What a fund actually asks first, and why the answer is never in the deck
Thirty-seven diligence items, in the order a credit committee reaches them, and which of them a connected loan tape closes without anyone typing.
It asks for your loan tape, one row per loan, and then rebuilds your delinquency, your yield and your leverage from it. That request arrives before yield, strategy or market size. A deck asserts numbers, and diligence recomputes them from loan-level records. The deck is the invitation. The tape is the meeting.
What each artefact is, and what it can be written from.
- The deck. Yield, strategy, market size, the growth chart.
- The management summary of delinquency.
- Any metric whose definition the lender chose.
A fund treats all of this as the control sample. It is what the numbers are checked against, not what they are taken from.
- The loan tape. One row per loan at a stated cut-off: origination date, original amount, current outstanding, term, rate, amortisation type, days past due, restructuring flag and date, product, borrower state, repayment method.
- The payment tape. One row per payment received, with date, amount and the split between principal and interest.
These two are what the analyst rebuilds your delinquency, your yield and your leverage from.
A deck can be written from memory, a plan or a hope. A loan tape and a payment tape cannot: each row has to come out of a system that recorded an actual contract and an actual payment. That is the whole reason diligence starts with the two documents nobody enjoys producing.
The analyst on the other side is paid to rebuild your numbers, not to read your narrative
A loan tape is the loan-level file: one row per loan at a stated cut-off date, carrying origination date, original amount, current outstanding, term, contractual rate, amortisation type, days past due, restructuring flag and date, product type, borrower state and repayment method. A payment tape is the separate record of every payment received, with its date, amount and allocation between principal and interest. They are the only two documents in a data room that cannot be written by somebody who is not looking at the loans.
The job specification for the person who reads them is published. One asset-backed fund asks for somebody able to «review diligence materials with a focus on evaluating and manipulating loan tapes to perform credit analysis» and to underwrite by «building collections and charge-off curves, measuring loss coverages». Their deliverable is your charge-off curve, built by them. Cascade Debt, which works with more than eighty originators and over two dozen institutional investors, calls the step tape cracking and places it before pricing, structuring or capital commitment, answering «a single diligence question: can we trust this tape to reflect real portfolio behavior».
A gap between the file and the deck is not discounted, it is exited: «Investors will be looking for the data to align with the statements in the pitch deck. If they don't, they will walk away». The deck is the control sample, not the evidence. And the tape turns contractual the day you close: in the 2025 Mexican transactions the column set is an annex to the trust deed, sitting beside the eligibility criteria, so what your systems can emit becomes a term of the facility rather than a courtesy to whoever writes the investor update.
The twelve fields we ask for before a package can go to a fund, filled in as a specimen.
One loan, at a stated cut-off date
Specimen values, drawn to show the shape of a row rather than a real borrower. The payment tape is the separate file: one row per payment received, with its date, its amount and the split between principal and interest.
Every number a fund quotes back at you is computed from a few hundred thousand rows shaped like this one. Twelve is not maximalist, it is the smallest set from which an outsider can rebuild a loss curve, and a lender who cannot emit these twelve cannot be read by anybody.
A credit committee works a fixed order, and it is not the order of your deck
The published record converges on six gates, each clearing before the next opens: can we lend to this legal person at all, what is in the book, how has the book behaved over time, who collects if you stop, how is the facility structured and what turns it off, and how will we know every month that nothing broke.
The calendar is a two-quarter operating project. Writing with six named private credit funds, a16z describes an introductory call built on a discovery template of fifteen to twenty questions, then a data room, an underwriting deep dive, a term sheet, and credit agreement drafting that «will take 2-3 months and can take longer depending on counsel familiarity», with successful raisers approaching about ten lenders. The Structured Finance Association's twelve-week best case is qualified by its own authors as realistic only for a well-organised, committed issuer with institutional support. Mexico offers one measured datapoint: on a development-bank warehouse line to a Mexican non-bank lender, board approval on 1 May 2020 was followed by signing on 17 September 2020, 139 days from approval alone, and of six Mexican and Latin American financial-institution deals at that institution two Mexican transactions were board-approved and never signed. Approval is not money.
The order a credit committee actually works, from the published record of how these facilities get done.
- Can we lend to this legal person at allEntity, registration, tax standing, ownership. Answered entirely in documents.
- What is in the bookThe loan tape, the payment tape, concentration against the caps, weighted averages.
- How has the book behaved over timeStatic pools by origination cohort, roll rates, cumulative net loss over the original balance.
- Who collects if you stopBackup servicer, its temperature, control over the collection accounts, the transfer clock.
- How is the facility structured and what turns it offBorrowing base, eligibility, advance rate, first-loss equity, the security package.
- How will we know every month that nothing brokeThe monthly package, the borrowing base certificate, the covenant test schedule.
