

              Generative AI arrives in the deal room, with governance close behind

              The five trends above predate the current wave of generative AI. It deserves its own treatment, because as of the mid-2020s large language models are moving from experiment to production in exactly the document-heavy corners of ECM, and because their risks are as distinctive as their gains.

              Where large language models actually help in a deal

              ECM is full of long documents and repetitive drafting, which is precisely what language models are good at. Three uses have moved fastest from pilot to real work.

              Drafting the first version of the prospectus. Much of a prospectus is structured boilerplate assembled from prior deals and the company's own filings. A model can produce a credible first draft of standard sections, leaving lawyers to do the high-judgement work rather than the blank-page work. The draft is a starting point, never the final word.

              Reviewing due-diligence documents. A model can read thousands of pages in a data room and surface what matters: unusual contract clauses, inconsistencies between documents, missing items. It does not replace the lawyers and bankers, but it points them at the ten pages out of ten thousand that need human eyes.

              First-draft research and marketing copy. Analysts and syndicate teams use models to draft research notes, investor summaries, and targeting rationales faster, then edit for accuracy and judgement.

              The risks are specific, and compliance treats them seriously

              A language model in a bank is not a neutral productivity tool; it is a new source of the exact risks the bank spends heavily to control.

              Confidentiality and MNPI. A model prompted with deal information is being fed material non-public information. If that data leaks into a shared model or a training set, it is a Chinese Wall breach with a new attack surface. Banks run models in controlled environments precisely to keep MNPI inside the wall.

              Hallucination and accuracy. A model can produce fluent text that is simply wrong, and in a prospectus or a research note a confident falsehood is a liability, not a typo. This is why every serious use keeps a human in the loop and treats the model's output as a draft to be verified.

              Explainability and model risk. Regulators expect a bank to explain how a decision was reached. A model that recommends an allocation or a targeting list but cannot show its reasoning collides with that expectation, which is why model governance, documenting, testing, and monitoring models, is now a discipline of its own.

              Use of GenAI The gain The governing risk
              Prospectus first draft Removes blank-page drafting Hallucination; disclosure liability
              Due-diligence review Finds the needle in thousands of pages Missed items; over-reliance
              Research and marketing copy Faster first drafts Accuracy; even-handedness
              Targeting and allocation support Sharper investor matching Explainability; model risk

              Model governance is now a regulatory regime, not an internal policy

              Banks have governed models for decades, because a mispriced derivative book taught the industry that a model is a liability as well as a tool. That discipline requires every model in production to be inventoried, independently validated before use, monitored for drift, and owned by a named person who is accountable when it misbehaves. Applying it to generative systems is straightforward in principle and awkward in practice, because a large language model resists most of what validation normally means. Its behaviour is not stable across versions, its outputs are not reproducible, and the vendor will not show you its weights.

              The European Union's AI Act, Regulation (EU) 2024/1689, adds a second, overlapping layer. It entered into force in August 2024 and phases in by tier: the prohibitions and the AI literacy duty from February 2025, general-purpose model obligations from August 2025, and the substantive high-risk regime from August 2026. It classifies systems by risk and attaches obligations to each tier. A handful of uses are prohibited outright. A larger set is designated high-risk and carries heavy requirements around data governance, technical documentation, logging, human oversight and accuracy. Most uses fall into a limited-risk tier where the main duty is transparency, telling people they are dealing with an AI system. General-purpose model providers carry their own obligations upstream.

              For the uses described above, the classification usually lands in the lighter tiers, since drafting a prospectus section or summarising diligence documents is not one of the regulated high-risk categories. That is not licence to relax. Two provisions bite regardless of tier. AI literacy obliges firms to ensure that staff deploying these systems understand what they are and are not. And the internal-use cases that touch employment decisions, which in a bank means recruitment and performance tooling, do land in the high-risk category, so the same firm will find itself operating under two very different regimes at once.

              The practical consequence is an inventory obligation that most firms discovered late. You cannot classify what you have not catalogued, and generative tooling entered banks through every door at once: vendor products that quietly added AI features, desk-level experiments, and staff using consumer tools on their own initiative. The first serious task under either regime is therefore the least glamorous one, which is finding out what is actually in use.

              Governance question Traditional model Generative model
              Can it be validated before use? Yes, against held-out data Only behaviourally, through evaluation suites
              Is the output reproducible? Yes Not reliably, and not across versions
              Can the logic be explained? Usually, from the specification Not in any meaningful sense
              Who owns it when it fails? The model owner The deploying business, not the vendor
              What changes without warning? Nothing outside a release The underlying model, at the vendor's discretion

              Full solutions are in the Answer Key.

