Synthetic SBA Form 413 Personal Financial Statement Data
Synthetic training data — no real PII, fully coherent identities
Generate synthetic SBA Form 413 Personal Financial Statements — a six-page personal balance sheet where summary totals on page one must reconcile against detail schedules three pages later. Training data for SBA lending and small-business underwriting automation.
148
Fields per document
6
Pages
Financial
Category
What this document is
SBA Form 413 is the Personal Financial Statement every principal with 20% or more ownership must file with a 7(a), 504, disaster or 8(a) application. It is a full personal balance sheet — assets, liabilities, source of income, contingent liabilities — followed by detail schedules for notes payable, stocks and bonds, and real estate owned. It is the single most information-dense document in the SBA loan file, and the one underwriters spend the most time keying by hand.
Why generate synthetically
Personal financial statements are the least shareable documents in commercial lending. They carry a named individual's Social Security number, home address, every account they hold, every debt they owe and their spouse's details, which means no lender can release them and no vendor can buy them. The result is that the automation built around them is developed against a handful of internal samples and tested against the same handful. Synthetic 413s remove the constraint entirely: coherent balance sheets, in volume, with ground truth, from people who do not exist.
What makes synthetic data useful
Each generated statement is one simulated principal's complete financial position, and the numbers hang together the way a real balance sheet does. Assets total to the asset line, liabilities total to the liability line, and net worth is the difference — median total assets $740,000 against median liabilities $323,000 for a median net worth of $418,000. The detail schedules agree with the summary they roll up into: the real-estate value on page one is the market value carried on the property schedule three pages later, and the mortgage balance matches too. Business entity type is drawn from a realistic small-business distribution rather than uniformly, so 41% of principals are LLC members and only 9% are partners.
Training challenges
The hard problem here is not reading a field, it is reconciling a document with itself. The same value appears in more than one place on different pages — an asset total restated on a later page, a mortgage balance that appears both as a summary liability and as a row on the property schedule — and an extractor that treats each page independently will happily return two different numbers for the same quantity with no signal that anything is wrong. On top of that the schedules are sparsely occupied: the notes-payable table carries three populated rows and two permanently empty ones, the stocks-and-bonds schedule is empty on every document, and two of the three property rows are unused. Empty table rows in a ruled grid are the classic false-positive generator for table extractors.
Generate synthetic SBA Form 413 Personal Financial Statement data
Start with 500 free credits. No credit card required.
Generate NowWho uses this data
SBA 7(a) and 504 lenders automating credit file intake, fintech small-business underwriting platforms, loan-origination-system vendors building document ingestion, and credit-analysis tools that spread a personal balance sheet into a model. Also used by lending-compliance teams who need to test reconciliation logic without exposing a real applicant's finances.
Document complexity profile
148 fields across six pages, four of which carry fields, with the density concentrated on the schedule pages rather than the summary. 71 currency amounts, 61 text, 11 checkbox targets, 3 dates and 2 signature regions, joined by 177 annotation relations. Computed totals roll up from their component lines, and several quantities are restated across pages, so the document imposes cross-page consistency constraints that a page-at-a-time extractor cannot see.
Key stats from our synthetic corpus
Quantitative characteristics of the SBA Form 413 Personal Financial Statement documents our generator produces.
| Metric | Value | Detail |
|---|---|---|
| Median net worth | $418,000 | Median total assets of $740,000 against median total liabilities of $323,000 across 1,000 synthetic principals. The interquartile range on net worth runs $294,000 to $536,000, with a long right tail to $911,000 — a realistic small-business-owner distribution rather than a uniform draw. |
| Business entity mix | 41% LLC | Principals are 40.8% LLC, 25.7% S-corporation, 14.1% C-corporation, 10.4% sole proprietor and 9.0% partnership. Entity type drives which downstream business return the applicant files, so the mix matters if you are training a document-bundle classifier rather than a single-form extractor. |
| Pages carrying fields | 4 of 6 | Field density is deeply uneven — a handful of fields at the top of page one, then 48 and 85 on the two schedule pages, then a short signature block. Two pages carry none. Uniform page-sampling during training over-represents the sparse pages. |
| Spouse block populated | 45% | 45.2% of statements carry a spouse name and Social Security number, matching the married rate in the simulated population. The conditional presence of a whole labelled block is a common source of spurious extractions on the 55% where it is absent. |
| Statements with unpaid taxes | 27% | 27.2% of principals report unpaid taxes, at a median of $5,000. Along with contingent endorser liability at 25.8%, these are the low-prevalence fields where an extractor's precision actually gets tested — most documents have nothing there. |
How this document co-occurs with others
Rates at which identities in our corpus that produce a SBA Form 413 Personal Financial Statement also produce other documents.
| Correlation | Rate | Detail |
|---|---|---|
| Schedule of Liabilities in the same loan file | 64% | 64% of principals who file a 413 also file SBA Form 2202, the Schedule of Liabilities. The two documents restate the same debts in different formats, which makes the pair the natural test of cross-document reconciliation inside a single application. |
| Personal return in the credit file | 100% | Every principal files a personal Form 1040. SBA underwriting cross-checks the income lines on the 413 against the return, so a bundle of the two is the minimum realistic input for an automated credit spread. |
| Partnership return where applicable | 9% | 8.6% of principals are partners in a business that files Form 1065. Low prevalence is the point: entity-specific document routing has to work on the minority case, and a corpus that oversamples it teaches the wrong prior. |
| Wage income backing the salary line | 64% | 64% of principals also draw a W-2 wage, which is what the salary line on the 413 reports. Pairing the statement with the wage statement tests whether a validator can tie a self-reported income figure to third-party evidence. |
Balance-sheet figures and prevalence rates above are corpus-derived: they were computed over all 1,000 identities in a local synthetic corpus generated by SymageDocs' World Simulation Engine at seed 20260421 — every adult identity is eligible for this form, so no sub-sampling applies. Field, page, type and relation counts come from the shipped SBA Form 413 definition in the SymageDocs form library. No real applicant, financial, or credit data was used at any stage.
Frequently asked questions
- What data format do synthetic SBA 413 documents include?
- Each generated identity produces a rendered six-page PDF plus a structured JSON annotation file with bounding boxes, field types, and ground-truth values for all 148 fields, plus 177 label-to-value relations. Fields are typed — 71 currency, 61 text, 11 checkbox, 3 date and 2 signature — so a model can be supervised on type as well as value. COCO, YOLO, FUNSD, and BIO/NER exports come from the same job.
- Do the balance-sheet numbers actually balance?
- Yes. Assets, liabilities and net worth are internally consistent on every document, and the detail schedules roll up to the summary lines they support. That is the property that makes the corpus useful for validation logic rather than only for extraction: you can train a model to flag a statement that does not balance, because every statement in the corpus does.
- Are the signature fields filled in?
- No. The applicant and spouse signature fields and their date lines are annotated but render empty on every generated statement, as does the free-text explanation block in Sections 5 through 8 and the four loan-purpose checkboxes at the top. If your pipeline depends on detecting a signature's presence, treat those regions as negatives rather than as a signature detector's training set.
- How much of the form is actually populated?
- The balance sheet and income sections are fully populated; the schedules are deliberately sparse. Unpaid taxes appear on 27% of statements, contingent endorser liability on 26%, and the spouse block on 45% — the share of simulated principals who are married. Everything unpopulated is still annotated, so blank is a supervised answer rather than missing data.
- Can I use this data commercially?
- Yes. Every name, Social Security number, address, account balance and creditor is synthetic, contains no real personal or financial data, and is licensed for commercial use including model training, benchmarking, and redistribution inside your own products.