The order is the instruction. Blocks one and four are procurement and can be finished inside a quarter, so they are never the reason a raise fails late. Block three cannot be bought at any speed, because it is a function of what a loan management system preserved on the day each loan was written.
Block one is not a licence check, because Mexico does not issue a lending licence
A SOFOM, sociedad financiera de objeto múltiple, is the standard Mexican non-bank lending company. Article 87-B of the Ley General de Organizaciones y Actividades Auxiliares del Credito, ultima reforma DOF 14 November 2025, states that credit, financial leasing and factoring may be carried on habitually and professionally by any person without federal government authorisation. What exists instead is a registration, and article 87-K adds the sentence with no United States analogue: an entity that loses it does not hold the character of a sociedad financiera de objeto múltiple.
The ENR and ER split decides how much of your reporting already exists. An entidad no regulada is inspected by the CNBV only for anti-money-laundering compliance and holds a dictamen técnico, a favourable technical opinion on those anti-money-laundering controls issued by an authorised auditor, under article 87-P, renewable every three years, as a condition of being a SOFOM at all. An entidad regulada is tied to a bank or regulated group and inherits that rulebook, including regulated loss provisioning and CNBV accounting criteria. If you are ENR, every metric you present is management-defined, which is why the fund does not read your definitions.
A good answer here is one file of six documents: the SIPRES record with its registration number, the 87-P dictamen with its renewal date, the acta constitutiva showing ENR or ER, three years of audited financial statements, the cap table with equity history, and a SAT opinion de cumplimiento dated within days of closing. The distance from that answer is measurable. A governance self-assessment run by IDB Invest with ASOFOM, AMSOFAC and ProDesarrollo across more than 150 Mexican non-bank institutions found fewer than 30% include independent board members and only one in three has a clear senior-management succession process.
What the statute actually requires of a Mexican non-bank lender, against what a foreign investor assumes it requires.
- Granting credit, habitually and professionally.
- Financial leasing.
- Factoring, the purchase of unpaid invoices at a discount.
- By any person, per article 87-B of the LGOAAC.
This is why a request to see the lending licence has no answer: the instrument does not exist.
- A registration, held with CONDUSEF and visible in SIPRES.
- A technical opinion under article 87-P, renewable every three years, as a condition of being a SOFOM at all.
- Article 87-K: an entity that loses the registration does not hold the character of a sociedad financiera de objeto múltiple.
Losing it is not a fine. It is the end of the corporate form, and that clause has no United States analogue.
There is no licence to lose and no licence to show, so the entity block is answered with a registration, a three-yearly technical opinion and a stack of signed documents. It is the fastest block to fix and the one most first-timers arrive without.
of the seven items in the entity block
Six of the seven are register printouts, notarised instruments or signed audit reports.
Block two is where the tape stops being a spreadsheet and becomes a schema
Our own published minimum, from the data request we send a lender before a package can go to a fund, is twelve fields. Twelve is not maximalist, it is the smallest set from which an outsider can rebuild a loss curve. Annex IV of Commission Delegated Regulation (EU) 2020/1224 runs to 101 loan-level fields plus 20 collateral fields for corporate and SME loans, and Annex VI carries 69 for consumer, per the Bank of England restatement effective 1 November 2024. SEC Schedule AL Item 3, the auto-loan schedule closest to an instalment tape, is 72 fields.
Mexico already has a loan-level schema and most unregulated SOFOMes never see it. CNBV report C-0430, alta de créditos comerciales, is 51 columns of origination data; C-0431, seguimiento, is 51 columns of monthly performance including field 47, número de días con atraso; C-0432, baja, is 16 columns recording the exit. It binds a SOFOM ER, which is why an ENR tape arrives thinner than expected and why the ER can reconcile its tape to a filing a third party already receives.
The shape mismatch costs more than the missing columns. The CNBV models a línea de crédito carrying one or more disposiciones, and for factoring directs the lender to assign one numero de disposicion to all invoices sharing term, rate and counterparty in a month. A fund's tape assumes one row is one loan. A Mexican system assumes one row is one drawdown. Reconciling those definitions is the commonest structural rework in a first Mexican raise. The two artefacts a first-timer never brings are the payment tape and the data dictionary, both named deliverables rather than niceties.
The honest way to declare a gap is a grammar, not a blank cell. Article 2 and Annex I of EU 2020/1224 define five no-data codes: ND1, never collected at underwriting; ND2, collected but not loaded into the reporting system; ND3, held on a separate system; ND4, available only later; ND5, not applicable. ND1 says no history of that variable exists and changes the credit assumption permanently. ND2 says the history exists, has a location and has a delivery date. An empty cell collapses both and is priced as the worse one.
The failure mode is not the empty column. It is the column that looks answered.