              5.1 Classify the failure

              For each scenario, name which of the six risk types is dominant, which step of the process it attaches to, and which control should have caught it. A deal is pulled after a journalist discovers undisclosed related-party transactions. A stock closes 14% below its offer price on day one despite a book covered four times. Settlement fails on 8% of allocations because investor standing instructions were stale. A regulator fines the bank because a research analyst spoke to investors during the blackout.

              Undisclosed related-party transactions found by a journalist. Dominant risk: liability and litigation, with reputational risk close behind. Attaches to Step 2, preparation and due diligence. The control that should have caught it is legal due diligence and the related-party disclosure process, backed by the new-business committee's willingness to decline the mandate.

              Down 14% on day one despite a 4× book. Dominant risk: execution. Attaches to Step 4, pricing and allocation. The control is book-quality analysis rather than cover: a book covered four times by price-sensitive and flip-prone orders is not a covered book in any meaningful sense, and order-type and investor-type analytics exist to reveal that before the price is struck.

              8% settlement failure from stale standing instructions. Dominant risk: operational and settlement. Attaches to Step 5. The control is reference-data governance, with settlement instructions held in a golden source, validated before allocation and re-confirmed with counterparties ahead of a deal rather than on the day.

              Analyst speaking to investors during blackout. Dominant risk: conflict of interest, with a direct regulatory consequence. Attaches to the research and investor-education phase. The controls are the blackout calendar enforced in the research publication system, entitlement revocation at blackout, and electronic communications surveillance.

              5.2 The KPI trap

              Two IPOs priced in the same month. Deal A rose 32% on day one and trades 6% below its offer price three months later. Deal B rose 3% on day one and trades 11% above its offer price three months later. The bank's marketing team wants to lead with Deal A in its next pitch. Argue the case for and against, using at least three of the quality indicators from this part, and state which deal you would put in the pitch book.

              The case for leading with Deal A. A 32% first-day rise is the most legible evidence of demand available, it makes the client look successful in the press, and issuers who have never listed find it reassuring. It also demonstrates that the syndicate could generate genuine heat.

              The case against. Three quality indicators point the other way. Aftermarket performance, the more meaningful measure, has Deal A 6% below issue while Deal B is 11% above. The first-day pop on Deal A represents money the issuer did not raise: roughly a third of the deal's value transferred to initial buyers, and the subsequent decline suggests that transfer was not buying a durable register. And a stock that pops and then fades is the signature of a book weighted to investors who sold into the opening, which book-quality and flip-rate analysis would confirm.

              Put Deal B in the pitch book. The audience is a chief financial officer choosing who will price their company. Leading with Deal A advertises that the bank prices deals cheaply and that its allocations do not hold, which is precisely the wrong claim to make to the person deciding. Deal B says the bank captured value for the issuer and built a register that stayed. If Deal A must appear, it belongs as evidence of distribution reach, not of pricing judgement.

              5.3 Triage the AI use cases

              For each proposed use of generative AI, state whether you would approve it, approve it with conditions, or refuse it, and name the governing risk and the specific control. Drafting the business-description section of a prospectus from company-supplied source documents. Summarising 4,000 pages of diligence documents to flag inconsistencies for a lawyer to review. Generating a shortlist of target investors from the bank's historical allocation data. Answering investor questions during a live bookbuild from a model trained on the deal documents. Drafting the connected research analyst's earnings forecast.

              Drafting the prospectus business description from company documents: approve with conditions. Governing risk is accuracy in a document carrying statutory liability. Controls: source-grounded generation with citations back to the supplied documents, no external model access to the material, and unchanged legal review. The lawyer's sign-off is the control, and it does not become lighter because a machine produced the draft.

              Summarising 4,000 pages of diligence to flag inconsistencies: approve. This is the strongest use case in the list. The risk is a missed inconsistency, and the control is that the output is a prioritisation aid for a lawyer rather than a conclusion. It fails safe: the model surfacing ten items of which three matter is a clear gain, because the alternative is a human reading everything and tiring.

              Generating a target investor shortlist from historical allocation data: approve with conditions. Governing risks are data protection, since investor contacts are personal data, and bias, since a model trained on past allocations will faithfully reproduce past favouritism. Controls: keep the data inside the perimeter, document the basis for processing, and have a human own the final list.

              Answering investor questions live during a bookbuild: refuse. Every risk converges here. An unverified statement to an investor during an offering is a potential misrepresentation and possibly selective disclosure, made at the moment of maximum consequence with no time for review. There is no control short of human review, and human review removes the only benefit.

