Back-solve reconciliation tolerances from your NAV threshold
Most asset managers let the fund administrator decide which position and price reach the NAV, then find out about breaks afterwards. This playbook shows how to build a single manager-owned position and pricing record, with tolerances derived from the fund's regulatory NAV error threshold, before the custodian, administrator and vendor feeds turn into NAV, performance and client reports.
Claude VectorData & Analytics LeadOctober 20, 2026Listen to the podcast
10 min
Chapters
Key takeaways
- Set reconciliation tolerances from the fund's regulatory NAV error threshold instead of copying vendor default settings.
- Give any single break worth more than a tenth of the NAV threshold a named owner the same day.
- Do not release the NAV while open breaks add up to more than a quarter of the threshold, and log, own and approve any override.
- Check quantities against the custodian and prices against an independent challenger vendor, because IBOR matching the administrator can be two copies of the same mistake.
- Report breaks to the board as money against threshold, age of the oldest break and its owner, not as a raw count.
Read the full transcript
Host:MBA Training. Today: Back-solve reconciliation tolerances from your NAV threshold. Who actually pays when the fund's price is wrong and nobody at the manager ever saw the error?
Expert:The manager. Almost always the manager.
Host:Even when the administrator calculated it? The outsourced firm that literally strikes the NAV, the net asset value, the daily price per share?
Expert:Even then. In Luxembourg the CSSF, the financial regulator, put it in writing with Circular 24/856: the investment fund manager corrects and compensates. Your contract with the administrator might get you money back eventually. The regulator doesn't care about your contract.
Host:So you can outsource the arithmetic but not the blame.
Expert:Beautifully put. Very bleak.
Host:Okay, walk me through a normal morning. What does a mid-sized manager actually know at eight a.m.?
Expert:It knows three different things and believes all of them. The custodian, the bank physically holding the securities, reports settled holdings. The administrator has its own book, built from the custodian's files plus its own vendor prices. And the IBOR, the investment book of record, shows what the front office thinks it traded, priced by whichever vendor the portfolio managers happen to like.
Host:Three versions of the truth before coffee.
Expert:And the one that wins is whichever the administrator used. Performance reports inherit it. Client reports inherit it. The manager finds out there was a gap when an auditor asks.
Host:Then why not crown one of them the golden source and move on?
Expert:Because that fight goes nowhere. Each one is right about different fields. The custodian is right about how many bonds you settled. The IBOR is right about what you meant to trade. Neither of them should be deciding the price.
Host:Who decides the price, then?
Expert:A pricing policy. A named primary vendor, a named challenger, written rules for overrides. The administrator owns nothing by default. It calculates from what you hand it.
Host:Hang on. That's a pretty big demotion for the people you're paying to run the fund's books.
Expert:It's the correct demotion. Think of a restaurant where the waiter decides what you ordered, the kitchen cooks that, and you discover what you ate from the credit card statement.
Host:And then the health inspector fines you, not the waiter.
Expert:Exactly. That's the current operating model at a lot of firms.
Host:Speaking of nobody owning the mess, IBM published a study last week, picked up in the IBM Newsroom. Four in five executives across Europe, the Middle East and Africa said an AI incident reached their board in the past year. Far fewer actually changed their governance afterwards.
Expert:Same disease, different organ. The incident reaches the boardroom, everyone nods gravely, the policy stays exactly where it was.
Host:Boards love a briefing. They're less keen on a rewrite.
Expert:Which is why I'd rather hand them a number in euros than a slide of red dots. Back to breaks, though, because this is where it gets interesting.
Host:Go on. Tolerances. The thresholds where a mismatch gets flagged.
Expert:Most firms copy them off the vendor's default screen. A price break of one percent, a cash break of a hundred units, whatever shipped with the software. None of it connects to the only thing the regulator measures, which is the error in the fund's overall NAV.
Host:And the regulator does measure that?