In the most regulated tape market in the world, in one quarter, current principal balance was reported as zero on 4,717,333 loans and current interest rate margin as not applicable on 3,495,901 loans.
The no-data grammar the European template defines, and what each code does to a credit assumption.
- Never collected at underwritingNo history of that variable exists. The credit assumption changes permanently, and no amount of engineering brings it back.
- Collected, not loaded into the reporting systemThe history exists. It has a location and it can be given a delivery date.
- Held on a separate systemThe same, one integration further away.
- Available only laterA timing statement rather than a gap. It tells the fund when to ask again.
- Not applicableThe field does not apply to this product. Nothing is missing at all.
The distance between the first code and the second is the whole point. One says the history does not exist and never will, which changes the loss assumption permanently. The other says the history exists, has an address and has a delivery date. A blank cell says both at once, so it is read as the first.
Loan-level field counts across the tape templates that actually exist, ordered ascending, with our own request marked.
The fund is not inventing its request. It is cutting down a template that already exists, and your file is measured against a schema you have never seen.
Block three decides your price, and a short history is not neutral, it is substitution
A static pool, also called a vintage or cosecha, groups every loan originated in the same period and follows only that group from month zero, so growth cannot dilute the loss rate and an underwriting decision sits in the same column as its consequence. The most precise published specification of this block is Mexican, free and in Spanish: HR Ratings, Metodología para Activos Financieros, October 2022, requires cohort analysis of the whole book, monthly periodicity, annual origination cohorts, five years back, split by product.
The consequence of a short history is not a decline, it is a substitution. Where an originator's own data does not reach far enough back, the agencies replace the missing period with marketwide performance covering the same span. Your loss assumption becomes the sector's, and every basis point of outperformance you cannot evidence is one you donate.
The number to learn to quote about yourself is your default rate by cohort, because the structure is built to withstand a multiple of it. One Mexican 2025 transaction shows a historical default rate of 2.3% against a maximum the structure absorbs of 27.2%, a ratio of 11.8 times. The denominator is the tell: charge-offs belong over the original pool balance rather than the current one.
Mexico even publishes the file-quality threshold as a number. HR Ratings draws its file review by formula, at a 90% significance level with a maximum admitted error of 5.0%. Above 10.0% inconsistent files a second sample is drawn, and if the inconsistencies persist the agency may apply a notch penalty and require a third-party audit within six months of issuance; above 20.0% it «evaluara la manera de continuar con el proceso de calificación». Both thresholds are free and public, so you can run the agency's own sample on yourself first.
What happens at each level of file inconsistency in a Mexican structured transaction, sampled at a 90% significance level with a maximum admitted error of 5.0%.
Share of sampled files found inconsistent
This is the only gate in the whole checklist whose grading scheme is published in advance and free to read. A lender can draw the same sample on its own files before anyone else does, and the cost of finding out at the second threshold rather than the first is a notch and a third-party audit inside six months.
the loss a Mexican structure absorbs, over the loss the book actually ran
A historical default rate of 2.3% against a maximum absorbed of 27.2%, in one 2025 transaction.
Block four is answered in signatures, not in dashboards
A substitute or backup servicer is a third party contracted in advance to collect the book if you stop, so somebody with the right and the systems to collect is already appointed on the day it is needed. S&P's operational risk framework, first published 9 October 2014 and republished 28 October 2025, names four conditions under which it will not issue or maintain a rating, and three are documentation or data failures rather than credit failures.
The temperature of that arrangement is worth real money and is defined purely as a data obligation. S&P puts a qualified backup servicer at up to six notches of uplift on a capped rating, hot at four to six, warm at three to four, cold at one to three, and defines hot as requiring upfront data mapping and testing, daily or weekly data file receipt, and verification of the servicer's report calculations. A lender whose tape lives in a spreadsheet cannot be hot at any price, because there is no system to map from, and the filed price of the arrangement is modest against what it buys: US$3,000 per month in one 2025 United States transaction.
Mexico is stricter for a first-timer. HR Ratings expects the administrador maestro to work without depending on the originator's operating capacity, with its own access to the files and to the pagarés, and applies notch penalties where that access does not exist. In the 2025 deals, failure to appoint a substitute servicer within 120 calendar days of a termination notice is itself an accelerated-amortisation event. Irreplaceability is a penalty rather than a moat: where the agency finds no viable substitution mechanism, the rating is restricted by the originator's own credit, which defeats the purpose of the vehicle.
Rating uplift by the temperature of the arrangement, and what each temperature requires of the lender's systems.
Backup servicer temperature, against uplift on a capped rating
Every rung of this ladder is defined by how often a file moves and whether somebody checked the arithmetic in it, not by the servicer's balance sheet. A lender whose tape lives in a spreadsheet cannot be hot at any price, because there is no system to map from. The filed monthly cost of the arrangement is small against six notches.
