What a loan tape is

A loan tape is the loan level file behind a credit portfolio: one row per loan at a stated cut-off date, with columns for identity, terms, borrower, collateral, balances and payment behaviour. Twenty-four columns will get a fintech a term sheet. The European consumer template runs to 69 fields, and the same word covers both.

A loan tape is one row per loan, and everything else in a data room is a claim about it

The file carries, at a minimum, origination date, original amount, current outstanding, term, contractual rate, amortisation type, days past due, restructuring flag and date, product type, borrower location and repayment method. The data request we send a lender asks for exactly that list before it asks anything about yield, and asks for it as a separate file rather than as a page inside a deck.

Authority in that file comes from two properties, and neither of them is the column count. The rows are emitted by a system that recorded an actual contract and an actual payment, and the whole file is stated as of a date, so every number computed from it can be recomputed by somebody else. A deck asserts. A tape can be checked.

A loan tape population is stated with the cut-off date, because the same book gives two different files

The population is the set of loans the file covers, and a tape carries one of two. Every loan outstanding at the cut-off date is a snapshot of the live book: it answers what the lender holds now. Every loan originated since a stated start date, repaid, written off and sold loans included, is an origination history: it is what a static pool and a loss history are rebuilt from, because a loan that leaves the file takes its loss with it.

Name the population in the data dictionary, next to the cut-off date, with the row count and the total outstanding that should tie to the balance sheet on that date. A reader who cannot tie the two cannot tell a missing loan from a repaid one, and reads every gap as the worse of the two.

A loan tape template: our twelve fields as a CSV, with the dictionary that has to travel with it

The five rows above are also a file. The template carries the twelve fields of the data request we send a lender, with the restructuring field split into a flag and a date, and the cut-off date stamped on every row so that two months of tape can be stacked without losing which month a row describes. The dictionary beside it states each column's unit and rule, and how an absent value is declared rather than left blank.

Download the loan tape template (CSV, 14 columns, 5 illustrative rows) and the data dictionary (CSV). Both are free to copy and adapt; they are version 2026-10-05.

There is no single loan tape standard, there are four layers, and a fund asks for a subset of one

Regulatory schemas sit at the top and they are the widest, because they are written to be enumerated in law rather than negotiated. The SEC asset level schedule under Regulation AB II lists 282 fields for a residential mortgage, 159 for a commercial mortgage and 72 for an auto loan, and the European templates in Commission Delegated Regulation (EU) 2020/1224 list 69 fields for a consumer exposure and 121 for a corporate one once collateral is counted.

Rating agencies sit one layer down, and they ask for less width and more history. European DataWarehouse built its extended template with Moody's and S&P Global Ratings to carry the fields the agencies still needed after the regulatory template dropped them, and HR Ratings expects originator specific cohort performance rather than a snapshot of today.

The working private credit tape is far narrower than either regulatory layer. Finley publishes a consumer template of 24 columns, our own published data request asks a lender for a twelve field minimum before anything else is discussed, and a warehouse credit agreement from 2002 goes narrower still, naming eight fields in the transmission that actually moves the money.

The local reporting layer is the one lenders forget they already have. A regulated Mexican lender files CNBV reports C-0430 and C-0431 every month, 51 columns each, on a stable credit identifier, which means the loan level data exists in the system long before anybody asks for a tape. It also exists through the eleven months between two audits, when the fund that financed the book is the one party not receiving it.

So the distance between a lender's current file and an institutional tape is almost never underwriting practice. About 102 columns of loan level data already sit inside a regulated Mexican book by obligation, per the CNBV instructivo updated on 23 January 2026, and the lender still cannot hand over one file with a dictionary attached. The gap is an export, and an export can be built inside a quarter. What a fund then does with that export, block by block, is a separate piece.

Fields differ by asset class, and the difference is wider than most lenders expect

A consumer tape and a commercial tape are not the same file with different labels on the columns. The European consumer annex asks for employment status, primary income, how that income was verified and whether repayment is assigned from a salary or a pension, while the corporate annex asks for industry code, enterprise size, revenue, total debt and EBITDA, per EU 2020/1224 as restated for the United Kingdom on 1 November 2024.

Asset class also decides how much history has to travel with the file. Fannie Mae publishes a single family loan performance layout of 108 fields as of 26 May 2026, and it is the only rating grade schema anyone can copy without a licence, which makes it the cheapest starting template for a lender who has to begin somewhere and has no counsel to ask.

A tape without a data dictionary is not a smaller tape, it is an unverified one

A dictionary is a named deliverable on a fund's request list rather than a courtesy. One published diligence checklist asks the lender to «Include a data dictionary explaining all codes, column headings, and non-obvious terms», and a second lists «Clear data dictionary and field definitions» beside loan level coverage and historical performance among the things to prepare before the raise.

Without one, the reader cannot tell an empty cell from a zero, a restructured loan from a renewed one, or days past due counted from the missed instalment from days past due counted from the last payment received. None of those is a credit question, and every one of them changes the delinquency the analyst computes from your file.

Absence has a grammar, and the grammar is worth adopting well outside Europe. Annex I of EU 2020/1224 defines ND1 to ND5: 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 blank is priced as ND1, because from outside there is nothing to tell the two apart.

Calculated fields do not travel, because the receiving side recomputes them from raw columns

Tape cracking is the name the market gives to the step before pricing, structuring or capital commitment, and its first move is «standardizing loan tapes into a consistent schema. This includes normalizing dates, balances, statuses, identifiers, and relationships across files that are often delivered in inconsistent or ad hoc formats».

The same firm, Cascade Debt, describes itself as one that «cleans and validates data tapes by recalculating values from raw inputs», across more than 80 originators and two dozen institutional investors, which is how an investor that never sees the book checks a certificate between audits. The practical rule for a lender follows directly: ship raw columns, ship the dictionary, and let the other side compute the averages. An average the reader cannot reproduce is not an average the reader disbelieves. It is one the reader never sees.

A weighted average is faked when there is nothing to weigh, and the fake is visible in the cell

In a screening package we read in July 2026, on an anonymised book of about MXN 124 million, the weighted average yield was a hardcoded formula sitting on a base that did not equal the portfolio, and a stated historical default rate of 0.02 per cent sat in the same workbook as a loss tab showing 1.41 per cent for 2023 and 1.34 per cent for 2024. That is why the data request we send now asks for the loan level file and the cohort loss curves before it asks a single question about yield.

The discrepancy is about seventy times, inside one package, and it is not primarily a credit finding. An analyst clicks the cell, sees the formula, and concludes that the loan level register the entire diligence depends on is probably not maintained. Reading a defect in the data as a defect in the operation is what the first hour of diligence is for.

How a fund checks a loan tape in the first hour, in the order it runs the checks

The checks run in a fixed order because each one is the precondition for the next. A tape whose control totals do not tie to the balance sheet is not analysed further, however good its columns are, and a delinquency figure is not read until it has been recomputed from the payment tape. The table is the order we see funds use, and the right hand column is where a tape most often fails at that step.

A tape that clears all eight is the one a fund starts to price. The data request we send a lender is written in the same order for the same reason: it asks for the file that makes the first check possible before it asks for anything that depends on it.

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. Every figure carries its source and the date of the version counted, so a reader can check it rather than take it. The anonymised case is our own reading of a package a lender supplied to us, at a 31 May 2026 cut-off.