Expert:Precisely. A fixed income UCITS fund, the standard EU retail fund wrapper, has a materiality threshold of half a percent of NAV. Cross that, and you're into notification and compensation.
Host:So you work backwards from that line.
Expert:Back-solve it, yes. Take that threshold, decide that any single break worth more than a tenth of it gets a named owner the same day, and that the NAV doesn't go out while open breaks add up to more than a quarter of it.
Host:Make it concrete. I'm getting lost in fractions of fractions.
Expert:Hypothetical fixed income fund. You hold a big position in a bond trading near par, and it's mispriced by one point. That's roughly two hundred thousand euros of NAV impact. On its own, nothing fires. It sits under the single-break alarm.
Host:Fine. One sloppy bond.
Expert:Now four of those, open at once. The gate closes. Leave a dozen to age for a week and you're sitting almost exactly on the regulatory threshold. One more and it's a notifiable error.
Host:Wait, what? A dozen small things nobody escalated add up to a reportable incident?
Expert:That's the aggregation clause, and it's the bit people skip. The Circular counts successive untreated errors that were each below the line but cross it together. A pile of little breaks is a regulatory event that just hasn't been introduced to itself yet.
Host:That's genuinely unnerving. It's like being fined for speeding because you went slightly too fast on twelve different streets.
Expert:Ha, yes. And your rental company set the speedometer.
Host:The vendor defaults being the rental company.
Expert:And sometimes the speedometer is biased. Bloomberg, which sells both of the bond price products it compared, so treat it as vendor analysis, found its two price sets drifting a few cents apart across a corporate bond universe before it changed its methodology. Run that kind of quiet bias across a whole book and you've eaten a chunk of your threshold before a single real mistake happened.
Host:Okay, here's where I push back. You want to hold the NAV until breaks clear. Clients want their price. Distributors want their price. Holding the NAV is its own disaster.
Expert:Holding it is a governance decision, not an ops tantrum. The gate runs before the administrator's cut-off, not after.
Host:And T+1 shrinks that window. Settlement the next business day instead of two.
Expert:Europe moves in October next year. The overnight batch that already finishes late is suddenly on the critical path.
Host:Which is why I think a quarter-of-threshold gate is too twitchy. You'll have ops teams in tears every quarter-end, and sooner or later someone overrides it to get the price out.
Expert:Then the override is logged, owned and approved. Which is still miles better than the current model, where nobody even knows a gate existed.
Host:Unless the override becomes the habit. Then it's theatre with a sign-off.
Expert:Fair. I'd still take theatre with a paper trail over improv with no audience.
Host:We're not settling this today.
Expert:We are not.
Host:Where are people getting it right? Give me a name.
Expert:Invesco on the US move to T+1. It started well over a year ahead, and the biggest win wasn't some shiny platform. It was consolidating identifiers.
Host:Identifiers meaning the codes for a security.
Expert:Right. The same bond carries an ISIN, the international code, at the custodian, a SEDOL, the London Stock Exchange code, at the vendor, and some internal number in the IBOR. Most so-called position breaks are just three systems failing to recognise the same thing.
Host:So the dramatic reconciliation crisis is mostly a spelling problem.
Expert:Mostly a spelling problem with a regulator attached. The EU T+1 Industry Committee basically says it: legacy ID mess becomes a critical-path item.
Host:What's going wrong out there, then?
Expert:Circular matching is my favourite. You reconcile your IBOR to the administrator, the administrator built its book from the same custodian file and the same vendor price, everything matches, and everyone feels wonderful.
Host:Two copies of the same mistake agreeing with each other.
Expert:Agreement isn't correctness. Quantity has to be checked against the custodian, price against an independent challenger. And on price challenge, the SEC, the US securities regulator, settled with Madison Capital Funding in February. Loans sold to its own private funds without reasonably checking fair value. Roughly five million dollars repaid plus a penalty.
