Aller au contenu
Corporate & Investment Banking Mon parcours English

Accueil / Capital Markets & Securities Services / Fonctions transverses / Module 30

30Client Reporting & Performance

How a bank shows each client what they own, what it is worth, and how it has performed — and the data and technology that make it accurate, timely and trusted

Montrer à chaque client ce qu’il détient et comment cela a performé.

Lecture
2 h 00
Sections
69
Exercices
1
Questions de rappel
33
Questions de QCM
16
Simulateurs
1

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

Two threads run through this module

The business thread follows trust. Client reporting is where a relationship is verified. A client hands a bank money and a mandate; for months afterwards, almost everything they experience of the bank is a statement, a valuation or a performance pack. If those are accurate and punctual, the relationship compounds. If a valuation is wrong, a statement is late, or a figure is restated, the damage is immediate and it spreads to every number the client has ever received. Reporting is therefore not administrative overhead. It is the visible surface of the entire business, and the surface is where trust is won or lost.

The technology thread follows data. A report is a join across positions, prices, transactions, reference data, corporate actions and client entitlements — assembled from many custodians, many asset classes and many currencies, then valued, calculated and delivered to a deadline. The report itself is the easy last step. The hard part, perhaps eighty percent of the effort, is getting the data ready, reconciled and validated before a single page is composed. Every serious reporting problem is a data problem wearing a document's clothes, which is why the modern answer is architectural: one reconciled source of truth, one calculation layer, many presentation channels.

How to read this module

Read it front to back the first time. Each part assumes the one before it. Part 1 establishes what reporting is and where it sits; Part 2 walks the production cycle end to end and is the longest and most important; Part 3 maps the people and teams; Part 4 opens the systems and data; Part 5 covers what goes wrong and where the business is heading. A one-page cheat sheet and a full glossary close the module.

Throughout, coloured boxes carry specific kinds of content. A Key insight box crystallises an idea worth remembering. A Going deeper box adds nuance for the reader who wants it. An IT & data lens box names the systems and data that carry a concept. A Watch out box flags a common trap. A Worked example box makes an abstraction concrete, usually through a single running case. An In a nutshell box recaps a part, and a Check your understanding box lets you test yourself before moving on.

Illustrative figures are flagged as such. Any number in this module that is not a cited real fact — a sample valuation, a performance figure, a fee — is illustrative only, present to make a mechanic tangible, not to state a market fact.

What you will be able to do

You will understand client reporting end to end

You will be able to follow a report from the moment data is cut off at period-end to the moment the client opens it, and explain what happens at every stage in between: reconciliation, valuation, performance and attribution, validation, composition and delivery.

You will be able to map every actor and team

You will know who produces reporting and who consumes it — the client, the front office, the reporting operation, the performance and risk team, fund administration, and the technology platform beneath them all — and where quality breaks at the handoffs between them.

You will understand where the numbers actually come from

You will understand that every figure in a report rests on a price, or mark, and be able to explain the pricing hierarchy, why stale and mistimed prices are the main source of error, and how time-weighted and money-weighted return answer two genuinely different questions.

You will be able to trace data front to back

You will be able to follow one object — a position, a price, a transaction — from the source system through the reconciled data warehouse and the calculation engines to the composed statement and the client channel, and see why an error upstream reaches every report downstream.

You will recognise the risk and transformation landscape

You will be able to name the ways reporting fails, recognise three real cases that map to three distinct failure modes, and judge which technology trends — digital-first delivery, a single data and calculation layer, and generative AI — actually change the economics or risk of reporting rather than merely its appearance.

1What Client Reporting Is and Why It Decides Trust

Before any mechanics, one idea: reporting is the moment the client sees the bank's work. This part builds the concept from zero — what a report is, where it sits in the chain, the forms it takes, the audiences it serves, and why accuracy and timeliness are not niceties but the product itself.

Reporting is the visible surface of the whole relationship

Client reporting tells each client, clearly and on time, what they own, what it is worth, and how it has performed — with the costs, income and risks made transparent. That one sentence is the entire discipline. Strip away the formats and the regulations and every report answers three questions a client always asks: what do I hold and what is it worth today; how has my money performed, and against what; what did it cost me and what did it earn. A report that answers those three cleanly has done its job.

The point that surprises newcomers is how much of a client's experience is the report. A client feels a trade for an instant. They live with the reporting for years. Most of what a relationship consists of, after the money is invested, is a sequence of statements, valuations and performance packs landing on a schedule. The bank may have executed brilliantly, managed risk carefully and settled every trade cleanly — but the client cannot see any of that directly. They see the report. The report is the interface to everything else, and an interface that is wrong or late makes the whole machine behind it look untrustworthy, however well it actually ran.

This is why reporting is strategic, not clerical. Transparency is how a client verifies the bank did what it said it would. Clear, accurate reporting is the evidence of a promise kept, and evidence delivered on a reliable cadence is the foundation of a relationship that lasts and grows. There is also a harder edge: much of what a report must contain is mandated by law. MiFID II costs-and-charges disclosure, the PRIIPs key information document, SFDR sustainability disclosures and fund rules dictate what must be shown and how. Reporting is therefore simultaneously a service the client values and a compliance obligation the bank cannot decline.

Errors are reputationally fatal, so accuracy and timeliness are the product

A wrong valuation or a missed statement erodes confidence instantly. Unlike most operational slips, a reporting error is seen — repeatedly, by the client and often by their advisers and auditors. A single wrong number does not just misstate one line; it casts doubt on every number the client has ever received, because the client now has to wonder what else was wrong. A late report, meanwhile, signals an operation that is not in control of its own calendar. Trust is slow to build and instant to lose, and reporting is where it is tested every single cycle. That is the sense in which being right and being on time are not qualities of the product — they are the product.

Reporting sits at the end of the chain, so it inherits every upstream error

Reporting is downstream of everything. A trade is executed, a position is booked, a price is struck, a corporate action is applied, a calculation is run — and only then, at the very end, is a report composed. That position in the chain is the single most important fact about reporting's difficulty. Because it comes last, reporting inherits every mistake made before it. A stale price, a missed dividend, an unbooked trade, a wrong exchange rate: each one flows quietly downstream and surfaces, in full view, on the client's statement. The report is the mirror the whole operation looks into, and the mirror cannot flatter what it reflects.

This is why an experienced reporting manager thinks first about data quality and controls, not about layouts and fonts. The presentation layer is genuinely the easy part. The reporting problem is really a data-quality problem: reconcile the positions, validate the prices, apply the corporate actions, and the report almost writes itself. Skip those, and no amount of design will save the output.

Client reporting is not regulatory reporting — do not confuse the two

Two disciplines share the same plumbing but point in opposite directions. Client reporting goes to the client: statements, valuations, performance packs. That is the subject of this module. Regulatory reporting goes to the authorities: transaction reporting under MiFIR, derivative reporting to trade repositories under EMIR, and similar obligations. Both draw on the same underlying positions and transactions, so they share data infrastructure and often the same reference data. But their audience, format, timing and rules are entirely different, and confusing them is a common beginner's error. When this module says "reporting", it means the client-facing kind, except where it explicitly notes a regulatory client report such as the MiFID II ex-post cost statement, which is client-facing but legally mandated — a hybrid we will return to.

Six families of report, one shared backbone

Everything a bank sends a client falls into six families, and all six are built from the same data. They differ in what they filter, value and format, not in what they are made of. That single-source principle — get the underlying data right once, and each report becomes a presentation of it — is the design goal of every modern reporting platform, and worth holding in mind from the start.

Report family What it answers The core content
Statements What is in my account? The account of record: holdings, cash and every transaction over the period — the client's canonical view
Valuations What is it worth now? Every position marked to market, with cost, unrealised P&L and accrued income at a point in time
Transaction reports What changed, exactly? A confirmed list of every buy, sell, corporate action, fee and cash movement — the audit trail of activity
Performance reports How did my money do? Returns over the period, versus a benchmark, with attribution and risk metrics
Tax reports What do I owe? Realised gains and losses, income and withholding, packaged for the client's tax jurisdiction
Regulatory disclosures What must I be told? MiFID II costs and charges, PRIIPs KIDs, SFDR statements and suitability reports — mandated content and format
Figure 1 Six data domains feed one reconciled backbone, from which all six report types are drawn

The common backbone is the reason the single-source principle works. Positions, prices, transactions and reference data feed all six families. A valuation is those positions priced; a statement adds the transactions; a performance report runs returns over them; a tax report reclassifies the same events by tax treatment; a disclosure applies a regulatory template. Solve the data once and the six report types stop being data problems and become formatting problems. Fail to solve it and you are fighting the same bad price in six different documents.

Three audiences, one platform, very different needs

A reporting platform serves three audiences that want almost opposite things, from the same underlying data. Understanding the differences early explains most of the design tension in the field.

Audience Who they are What they want What governs them
Institutional Pension funds, insurers, corporates Detailed, data-rich reporting, custom cuts, machine-readable feeds MiFID II professional-client rules; often feeds their own regulatory reporting
Wealth / private bank Individuals and families Clear, personalised, well-presented statements and consolidated views MiFID II retail protections, PRIIPs, suitability
Fund clients Funds the bank administers NAV, holdings and investor-level reporting to strict deadlines Fund rules, audit, regulator-facing

Institutional clients treat reporting as an input to their own machinery. A pension fund or insurer often feeds the bank's valuation straight into its own regulatory reporting — for example, an insurer's Solvency II asset reporting — so it wants granularity, machine-readable feeds and rock-solid identifiers, and it will notice a missing ISIN before it notices a nice chart. Wealth clients want the opposite: clarity, personalisation and a well-designed statement that makes a complicated portfolio understandable, with the stronger protections that apply to retail investors. Fund clients care above all about the net asset value being struck accurately and on time every business day, because a late or wrong NAV stops investors trading the fund. One platform must satisfy all three, which is why the shared foundation matters so much and the presentation layer must flex so widely.

Reporting is designed at onboarding, not at month-end

The quality of a client's reporting is largely decided before the first report is ever produced, when the relationship is set up. At onboarding, a set of decisions is captured that will govern every future pack: the client's classification (retail or professional, which drives their protections and mandated reports), their base reporting currency, their reporting frequency and deadlines, which benchmarks apply to which mandates, their delivery channels and language, their entitlements, and the custodians and accounts that will feed the picture. Get these right and the reporting factory has clean instructions to run on. Get them wrong — a misclassified client, a missing benchmark mapping, an account the platform does not know to include — and every cycle inherits the error until someone traces it back to the setup. This is why experienced operations treat onboarding data as reporting data: the client static captured at the start is the reference layer that the six data domains all join against, and a defect there is not a one-off but a recurring misstatement. Reporting, like so much of the discipline, rewards getting the data right once, as early as possible.

A valuation is what the client actually sees — so read one closely

Make it concrete: a monthly institutional valuation statement is a table of holdings, marked to market, totalled and reconciled. The extract below is illustrative, drawn from the Rhône Valley pack, but every real valuation has this shape.

Holding Quantity Price Market value (EUR) Unrealised P&L
EU Govt 2.5% 2031 5,000,000 101.35 5,067,500 +117,500
Equity ABC SA 40,000 72.40 2,896,000 +386,000
Corp Bond XYZ 4% 3,000,000 98.10 2,943,000 -72,000
Fund GHI (units) 15,000 128.55 1,928,250 +228,250
Cash EUR - - 512,340 0
Total portfolio 13,347,090 +659,750

The arithmetic teaches the concepts. Market value equals quantity times price, and the price — the mark — is the crux of everything, because it is the one number in each row that is a judgement rather than a fact. Cost is what the client paid; market value minus cost is unrealised P&L, here +659,750 EUR or +5.20% against a cost basis of 12,687,340 EUR — profit that exists on paper but has not been crystallised by a sale. Accrued income, here 76,240 EUR of coupons and interest, is money earned but not yet paid, owed to the client at the valuation date. And two dates must always be explicit and consistent: the as-at date the value refers to (30-Jun close) and the priced-at moment it was actually struck (01-Jul, 06:00). Get those two out of step and the total stops meaning anything.