Block five is arithmetic that somebody else will redo on your book
A borrowing base is the running calculation of how much you are allowed to have drawn right now: eligible receivables, less exclusions and concentration excesses, multiplied by the advance rate, less reserves. The advance rate is the percentage of eligible collateral the fund will lend against, and the remainder is your own equity in first-loss position, absorbing losses before the fund loses a peso. Eligibility criteria are the per-loan tests a receivable must pass to count at all. A concentration limit caps how much of the pool one obligor, sector, state or tenor bucket may represent. Mexico quotes the same machine upside down: aforo, or overcollateralisation, is a coverage multiple of collateral over debt, so an aforo of 1.33x is an advance rate of about 75%, and confusing the two in a first call is a tell.
The Mexican levels are public and carry a 2025 vintage. Across four transactions rated between March and November 2025, single-obligor caps run 3.0% to 4.0%, top five at or below 18%, top ten at or below 35%, sector 20% to 30%, per state 12% to 15% with Mexico City allowed 25%, and restructured exposure at or below 5%. The same agency publishes the point at which a pool stops being structured finance and becomes corporate credit: a portfolio is pulverizada when no client exceeds 5.0% and the top ten stay below 35.0%.
Inverting the published aforo multiples of those 2025 deals gives an implied advance rate of about 75% to 83%, so 19% to 25% of every peso originated stays as the lender's own first-loss equity, and a US$10M book needs about US$1.9M to US$2.5M of your own money standing in front of the fund. These are implied by inversion and are never quoted as advance rates. The 95.0% advance rate disclosed by a Nasdaq-listed United States consumer lender, stepping to 92.0% on triggers, is not a benchmark for a Mexican first-timer and should never be quoted as one. Nor is size: initial institutional facilities rarely exceed US$25 million.
The most actionable published warning is that the effective advance rate may be lower than the stated rate in the documents, so model your borrowing base on your current portfolio, because the highest stated rate is not always the best deal. In one 2025 Mexican transaction the pledged pool itself breached the single-obligor, top-five and top-ten tests, and P$9.6 million of a P$252.5 million deal, 3.8%, was struck from the base on day one. Nothing had gone wrong with the loans. The pool was pledged in a shape the documents did not accept.
Concentration limits across four Mexican transactions rated between March and November 2025. Where a limit runs as a range, the bar spans it.
Below 5.0% per client with the top ten under 35.0% is the point at which the same agency calls a pool pulverizada, and treats it as structured finance rather than corporate credit.
Mexico City is allowed 25%.
Advance rates in the same deals, implied by inverting the published aforo multiples, run about 75% to 83%. That leaves 19% to 25% of every peso originated standing as the lender's own first-loss equity.
None of these is a credit test. A lender with a perfect payment record and three large clients fails the first row on arithmetic alone, and the balance comes straight off the borrowing base. Run them against your own pool before the pool is pledged, because the day it is pledged the shape is fixed.
of a pledged Mexican pool struck from the borrowing base on day one
A concentration test, not a credit test: P$9.6 million removed from a P$252.5 million deal.
Block six is a covenant, and it is the covenant a healthy lender trips
The floor is published precisely: at a bare minimum lenders ask for monthly reporting packages consisting of loan tape metrics, consolidated and sometimes audited financial statements, and accounts payable and receivable, within thirty days of month end. The format you would otherwise have to invent is published too, since the European investor report template carries, per trigger, the threshold level, the actual value, the status, the cure period and the consequence for breach (EU 2020/1224 Annex XII).
Lateness is not a lesser category of breach. In the 2025 Mexican deals a master-servicer report delivered late and not cured within ten business days, and file-audit exceptions above 10.0% of the files validated, are each accelerated-amortisation events in their own right, sitting in the documents on the same footing as a portfolio breach. A default of any kind is leverage: it reopens pricing and terms in the counterparty's favour.
Here the causality runs opposite to what founders assume. A rating and a listing are lost because reporting stopped, not because reporting stopped after losses appeared. S&P withdrew its ratings on one large Mexican non-bank lender expressly «por falta de informacion» on 1 December 2022, and the listing followed. Moody's writes the same rule as an obligation, a prohibition on rating with insufficient information under which employees «must refuse to provide a rating when there is a lack of reliable data, or the quality of information available is not satisfactory» (Exhibit 2 to Form NRSRO, retrieved the quality of information available is not satisfactory».
The shape of one row of the investor report template, with a Mexican portfolio covenant filled in as a specimen.
One line of the monthly package
Specimen values, drawn to show the shape of the line rather than a real portfolio. The five columns are the ones the European investor report template requires per trigger: threshold level, actual value, status, cure period and the consequence for breach.