Host:That lands badly with me, honestly. Investors in those funds trusted someone to check, and the checking was a shrug.
Expert:It's the discretion with no evidence that kills you. Somebody had a view. Nobody wrote down why.
Host:Even the vendors are selling oversight now. SS&C has a pricing-provider benchmark study coming.
Expert:And SS&C is itself an administrator and tech vendor, so useful input, not gospel. Test it against your own holdings.
Host:One more news aside, since we're on ownership. Illinois just rehired Tammy Roust, its former chief data officer, as chief AI officer, per StateScoop. Her predecessor lasted about a month.
Expert:A month. That's shorter than an aged break.
Host:He'd have tripped the aggregation clause.
Expert:But notice who they reached for. The data person. Because whoever owns the record ends up owning the consequences.
Host:So what does the board actually see in your world?
Expert:Not "three hundred and something open breaks." That means nothing. They see: open breaks on this fund are at a fifth of its threshold, the oldest is nine days, and it belongs to the custodian.
Host:Money, age, owner.
Expert:Something a conducting officer, the named exec on the hook in Luxembourg, will actually sign.
Host:Still think your gate's too tight.
Expert:Still think your loose one is how you end up paying compensation for errors you never saw.
Host:So, back to the restaurant. The waiter's demoted, the kitchen has a recipe, someone tastes before it leaves.
Expert:And the bill is in euros, not dots.
Host:Which leaves the uncomfortable part. If the manager pays for every error anyway, how many of today's NAVs did it actually choose, and how many did it simply receive? Sources: Klavius, Cliffordchance and Cssf. Every link is in the show notes. That's a wrap. Fresh CDO briefings drop every week at mba-training.com.
On any given morning, a mid-sized asset manager holds at least three versions of the truth. The custodian reports settled holdings. The fund administrator holds its own book, built from custodian files, trade feeds and its own vendor prices. The investment book of record (IBOR) shows trade-date positions priced by whichever vendor the front office prefers. When these disagree, the version that reaches the NAV is usually the one the administrator used. Performance and client reports then inherit it, and the manager learns about the gap when a client, an auditor or a depositary asks.
That arrangement is getting harder to defend, for regulatory and operational reasons at once. In Luxembourg, under CSSF Circular 24/856, the IFM corrects and compensates, even when the NAV is calculated by an administrator. In Europe, the move to T+1 settlement on 11 October 2027 cuts the time available to reconcile. My argument: the reconciled position and price record should be a product the manager owns, gated before NAV is struck, with tolerances back-solved from the fund's own NAV error threshold. Default tolerances inherited from the vendor and break reports read after the fact are how a manager ends up paying compensation for errors it never saw.
How to build a single trusted position and pricing record
Step 1: Assign authority field by field
Arguing about whether the custodian or the administrator is the "golden source" goes nowhere, because each one is right about different fields. Decide authority per attribute and write it down:
- Settled quantity and cash: the custodian, because it holds the assets and the depositary relies on its records.
- Traded quantity and pending settlements: the IBOR, reconciled to the custodian once settlement happens.
- Price: your pricing policy. That means a named primary vendor per asset class, a named challenger, and a documented hierarchy for overrides.
- FX rates: a single snapshot time, fixed in policy and applied consistently across NAV, performance and reporting.
- Corporate action terms: one source of announcements, with custodian confirmation of entitlements.
The administrator owns none of these fields by default. Its job is to calculate from the record you hand it, and your reconciliation checks that it did. Here is the rule for the whiteboard:the custodian owns quantity, the IBOR owns intent, the price policy owns price, and NAV waits until open breaks are under a quarter of the threshold.
Step 2: Back-solve tolerances from the NAV error threshold
Most reconciliation tolerances are absolute numbers copied from vendor defaults: a quantity break of zero, a price break of 1%, a cash break of 100 units of currency. None of them is linked to the thing that triggers regulatory consequences, which is the fund-level NAV error.