Valuation and pricing are the source of truth

Every number in every report rests on a price, so where the mark comes from is the whole game. Prices are not equally easy to obtain, and the difficulty rises sharply as liquidity falls.

Figure 2 The pricing hierarchy: certainty of the mark falls as liquidity falls

For liquid, listed assets — most large-cap equities, major government bonds — a clean market close from an exchange or a vendor feed gives an unambiguous mark. That is the easy case. For less liquid assets, and many corporate bonds fall here, there is no continuous market price, so the mark comes from an evaluated or vendor price, or a dealer quote — a considered estimate, not a trade. For genuinely illiquid or private assets, valuation is a model or an appraisal, updated infrequently, and it is the hard case where judgement dominates. A documented pricing hierarchy defines which source to use, in what order, so that the mark is a governed decision rather than a matter of whoever grabbed a number first.

Getting pricing wrong is the main way reporting fails

The four classic pricing failures are worth naming, because each has a dedicated control. A stale price is a mark that has not updated — a holiday, an illiquid name, a feed gap — and it silently misstates value until someone notices; stale-price detection is a core control, not an optional check. A bad or fat-finger price is an erroneous vendor figure that flows straight into the client's valuation unless tolerance checks flag the outlier before publication. An FX or timing mismatch occurs when positions are priced at different moments, or converted at the wrong rate, so the portfolio total quietly breaks. Governing all of this is a pricing or valuation committee that owns the policy, the sources, the overrides and any fair-value adjustments — an audited control function, because the mark is where the largest, most invisible errors live.

Regulation turns transparency into a computation-and-document challenge

A large part of what a report must contain is not chosen by the bank but mandated by regulation, and two rules in particular turned transparency into a hard engineering problem. They are worth meeting now, because they recur through every later part.

MiFID II costs and charges — no hidden cost

MiFID II requires firms to disclose all costs and charges — product costs, service costs and any third-party payments — aggregated and in cash terms, not just as percentages. The intent is blunt: the client should see the true, all-in cost of investing, with nothing buried. The rule has two halves. Ex-ante disclosure is an estimate of total costs shown before the client invests, so the cost is visible up front. Ex-post disclosure is an actual annual statement of what the client really paid over the year, including the drag those costs put on the return. Producing an ex-post figure sounds trivial and is not, because the cost components are scattered across product, custody and fee systems that were never designed to produce a single aggregated, per-client, cash-terms number. As Part 5 will show, building that aggregation capability was a multi-year programme for many European firms.

The ten-percent depreciation alert — a real-time obligation

For discretionary portfolios or leveraged positions, if the value falls ten percent from the last reported figure, the firm must tell the client by the end of the same business day — and again for each further ten percent. This is not a periodic report; it is a hard, time-critical trigger. Meeting it needs a daily valuation of every affected client, a per-client baseline to measure the fall against, and an automated alerting channel — in other words a real-time monitoring system, not a monthly pack. The data burden is severe: a wrong price triggers a false alert or, worse, suppresses a real one, so the same pricing discipline that governs the monthly valuation must run every single day for these clients.

PRIIPs KID — a standardised document, industrially produced

The PRIIPs Key Information Document is a short, standardised disclosure that must be given to a retail investor before they buy a packaged product — a structured product, a fund, a unit-linked insurance policy. It has a fixed content set: a summary risk indicator on a one-to-seven scale, performance scenarios from favourable to stress, costs over time expressed as a reduction in yield, a recommended holding period, and how to complain. It is capped at three pages with the same layout everywhere, precisely so a retail client can compare products like for like.

MiFID II costs and charges PRIIPs KID
Discloses all costs, aggregated, in cash terms Standardised pre-sale disclosure for packaged products
Ex-ante estimate before investing; ex-post actual annually Risk indicator, scenarios, costs, holding period
Plus the same-day 10% depreciation alert Maximum three pages, identical layout everywhere
A data-aggregation challenge across many systems A mass-production challenge: thousands of current documents

What makes the KID a reporting challenge is not the content but the scale and the calculation. The risk indicator, the scenarios and the costs use tightly specified methodologies — reproducible, auditable computation, not free text — and a large issuer may need thousands of KIDs, each product-specific and kept current, fed by historical returns, volatility and cost data through a rules engine. It is an industrial document-generation task sitting on top of a genuine data-and-calculation pipeline, and it applies to retail clients who get stronger protections than institutions: appropriateness and suitability tests, and the clear-fair-not-misleading standard on everything shown.

The reporting cycle is a calendar-driven factory

Reporting runs on a relentless calendar of deadlines, and each frequency serves a different need. Daily reporting — fund NAVs, institutional position and cash files, portal refreshes — runs every business day to tight, non-negotiable deadlines. Monthly reporting is the workhorse: the valuation, transactions, income and often performance pack that is the standard institutional and wealth cadence. Quarterly reporting adds fuller performance and attribution reviews, ESG reporting and often the formal client-review meeting pack. Annual reporting brings year-end statements, tax packs and, for funds, audited reporting. Alongside these scheduled cycles sit two triggered kinds: event-driven reports such as the MiFID II ten-percent depreciation alert, corporate-action notices and contract notes, which fire on an event rather than a date; and ad-hoc or on-demand cuts, whether a bespoke client request or a self-service portal query, which are increasingly the norm.

The discipline that matters is meeting the cycle, every time. A daily NAV that is late or wrong stops investors trading a fund. A monthly pack that slips damages trust. Much of the reporting operation is not about producing anything clever; it is about hitting the deadline reliably, which is exactly why automation, data readiness and exception handling dominate the engineering agenda. The factory metaphor is precise: the same product, produced to a schedule, where a missed slot has real consequences and cannot be recovered by working harder tomorrow.

Why this is a permanent and growing workload, not a one-off build

Reporting is not a system you build once; it is a capability you rebuild continuously, driven by two waves that never stop. The first wave is regulatory. MiFID II forced full cost transparency across European firms from 2018; PRIIPs mandated the standardised retail KID; SFDR and the EU Taxonomy added sustainability disclosure; the EU Retail Investment Strategy now proposes to standardise cost and value-for-money disclosure further. Each wave lands a new build-and-change programme on the reporting stack, and because the rules keep evolving, the backlog never clears. The second wave is client expectation. Delivery has moved from posted paper to secure PDF to interactive portals to machine-readable API feeds, and clients increasingly expect on-demand, personalised, mobile access to the same governed numbers. Together these two waves make reporting one of the largest permanent IT workloads in a bank — a business that is cyclical in its deadlines and structurally growing in its demands. That combination is why the technology under reporting is a strategic asset, not a cost centre, and why the rest of this module spends so much time on the data and systems beneath the documents.

The scale and economics of client reporting

Client reporting is not a niche back-office task; it sits on top of one of the largest industries in finance, and its economics decide how it is run. The custody and asset-servicing industry that produces the underlying data operates at enormous scale: global assets under custody and administration run into the hundreds of trillions of dollars, and the largest custodian and asset-servicing banks — the likes of BNY, State Street and J.P. Morgan, and in Europe BNP Paribas and CACEIS — each hold or administer assets measured in the tens of trillions. Every one of those positions must be valued and reported to someone, on a schedule, which is why reporting is a permanent, industrial-scale workload rather than a cottage activity.

Two economic facts shape every reporting operation. First, the cost to serve is dominated by fixed costs — the platform, the data, the controls — so unit cost falls sharply with scale, which is exactly why asset servicers exist and why the make-versus-outsource decision of Part 3 turns on volume. Second, the cost of an error is wildly asymmetric to the cost of preventing it: a tolerance check that catches a stale price costs minutes, while the error it prevents can cost a restatement, a compensation payment, a regulatory fine, or a lost mandate. A single wrong daily NAV that forces investor compensation can cost more than a year of the reporting team that should have caught it (illustrative). That asymmetry is the entire economic justification for spending on reconciliation, validation and resilience.

Reporting is booked as a cost but competed on as a product. For most of its history it was treated purely as a cost centre — a compliance obligation to be minimised. That framing is now wrong. In private banking and institutional asset servicing, the quality of reporting — its clarity, timeliness, digital delivery, and consolidated and look-through views — is a decisive factor in winning and keeping mandates, and it appears explicitly in the scoring when a client chooses a manager or servicer. A firm that reports badly loses assets it never lost on performance. So the honest way to read the reporting budget is neither pure cost nor pure revenue: it is the price of retention, and the transformation business case rests on that, not on cost saving alone.

  1. In one sentence, what does client reporting tell a client, and why is it the "moment of truth" for trust?

  2. Why does reporting's position at the end of the chain make data quality its central problem?

  3. Name the six report families and the one backbone they share.

  4. How do institutional, wealth and fund clients differ in what they want from reporting?

  5. What is the "mark", and why is pricing the crux of reporting?

  6. Why is client reporting a permanent and growing workload rather than a one-off build?

1

  1. Why is data quality reporting's central problem?

  2. Why is client reporting the “moment of truth” for trust?

  3. What do the six report families share?

2The Reporting Production Cycle, End to End

This is the core of the module. We follow a monthly institutional pack from the moment data is cut off to the moment the client reads it, stopping at each stage to see what happens, who leads it, the technology that carries it, and where it breaks. Along the way we open the two hardest topics in reporting — valuation and performance — down to worked numbers.

The whole cycle at a glance

A monthly pack is produced in seven stages, and the effort is not spread evenly across them. The table below is the map; the rest of the part is the territory. Timings are expressed in business days after month-end (BD), using the Rhône Valley pack's BD+5 deadline as the reference; they are illustrative.

# Stage Timing Output Who leads
1 Data cut-off BD 0 Loaded positions, prices, transactions, corporate actions Reporting operations
2 Reconcile BD 1-2 Positions and cash agreed to custodians; breaks cleared Reporting operations
3 Value and mark BD 2-3 Portfolio marked; income accrued Operations + pricing
4 Performance and attribution BD 3 Returns, benchmark comparison, attribution, risk Performance and risk team
5 Validate and review BD 4 Tolerance checks passed; four-eyes sign-off; commentary Operations + reviewer
6 Compose BD 4-5 The pack assembled from templates Technology + operations
7 Deliver BD 5 PDF and portal published; client reads Technology + front office

The effort curve: most of the work happens before the report is written

The single most important fact about the cycle is where the effort concentrates. It is not at the end, in composition and delivery, which are largely automated. It is in the middle-front — cut-off, reconciliation, valuation and validation — the stages that get the data right. Roughly eighty percent of the effort in a reporting cycle is data readiness, reconciliation and validation before a single page is composed. The report is the easy last step; the work is everything that makes the report trustworthy.

Figure 3 The reporting production cycle, with effort concentrated in data readiness rather than composition

This shape explains why reporting operations are built the way they are. If the hard part were the document, you would invest in design tools. Because the hard part is the data, you invest in reconciliation engines, tolerance checks and exception-handling workflows — the machinery that surfaces a bad price or an unreconciled position early, when there is still time to fix it, rather than late, when the deadline has arrived and the only choices are to publish something wrong or to miss the slot. Everything in this part is organised around that reality.

Step 1 — Data cut-off starts an unforgiving clock

At cut-off the period closes and the inputs are frozen, and from that instant the deadline is fixed. Month-end positions, prices, transactions and FX rates are loaded, and corporate actions falling in the period are applied. The clock starts here because the deadline — BD+5 for the Rhône Valley pack — counts from the close, and nothing downstream can begin until the inputs are in. A single late feed at this stage cascades: if a custodian's file arrives on BD+2 instead of BD+1, every later stage compresses, and the risk of a late or rushed pack rises. Cut-off is therefore less an event than a coordination problem, ensuring that many independent sources all deliver a consistent, complete picture as at the same moment.