              Drafting the connected analyst's forecast: refuse. This is not a model risk question at all. Research independence requires that the analyst's forecast be the analyst's own work, and outsourcing it to a system trained on deal material breaches the barrier the whole structure of Part 3 exists to maintain.

              5.4 Does it move cost or risk?

              Apply the filter from this part to three proposals: replacing overnight batch settlement feeds with event streaming; adding a real-time demand dashboard for the issuer during bookbuilding; migrating the allocation engine from an on-premises data centre to public cloud with no functional change. For each, state whether it moves economics, moves risk, moves both, or moves only appearance, and defend the answer.

              Batch to event streaming: moves both, decisively. Economics improve because reconciliation staff and the breaks they investigate largely disappear rather than being made faster. Risk falls because the window in which two systems hold different versions of the truth closes. This is the clearest genuine transformation in the list, and its benefit is invisible in any demonstration.

              Real-time demand dashboard for the issuer: moves risk, modestly, and mostly moves appearance. The book is already visible to the syndicate and the issuer is already briefed continuously. What changes is that the issuer sees it directly, which improves the pricing conversation and reduces the sense of asymmetry. That is worth something. It is not worth what it is usually priced at, and it introduces a new risk: an issuer watching a live book may push for a price the syndicate knows the demand curve will not support.

              Lift-and-shift of the allocation engine to cloud with no functional change: moves appearance, mostly. Nothing about the deal changes. Costs frequently rise rather than fall, because a system designed for fixed infrastructure runs inefficiently on metered infrastructure. There is a real resilience gain if failover is genuinely rebuilt, and a real concentration risk added under the third-party rules of DORA. The honest verdict is that this is an enabling step whose value depends entirely on what follows it, and it should be defended as such rather than as a transformation in its own right.

              5.5 Carry it forward

              In a single paragraph each, connect one idea from Part 4 to one failure in Part 5, and one idea from Part 2 to one transformation trend in Part 5. The connections should be specific enough that someone who had read only one of the two parts could not have made them.

              Part 4 to Part 5. The data lineage principle explains the settlement-failure scenario completely. Stale standing settlement instructions are a reference-data defect at the top of the chain, and the chain converts a quiet data-quality problem into an 8% settlement failure on the one day the company is watching. What makes it a Part 5 risk rather than a Part 4 inconvenience is timing: the same defect on an ordinary trading day produces a break that operations clears by lunchtime.

              Part 2 to Part 5. Order types and the demand curve explain why digital bookbuilding is a genuine transformation rather than a faster interface. The syndicate's judgement in Step 3 rests on reading the composition of the book, not its total, and doing that manually across hundreds of orders arriving over several days is where errors and blind spots live. A platform that classifies orders by type, investor and timing as they arrive changes the quality of the pricing decision itself, which is why this trend attaches to Step 3 and pays for itself there.

              1. Match each case to its risk type: WeWork, Facebook, the 2022-23 drought.

              2. Which of the six risk types is endured rather than managed, and how is it defended?

              3. Why is the first-day pop a poor sole measure of a deal?

              4. Name a transformation trend that changes risk rather than just appearance, and say why.

              5. Carrying forward from Part 4 — name one control from the systems part that would have prevented one of the three failure cases, and say why it would not have prevented the other two.

              6. Carrying forward from Part 1 — why does the market cycle make technology investment harder to sustain than headcount?

              5

              1. Match the case to its risk type: Facebook's IPO.

              2. Which risk type is endured rather than managed?

              3. Why is the first-day pop a poor sole measure of a deal?

              4. Which transformation actually changes risk rather than appearance?

              Module Summary — The Six Ideas That Matter Most

              If you retain only a handful of things, retain these. Each compresses a whole part into a single, load-bearing idea.

              The product defines the platform, not the size

              ECM bridges investment banking, capital markets, and investor relations. Its products range from six-month IPOs to overnight ABBs, and the technology they need can differ by an order of magnitude, so the product, not the size, defines the platform. The convertible proves the rule by breaking it: an equity-linked deal is priced on a volatility surface and a credit spread rather than a book of orders, settles as a bond while its hedge settles as equity, and therefore runs on infrastructure the equity stack does not contain.

              Data cascades, so errors are never contained

              The end-to-end flow has clear system dependencies. An order entering the bookbuilding platform carries a handful of fields — investor identity, quantity, price type, investor classification — and every downstream object depends on them. An error at entry is not a bookbuilding problem; it is a bookbuilding, allocation, confirmation, settlement, and regulatory-reporting problem, simultaneously. The corollary is that the only cheap place to fix a data error is where it was created, and that reference data held in a genuine golden source is worth more than any amount of downstream reconciliation.