Nothing in this row is an opinion, and nothing in it can be assembled the week it is due. The cure period is the only forgiving column, and in the 2025 Mexican deals a master-servicer report delivered late and not cured within ten business days is an accelerated-amortisation event in its own right, on exactly the same footing as a portfolio breach.
Four defects in one real package, and not one of them was about credit
In July 2026 we did a fund analyst's job on the screening pack of a Mexican lender with a book of about MXN 124 million, at a cut-off date of 31 May 2026. The asset was attractive. The package was unfundable as delivered, and the reason was never the credit.
Each finding with the figure the package presented, the figure a recomputation produced, and the method that produced it.
| Defect | As presented | As recomputed | Method |
|---|---|---|---|
| Netting hid the leverage | Debt to equity 2.68x | 7.59x, understated by 2.8 times | MXN 71,336,593 of customer loans was netted against an equal debt collateral line; regrossing the two restored debt of MXN 110.36m against the MXN 39.03m shown |
| The tape contradicted the claim | Historical default and charge-off rate 0.02% | 1.41% in 2023 and 1.34% in 2024 | Reading the loss history tab of the same workbook. The contradicting number travelled in the same file |
| Headline metrics were typed, not computed | Weighted-average yield 102% | About 70.7% per annum | Recomputing from the loan-level data. The figure was stale rather than false, and one hardcoded cell ends trust in every other cell |
| No loss provisioning anywhere | Annual profit about MXN 4.5m | About MXN 0.7m normalised | Applying reserve scenarios: 1.5% consumes 42% of annual profit, 3% consumes 84%, 5% turns the year into a loss |
Not one of the four needed an outside document to find. Each was arithmetic inside a file the lender had already handed over, which is why no deck could have prevented any of them and why a recomputation found all four in an afternoon.
Alongside these sat 42 senior secured creditors against apparently the same asset pool, 37 of which recorded collateral as «NA», and about 30 facilities with no stated maturity.
The tape is where the sector's memory is stored, and the market reprices the discovery
Sorting a tape descending by exposure is a one-minute operation that no summary survives. One Mexican lender described a book averaging about US$175,000 per loan while carrying a single US$35 million exposure, and about half the value of its stated loan portfolio, about US$1.1 billion, turned out to be unpaid interest the company did not break out, against a reported non-performing ratio of about 1.5%.
A definition chosen in good faith can make a portfolio unrecognisable to an outsider, which is why a fund does not ask for your non-performing loan ratio at all. It asks for raw payment history and computes the ratio itself, under its own definition, on every loan. A zero is not a better answer either, because «if NPL is zero, either you are not taking enough risk, or you are lying».
The supervisory metric has moved from the ratio to the flag. The Financial Stability Board's recommended borrower credit-quality measure is the share of loans with «covenant breaches, waivers, restructurings, amend-extends, distressed exchanges, payment deferrals, PIK activations, or defaults» in the last twelve months. A book reporting 2% at ninety days past due and 25% touched in twelve months has a 25% problem, and the touched number lives only in the restructuring flag on the tape. That is also the good news, because uncertainty is the one input a lender can reduce at will. The gap between two audit dates is where that uncertainty is priced.
What the supervisory measure of borrower credit quality has moved to, against the ratio a lender volunteers.
- A non-performing ratio, computed under a definition the lender chose.
- Loans past ninety days, as a share of the current book.
- One number, at one date.
A zero here is not a better answer: if the ratio is zero, either the risk is not being taken or the number is not real.
- Covenant breaches and waivers.
- Restructurings and amend-extends.
- Distressed exchanges and payment deferrals.
- Interest capitalisations and defaults.
The share of loans touched by any of these in the last twelve months, which is the Financial Stability Board's recommended measure of borrower credit quality. It is routinely a multiple of the ratio beside it.
The ratio counts what is late today. The flag counts what was touched in a year, and every item on the right lives in one column of the loan tape that no summary carries. Sorting a tape descending by exposure is a one-minute operation, and it is the operation no summary survives.
The distribution is the finding, not the total
Of the 37 items, a connected feed closes 18 outright, 4 partially and 15 not at all. That is the headline, and it is the least interesting part. Where those 18 sit is the finding.
Count of items closed outright, partially and not at all, by block, summing to 37.
Buying a data layer to solve the entity block is a category error, and refusing one because the entity block exists is the same error inverted. The two blocks that go almost entirely violet, vintage and reporting, are the two a first-timer cannot assemble out of documents that already exist.
Running total of items reached against items a connected feed closes outright, in the order a credit committee works through them.
Blocks in committee order, left to right, numbered as they are numbered in the checklist below. Both totals are counted from that checklist rather than entered by hand.
The violet line is flat for the whole of block one, so a committee is seven items into the list before a data connection closes anything at all, and what stalls a first raise there is paperwork rather than data. The two lines never converge: nineteen items separate them at the end, and those nineteen stay the lender's own work.
