Fails to deliver: read balances without summing daily stocks
The SEC fails-to-deliver file reports a settlement-date outstanding balance, not the day’s newly failed deliveries. Adding successive balances can count the same unresolved shares repeatedly. A useful reconstruction preserves the security identifier, reporting system, date and missing-record rules before discussing a trend.
This guide uses the SEC’s Fails-to-Deliver Data explanation as read on October 25, 2026. It concerns aggregate net balances in NSCC’s Continuous Net Settlement system, not every unsettled obligation everywhere. All numerical rows below are hypothetical, not observations about a named security or market participant.
Define the data object before importing it
The SEC’s “What You Should Know About the Data” describes aggregation over NSCC members, the settlement-date balance and important interpretation limits. The public record does not identify the individual failing participant or beneficial investor.
The layout contains settlement date, CUSIP, symbol, description, quantity and a price field. Import the pipe-delimited file using explicit types. Preserve the nine-character CUSIP as an identifier, including leading zeros, rather than allowing spreadsheet number conversion to alter it.
Use CUSIP and date as key inputs to the security reconstruction, with a documented mapping when identifiers change. A ticker alone can change or be reused. A similar issuer description does not establish that two rows represent the same security and units.
Keep source file, download date and any later replacement version. Reading a formatted chart without its underlying date definition makes it easy to confuse settlement dates with trade dates or publication dates.
Work three outstanding balances
Assume three consecutive valid settlement dates S1, S2 and S3 for the same identified security, with no unit or identifier changes. Outstanding balances are 1,000, 1,200 and 900 shares.
| Settlement date | Outstanding balance | Change from previous date | What is observed |
|---|---|---|---|
| S1 | 1,000 | Unknown without earlier balance | Stock at S1 |
| S2 | 1,200 | +200 | Net increase |
| S3 | 900 | −300 | Net decrease |
The sum 1,000 + 1,200 + 900 = 3,100 is not the number of newly failed shares or distinct shares over the period. At most, a deliberately defined sum of daily stocks can describe a stock-time measure across the specified dates. Its unit and interpretation must be stated; it is not a unique-share count.
The sequence rises 20% from S1 to S2, falls 25% from S2 to S3, and ends 10% below S1. These percentages describe changes in outstanding balances. They are not the security’s price returns or the percentage of its shares sold short.
Show why new failures and settlements are not identified
For the simplified balance identity, let N be new failures and C be failures settled during the interval:
Closing balance = opening balance + N − C.
The S2 increase of 200 is compatible with N = 200 and C = 0, N = 500 and C = 300, or N = 1,200 and C = 1,000. Each gives 1,200 from an opening 1,000.
For S3, the fall of 300 is compatible with N = 0 and C = 300 or N = 900 and C = 1,200. Both end at 900 from an opening 1,200. The second path could replace the entire opening balance with new failures, while the first could leave 900 older shares outstanding.
These are illustrative paths, not reconstructed transactions. Two stock observations supply a net difference, not the two separate flow amounts or their owners. Additional private settlement information would be needed to distinguish them.
The SEC explanation specifically cautions that the age and underlying source of outstanding failures cannot be determined from these aggregate numbers. A run of positive balances does not establish that the same shares or participant remained failed throughout.
Apply the correct missing-record rule
The SEC’s historical threshold discussion distinguishes dates before September 16, 2008 from dates on or after it. Earlier records included only securities meeting a 10,000-share balance threshold; later data includes securities with outstanding balances and omits zero balances.
Therefore, a missing pre-change row can coexist with a nonzero balance below 10,000. It cannot universally be coded as zero. The later zero interpretation also depends on having the correct, complete file for a valid settlement date and correctly matching the security.
A file not yet published, a failed download, a non-settlement date or an identifier mismatch is not the same thing as a documented missing security row in the relevant file. Keep “unavailable,” “not applicable” and the eligible zero state separate.
For example, if a complete post-change file for a valid settlement date contains no row for the verified CUSIP, the documented rule can support a zero balance within this dataset. If the entire half-month file is absent, it cannot.
Respect publication timing and the price field
The page describes first-half files as available at month-end and second-half files around the fifteenth of the following month, while explicitly not guaranteeing a particular posting date. Historical data before July 2009 was packaged differently. Preserve these distinctions when building an information-available-at-the-time analysis.
A backtest should use when the file was actually available, not pretend the settlement-date value was known that same day. Record a publication or retrieval timestamp and retain the version used.
The SEC also explains that the price field concerns the previous day’s closing price when available under its stated condition; a dot can denote unavailable price. A dot is not a zero price. The page cautions about price accuracy and comparability to other sources.
Multiplying 900 shares by an assumed price of 10 gives a proxy amount of 9,000. That is not a measured loss, settlement invoice, current market value of a particular owner’s position or proof of damages. Label the price date and the proxy’s limitations.
Separate failures from short-sale conclusions
The SEC expressly notes that failures can arise from both long and short sales for multiple reasons and are not by themselves evidence of abusive or naked short selling.
A balance of 900 therefore does not identify 900 newly opened short shares, a particular fund’s position or its motive. Combining the file with a ticker’s short-volume statistic does not supply the missing identity link.
An issuer’s shares outstanding can provide a separate scale denominator only if security class, dates and units align. Even a correctly labeled balance-to-shares ratio would describe this dataset’s outstanding failures relative to that denominator, not the fraction of the company owned by a failing party.
Execute the reconstruction checks
| Check | Input to retain | Error prevented |
|---|---|---|
| Identity | CUSIP, symbol and mapping | Different securities stitched together |
| Date | Settlement date and calendar | Trade or publication date substituted |
| Version | File period and retrieval record | Later revision used unknowingly |
| Missingness | Regime, complete file and valid date | Every absence called zero |
| Measure | Outstanding balance and net change | Stock sum called new failures |
| Inference | Unknown inflows, settlements and age | Unidentified flows invented |
| Price | Field meaning and missing marker | Dot or prior close treated as current value |
If rows conflict across vendors, compare the original SEC file, extraction types, identifier mapping and later replacements before choosing a preferred chart. Do not fill a gap by copying the previous balance unless it is explicitly a modeling assumption.
Update the reconstruction when the SEC replaces a file, an identifier mapping is corrected, a valid zero state is established or an error in the import is found. Preserve the earlier version and explain which descriptive conclusion changed. The result should report balances and limits, not manufacture an ownership or misconduct narrative.
Test import behavior with a controlled missing row
Before charting a long history, create a small test import containing a valid zero-state absence, an unavailable file and a price dot. Confirm that these become three different statuses. A charting library that silently carries the prior balance forward can manufacture persistence even when the import table is correct.
For a split or identifier change, retain the original quantities and make any adjusted series a separate output with its conversion factor. A jump caused by comparing unlike share units is not evidence of new settlement failures. Re-run both the balance differences and the missingness classifications after correcting the mapping.
FAQ
Are 1,000, 1,200 and 900 equal to 3,100 new failed shares?
No. They are successive outstanding balances and can include overlapping unresolved shares.
Does a persistent balance identify the age of its shares?
No. New and settled failures can replace one another within the same aggregate stock.