              Bookbuilding is where the price is really made

              It is the critical step, because it is where price discovery happens. Valuation sets the range, but investor demand, weighted by investor type as much as quantity, determines the price. A book five times covered by long-only conviction money is worth more than a ten-times book of hedge-fund limit orders, which is why the shape of the demand curve matters more than its height. The expectation that curve resolves against was formed weeks earlier, during research and investor education, by analysts the issuer never briefed. And the discount left on the table at pricing is structural rather than accidental: it pays for participation, for honest bidding, and for a first day that does not damage the register. The IPO is the first deal in a company's equity life, not the last; lock-up expiry, sell-downs, and follow-ons replicate the same machinery for years.

              Compliance is core architecture, not an afterthought

              Chinese Walls, six regulatory regimes, and the messaging standards that carry every instruction shape the architecture from day one. FIX moves front-office orders in real time; ISO 20022 moves post-trade instructions to the CSD. Beneath both sits the fact that gives the architecture its shape: the bank holds price-sensitive information for months, and the most dangerous person holding it has a valid login, so entitlements are granted per deal, wall-crossings are logged, communications are surveilled and documents are watermarked. The systems that run the critical steps are held to an availability standard with redundancy, automatic failover, rehearsed disaster recovery and stated recovery objectives, because on listing day there is no second chance — and the cheapest control of all is deploying nothing that week.

              League tables quietly distort behaviour

              They drive competitive behaviour, sometimes to the point of distorting pricing, because banks compete for ranking as well as for fees. Reading these incentives explains a great deal of what banks actually do, including taking marginal deals or pricing them cheaply to capture volume credit. The same logic operates one level down, inside the syndicate: the titles on a prospectus cover are partly diplomacy, while the selling concession is settled on evidence of demand delivered, so the fee split rather than the cover tells you which desks genuinely moved the stock.

              Real transformation moves cost or risk

              The five core trends, AI-assisted investor targeting, digital bookbuilding, RegTech automation, cloud and API-first architecture, and straight-through processing, each attach to a step you already know, and each earns its place the same way: by shifting the cost curve or the risk profile rather than putting a nicer surface on the same manual work. That filter is the one to carry into any technology proposal. Generative AI belongs on the list under exactly the same test, and it now arrives with a governance regime attached rather than an internal policy, because the risk that escapes detection is not the inaccurate output, which review catches, but the confidential input pasted into a system outside the bank's perimeter.

              Process Cheat Sheet — The Whole Deal on One Page

              A single reference for the entire IPO flow. Print this. Everything below is developed in full in Parts 2 to 4.

              Stage Timing Core activity Key output Lead and systems
              1. Origination Ongoing Pitch, win the mandate, agree the fee Engagement letter Origination · CRM, pitch tools
              2. Preparation 8-12 wks Due diligence, prospectus, valuation Approved prospectus, price range Legal, auditors · doc mgmt, data room
              2½. Research & PDIE 3-4 wks Analyst presentation, connected research, investor education Indicative price range Research, syndicate · publication, entitlements
              3. Marketing 2-3 wks Pilot fishing, cornerstones, roadshow Investor feedback, indications Sales, syndicate · investor DB
              4. Bookbuilding 5-10 d Collect graded demand, build the book Order book, demand curve Syndicate, sales · bookbuilding platform
              5. Pricing & allocation 1-2 d Fix the price, allocate scarce shares Final price, allocation table Syndicate, ECM · allocation engine
              6. Settlement & listing T+1 / T+2 Create shares, DVP, admit to trading Settled shares and cash Operations, CSD · settlement systems
              7. Stabilisation ~30 d Support the stock, greenshoe, research Orderly aftermarket, coverage Trading, research · surveillance

              Key numbers to remember

              Metric Rule of thumb
              IPO timeline 4-6 months, mandate to listing
              Gross spread IPO 3-7% · follow-on 2-3% · ABB 1-2%
              Healthy oversubscription 3-10x (under ~1.5x = at risk)
              Target first-day pop ~10-15%
              Greenshoe Up to 15% over-allotment
              Lock-up Typically 90-180 days
              Research cooling-off ~25-40 days after listing
              Settlement US T+1 (since May 2024); EU/UK/CH T+2, moving to T+1 on 11 October 2027
              Fee split inside the spread Management 20% · underwriting 20% · selling concession 60%
              Syndicate economics Global coordinator ~40% · joint bookrunner ~30% · co-lead ~20% · co-manager ~12%
              Free float minimum Commonly 10-25% of shares, depending on venue and segment
              Conversion premium (equity-linked) 25-40% above the share price at issue
              Recovery objectives, live book RTO measured in minutes · RPO effectively zero
              Wall-crossing population Hundreds on a large IPO, mostly outside the bank