Step 2 — Reconciliation is the primary control

Before anything is valued, positions and cash are reconciled to the custodians and counterparties, and this is the single most important control in reporting. Reconciliation means independently checking the bank's own book of record against the custodian's record of what is actually held, position by position and currency by currency, and investigating every difference — every break — until it is explained and cleared. A break might be a trade booked on one side but not the other, a corporate action applied by one party and not the other, or a timing difference. Whatever the cause, an unreconciled position must not be reported, because reporting a holding the custodian does not confirm is reporting a fiction.

Reconciliation is to reporting what the internal deal-governance gate is to a capital-markets transaction: the control that must be passed before the bank commits its name to a number. It is deliberately placed before valuation, because there is no point pricing a position you cannot confirm exists. In the Rhône Valley pack, reconciliation runs across both custodians and all three currencies, and the rule is absolute — breaks are investigated and cleared before valuation begins, not patched afterwards.

Step 3 — Valuation marks the portfolio and accrues its income

With positions confirmed, each is marked to market and its income accrued, turning a list of holdings into a value. This is where the pricing hierarchy from Part 1 does its work: liquid names take an exchange close, less liquid names take an evaluated or vendor price, illiquid names take a model or appraisal, and every mark is drawn from the documented source of truth in the agreed order. Income is accrued in parallel — coupons and interest earned but not yet paid are calculated up to the valuation date, so the client sees what they are owed, not only what has been received.

Consistency of timing is as important as the marks themselves

A portfolio total is only meaningful if every position is priced as at the same moment and converted at the same FX snapshot. This is subtle and it is where totals quietly break. If the equities are marked at a European close but a US holding is marked at the previous US close, and the FX rate used is neither day's, the total is a number that corresponds to no actual moment in time. The valuation policy therefore fixes a single as-at point and a single FX snapshot, and every mark is struck consistently against them. In cross-currency portfolios like Rhône Valley's, this timing discipline is the difference between a total that reconciles and one that does not.

Step 4 — Performance and attribution answer "how did my money do"

A valuation says what the portfolio is worth; performance says how it got there, and it is the analytical heart of the pack. This stage is owned by the performance and risk team, who compute returns, compare them to a benchmark, decompose them by attribution, and add risk metrics. It is the most methodologically demanding stage in reporting, and it starts with a distinction that causes more client disputes than any other: there are two legitimate ways to measure return, and they can give different answers.

Two returns: time-weighted judges the manager, money-weighted describes the client

Figure 4 Time-weighted and money-weighted return answer two different questions about the same portfolio

Time-weighted return (TWR) measures the return of the portfolio itself, stripping out the effect of the client's own cash flows. It is computed by chain-linking the return of each sub-period between cash flows, so that a deposit or withdrawal does not distort the figure. TWR answers a specific question: how well did the manager do, independent of when the client happened to add or remove money? Because it removes the client's timing decisions, TWR is comparable across managers and against benchmarks, which is why it is the industry standard for reported performance and the basis of the GIPS standards we come to shortly.

Money-weighted return (MWR), also called the internal rate of return, measures the return the client actually experienced, including the size and timing of their cash flows. It is the discount rate that sets the net present value of all the flows — deposits, withdrawals and the final value — to zero. MWR answers a different question: what did I, the client, actually earn on my money? The two diverge whenever cash-flow timing matters. A large deposit made just before a market rally boosts the client's MWR, because more money was invested during the good period, but it does not change TWR, because TWR deliberately ignores when the money arrived.

A performance summary makes the concepts tangible

A quarterly performance summary puts returns, benchmark and attribution on one page — read it closely. The extract below is illustrative, from the Rhône Valley pack's global-balanced mandate for Q2 2026.

Returns Portfolio Benchmark Relative
Period return (Q2) +3.42% +2.90% +0.52%
Year to date +6.18% +5.55% +0.63%
1 year +9.74% +8.90% +0.84%
3 year (annualised) +7.10% +6.65% +0.45%
Volatility (1y, annualised) 8.9% 9.4% -0.5%
Sharpe ratio (1y) 0.94 0.82 +0.12

Relative return is portfolio minus benchmark, here +0.52% for the quarter — the portfolio beat its yardstick. But the return is only half the story. The risk figures show the portfolio achieved it with lower volatility (8.9% against 9.4%) and a higher Sharpe ratio (0.94 against 0.82), meaning more return per unit of risk taken — a better result than the headline +0.52% alone conveys. And the +0.52% can itself be decomposed, which is the job of attribution.

Attribution decomposes the relative return into decisions

Attribution answers *why* a portfolio beat or lagged its benchmark, by splitting the relative return into the decisions that produced it. The classic decomposition has two main effects. The allocation effect captures whether over- or under-weighting whole sectors or asset classes helped — a top-down bet on which parts of the market to hold. The selection effect captures whether the specific securities picked within each group beat that group's average — a bottom-up bet on names. A third interaction and other term captures the combined effect plus currency and timing effects in multi-asset portfolios. The extract below decomposes the same +0.52%.

Attribution (Q2) Allocation Selection Total
Equities +0.18% +0.12% +0.30%
Fixed income +0.06% +0.09% +0.15%
Alternatives -0.02% +0.05% +0.03%
Cash / FX +0.01% +0.03% +0.04%
Total relative +0.23% +0.29% +0.52%

The parts must sum to the total — a built-in reconciliation check. Allocation contributed +0.23% and selection +0.29%, and they add to the +0.52% relative return, so the model reconciles. This is not a formatting nicety; attribution that does not add up to the measured relative return signals a data or model error, and a good performance analyst treats a failed reconciliation as a stop sign, not a rounding issue. Attribution turns a bare number into a narrative about skill and decisions, which is exactly why clients value it — and why it is hard, because it needs holdings and returns at the security level, aligned to benchmark constituents, with currency effects separated cleanly.

The attribution arithmetic, in one line each

The two effects are not magic; each is a single multiplication, and seeing the arithmetic once demystifies the whole decomposition. Take one segment — say equities — with a portfolio weight w_p and a benchmark weight w_b, a portfolio return inside the segment r_p and a benchmark segment return r_b, and an overall benchmark return R. The allocation effect is the weight you bet times how that segment did against the whole benchmark: (w_p − w_b) × (r_b − R). Over-weight a segment that beat the overall benchmark and the allocation effect is positive. The selection effect is the benchmark weight times how much better your securities did within the segment: w_b × (r_p − r_b). Pick better names than the index inside a segment and the selection effect is positive. Sum both across every segment and you recover the total relative return — which is exactly why the columns must add up. So the +0.30% on the equities row is not a black box: it is a modest over-weight of a segment that beat the benchmark, plus security picks a little better than the segment index, each a single multiplication and then added. (Figures illustrative.)

Deep dive — benchmarking, GIPS and risk make performance credible

A return means little without a benchmark

A benchmark is the yardstick that gives a return meaning, and choosing and maintaining it is a data discipline in itself. Benchmarks come in a few forms: a published market index for a single asset class; a blended or composite benchmark, such as 60% equities and 40% bonds, matching a multi-asset mandate; a custom benchmark, a bespoke basket agreed with the client to match a specific strategy; and a peer group or absolute target, such as cash plus 3%. The discipline behind benchmarking is exacting. Index constituents and returns are licensed vendor data, so cost, entitlements and correct usage all matter. Comparisons must be like-for-like — same currency, same period, same treatment of income (a total-return index versus a price-return one is a common and serious mismatch). Custom benchmarks must be rebalanced and recomputed consistently, a standing calculation the bank owns. And benchmark changes must be tracked and disclosed, with historical comparisons using the benchmark in force at the time.

GIPS makes reported performance comparable across firms

The Global Investment Performance Standards (GIPS), from the CFA Institute, are the voluntary global rules for how firms calculate and present performance. Their purpose is to stop firms flattering their numbers — cherry-picking their best accounts, choosing convenient periods, or presenting returns in incomparable ways — so that a prospective client can compare managers fairly. The core mechanism is the composite: a firm must group all similar portfolios into a composite and show the composite's return, so it cannot showcase only its winners. GIPS then prescribes consistent methodology for return calculation, valuation frequency, fee treatment and cash flows, and requires full disclosure of dispersion, the number of portfolios, the benchmark and the fee basis alongside every compliant presentation. Firms can be independently verified as GIPS-compliant, and that verification is a strong signal of data and process integrity — which is why claiming GIPS compliance is a firm-wide commitment, prominent in institutional and fund marketing.

Risk metrics show what was taken to earn the return

Clients increasingly want to know the risk taken to earn a return, not just the return, so a modern performance report carries a standard set of risk measures. Six recur often enough to know by name. Volatility (standard deviation) is the headline risk number — how much returns fluctuate — where higher volatility means a bumpier ride for the same average return. The Sharpe ratio is return earned per unit of risk, letting clients compare risk-adjusted performance across portfolios. Maximum drawdown is the largest peak-to-trough fall over the period — the worst loss the client would actually have lived through, and very tangible. Value at Risk (VaR) is an estimated worst-case loss at a confidence level over a horizon, such as a 95%, one-day VaR — a forward-looking gauge. Tracking error is the volatility of the relative return, measuring how closely the portfolio follows its benchmark, and it is central to benchmarked mandates. And beta and related factor exposures show sensitivity to the market and to factors such as rates, credit and style — revealing where the risk actually sits. Each of these needs a clean time series of returns and exposures, so risk reporting, like everything else, is a data-history and computation problem, and consistency of method across clients is what makes the numbers comparable.

Deep dive — gross versus net, and the commentary that explains it

Returns are quoted gross or net of fees, and the difference matters

A return figure is incomplete until you know whether it is gross or net of fees, because the same performance can look materially different depending on which is shown. A gross return is the portfolio's performance before the manager's fees and costs are deducted; a net return is what the client actually keeps after those costs. The gap between them is precisely the drag that MiFID II costs-and-charges disclosure exists to make visible, and it compounds over time, so a strategy that beats its benchmark gross can lag it net. Reporting must be explicit and consistent about which basis it shows — mixing gross returns with a net benchmark, or switching basis between periods, produces comparisons that are not merely confusing but wrong. For a like-for-like comparison the portfolio and the benchmark must be on the same basis, and where both gross and net are shown, each must be labelled unambiguously. This is one more instance of the module's recurring theme: a number without its definition — its basis, its currency, its as-at moment — is not information, it is a trap.

Commentary is the narrative layer, and it carries its own risk

Most performance packs include written commentary — the "what happened this quarter" narrative — and although it is words rather than figures, it is a governed part of the report. Good commentary turns a table of returns and attribution into a story the client can follow: which decisions helped, which hurt, what the manager expects. It is valued precisely because clients want to know why, not just the number. But commentary carries risks the numbers do not. It must be consistent with the figures it describes, because a narrative that contradicts the attribution undermines both. It must be clear, fair and not misleading, the same standard that governs every retail disclosure, so it cannot spin a bad quarter into a good story. And it must not leak material non-public information or stray across the confidentiality wall into things the client is not entitled to know. This is why commentary, like every figure, passes the four-eyes gate before it goes out — and why the arrival of generative AI as a commentary-drafting tool, which Part 5 examines, is both an obvious efficiency and a genuine governance question.

Deep dive — private markets and performance-fee reporting

Two fast-growing corners of reporting speak a different language from listed portfolios, and both are worth meeting: private-markets funds and performance fees.

Private-markets reporting uses money multiples, not just returns

Private-markets funds — private equity, private credit, infrastructure, real assets — report on a different vocabulary, because the money moves and is valued differently. An investor does not hand over the full amount up front; they make a commitment, which the fund calls in tranches (capital calls) as it invests, and returns as distributions as it exits. There is no daily market price for an unlisted company, so the fund's value is the general partner's periodic mark, usually quarterly and lagged. That changes the metrics the report must carry.