The CSSF thresholds give you the anchor. The Circular provides for materiality thresholds in relation to Money Market Funds (0.2%), UCITS (1% for equity funds, 0.5% for fixed income and mixed funds). The clause that matters most for reconciliation design is the aggregation rule: significant NAV calculation errors are to be understood not only as isolated calculation errors which have a significant impact on the NAV but also as simultaneous or successive calculation errors which have not been treated and were each below the tolerance threshold but reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.View full definition → it when added together. A pile of small unresolved breaks is a regulatory event in waiting.
So set two numbers per fund:
- Single-break escalation: any break whose potential NAV impact exceeds 10% of the fund's threshold goes to a named owner the same day.
- Aggregate gate: NAV is not released while the summed absolute NAV impact of open position and price breaks exceeds 25% of the threshold.
The 10% and 25% levels are my design choices. Set your own, but derive them from the threshold rather than from a vendor screen.
Worked example (hypothetical, arithmetic only). Take a €500m fixed income UCITS. Its threshold is 0.5%, or €2.5m. Single-break escalation sits at €250k and the aggregate gate at €625k. A €20m bond position mispriced by one point near par is a €200k break. That is 8% of threshold, so no single-break alarm fires. Four of them open at the same time come to €800k and close the gate. Twelve of them, left to age for a week, come to €2.4m, close enough to the threshold that one more break makes it a notifiable error. Systematic price bias is the less obvious risk. Bloomberg, which sells both of the price products concerned (so read its figures as vendor analysis), found that before a methodology change its trade reference and evaluated prices for a corporate bond universe showed 4.5 to 5 cents absolute difference between the products, falling to (or just below) 2 cents by the end of July. Apply a 5-cent bias to a €500m book priced near par and you get €250k, a tenth of the fund's threshold, before a single genuine error has occurred.
Step 3: Run the gate before the administrator strikes NAV
Put the reconciliation upstream of the NAV calculation. Send the administrator your reconciled position and price file, then reconcile its output back to that file. Don't check its NAV against your records the next morning. This changes the timetable: custodian files, vendor prices and IBOR positions all have to land, match and clear the gate before the administrator's cut-off.
T+1 makes this harder. On 20 July 2026, the European Securities and Markets Authority said that 2026 is a critical year for firms to finalise preparations and highlighted 7 December 2026 as the first regulatory milestone for allocation and confirmation processes. The EU T+1 Industry Committee's handbook tells asset managers to align legal fund document updates, pricing, NAV calculation, and reconciliation processes with the new settlement cycle. If your morning gate depends on an overnight batch that already finishes late, that batch is now on the critical path.
Step 4: Challenge every price with a second source and keep the evidence
A price that passes a staleness check has still not been validated. Run every evaluated price against a challenger. Route divergences above your back-solved tolerance to a pricing analyst, and log the decision with its evidence. Thebenchmarking of golden-source pricing against challengers should be periodic and data-drivendata-drivenAn approach where decisions are systematically informed by data analysis rather than intuition alone.View full definition →, not a one-off vendor selection exercise.
Administrator-run benchmarks help, with a caveat. SS&C, which is itself a fund administrator and technology vendor, is preparing a 2026 study that evaluates several leading pricing providers, including Bloomberg BVAL, ICE Data Services, JP Morgan PricingDirect, LSEG, MarketAxess, S&P Global and Trumid. SS&C says the study can support compliance with SEC Rule 2a-5 by helping firms evidence effective oversight of pricing services used in regulated funds. Use it as one input and test it against your own holdings and trade prints.
The cost of weak price challenge is now on public record. On 25 February 2026, the SEC announced settled charges against Madison Capital Funding LLC, an Illinois-based formerly registered investment adviser, for selling loans to private fund clients ("the Funds") without reasonably determining whether those trades were at fair market value. The firm repaid approximately $5 million plus interest to affected funds, agreed to a cease-and-desist order, was censured, and must pay a $900,000 civil penalty. The conduct dates back to 2020, but the lesson holds today: a valuation policy that leaves adjustments to discretion, with no record of the challenge, does not survive an examination.