All 37 diligence items, in the order a credit committee reaches them. Each block header carries its own tally and one line on what a live connection does about that block.
What a connected feed doesNothing at all. Every item here is a register printout, a notarised instrument or a signed audit report, and no data connection produces one. The only thing that helps in this block is knowing which six documents belong in one file before the first call, and assembling them is procurement rather than engineering.
- 1The company legally is a SOFOM and the registration is liveCONDUSEF SIPRES record, dated printout with the registration number
- 2The compliance gate that keeps the registration aliveCNBV art. 87-P dictamen técnico favorable, with its renewal date
- 3Corporate form and objeto socialActa constitutiva and Registro Público de Comercio inscription showing ENR or ER
- 4Three years of audited financial statementsSigned audit reports, three consecutive years
- 5Ownership, control and equity historyCap table, ownership breakdown, equity invested, tax returns since inception
- 6Tax good standingSAT opinión de cumplimiento 32-D positiva dated within days of closing, plus 69-B searches
- 7Credit-bureau membership and live reportingSIC contract plus recent transmission acknowledgements
What a connected feed doesPart of it. The loan tape, the payment tape, the concentration snapshot against the caps and the weighted-average characteristics can be computed straight from a loan management system every month, so they arrive reconciled rather than typed. The data dictionary and the assignability opinion stay documents somebody writes.
- 8The loan tape, one row per loan, full field coverageLoan-level file at the cut-off date, mapped against the disclosure templates
- 9The payment tape, a separate artefactEvery payment received, with date, amount and allocation to principal and interest
- 10The data dictionaryDocument defining every code, column heading and non-obvious term
- 11Concentration snapshotSingle obligor, top five, top ten, sector, state and tenor, against the caps
- 12Weighted-average characteristics computed from the tapeYield, tenor, seasoning and bureau score reconciled to the loan-level file
- 13The receivable is legally assignable and pledgeableLoan contract template, pagaré and RECA entry with the assignment clause present
What a connected feed doesAll of it except the definitions. Static-pool loss curves, vintage default rates, the roll-rate waterfall, cumulative net loss over the original balance and recovery history are all rebuildable from the same monthly extract, provided the monthly extracts were kept. The charge-off, fraud and recovery policy is a document only the lender can sign.
- 14Static-pool loss curves by origination cohortCumulative charge-offs over originated principal at each month on book, five years, by product
- 15Vintage default ratesDefault rate per cohort over time, newest cohorts visible against oldest
- 16Roll-rate waterfall by vintageMigration between delinquency buckets, per vintage, last twelve months
- 17Cumulative net loss on the original pool balanceNet of recoveries, over original rather than current balance
- 18Recovery historyRecoveries after default over principal defaulted, by cohort, function named
- 19The definitions behind the numbersWritten charge-off, fraud and recovery policy
- 20Restructuring and renewalPolicy document, plus count and balance restructured in twelve months against the cap
What a connected feed doesOne item outright, and part of a second. A verified daily file is what lets a backup servicer be contracted hot rather than cold, which is the difference S&P prices at up to six notches of uplift, and a payment feed evidences where remittances actually land. The policy, the vendor-named cash diagram, the backup servicing agreement, the account control agreement and the org chart are signatures and procurement.
- 21Servicing and collections policyThe policy document and the collections waterfall
- 22Where the cash physically goesDiagram of disbursement and remittance by entity, with the vendors named
- 23Somebody can collect this book without youBackup servicer agreement, its temperature stated, and the transfer clock
- 24Control over the collection accountsAccount control agreement or the Mexican trust-account equivalent
- 25The peopleOrg chart, servicing and risk team structure, head of risk background
- 26The daily file that makes hot backup possibleLive data connection with verified report calculations
What a connected feed doesThe arithmetic, not the paper. Eligibility criteria can be run across every loan, the borrowing base computed on the current portfolio at the offered advance rate, the excluded balance shown in pesos and percent, and first-loss equity tested with FX sensitivity. The RUG searches, the trust deed and the facility schedule stay with counsel.
- 27Every existing facility, in fullSchedule of all facilities with amount, maturity, rate, specific collateral and priority
- 28The borrowing base modelled on your bookBase computed on the current portfolio at the offered advance rate, effective against stated
- 29How much of the book is actually ineligibleEligibility criteria run across every loan, excluded balance in pesos and percent
- 30The book is not already pledgedRUG search on the company's own folio at signing and re-run at closing
- 31The security package worksTrust deed with the signed extrajudicial-sale section and named ejecutor, plus RUG receipts
- 32There is enough first-loss equityNet worth against the haircut the Mexican advance rate implies, with FX sensitivity
What a connected feed doesAll of it, and this block is the one that is pure mechanics. The monthly package, the borrowing base certificate computed on the loans rather than asserted, the covenant test schedule with threshold, actual, status and cure period, the tape reconciled line by line to the balance sheet, and the quarterly cohort refresh with a record of on-time months.