Metric What it means
Commitment / called / distributed How much was promised, drawn and returned
TVPI (total value to paid-in) (Distributions + residual NAV) ÷ capital called — the total money multiple
DPI (distributions to paid-in) Cash actually returned ÷ called — how much is realised
RVPI (residual value to paid-in) Unrealised NAV ÷ called — how much is still on paper
IRR The money-weighted return on the actual cash-flow dates — the right measure here
Figure 5 The J-curve: a private-markets fund reports losses before gains

The J-curve is the shape every private-markets report shows early on. Because fees and drawn capital go out before any investment is sold, a fund reports losses in its first years and gains only later, tracing a J. A client who reads the early NAV as failure has misread the instrument, which is why private-markets reporting is as much about explaining the shape as about the numbers. Note which return metric matters: because the general partner, not the investor, controls the timing of calls and distributions, the money-weighted IRR is the honest measure here, not a time-weighted return. And the valuation caution from the H2O case lives precisely here — RVPI is the part of the multiple that rests on a lagged, estimated, GP-supplied mark, so the report must be explicit that a large share of the reported value is unrealised judgement, not cash.

Performance fees turn the report into a fee calculation

Where a manager charges a performance fee, the report must compute and disclose it correctly, and the mechanics have sharp edges. A management fee is a flat percentage of assets, simple to show. A performance fee — also called an incentive fee, or carried interest in private markets — is a share of the gains, but only of gains that clear two protections for the client. The hurdle rate is a minimum return the manager must beat before earning any performance fee, so a fund returning 5% against an 8% hurdle earns none. The high-water mark is the highest value the fund has previously reached; the manager earns a performance fee only on gains above that mark, so a client who suffered a loss is not charged twice to recover ground already paid for.

Step 5 — Validation and four-eyes review is the last gate before the client

Nothing leaves without passing automated checks and a human sign-off — the maker-checker gate. Validation runs tolerance and completeness checks across the whole pack: stale-price detection, outlier and large-move flags, missing-data checks, and the attribution reconciliation just described. Anything that trips a check becomes an exception to investigate, not a number to publish. On top of the automated layer sits four-eyes review: a second qualified person, independent of whoever produced the pack, reviews the figures and formally signs off, and the client commentary — the "what happened this quarter" narrative — is drafted and reviewed here too. The principle is simple and non-negotiable: no client report leaves the building without a control step, because this is the last moment an error can be caught before it becomes visible.

Step 6 — Composition assembles the pack from templates

With validated data, composition is the assembly step where numbers flow into branded, personalised templates — and it is deliberately the easy stage. A document-composition tool merges the reconciled, calculated data into the pack's templates: the valuation table, the transaction list, the performance summary, the commentary, the regulatory disclosures, each laid out to the client's preferences for language, currency and level of detail. Because everything upstream has been reconciled, valued, calculated and validated, composition is mechanical — which is the whole point of the architecture. If composition is doing anything clever with the numbers, something has been designed wrong; the cleverness belongs in the data and calculation layers, and the presentation layer should be a thin, reliable skin over them.

Step 7 — Delivery publishes the same numbers across channels

The finished pack is delivered by the channels the client prefers, and the discipline is that every channel shows identical numbers. For the Rhône Valley pack, that is a secure PDF and a portal refresh by BD+5; for other clients it may add a machine-readable feed into their own systems. The same governed data now has to serve documents, screens and feeds at once — one source, many channels — and the numbers on the PDF, the portal and the feed must agree exactly. This is only possible because they are all reading from the same calculation layer; the moment a channel computes its own version of a number, divergence and disputes follow. Delivery is also where security bites: digital channels raise authentication, entitlement and data-protection stakes, because convenience must never weaken the control that each client sees only, and exactly, their own data.

Deep dive — consolidated and look-through reporting

Clients rarely hold everything in one neat account, so two advanced reporting capabilities assemble the fuller picture — and both are data-availability problems. Consolidated reporting brings a client's whole financial position together: multiple accounts and mandates into one household or entity view, assets across several custodians and banks, even held-away assets where the data is supplied — one total, one asset allocation, one performance, the client's whole picture on a page. It is prized in private banking, but it is only ever as good as the completeness of the data fed in; a held-away account the client forgets to supply is simply invisible. Look-through reporting goes the other way, seeing through funds and wrappers to the underlying holdings, so that a fund-of-funds shows real exposures rather than fund names. Clients want it because true asset allocation, sector and geographic exposure, and concentration are all masked if you only see the wrapper — and the same underlying name held via several funds must be netted, because real risk lives at the look-through level. It is hard because it needs each fund's underlying holdings data, which is often lagged and licence-restricted; data availability, not calculation, is the binding constraint.

Deep dive — corporate actions, the quiet source of error

Corporate actions are the events that change a position or its value between periods, and they are one of the most persistent sources of reporting error precisely because they are easy to miss. A dividend is paid, a bond coupon falls due, a stock splits two-for-one, two companies merge, a rights issue offers new shares: each of these silently alters what the client holds or what it is worth, and unlike a trade, no one at the bank actively initiated it. The event happens in the market and the reporting platform must detect it, interpret it correctly, and apply it to every affected position — or the client's report will be wrong with no obvious symptom.

Corporate action What it changes The reporting risk if missed
Cash dividend / coupon Adds income and cash Understated income and cash balance
Stock split / consolidation Changes quantity, not value A doubled or halved position, wildly wrong market value
Rights issue Offers new shares at a price Missed entitlement; wrong position and cash
Merger / takeover Replaces one holding with another A stale holding in a company that no longer exists
Return of capital Reduces cost basis Wrong unrealised P&L and tax figures

The two hard parts are mandatory-versus-voluntary handling and the effective-date timing. A mandatory action, such as a split or a cash dividend, happens automatically and the platform must simply apply it correctly. A voluntary action, such as a rights issue or a tender offer, requires the client to make an election, so the platform must capture the decision and reflect the chosen outcome. And every action has key dates — ex-date, record date, pay-date — that determine exactly when the position and the value change, so applying an action a day early or late misstates the report. A split applied to quantity but not to the cost basis, or a dividend booked to income but not to cash, breaks the internal consistency the client's statement depends on. This is why a dedicated corporate-actions data source and a controlled application process sit at the heart of the data-readiness stage: a single missed or mis-dated action is enough to make an otherwise perfect pack wrong.

The daily NAV cycle — the same stages, compressed into hours

The monthly pack is the fullest version of the cycle, but for fund clients the same stages run every single business day, compressed into hours, and that compression changes everything. A fund's net asset value — the per-unit value at which investors buy and sell — must be struck accurately by a hard daily deadline, because investors transact on it and a late or wrong NAV freezes or misprices their trades. The stages are the same as the monthly pack: the fund's positions and cash are captured, reconciled, priced, the NAV is calculated, validated by four-eyes and published to the transfer agent, the administrator and investors. But there is no slack. Where the monthly pack has five business days to absorb a late feed or investigate a break, the daily NAV has hours, so automation is not a convenience but a precondition, and exception handling must be near-instant.

This is why fund reporting drives the most demanding operational resilience in the whole discipline. A monthly pack that slips a day is embarrassing; a daily NAV that is not struck stops a fund trading and can force compensation to investors who transacted on a wrong or estimated price. The daily cycle therefore concentrates the same effort curve — most of the work in data readiness and reconciliation, little in composition — into a window so tight that any manual step is a risk. It is the clearest illustration of the module's central point: the report itself, the published NAV number, is trivial to produce once the data is right, and the entire challenge is getting the data right, reconciled and validated, before a deadline that does not move.

The post-delivery life of a report

A report is not finished when it is delivered — it lives on as a record, a baseline and sometimes a dispute. Once published, every figure's source, calculation and any later correction must be traceable, because reports are referenced for years: by the client and their advisers, by auditors signing off, by regulators checking a disclosure was made, and occasionally by lawyers in a dispute. A published valuation also becomes the baseline for the next period's event-driven checks — most importantly the MiFID II ten-percent depreciation alert, which measures a fall against the last statement, so last month's number is this month's reference point. And if an error is found after publication, the process is not a quiet edit but a governed restatement: the correction is made, versioned, explained and re-issued, because a silent change to a number the client has already seen is worse than the original error. The audit trail and versioning that make restatement possible are therefore not back-office housekeeping; they are what let a reporting operation stand behind every number it has ever sent.

  1. Why is roughly 80% of the effort in a reporting cycle spent before the report is composed?

  2. What is reconciliation, and why must it precede valuation?

  3. How do time-weighted and money-weighted return differ, and when do they diverge?

  4. In attribution, what do the allocation and selection effects capture, and why must they sum to the total?

  5. Why must every position be priced at the same as-at moment and FX snapshot?

  6. Why are corporate actions a persistent source of reporting error?

  7. How does the daily NAV cycle differ from the monthly pack, and why is automation a precondition?

  8. Why is a post-publication error handled as a governed restatement rather than a quiet edit?

2

  1. Why must reconciliation precede valuation?

  2. TWR versus MWR: which describes the client's actual experience?

  3. Where is about 80 % of the effort in a reporting cycle?

3Who Makes Reporting Happen

Reporting is a team sport, and quality breaks at the handoffs between teams. This part maps the actors — the client, the front office, the reporting operation, the specialists and the platform beneath them — then goes deep on the two things that shape everything: the different audiences a report must serve, and the confidentiality and entitlement wall that decides who may see what.

Four camps, with the reporting operation in the middle

Reporting spans four camps, and the reporting operation sits at the centre of every interaction between them. The first camp is the client — institutional, wealth or fund — the consumer of reporting and the party whose trust is on the line. The second is the front office: sales, relationship managers and client-service teams, who own the relationship, agree the reporting needs and field the questions. The third is the internal production function: the reporting operations team, the performance and risk specialists, and fund administration, who actually manufacture the reports. The fourth is the platform: the data and technology teams who build and run the warehouse, the calculation engines, the portal and the feeds. Around all four sit external parties — custodians and data vendors supplying the raw inputs, and the client's own advisers and auditors consuming the output.

Figure 6 The reporting operation sits between the client and every internal and external source

The reporting operation is one party to almost every interaction, and that is where the risk concentrates. It receives data from custodians and vendors, agrees requirements with the front office, hands calculations to and from the performance team, coordinates with fund administration on NAVs, depends on technology for the platform, and answers to the client through the relationship manager. Because it touches everyone, it is also where every handoff can fail. Quality does not usually break inside a team; it breaks in the gaps between them — between operations and performance, between fund admin and technology, between the front office's promise and the platform's capability. A clear operating model, shared data and unambiguous ownership of each figure are what keep a report accurate and on time across all those hands.

The information and instruction flows

Every major interaction can be written as a flow — from one party, to another, at a stage of the cycle. The table makes the hub-and-spoke concrete.

From To What flows Stage
Custodians and vendors Reporting operations Positions, cash, prices, corporate actions Cut-off
Reporting operations Custodians Reconciliation queries on breaks Reconcile
Reporting operations Performance and risk team Reconciled positions and cash flows Performance
Performance and risk team Reporting operations Returns, attribution, risk metrics Performance
Fund administration Reporting operations Struck NAVs for fund clients Value
Reviewer Reporting operations Four-eyes sign-off and exceptions Validate
Technology platform All internal teams Warehouse, calculation, composition, delivery All stages
Reporting operations Front office / RM The finished pack and commentary Deliver
Front office / RM Client Delivery, explanation, query handling Deliver
Client's advisers and auditors Client / bank Checks, advice, sign-off Post-delivery

The audience taxonomy — one platform, three clients who want opposite things

The deepest fact about reporting's design is that a single platform must serve three audiences whose priorities barely overlap. Part 1 introduced them; here we go deep, because the audience type drives everything about a report — its format, its frequency, its protections and its delivery channel. This is reporting's equivalent of the buy-side taxonomy in a capital-markets deal: knowing the client type tells you what the report must be.