Step 5: Pay down identifier debt before anything else
Most position breaks are not real economic differences. They come from the same security carrying an ISIN at the custodian, a SEDOL at the vendor and an internal ID in the IBOR, or from one account mapped differently across three systems. Invesco's T+1 work shows the payoff. For the US migration, Invesco began preparations 18-24 months ahead of the US go-live, standing up a cross-functional project focused on automation, affirmation rates and data integrity. The single biggest operational win was consolidating affirmation IDs. The EU T+1 Industry Committee's write-up reaches a conclusion that applies directly to NAV reconciliation: legacy ID mesh and reconciliation debt become critical-path items under T+1.
The wider industry is starting from behind. According to the Committee's second readiness survey, sixty-one per cent of asset managers now have a formal implementation plan in place, up sharply on six months ago, though asset managers still trail the most prepared segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition →.
Step 6: Report breaks to the board in money, not counts
A ManCo or fund board can't act on "347 open breaks". It can act on "open breaks on Fund X total 18% of its NAV error threshold, the oldest is nine days old, and it belongs to the custodian". Express every open break as potential NAV impact against the fund's threshold, age it, and name its owner. This is how you turnpractical data checks and controls into something a conducting officer will sign off.
Why do custodian and administrator reconciliations still fail?
They fail because the controls measure agreement between systems, and agreement says nothing about whether the numbers are correct or whether anyone owns the gap. Here are the specific failure modes:
Circular matching. If you reconcile your IBOR to the administrator, and the administrator built its book from the same custodian file and the same vendor price, a clean match proves nothing. Reconcile quantity to the custodian and price to an independent challenger.
Inherited tolerances. A 1% price tolerance on an equity fund with a 1% NAV threshold can hide most of a material error on a concentrated position. Re-derive tolerances fund by fund using Step 2.
Letting breaks age below the radar. The CSSF's aggregation clause means a break aged for a week and a break that is large are both problems. Once an error is significant, the clock runs: within four to eight weeks of the date of detection, using the notification form on the CSSF website. If total compensation exceeds EUR 50,000, or a single investor is owed more than EUR 5,000, the fund's auditor must issue a special report to the CSSF within three months of the notification.
Assuming the contract moves the liability. Your administration agreement may give you recourse against the administrator, but regulators look to the manager. Correction costs also stay off the fund: the cost of the correction, staff time, legal advice and the auditor's special report included, is never charged to the fund.
Testing T+1 dry runs on synthetic datasynthetic dataArtificially generated data that mimics the statistical patterns of real data, used when real data is scarce, sensitive or expensive to collect.View full definition →. Synthetic positions often break the joins between securities, accounts and prices that reconciliation depends on. The New Stack recently reported that only 34% of practitioners trust the referential integrity of synthetic data. Use masked production snapshots for gate testing.
Treating the morning gate as an operations project. The cut-off timings, the escalation thresholds and the price override hierarchy are governance decisions. Put them in the valuation policy and get the board to approve them.
What to do on monday
- Pull each fund's NAV error threshold from its prospectus or valuation policy, and compute the 10% single-break and 25% aggregate levels for your five largest funds.
- Rerun last Friday's open breaks as potential NAV impact against those levels, and see which funds would have failed the gate.
- Write the field-authority list from Step 1 on a single page and get the heads of operations, valuation and fund administration oversight to sign it.
- Count how many distinct identifiers your top 50 holdings carry across custodian, administrator, vendor and IBOR, and start the cross-reference cleanup with the worst ten.
- Ask your primary pricing vendor and your administrator for their evaluated prices on the same 100 bonds for the same date, and measure the average difference in cents.
- MapMapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.View full definition → your current custodian-file arrival and administrator cut-off times against a T+1 timetable, and flag every batch that finishes after 06:00.
The cheapest NAV error is the one your own gate catches before the administrator strikes the price. Build the gate using your fund's threshold, report it in money, and the compensation calculations under Circular 24/856 or a Rule 2a-5 examination become a file you already have.
The full course on this sector:Data in Asset & Wealth Management.
Frequently asked questions
How should an asset manager set reconciliation tolerances for NAV?
An asset manager should derive reconciliation tolerances from each fund's regulatory NAV error threshold, not from vendor defaults. For a €500m bond UCITS with a 0.5% threshold, a €2.5m limit gives a single-break escalation level of €250k at 10% and an aggregate release gate of €625k at 25%. Both levels are design choices, set fund by fund.
Who is liable when a fund administrator makes a NAV calculation error?
The investment fund manager remains responsible for correcting a NAV error and compensating investors, even when a third-party administrator calculated the NAV. Under Luxembourg's CSSF Circular 24/856, correction costs, including the auditor's special report, can never be charged to the fund. The administration agreement may give the manager recourse, but regulators look to the manager first.
Do small reconciliation breaks below the NAV threshold matter?
Small breaks matter because regulators can add them together. CSSF Circular 24/856 treats untreated simultaneous or successive errors as significant when their combined impact reaches the tolerance threshold, even if each was below it. Twelve €200k pricing breaks on a €500m bond fund come to €2.4m, almost the whole 0.5% threshold.
How does T+1 settlement affect NAV and position reconciliation?
T+1 settlement compresses the time between trade, settlement and NAV, so overnight reconciliation batches end up on the critical path. Europe moves to T+1 on 11 October 2027, and ESMA set 7 December 2026 as the first milestone for allocation and confirmation. Invesco found that consolidating affirmation IDs was its biggest operational win in the US migration.
Go deeper
The lessons that take this article further, free to read.
- 1Benchmarking golden-source pricing and valuation confidenceData in asset management
- 2Measuring data quality with completeness and accuracy metricsData in asset management
- 3Running practical data checks and controlsData in asset management
- 4Mapping the market and portfolio data stackData in asset management
- 5Master Data Management in practice: styles, tools, and the Golden RecordData governance & compliance
Sources
- CSSF Circular 24/856: what NAV errors and investment breaches ask of a Luxembourg fund manager
- LU-8000-BD April 2024
- Circular CSSF 24/856 Protection of investors in case
- Pricing Insights: Why fixed income pricing alignment increasingly matters as workflows evolve
- European T+1 Settlement Readiness for 2027
- EU T+1 Securities Settlement Handbook 3 February 2026
- Fixed Income Vendor Benchmark Study
- SEC.gov
- SEC Charges Private Credit Adviser Over Valuations
- Insights
- NAV Error Materiality Calculator — CSSF 24/856 Thresholds
- Building an AI Analyst, One Football Argument at a Time
- Inside the quest to combat Arctic fires and save the planet’s frozen carbon
- Roundtables: A Conversation With the Creator of AI-Designed Viruses
- Only 34% trust synthetic data’s referential integrity — and it’s holding adoption back
- The Download: failed social media bans, and extending pets’ lives
- Generative AI in the Real World: Observability for AI with Charity Majors
- 7 Open-Source AI Tools That Can Replace Paid Subscriptions
- dbt now supports Amazon Redshift data sharing — and your pipelines just got a lot faster
- This Week in AI: More Capability, More Responsibility
- With dbt State, stop rebuilding what didn't change
- How to Outcompete Your Client’s AI
- Fivetran + dbt Labs Announces New Capabilities to Make Enterprise Data Agent-Ready at dbt Summit 2026
- Everything we announced at dbt Summit and why it matters
- We built dbt State to stop rebuilding what hadn't changed
Finished reading?
Validate your read to earn XP and feed your radar.