- 33The monthly packageLoan tape metrics, financial statements and payables and receivables, within thirty days
- 34The borrowing base certificateCertificate with each funding request, computed on the loans rather than asserted
- 35The covenant test schedulePer test: threshold, actual value, status, cure period and consequence for breach
- 36The tape reconciles to the financialsTape total tied line by line to the balance sheet, gross and un-netted
- 37Vintage refresh and reporting disciplineCohort curves refreshed at least quarterly, plus a record of on-time months
Colour is the answer, and it clusters rather than spreading. A feed closes 18 of the 37 items outright and 4 in part, and every one of the 15 it leaves is a register printout, a notarised deed, a countersigned agreement or a written policy. Block six has no hollow square in it. Block one has no filled one.
Ten of the twelve items in the vintage and reporting blocks are closed outright and one partially, and the single exception is a written policy document. Those are exactly the two blocks a first-time raiser cannot assemble from documents that already exist, because both rest on monthly snapshots that were either archived or were not.
The feed closes nothing in entity and licensing and almost nothing in servicing, where six of seven entity items and four of six servicing items are register printouts, notarised deeds and countersigned agreements. The instruction follows from the order: blocks one and four are procurement and can be signed inside a quarter, while block three cannot be bought at any price in month one, because it is a function of what your loan management system did or did not preserve on the day each loan was written.
Three kinds of work, and only one of them can be bought in a quarter.
- 15 are signatures
- A SIPRES printout, an 87-P dictamen, an acta constitutiva, three audit reports, a SAT opinion, RUG searches, a trust deed, a backup servicing agreement, an account control agreement and four written policies. No data connection produces any of them, and any software sold as producing them is selling the wrong thing. They are procurement, and procurement can be finished inside a quarter.
- 4 are both
- Bureau transmission, restructuring and renewal, the schedule of existing facilities and where the cash physically lands. The count and the balance are computed from the loans; the policy that governs them, and the diagram that names the vendors, are written.
- 18 are arithmetic
- The loan tape and the payment tape, the concentration snapshot against the caps, the weighted averages, five cohort views including static-pool loss curves and cumulative net loss over the original balance, the borrowing base with its ineligible balance, and the monthly package with its covenant test schedule and the tape tied to the financials. Every one is recomputable from loan-level records on any month those records survive for.
The one thing no connection can do is return a month that was never archived. That is why the vintage block is the reason to start early rather than the reason to buy late, and it is the single most common way a first raise arrives a year too soon.
Of the 37, 15 are documents somebody signs and 18 are arithmetic on the loans. The first group is procurement and moves fast. The second is a function of what a loan management system preserved, which is why it cannot be started in the month the raise starts.
What the same book looks like when it is read continuously
Eighteen of the thirty-seven are arithmetic on the loans, so what they look like once a book is read continuously rather than assembled for a data room is a fair question. These are those readings, on demonstration data.
Outstanding
1,246.8M
MXN, rebuilt from the loan management system rather than submitted
- Borrowers
- 1,842
- Unmeasured
- 173
- Last read
- 01 Aug
One borrower in ten is unmeasured rather than late, and priced as if it were both.
Grey is not a softer red. A grey borrower is current and unmeasurable, and a fund that cannot separate the two prices both as the worse one.
Growth with falling arrears prices a facility. Growth with rising arrears reprices it, and a quarterly summary shows neither.
The covenant is drawn as a line rather than reported as a number, so the distance to a breach is read instead of computed.
Every figure, borrower and counterparty in this block is demonstration data created for illustration. It is not a real portfolio. The instruments are the ones on our analytics page, shown here because the article above is about what cannot be seen between two audits.
Effect is not a bank, a lender, a broker-dealer, an investment adviser or a credit rating agency. Nothing here is investment, legal or tax advice, and no outcome is promised. Every figure carries its source and its as-of date so a reader can check it rather than take it. Mexican advance rates are implied by inverting published aforo multiples rather than quoted directly. The anonymised case is our own reading of a package a lender supplied to us, as of a 31 May 2026 cut-off.
Where every figure on this page comes from28 sources
- 01i80 Group, job postings for asset-backed credit analysts, from December 2022.
- 02Cascade Debt, on tape cracking as the step before pricing, 16 January 2026.
- 03Cascade Debt, on the tape aligning with the pitch deck and on a zero NPL, 31 January 2024.
- 04a16z, on raising a first debt facility, written with six named private credit funds, 4 October 2023.
- 05Structured Finance Association, twelve-week issuance timeline, October 2025.