Institutional clients want data, not decoration

Pension funds, insurers and corporates treat reporting as raw material for their own processes. They want detailed, data-rich reporting, custom cuts and machine-readable feeds, because the bank's valuation frequently becomes an input to their regulatory reporting — an insurer feeding Solvency II asset data, a pension fund feeding its own trustees and regulators. Their overriding priorities are granularity and clean identifiers. They will forgive a plain layout and never forgive a missing ISIN, a stale price or a feed that arrives late, because those break their downstream machinery. They are typically classified as professional clients under MiFID II, so they receive fewer protective wrappers and more data.

Wealth and private-bank clients want clarity and a consolidated picture

Individuals and families want the opposite: clear, personalised, well-presented statements and a consolidated view of a portfolio that may be spread across accounts, custodians and asset types. Presentation is not vanity here — a good statement makes a complicated set of holdings genuinely understandable to someone who is not a professional investor, which is itself a form of service. Wealth clients are usually retail clients under MiFID II, so stronger protections apply: suitability, PRIIPs disclosure for packaged products, and the clear-fair-not-misleading standard on everything they are shown.

Fund clients live and die by the NAV deadline

For funds the bank administers, the priority is the net asset value: struck accurately and on time, every business day, to a deadline that does not move. A late or wrong NAV is not a service lapse; it stops investors from trading the fund and can force compensation. Fund reporting adds holdings and investor-level reporting, is audited, and is regulator-facing, so its discipline is deadline reliability above all else.

Advisers and auditors are the fourth reader

Sitting behind all three are the consultants, custodians and auditors who also consume the reporting — to check it, advise on it, or sign it off. They are not the primary audience, but they are an expert one, and they read the numbers adversarially. A report has to withstand not only the client's eye but the auditor's.

The four differences that follow from audience type are worth memorising as a grid.

Dimension Institutional Wealth / private Fund
Top priority Granularity and data feeds Clarity and presentation NAV accuracy and deadlines
Frequency Daily / monthly Monthly / quarterly Daily (NAV)
Regulation MiFID II (professional) MiFID II retail + PRIIPs + suitability Fund rules + audit
Delivery Feeds / API + PDF Portal / app + PDF Data files to TA, admin and investors

The confidentiality and entitlement wall

Reporting is built on a hard rule — each client sees only, and exactly, their own data — and enforcing that rule is a technology problem, not just a policy. A single reporting platform holds the positions, valuations and performance of thousands of clients at once. The institution therefore knows a great deal that it must never let leak from one client to another, or from one function to another, and the wall that prevents this runs through the whole platform. Three separations matter most.

The first is between clients: entitlements must guarantee that a client, logging into a portal or receiving a feed, can reach their own data and no one else's. This sounds obvious and is the single most dangerous failure in digital reporting, because a cross-client data leak is both a confidentiality breach and, under GDPR, a reportable personal-data incident. The second is between advice and reporting: where the bank gives investment advice, the reporting function must reflect what happened, not what the advice desk wishes had happened, and the suitability report that justifies a recommendation is a controlled, separate artefact. The third is around material non-public information: commentary and narrative in reports must not reveal, or be contaminated by, information the client is not entitled to — a live risk now that AI drafts narrative from data.

The six internal teams, and how they map to the cycle

Six teams manufacture and run reporting, each owning a stage and a system, with the confidentiality wall running through all of them. Knowing who does what — and where the business analyst, data engineer and developer fit — is essential for anyone building or assessing a reporting platform.

Team Function Core role Primary technology
Client reporting operations Operations Runs the factory: cut-off, reconciliation, production, review, delivery Reconciliation and reporting workflow
Performance and risk Middle office Computes returns, attribution and risk; owns methodology and GIPS Performance and attribution engines
Fund administration Operations Strikes NAVs; produces fund and investor-level reporting Fund-accounting and TA systems
Data and reference-data stewardship Data Maintains the security master, prices and client static Reference-data and pricing platforms
Client service / RM Front office Agrees needs, explains figures, handles queries and disputes CRM and client portal
Technology and change Technology Builds and runs warehouse, engines, portal and feeds The whole reporting stack

Within the technology and change function sit four roles worth naming individually. The business analyst translates client and regulatory reporting needs into requirements, data mappings and report specifications — the bridge between what a rule or a client demands and what the platform must do. The data engineer builds the pipelines that ingest, reconcile and serve positions, prices and transactions to the reporting layer — the plumbing that makes data readiness possible. The report or platform developer builds the report templates, the calculation logic, the portal and the feeds — the delivery machinery. And production support and control runs the daily and monthly cycle, monitors jobs and exceptions, and owns the operational-resilience story that Part 4 will show is decisive on the one irreversible day.

On the business and operations side, four roles complete the picture. The client reporting analyst produces and checks packs, clears breaks and meets the cycle. The performance analyst computes returns, attribution and risk, and owns the methodology and GIPS compliance. The data or reference-data steward maintains the security master, prices and client static — the keys everything joins on. And the client-service manager or relationship manager agrees reporting needs, explains numbers, and handles queries and disputes, standing between the operation and the client.

Queries and disputes — where the relationship is tested after delivery

Delivering the report is not the end of the interaction; the client's questions and occasional disputes are, and how they are handled is a real part of the operating model. A client who does not understand a figure, disagrees with a valuation, or spots what they think is an error raises a query, usually through the relationship manager. That query has to be triaged — is it a misunderstanding to explain, a genuine data issue to investigate, or a disagreement over methodology to escalate — and answered against the audit trail, because the ability to show a figure's source, calculation and any correction is exactly what lets the operation resolve a dispute with evidence rather than assertion. Most queries are explanatory: the client asks why performance differs from a headline index, and the answer is the difference between gross and net, or a benchmark mismatch, or a cash-flow-timing effect. Some are genuine catches — a client is often the last, sharpest validation check, and a query that uncovers a real error must trigger the governed restatement process, not a quiet fix. The volume and nature of queries is itself a signal: a rising tide of "I don't understand this" queries points to a clarity problem in the report, while a rising tide of "this looks wrong" queries points to a data problem upstream. A mature operation treats its query log as free feedback on the quality of both its data and its design.

The control framework — who owns each number

Because reporting spans so many teams, the discipline that holds it together is unambiguous ownership: every figure has one owner, and every control has one accountable team. The recurring failure in reporting is not that no one checks a number but that everyone assumes someone else did. A clear control framework — often written as a responsibility matrix — removes that ambiguity by naming, for each stage, who does the work, who reviews it, and who is accountable if it is wrong.

Control Owned by Reviewed by Accountable
Reconciliation to custodian Reporting analyst Team lead Head of reporting operations
The mark / pricing policy Pricing team Valuation committee Chief valuation officer
Performance methodology Performance analyst Performance head Head of performance and risk
Four-eyes sign-off on the pack Independent reviewer - Head of reporting operations
Entitlements and data access Technology Information security Chief information security officer
Regulatory content and timing Compliance Second-line risk Chief compliance officer

The framework matters most at the seams, where the running example showed quality breaks. A reconciliation break that operations believes performance will catch, a stale price the pricing team assumes the reviewer will flag, a regulatory deadline each team thinks another owns — these are the gaps a control framework closes, by making the owner of each number explicit and the sign-off a named, accountable act rather than a diffuse hope. In a well-run operation, if you point at any figure in a client pack, someone can tell you immediately whose number it is.

The operating model — build it, outsource it, or both

A bank does not have to run its own reporting factory, and the choice of operating model is one of the biggest structural decisions in the field. Three models coexist, and most large institutions end up with a blend.

The in-house model keeps the whole chain — data, calculation, composition and delivery — inside the bank, run by its own operations and technology teams. It offers the most control and the greatest ability to differentiate, and it makes sense where reporting is a competitive product, as it often is in private banking. The outsourced model hands the production to a specialist asset servicer — a custodian bank or fund administrator that reports for many clients at scale. In Europe this is a substantial industry: firms such as BNP Paribas, Société Générale's securities-services arm, CACEIS and State Street run reporting and fund administration for other institutions, achieving a scale and a spread of fixed regulatory-change cost that a single mid-sized bank cannot match. The hybrid model, the commonest in practice, keeps the client-facing and differentiating layers in-house — the portal, the private-bank statement, the client relationship — while outsourcing the commoditised production of NAVs, valuations and reconciliation to a servicer.

Model What the bank keeps What it gains What it gives up
In-house The whole chain Control, differentiation, data ownership Cost, and carrying every regulatory-change build itself
Outsourced The client relationship only Scale, shared change cost, specialist expertise Direct control; dependence on a servicer's platform and timeliness
Hybrid Client-facing and differentiating layers Control where it matters, scale where it does not Complexity at the seam between in-house and servicer

Reporting as a commercial commitment — RFPs and SLAs

Because clients now choose partly on reporting quality, the reporting operation carries hard commercial commitments written into contracts. When an institutional client or a fund selects a manager or an asset servicer, the reporting capability is scored in the request for proposal alongside performance and fees: the frequency and timeliness offered, the delivery channels, the consolidation and look-through available, the data-feed formats, and the ability to feed the client's own regulatory reporting. Winning the mandate then commits the firm to a service-level agreement — a contractual promise on what will be delivered and by when.

A reporting SLA turns a deadline from a service goal into a contractual liability. It typically fixes the delivery deadline (a NAV by a stated time, a monthly pack by BD+n), an accuracy standard, and the consequence of missing it: service credits or penalties that reduce the fee, and, for a repeated or serious failure, the client's right to leave. This is why the resilience and automation of Part 4 are not engineering luxuries but the means of meeting a promise the firm has been paid to keep. And the operating-model choice compounds it: outsource production to an asset servicer and the firm still owes its own client the SLA, so the servicer's timeliness becomes the firm's contractual exposure — accountability that, as we saw, does not transfer.

  1. Why does reporting quality tend to break at the handoffs between teams rather than inside them?

  2. How do the priorities of institutional, wealth and fund clients differ?

  3. Why is "each client sees only their own data" a technology problem, not just a policy?

  4. What does an explicit control framework add, and where does it matter most?

  5. Give two of the four technology-and-change roles and what each does.

  6. In the outsourced operating model, what transfers to the asset servicer and what does not?

3

  1. Where does reporting quality usually break?

  2. What do wealth clients prioritise?

  3. Why is “each client sees only their data” a technology problem?

4Systems, Data and the Technology That Carries Reporting

Reporting is a data-and-technology problem more than anything else, so this part opens the machine. We start with the clean question of what functions any reporting platform needs, then name who builds them, follow one object from source to screen, dissect the data a report joins, and close with the identifiers, integration patterns, messaging, resilience and regulation that decide whether the whole thing holds together on the day it matters.

The functional architecture — three layers, computed once

Before naming a single vendor, ask what functions any reporting platform must perform — and the answer is a chain of six blocks over a foundation of shared controls. Data must be ingested and reconciled; stored in a single golden source; valued and priced; run through performance, attribution and risk; composed into documents; and delivered across channels. Beneath all six sit cross-cutting controls that every block depends on. The functional view matters because it is stable — the functions do not change when the vendors do — and because it reveals the one architectural idea that organises everything else.

Figure 7 The reporting stack as three layers — data, calculation, presentation — over cross-cutting controls

The decisive principle is to separate the data layer, the calculation layer and the presentation layer, and to compute each number exactly once in the middle. Data is ingested, reconciled and stored once. Each figure — a valuation, a return, a risk metric — is computed once, in the calculation layer, from that single store. And many presentation channels — the PDF, the portal, the API feed — each read the already-computed number and merely format it. This is what makes reporting consistent (every channel shows the same figure because there is only one), auditable (every number has one lineage), and cheap to extend (a new channel is a new skin, not a new calculation). It is the target state of almost every reporting transformation, and everything else in this part is in service of it.