              The one-sentence version

              An equity deal wins a mandate against a possible trade sale, makes the company listable, builds an approved prospectus and a valuation range, lets analysts form the price expectation before management ever markets, discovers the price by collecting graded demand, prices at a deliberate discount, allocates scarce shares to the holders the issuer wants, settles shares against cash through DVP, and supports the new stock for a month, with data cascading through one connected chain of systems the whole way and a confidentiality perimeter around all of it.

              Sources and further reading

              Regulatory positions and market data stated in this module were checked in July 2026. Regulation in this field changes on a scale of months, so treat every date below as a fact with a shelf life and verify it before relying on it in client work.

              How to read this section

              Where the module states a legal obligation, it is naming a specific instrument, and the reference is given below so it can be checked at source rather than taken on trust. Where the module gives a number that is not a market convention, its origin is identified. Figures explicitly flagged as illustrative in the text — the TechCo case throughout, and the fee and allocation examples — are invented to teach a mechanic and have no source, by design.

              Primary legislation referenced

              Short name used in the module Instrument Relevance
              Prospectus Regulation Regulation (EU) 2017/1129 Content, approval and passporting of the prospectus
              Market Abuse Regulation (MAR) Regulation (EU) No 596/2014 Inside information, insider lists, market soundings, stabilisation safe harbour
              MiFID II Directive 2014/65/EU Conduct, research and inducements, product governance
              MiFIR Regulation (EU) No 600/2014 Transaction reporting and transparency
              GDPR Regulation (EU) 2016/679 Investor and contact data as personal data
              CSDR Regulation (EU) No 909/2014 Settlement discipline and the settlement cycle
              DORA Regulation (EU) 2022/2554 ICT risk, incident reporting, resilience testing, third-party risk
              AI Act Regulation (EU) 2024/1689 Risk tiers, transparency, AI literacy, high-risk obligations

              Dated facts and where they come from

              DORA applies from 17 January 2025. The regulation was adopted on 14 December 2022 and entered into application on that date, with no transition period. Supervisory material is published by ESMA and the other European supervisory authorities.

              The AI Act phases in between 2024 and 2027. It entered into force on 1 August 2024. Prohibited practices and the AI literacy obligation applied from 2 February 2025; general-purpose AI model obligations and the governance and penalty provisions from 2 August 2025; the substantive high-risk regime from 2 August 2026. The European Commission maintains the authoritative timeline.

              The EU, UK and Switzerland move to T+1 settlement on 11 October 2027. The amending regulation adjusting the CSDR settlement cycle was published in the Official Journal on 14 October 2025 for application on that date, and the UK has legislated to the same date. The United States, Canada and Mexico moved to T+1 in May 2024.

              SWIFT retired legacy MT messages for cross-border payments on 22 November 2025. This ended the CBPR+ coexistence period for the MT 1xx, 2xx and 9xx families. It did not cover the MT 54x securities settlement messages discussed in Part 4, which remain ISO 15022 and are migrating separately.

              Global IPO proceeds by year. The cycle chart in Part 1 uses EY Global IPO Trends, which reports approximately US$272bn in 2020, US$460bn in 2021, US$184bn in 2022 and US$126bn in 2023, with 2024 broadly flat against 2023. Figures from different providers differ by several per cent because they classify listings and count proceeds differently; the chart is drawn to convey the shape of the cycle, not to be read off.

              Where to go deeper

              For the regulatory material, the consolidated texts on EUR-Lex are authoritative and the supervisory authorities' own guidance is more usable: ESMA for markets and settlement, the EBA for prudential and ICT matters, and the national competent authority for anything venue-specific. For the messaging standards, the SWIFT standards documentation and the ISO 20022 message catalogue are the primary references, and both are public.

              For market data, the quarterly IPO reviews published by the large accounting firms are free, consistent over time and adequate for orientation. Anything used in client work should come from the bank's own subscription data rather than from a public summary, because the classification rules behind league-table and volume figures matter more than the headline numbers.

              On the theory behind underpricing in Part 2, the three explanations given there correspond to established strands of the academic literature on new-issue pricing: adverse selection and the winner's curse, information revelation during bookbuilding, and reputational or insurance-based accounts. Any corporate finance text covering initial public offerings will set out the models in full.