- 06IDB Invest, project 13057-01, warehouse line to a Mexican non-bank lender: board approval 1 May 2020, signing 17 September 2020, read 5 August 2026.
- 07Ley General de Organizaciones y Actividades Auxiliares del Crédito, articles 87-B, 87-K and 87-P, ultima reforma DOF 14 November 2025.
- 08IDB Invest with ASOFOM, AMSOFAC and ProDesarrollo, governance self-assessment across more than 150 Mexican non-bank institutions, 23 February 2026.
- 09Warehouse loan and security agreement, amended and restated to 21 March 2002, advance field minimum.
- 10Effect lender data request, item B.1, our published twelve-field minimum, July 2026.
- 11Finley, consumer loan tape template, 30 March 2022.
- 12CNBV, instructivo for reports C-0430, C-0431 and C-0432, updated 23 January 2026.
- 13Commission Delegated Regulation (EU) 2020/1224, article 2 and Annexes I, IV, VI and XII, per the Bank of England restatement effective 1 November 2024.
- 1417 CFR 229.1125, SEC Schedule AL Item 3, retrieved 5 August 2026.
- 15Fannie Mae, single-family loan performance data file layout, 26 May 2026.
- 16ABF Journal, on payment tapes, data dictionaries, facility size and effective advance rates, 24 November 2025.
- 17ESMA, Securitisation Repository Insights, on reported zero balances and ND5 fields in the fourth quarter of 2023, March 2024.
- 18HR Ratings, Metodología para Activos Financieros, cohort requirements and file-review thresholds, October 2022.
- 19HR Ratings, SERFICB 25 presale, 24 March 2025; ARRENCB 25, 28 August 2025; LFUCB 25, 20 November 2025; CAPEMCB 25X final characteristics, 2025.
- 20S&P Global Ratings, operational risk framework for structured finance servicers, first published 9 October 2014 and republished 28 October 2025.
- 21Academy Securities, filed backup servicing fee in a 2025 United States transaction, 15 October 2025.
- 22Nasdaq-listed United States consumer lender, current report on Form 8-K disclosing a 95.0% advance rate stepping to 92.0%, 2 April 2025.
- 23S&P Global Ratings, withdrawal of ratings on a large Mexican non-bank lender for lack of information, 1 December 2022.
- 24Moody's, Exhibit 2 to Form NRSRO, prohibition on rating with insufficient information, retrieved 5 August 2026.
- 25The Wall Street Journal, on a Mexican lender's single large exposure and unpaid interest, 15 July 2022.
- 26Financial Stability Board, Report on Vulnerabilities in Private Credit, borrower credit-quality measure, 6 May 2026.
- 27Effect credit memo on an anonymised Mexican lender, book about MXN 124 million, cut-off 31 May 2026, analysis dated 17 July 2026. The source for the four defects.
- 28Effect evidence pack, Part C, 5 August 2026, built from published fund, rating-agency and development-bank criteria and from our own lender data request. The source for the 37-item checklist and its coverage counts.
Questions this raises
- What does a fund ask for before it asks about yield?
- The loan tape at a stated cut-off date, one row per loan, and the payment history behind it, then a reconciliation of that tape to the audited financial statements, gross and un-netted. Yield only becomes meaningful once the denominator is agreed, which is why the first hour is arithmetic rather than strategy.
- We are a SOFOM ENR and do not file the CNBV R04 C reports. Does that disqualify us?
- No, but it removes a shortcut. A regulated SOFOM can reconcile its tape to a 51-column monthly filing a third party already receives, while an unregulated one has to build that reconcilability itself, and the fund will spend its diligence budget on the bridge you cannot point to.
- What do we do about fields we simply do not have?
- Declare them rather than leave them blank, using the European ND1 to ND5 grammar in EU 2020/1224. ND1 means the field was never collected at underwriting and changes the credit assumption permanently, while ND2 and ND3 mean the data exists elsewhere and needs only a delivery date. An undeclared empty cell is priced as ND1 whether or not it is one.
- Our company is three years old and our data is thin. What can we fix before the raise, and what can we not?
- Of the 37 items a committee works through, a connected data feed closes 18 outright and 4 in part, and the 15 it leaves are register printouts, notarised deeds and countersigned agreements you can procure inside a quarter. What cannot be fixed late is cohort history. The rating agencies expect originator-specific performance for the longer of five years or a full economic cycle, and where that is unavailable they substitute marketwide data covering the same span, so your outperformance stops counting and the sector average is used in its place.
- How long should we budget for a first institutional facility?
- Twelve weeks is the published best case for a well-organised issuer and three to six months is the realistic range, with credit agreement drafting alone taking two to three months. The one measured Mexican datapoint took 139 days from board approval to signing, and disbursement remains conditional after signing.