Layer Functional blocks What it owns
Data Ingestion and reconciliation; golden data warehouse One reconciled source of positions, prices, transactions, reference data
Calculation Valuation and pricing; performance, attribution, risk Every client-facing number, computed once
Presentation Report composition; delivery channels Formatting the computed numbers for each audience and channel
Cross-cutting Security master, reconciliation, entitlements, audit The controls every layer depends on

The vendor landscape — mostly buy, selectively build

Most banks buy the core engines rather than build them, because a rich specialised vendor market exists — but where a firm chooses to build is a map of where it thinks its advantage lies. The buy-versus-build pattern is worth reading closely: the commoditised functions are bought, and the functions a firm believes differentiate it are built. The names below are illustrative of the market, not endorsements, and the landscape consolidates constantly.

Functional block Typical vendors (illustrative) Buy or build
Investment book of record SimCorp, BlackRock Aladdin, Charles River Buy
Performance, attribution, risk FactSet, SimCorp (Axioma), Aladdin, Bloomberg PORT Buy
Market and reference data Bloomberg, Refinitiv/LSEG, ICE, index and ESG vendors Buy
Report composition OpenText, Quadient-style, Vermilion-type tools Buy
Client reporting suites Vermilion, Kurtosys, InvestCloud Buy
Client portals and apps Digital-experience platforms and bespoke builds Build or hybrid
Fund admin / transfer agency Specialist TA and fund-accounting systems Buy
The reconciled data warehouse Often proprietary Build

The pattern tells a story about competitive advantage. Performance engines, market data and composition tools are bought because they are shared industry problems where a vendor achieves scale no single bank can. But two things are frequently built: the reconciled data warehouse that joins everything, and the client-facing digital experience. Those are precisely the two places a bank can differentiate — the quality of its single source of truth, and the quality of what the client actually touches. Many European banks and asset servicers — BNP Paribas, Société Générale, CACEIS, Amundi and others — run a mix of these vendor platforms plus in-house builds, and integrating across them is the real, permanent engineering job. The vendor is rarely the hard part; the seams between vendors are.

Data lineage — following one object from source to screen

Follow a single position from where it is born to where the client reads it, and you see why an error upstream reaches every report downstream. The cascade principle is the whole reason reporting is a control discipline: because reporting is last, every stage inherits the stages before it, and a mistake at the top flows all the way to the bottom in full view.

Figure 8 Data lineage: an error at the source cascades all the way to the client statement
Stage System Data object created
Trade and hold Books and records, custody feeds Position and transaction records
Aggregate Golden data warehouse (IBOR) One reconciled, identifier-mapped source
Value Valuation and pricing engine Marked position, accrued income
Measure Performance and risk engine Return, attribution, risk metric
Compose Reporting engine Statement, valuation, performance pack
Deliver Portal, PDF, API The client-facing report

The cascade cuts both ways, which is the practical lesson. A bad price entered at the valuation stage misstates the market value, then the unrealised P&L, then the performance, then every channel that shows any of them — one error, many symptoms, all visible to the client at once. But the same structure means a fix applied once, upstream, corrects every downstream report automatically. This is the argument for solving data problems at the source and in the warehouse, rather than patching them in individual reports: a patch fixes one document while the error persists everywhere else, whereas an upstream fix propagates the correction through the whole chain.

The anatomy of the data a report joins

A report is a join across six data domains, and each is a data-quality battleground. Part 1 named them; here we open them, because understanding the fields and their failure modes is what separates someone who can build a reporting platform from someone who can only describe one.

Figure 9 A report is a view joined across six data domains — it is only as good as the weakest join

Positions and holdings are the book of record: what each client owns, in what quantity, in which account. This domain must reconcile to the custodian daily, and its core failure is a position that does not agree — a phantom or a missing holding. Prices and market data provide the mark for every instrument, and this is the single biggest source of error, for all the reasons Part 1 gave: staleness, bad ticks, FX and timing. Transactions and cash are the flow — every trade, fee, coupon, dividend and cash movement — that explains how holdings changed between periods, and it must tie out to the position moves. Reference and static data are the keys everything joins on: instrument identifiers, asset classification, client static, benchmarks; a wrong key misjoins silently. Corporate actions — dividends, splits, mergers, coupons — are events that change positions and value, and missing one misstates the report without any obvious error. Client and entitlement data — who the client is, what they may see, their classification, currency and delivery preferences — are the personalisation and control layer.

The core data object, field by field

Take one object — a valuation record for a single holding — and see what each field drives downstream. This field-level view is where abstraction becomes engineering.

Field What it holds What it drives downstream
Instrument identifier (ISIN) The security's global code Every join to price, reference and corporate-action data
Account / client identifier Whose holding this is Entitlement, consolidation, delivery
Quantity Units or nominal held Market value, income accrual
Price and price source The mark and where it came from Market value; audit of the pricing hierarchy
Price and as-at timestamps When the mark was struck and refers to FX consistency, stale-price checks
Cost basis What the client paid Unrealised P&L, tax reporting
Accrued income Income earned, not yet paid The income line; total return
Base and local currency Reporting and instrument currency FX conversion, portfolio total

Deduplication is the non-trivial, mission-critical problem hiding in this table. The same instrument arrives from different feeds under different codes; the same position may appear in two systems; the same corporate action may be delivered twice. If the platform cannot reliably decide that two records are the same security, holding or event, it will either double-count or miss — both of which misstate the client's report. Deduplication depends entirely on a reliable identifier strategy, which is why identifiers get their own treatment next.

Identifiers — the keys that make the join possible

Three identifiers do most of the work of tying reporting data together, and each bites in a different place. Without them, deduplication fails and the six-way join collapses.

Identifier What it names Where it bites
ISIN The security (International Securities Identification Number) Joins positions to prices, reference and corporate-action data
LEI The legal entity (Legal Entity Identifier) Identifies the client or counterparty; regulatory reporting
FIGI The instrument, vendor-neutral (Financial Instrument Global Identifier) Maps the same instrument across feeds that use different codes

The ISIN is the workhorse of the security join — nearly every downstream lookup keys on it — but it is not universal, which is where the FIGI and older national codes such as CUSIP and SEDOL come in, and where a golden-security master becomes essential: a single mapping that knows all the codes for the same instrument and joins the feeds that disagree. The LEI names legal entities rather than securities, and it matters most where reporting meets regulation — an institutional client feeding its own regulatory reporting needs consistent LEIs across the whole chain. A reporting platform that treats identifiers casually will produce reports that look fine and are subtly wrong; a platform that manages a golden master treats identifiers as the load-bearing keys they are.

Integration patterns — batch, API and event streaming

Three ways of moving data between systems coexist in reporting, and the migration between them is more than a speed upgrade. Batch is the legacy default: files delivered on a schedule — the overnight custody file, the end-of-day price file — simple and robust, but latent, because nothing is current until the next batch runs, and it is the reason so much reporting is inherently periodic. API is the modernisation: data fetched on demand, so a portal can show a value now rather than waiting for tonight's batch. Event streaming — Kafka-style push — is the frontier: changes are pushed the instant they happen, so downstream systems stay continuously current.

Immutable audit and the proprietary calculation engine

Two design rules recur across every serious reporting platform: compliance-critical records must be append-only, and the calculation engine is always proprietary in its logic. An audit trail that could be quietly edited would be worthless — the whole point is to prove what was reported, to whom, when, with what content, and any later correction. So the records that underpin restatements, disputes and regulatory evidence are immutable: appended to, never overwritten, so history cannot be rewritten. Governance and reporting platforms exist largely to guarantee this append-only property and to retain the records for the years regulation demands. The calculation engine, by contrast, is where a firm's specific methodology lives — its exact return calculation, its attribution model, its risk definitions — and while the engine may be bought, the configuration and logic are proprietary and governed, because they must be consistent across every client and defensible to an auditor. The engine is not a black box the firm cannot explain; it is a controlled, documented, versioned asset, because a number it produces may have to be justified in a dispute years later.

IBOR, ABOR, and time as a first-class dimension

The investment book and the accounting book are not the same book

A reporting platform usually needs two books of record, and confusing them is a classic error. The investment book of record (IBOR) is the near-real-time view of positions a portfolio manager and a client valuation want: it reflects trades as soon as they are done, includes pending and expected activity, and answers "what do I effectively own right now?". The accounting book of record (ABOR) is the accounting-grade view: it reflects settled transactions under accounting rules, is the basis for the official NAV and the fund's financial statements, and answers "what is confirmed and booked?". The two can legitimately differ on a given day — a trade done but not yet settled sits in the IBOR before it reaches the ABOR — and a good platform knows which book a given report should read from. A client's intraday portal valuation may sensibly use the IBOR; the fund's official NAV must use the ABOR. Reporting from the wrong book produces numbers that are each internally correct yet answer the wrong question.

Reporting has two time axes, not one

The subtle idea that underpins every restatement and every audit is that reporting data is *bitemporal* — it lives in two dimensions of time at once. One axis is valid time: the date a fact refers to — the as-at date of a valuation, the effective date of a corporate action. The other is transaction time, or knowledge time: the date the firm actually recorded or learned the fact. These come apart constantly. A price correction received today for last month's valuation has a valid time of last month but a knowledge time of today. Storing both is what lets a platform answer the two different questions an auditor or a client dispute demands: "what is the correct value for 30 June?" and, quite separately, "what did we report, and believe, on 5 July?". A system that keeps only the latest values can answer the first and not the second, which makes a restatement impossible to explain — you can no longer show what you knew when. Bitemporal storage is therefore not a database nicety; it is the technical foundation of the audit trail, the restatement process and the ability to defend a figure years later.

Testing is where a silent error is born or caught

A reporting change is uniquely dangerous because it can break a number without breaking anything visible, so testing is a first-class discipline. Change a template, a calculation rule, a data mapping or a benchmark, and the report may still produce a perfectly formatted output that is now subtly wrong — no error, no crash, just a different number. The defence is regression testing against a golden copy: run the changed platform over a known period and compare every figure, cell by cell, against the previously validated output, investigating every difference before release. This, plus staged rollout and the ability to roll back, is what stops an improvement from becoming an incident. It is the software-engineering face of the principle that runs through the whole module: in reporting, a change you cannot prove is safe is a change that will eventually misstate a client's number.

The aggregation challenge and the messaging beneath it

The single hardest problem in reporting is assembling one correct picture from fragmented data — and it is fought over the messaging standards that carry that data. Assets sit at multiple custodians, each with its own file format, timing and identifiers. Multiple asset classes — equities, bonds, funds, derivatives, cash, private assets — are each valued differently. Positions live in one system, transactions in another, reference data in a third. Multiple currencies need consistent FX rates and timing. The same instrument carries different codes in different feeds. And feeds arrive at different times, so a report can only be as complete as its slowest still-missing source.

Figure 10 Reporting quality is aggregation quality — solve it once, centrally, into one golden source

The winning architecture solves aggregation once, centrally, into a single reconciled warehouse — the golden source every report reads from. The alternative, patching aggregation inside each report, guarantees that the same fragmentation is fought repeatedly and inconsistently. Solve it centrally and every report inherits a clean, joined, identifier-mapped picture; this is the central design decision of the whole systems landscape.

The messaging standards that carry the data

Reporting rides on defined messaging standards, and they split into a front-office data-feed world and a post-trade custody-message world. Knowing the specific message types is part of speaking the language.

Layer Standard Carries Timescale
Market data Vendor feeds, FIX for prices Prices, quotes, FX, index levels Intraday to end-of-day
Custody statements SWIFT MT535 / MT536 Statement of holdings; statement of transactions End-of-day / periodic
Settlement SWIFT MT540-MT548 Settlement instructions and confirmations Trade-date to settlement
Modern post-trade ISO 20022 (semt / sese) The same custody and settlement content, richer and structured Migrating from MT

The custody feed that begins a reporting cycle typically arrives as SWIFT securities messages — the MT535 statement of holdings and MT536 statement of transactions are the workhorses that tell the bank what the custodian says the client holds and what moved. These, and the settlement messages around them, are migrating to ISO 20022, the richer, structured successor, in the same way the rest of the industry's post-trade plumbing is. Market data arrives through vendor feeds and, for prices, sometimes FIX.

Operational resilience — the one day that cannot be repeated

Some moments in reporting have no second chance, and resilience is the discipline of surviving them. A daily NAV strike has a hard cut-off after which investors trade on it; a MiFID II ten-percent depreciation alert must reach the client by the end of the same business day the threshold is breached. These are irreversible: you cannot re-run yesterday's NAV, and a same-day alert sent the next day is a breach, not a late delivery. Resilience is what guarantees the platform is standing at exactly those moments, and it is built from precise, distinct concepts.

Concept What it guarantees Why it matters on the critical day
Redundancy No single point of failure A failed server or feed does not stop the NAV strike
Automatic failover Instant switch to a standby The cut-off is met even as a component dies
Disaster recovery A tested recovery to a second site A site outage does not lose the day's reporting
Business continuity People and process keep running The operation continues through disruption

The regulatory stack read as a system specification

A stack of rules dictates what reporting must contain, in what form and by when — and each rule is best read not as a legal summary but as a specification the platform must implement. This is a predominantly EU-shaped stack, supervised by ESMA and by national regulators such as the AMF and ACPR in France, and post-Brexit the UK is diverging on several points, so pan-European firms must run both regimes at once.

Figure 11 The regulatory stack shaping client reporting — predominantly EU, supervised by ESMA, AMF and ACPR

MiFID II costs and charges specifies ex-ante and ex-post cost disclosure in cash terms, the ten-percent depreciation alert, periodic statements and suitability reports — the platform must aggregate all costs per client and monitor daily thresholds. PRIIPs specifies the standardised retail KID — a summary risk indicator on a one-to-seven scale, performance scenarios, costs over time and a recommended holding period, at most three pages in a fixed layout — which the platform must mass-produce, product by product, from a reproducible calculation pipeline. SFDR and the EU Taxonomy specify sustainability disclosures — product classification under Articles 6, 8 and 9, principal adverse impact indicators, and taxonomy alignment — which the platform must source, join to holdings at the security level, and aggregate. GIPS and the Benchmarks Regulation specify how performance is calculated and presented, and how benchmarks are governed and used. CRS and FATCA specify cross-border reporting of client tax data to authorities. And GDPR and record-keeping rules specify how client personal data is protected and how long reports and communications must be retained and be evidenced.

The data feed as a product in its own right

For institutional clients, the machine-readable data feed is not a byproduct of reporting but a product with its own design, contract and support obligations. A pension fund or insurer that ingests the bank's valuation into its own systems does not want a PDF; it wants a structured file or an API that lands reliably, in an agreed format, to an agreed schedule, with agreed identifiers. That turns the feed into something closer to a software product than a document: it has a schema that must be versioned so a change does not silently break the client's ingestion, a delivery service-level agreement, a defined behaviour when data is late or incomplete, and a support relationship for when the client's systems reject a file. The same governed numbers that compose the human-readable statement flow into the feed, but the engineering discipline is different — a feed consumed by a machine has no reader to notice that a column moved or a date format changed, so backward compatibility and schema governance matter far more than they do for a document a person reads.

Data quality as a managed discipline

Because reporting inherits every upstream error, data quality is not a hope but a managed discipline with its own metrics, and it is worth naming the dimensions along which data can be wrong. Data can fail on completeness (is every expected position, price and transaction present?), accuracy (is each value correct?), timeliness (did it arrive before the deadline?), consistency (do the same facts agree across systems?), validity (does each value fall within its allowed range and format?) and uniqueness (is anything duplicated?). A mature reporting platform measures each of these continuously, because a report can only be as good as the weakest dimension on the day it is produced.

Reconciliation is not one control but several

"Reconciliation" is shorthand for a family of distinct checks, and knowing which is which is part of understanding where data can be wrong. Position reconciliation agrees what is held; cash reconciliation agrees the money; transaction reconciliation agrees what moved; and for fund clients, NAV reconciliation agrees the struck value against an independent calculation. Each catches a different class of error, and a platform that runs only one is blind to the others.

Reconciliation type Agrees Catches
Position Holdings vs custodian Phantom, missing or mis-sized positions
Cash Cash balances vs custodian Missed income, fees, unbooked movements
Transaction Activity vs custodian / counterparty Unbooked or duplicated trades and corporate actions
NAV (funds) Struck NAV vs independent calc Pricing, accrual and calculation errors before publication

The reconciliation break lifecycle

Figure 12 The reconciliation break lifecycle — the recurring breaks are the early-warning system

A reconciliation break is not a nuisance to be cleared and forgotten; it has a lifecycle that, managed well, is the early-warning system of the whole operation. A break is detected when the automated reconciliation flags a difference between the bank's record and the custodian's. It is classified — a position break, a cash break, a price difference, a timing difference — because the classification points to the cause. It is investigated and assigned to whoever can resolve it. It is resolved, with the fix applied at the source so it does not recur. And it is analysed: a recurring break of the same type is a signal of a systemic problem — a feed that is always late, a corporate action a system handles wrongly — that should be fixed once rather than cleared every cycle. The metric that matters is not just how many breaks occur but how quickly they clear and whether the same ones keep returning; a rising or slow-clearing break rate is the surest leading indicator that a reporting operation is losing control of its data.

Cloud, and why reporting is migrating to it

Reporting workloads are moving to the cloud, and the reasons are specific to reporting's shape rather than generic IT fashion. Reporting demand is spiky — month-end, quarter-end and year-end concentrate enormous compute into a few days while the rest of the month is quiet — and cloud's elastic capacity fits that shape far better than a fixed on-premise estate sized for the peak and idle the rest of the time. Cloud also makes change faster, which matters in a discipline where the regulatory backlog never clears, and it makes API-first, machine-readable delivery natural rather than bolted on. The migration is not free of tension: client data in the cloud raises data-residency and GDPR questions that European firms must answer carefully, often keeping data within EU regions and under strict access control. But the direction is set, because the economics of elastic compute against a spiky, deadline-driven workload are compelling, and because the modern data-and-calculation layer — the single reconciled source that everything else depends on — is far easier to build and scale on cloud data platforms than on the fragmented legacy estates it is meant to replace.

  1. What is the three-layer architecture, and what does "compute each number once" buy you?

  2. Why are the data warehouse and the client experience often built while the engines are bought?

  3. Explain the cascade principle and why it argues for fixing data upstream.

  4. What do ISIN, LEI and FIGI identify, and why does a golden-security master matter?

  5. Why does moving from batch to event streaming remove reconciliation rather than just add speed?

  6. Why is redundancy alone insufficient for resilience, and what must be added?

  7. What does it mean to read the MiFID II 10% alert as a system specification rather than a rule?

4

  1. What does “compute each number once” buy you?

  2. What do firms usually build?

  3. Why fix data upstream rather than patch a report?

5Where Reporting Fails, and Where It Is Heading

Now the honest part. Reporting fails in a small number of characteristic ways, three of which have played out publicly and expensively. This part maps the risks, tells three real failure stories that map to three distinct failure modes, weighs the transformation trends — including generative AI — against a single test, and closes with the six ideas from the whole module worth keeping.

The six risks, mapped by who controls them

Reporting risk is not a long list; it is six types, and arranging them by controllability turns a catalogue into a map. Some of these a reporting operation controls almost entirely; others it can only defend against.

Risk type What it is Who owns it Main defence
Data and valuation error A wrong position, price or corporate action reaches the client Reporting operations Reconciliation, tolerance checks, pricing governance
Timeliness failure A NAV, pack or alert is late Production support Automation, monitoring, resilience
Disclosure and compliance breach A mandated report is missing, wrong or late Compliance and operations Regulatory rules engine, retention, evidence
Conduct and misstatement Figures are misrepresented or misleading Front office and senior management Four-eyes, independent review, governance
Confidentiality and data protection A client sees another's data, or data leaks Technology and data Entitlements, access control, encryption, audit
Operational and systems resilience A platform or vendor failure halts reporting Technology and production Redundancy, tested failover, disaster recovery

Telling a good reporting operation from a lucky one

A clean month proves nothing on its own; you judge a reporting operation by leading indicators, not by the absence of complaints. The trap is the same one that flatters a trader after a lucky year: no news is taken for good process, when it may just be good fortune. Six metrics tell the real story.

Metric What it measures Why it matters
Accuracy / error rate Reported figures later found wrong The core quality signal
On-time delivery rate Packs and NAVs met to deadline The factory's reliability
Reconciliation break rate Breaks per cycle, and time to clear The health of the data foundation
Restatement rate Reports re-issued after publication The cost of errors that escaped
Query and complaint volume Client questions and disputes The client's felt experience
Straight-through / automation rate Share of the cycle without manual touch Scalability and resilience

Three real cases — the three ways reporting fails

Three real failures, each a different failure mode, show what these risks look like when they escape the controls. They are not random; together they cover the three ways reporting goes publicly wrong — a bad valuation reaching the client, a report that misstates performance and risk, and a systems failure that stops reporting altogether.

Case 1 — H2O Asset Management: when the valuation itself is the problem

H2O Asset Management shows the first failure mode — an illiquid, hard-to-value holding that compromises the numbers a client relies on. From around 2019, H2O, then a Natixis affiliate, held large positions in illiquid private bonds linked to the German financier Lars Windhorst across several of its funds. Because those holdings were illiquid, their valuation was uncertain and investors could not reliably exit at the stated values. In August 2020 France's markets regulator, the Autorité des Marchés Financiers, required H2O to suspend subscriptions and redemptions in three funds, citing significant exposure to, and valuation uncertainty in, those investments; the affected holdings were later moved into "side pockets" so investors could not access them. In January 2023 the AMF's enforcement committee imposed a record fine of €75 million, criticising H2O's "deliberate choice" to select illiquid investments in breach of the rules and compromising investors' ability to redeem. The UK's Financial Conduct Authority pursued the matter as well. The lesson is Part 1's pricing point at industrial scale: when the mark is a judgement on an illiquid asset, valuation is not a back-office step but the very thing that can fail — and when it fails, the client's reported value is a fiction they cannot act on.

Case 2 — Allianz Global Investors Structured Alpha: when the report lies

The Allianz Structured Alpha case shows the second, gravest failure mode — the report itself is falsified to conceal risk and performance. Between roughly 2016 and March 2020, portfolio managers on the Structured Alpha team at Allianz Global Investors U.S. — funds holding about $11 billion for institutional investors including pension funds — manipulated the risk and performance reports sent to clients to hide how much downside risk the strategy actually carried. The detail that makes the case unforgettable is how crude the manipulation was: in one risk report a projected loss under a market-crash scenario was reduced from negative 42.15% to negative 4.15% simply by dropping a digit, and performance figures were "smoothed" by altering reported losses. When markets crashed in early 2020, the concealed risk became real and the funds collapsed. In May 2022 Allianz Global Investors U.S. pleaded guilty to criminal securities fraud, the SEC found violations of the securities laws, and the total cost — restitution, disgorgement and penalties — exceeded $6 billion, one of the largest such resolutions on record. This is the failure mode reporting exists to prevent: not a data error that slips through, but a deliberate corruption of the numbers a client trusts. It is why independent review, four-eyes sign-off and the separation of the calculation layer from anyone with an incentive to flatter it are not bureaucratic overhead but the core control.

Case 3 — BNY Mellon / SunGard: when the machine stops

The 2015 BNY Mellon NAV outage shows the third failure mode — an operational and systems failure that halts reporting entirely, with no fallback. In August 2015, BNY Mellon relied on a third-party accounting system, SunGard's InvestOne, to calculate net asset values for a large book of fund clients. During a scheduled upgrade around 22 August, the platform's data became corrupted — and critically, SunGard's backup environment was corrupted too, so failover could not happen. As a result, on 24 August BNY Mellon could not strike accurate same-day NAVs for roughly 1,200 funds across some sixty-six clients, forcing the use of estimated and late values while the system was recovered over several days. BNY Mellon later agreed a $3 million settlement with Massachusetts for failing to calculate the NAVs, with regulators noting the absence of an adequate backup plan for a vendor failure. The lesson maps exactly onto Part 4's resilience point: redundancy is not resilience unless the failover is tested and actually works. A backup that is itself corrupted is no backup at all, and the one irreversible moment — the daily NAV strike — is precisely when that gap is exposed.

Case Failure mode Maps to stage The lesson
H2O Asset Management (2020-23, AMF €75m) Valuation of illiquid assets misstates the client's value Valuation When the mark is a judgement, valuation is where reporting fails
Allianz Structured Alpha (2016-22, >$6bn) The report itself falsifies risk and performance Validation and review Independent review and a governed calc layer exist to prevent exactly this
BNY Mellon / SunGard (2015, $3m + disruption) Systems failure halts reporting, no working failover Delivery / resilience Redundancy without tested failover is not resilience

Generative AI in reporting — augment, don't author

Generative AI has genuine, specific uses in reporting, and an equally specific set of risks — and the line between them is the line between drafting words and producing figures. Used well, it augments; used carelessly, it corrupts the one thing reporting cannot afford to get wrong.

Where large language models actually help. First, natural-language commentary: an LLM can draft the "what happened this quarter" narrative from the numbers, for a human to review — turning a slow, manual writing task into a fast, reviewed one. Second, personalisation at scale: tailoring content, highlights and explanations to each client's holdings and interests, which is otherwise impossible to do by hand across thousands of clients. Third, data-quality and anomaly detection: models flag stale prices, outliers and breaks before they reach the client, strengthening the validation gate. Fourth, conversational queries: clients or relationship managers ask questions of a portfolio in plain language rather than waiting for a cut or a cycle.

Use of GenAI The gain The governing risk
Drafting client commentary Fast, consistent narrative at scale Hallucination; must be reviewed and owned before it goes out
Personalising content Relevant, tailored reports across thousands of clients Entitlement and MNPI leakage across the confidentiality wall
Anomaly and data-quality detection Catches errors before the client sees them False confidence; the model is only as good as the data
Conversational portfolio queries Self-service, plain-language access Wrong or unexplained answers on regulated numbers

The caveats are non-negotiable, and they all reduce to one rule: the figures never come from the model. Reporting has no tolerance for hallucination, so AI may draft the words but the numbers come from the governed calculation layer, full stop. Accountability stays human: a regulated client report needs sign-off, and AI-drafted commentary must be reviewed and owned before it goes out — the Allianz case is a reminder of what is at stake when the numbers a client sees are not independently governed. AI on messy, unreconciled data amplifies errors rather than fixing them, so the data foundation is the prerequisite, not an optional extra. And explainability matters: clients and regulators need to know how a figure or statement was produced, and a black box is a hard sell in a discipline built on evidence and audit.

The target operating model — where all this is heading

Pull the transformation threads together and a single target operating model emerges, the state almost every serious reporting programme is trying to reach. It has four features. There is one reconciled golden source of positions, prices, transactions and reference data, so aggregation is solved once rather than per report. There is one calculation layer that computes each valuation, return and risk metric once, consistently, for every client and channel. There are many thin presentation channels — statement, portal, feed — that format the same computed numbers without recomputing them. And there is pervasive automation with exception-based control, so the clean majority of the cycle runs untouched and human effort concentrates on the genuine breaks and the four-eyes sign-off. Everything in this module points at this model, because it is the only shape that makes reporting simultaneously accurate, consistent, auditable, resilient and cheap to extend as regulation and client expectation keep changing.

5

  1. Which case shows the report itself falsifying risk and performance?

  2. What does BNY Mellon / SunGard 2015 teach?

  3. How should generative AI be used in reporting?

  4. Which trend is the foundation for the others?

Module summary — the six ideas that matter most

Idea 1 — Reporting is the visible surface, so trust is the product

Statements, valuations and performance are where the client actually sees the bank's work. Everything else the bank does is invisible to the client; the report is the interface to all of it. Accuracy and timeliness are therefore not qualities of the product but the product itself, and reporting is fundamentally a control function.

Idea 2 — Data cascades, so errors are never contained

Because reporting sits at the end of the chain, it inherits every upstream error, and an error at the source reaches every downstream report at once. The core position-and-price record is the atom; every report is built from it, so a bad atom corrupts every molecule. This is why the data model, not the document, is the real product, and why problems are fixed upstream rather than patched per report.

Idea 3 — Pricing is the crux, so the mark is where reporting is really made

Every number rests on a price, and stale, bad or mistimed marks are the main source of error. The pricing hierarchy — exchange close, evaluated price, model valuation — governs where the mark comes from, and pricing governance is the central control, because the largest and most invisible errors live in the mark, as the H2O case showed.

Idea 4 — Aggregation into one golden source is the decisive challenge

Multi-custodian, multi-asset, multi-currency data joined into one reconciled source is the hard, decisive problem of reporting. Solve aggregation once, centrally, and every report inherits a clean picture; patch it per report and you fight the same fragmentation forever. Reporting quality is aggregation quality.

Idea 5 — Three layers win: compute once, present many

Separate the data, calculation and presentation layers, compute each number once, and let many channels present it. This is what makes reporting consistent, auditable and cheap to extend, and it is the target state of almost every reporting transformation. A portal is a skin over this foundation, not a substitute for it.

Idea 6 — Regulation is core architecture, resilience is measurable, and AI augments but never authors

Compliance is designed in, not bolted on: MiFID II, PRIIPs, SFDR, GIPS/BMR, CRS/FATCA and GDPR are the platform's requirements, resilience is where "mission-critical" becomes a measured guarantee against the one irreversible moment, and generative AI is worth deploying only where it changes economics or risk — governed so that it drafts words and checks data but never produces the figures.

  1. Why is arranging the six reporting risks by controllability more useful than listing them?

  2. What is the "no complaints" fallacy in judging a reporting operation?

  3. Name the three real failure cases and the distinct failure mode each represents.

  4. What single test should every transformation trend pass?

  5. Where does generative AI genuinely help in reporting, and what is the absolute boundary?

  6. How does the BNY Mellon case illustrate the Part 4 resilience lesson?

Process Cheat Sheet — The Whole Reporting Cycle on One Page

A single reference for the entire client-reporting flow, from data cut-off to the client's inbox.

The production cycle, stage by stage

Stage Timing Core activity Key output Lead and systems
Data cut-off BD 0 Freeze inputs; load positions, prices, transactions, FX, corporate actions Complete, consistent inputs Operations · orchestration, feeds
Reconcile BD 1–2 Agree positions and cash to custodians; clear breaks Reconciled book of record Operations · reconciliation engine
Value and mark BD 2–3 Mark to the pricing hierarchy; accrue income; one as-at, one FX Valued portfolio Operations + pricing · valuation engine
Performance BD 3 TWR/MWR, benchmark, attribution, risk Performance figures Performance and risk · attribution engine
Validate and review BD 4 Tolerance checks; four-eyes sign-off; draft commentary Approved, checked data Operations + reviewer · validation rules
Compose BD 4–5 Merge data into branded, personalised templates Assembled pack Technology · composition tool
Deliver BD 5 Publish PDF, refresh portal, push feed The client-facing report Technology + front office · portal, PDF, API

Key numbers and rules of thumb

Metric Rule of thumb
Effort split ~80% data readiness, reconciliation and validation; ~20% composition and delivery
Monthly institutional pack Delivered around BD+5 after month-end (illustrative)
Fund NAV Struck accurately every business day, to a hard, non-negotiable deadline
MiFID II 10% depreciation alert By the end of the same business day the 10% fall is breached, and each further 10%
Ex-post costs and charges Annual statement of what the client actually paid, in cash terms
PRIIPs KID Maximum three pages, fixed layout, risk indicator on a 1–7 scale
Attribution check Allocation + selection + other must sum exactly to the relative return
Two returns TWR judges the manager (strips flows); MWR describes the client (includes flows)
The six data domains Positions, prices, transactions, reference/static, corporate actions, client/entitlement
The three layers Data (warehouse) · Calculation (compute once) · Presentation (many channels)
Golden identifiers ISIN (security) · LEI (entity) · FIGI (vendor-neutral instrument)
Custody messages SWIFT MT535 (holdings), MT536 (transactions), migrating to ISO 20022

The whole cycle in one sentence

Freeze the data at cut-off, reconcile it to the custodians, mark it to one consistent price and time, measure and attribute the performance, validate and sign it off, compose it into the client's personalised pack, and deliver the same governed numbers across every channel — right and on time, because in reporting that is the product.

Practice — Diagnose the Broken Report

The real work of reporting is not producing a report but catching the one that is wrong. These exercises put you in the analyst's seat: each describes a symptom, and the task is to name the likely cause and the control that should have caught it. Work them before reading the answers.

Spot the error

x7.1

A client's monthly portfolio total is 3% lower than expected, yet every individual holding's price looks right in its own currency. Likely cause, and the control that should have caught it? — An FX-and-timing mismatch: positions priced at different moments, or converted at inconsistent rates. The control is a check that every price and FX rate carries the same as-at timestamp, plus the one-as-at-moment, one-FX-snapshot valuation policy. 2. A corporate-bond position shows the same price for three weeks while comparable bonds have moved with rates. What has probably happened? — A stale price: the evaluated-price feed has dropped the illiquid name. Stale-price detection — comparing each mark to its last good value and to peers — should have flagged it before publication. 3. A performance report shows the portfolio beating its benchmark by +0.5%, but the attribution table's allocation and selection effects sum to +0.7%. Which figure do you trust? — Neither yet — the mismatch is itself the signal. Attribution must reconcile to the measured relative return; a residual means a data or model error (a misaligned benchmark constituent, a mishandled currency effect), so both must be investigated before either is sent. 4. A client says their statement double-counts a holding they own through two different funds. What underlying problem does this point to? — A deduplication or identifier failure — the same underlying security is not recognised as one across feeds, or look-through is aggregating without netting. The fix is a golden-security master and netting at the look-through level. 5. A daily NAV was published on time but is later found wrong because a dividend was missed. Where did the process fail, and what happens now? — Data readiness failed: a corporate action was not applied before cut-off, so reconciliation or the corporate-actions control missed it. Because the NAV was published and traded on, the fix is a governed restatement — versioned, explained, re-issued — not a quiet edit, and investors who traded on the wrong NAV may need compensation. 6. An intraday portal valuation and the official end-of-day NAV disagree for the same fund on the same day. Is this necessarily an error? — Not necessarily: the portal may correctly read the near-real-time IBOR (including a done-but-unsettled trade) while the official NAV reads the settled, accounting-grade ABOR. Each answers a different question; the error would be reporting the wrong book for the purpose.

A capstone — the month everything is late

What the practice teaches

The pattern behind every exercise is the same: the visible symptom is downstream, and the cause is upstream in the data. A wrong total is an FX-and-timing cause; a frozen price is a stale-feed cause; a non-reconciling attribution is a data or model cause; a double-count is an identifier cause; a wrong NAV is a data-readiness cause; two disagreeing valuations may be an IBOR-versus-ABOR distinction, not an error at all. Learn to read a reporting symptom backwards to its upstream data cause and you have learned the core skill of the discipline — because in reporting, the report is never really the problem. The data is.