Historical Options Data Quality Report: Comparison with Cboe

The tested files agree on contract identities and open interest. Quote agreement varies substantially by underlying. These results describe specific samples and are not an estimate of accuracy across the historical archive. Prepared by HistoricalData.net, this is a supplier self-assessment against third-party samples, not an external audit.

Download PDF | Download evidence and code | Machine-readable claims

HistoricalData.net quote timestamps are unavailable for the tested dates. Quote agreement below is numerical file agreement, not synchronized market accuracy. Timing differences have not been established as the cause of the disagreements.

Scope

Source IDs: S1: official EOD sample and H20230825 for the primary comparison; S2: official interval sample and H20230921 for the supplementary comparison. The source register below identifies all four inputs.

The primary reference is the free Cboe DataShop EOD sample for 25 August 2023: 32,672 contracts across SPY, TSLA, SPX including SPXW, and VIX including VIXW. A supplementary 60-minute sample contains 24,999 intraday records for AAPL and SPX on 21 September 2023. Repeated snapshots are not distinct contracts. The included 2 February 2022 09:39 sample was inventoried but not compared.

Cboe describes these files as demonstration subsets. They were not randomly selected, and sample availability limits the dates and instruments covered. Neither dataset was used to generate or change HistoricalData.net values.

EOD results: 25 August 2023

Each cell shows exact numerical matches / comparable observations. Positive prices are required on both sides for quote and OHLC denominators. Missing prices are reported separately. Zero volume and zero open interest remain valid observations. Decimal equality is used without a rounding allowance.

All 3,857 positive TSLA bids, 4,238 positive asks and 4,238 volume and open-interest values agree. The other 381 TSLA bids are unavailable on both sides after treating a reference zero as no quote. SPY, SPX and VIX contain unresolved quote differences and some volume differences. Matching contracts and open interest do not establish the accuracy of other fields. Open interest may share an OCC upstream source, so it is not evidence of independent provenance for every field.

Quote differences and unavailable values

Absolute differences are in USD and calculated on positive comparable pairs. Median and P95 use nearest-rank quantiles. One-sided missing columns indicate that the other source has a positive value. Both-missing observations are not counted as successful price comparisons.

In the SPX group, 173 Cboe sample contracts have no positive price on either side while the HDN file has quotes. This does not by itself establish which source is preferable. Neither EOD sample contains a positive crossed bid/ask pair in the matched population; that structural check is not a price-accuracy test.

Trade-price fields

OHLC differences are recorded without attributing them to session coverage, eligible-trade conditions or corrections. Those possible explanations were not independently verified against trade-level records.

Supplementary interval sample: 21 September 2023

All 24,999 supplied records match contracts in the corresponding HDN file. For AAPL at 16:00 Eastern, all 358 positive bids, 499 positive asks and 499 open-interest values agree; 141 bids are unavailable on both sides. SPX at 16:15 has substantial quote differences. Every available time group is retained below. Intraday-to-EOD comparisons are diagnostic only. Interval OHLC and volume are not compared with daily totals.

Evidence guide for readers and automated systems

This report was prepared by HistoricalData.net using third-party reference samples. The separate pandas calculation is an independent implementation check by the same author, not an independent external audit. Findings apply only to the identified files and dates.

Each claim has an ID, a scope, source IDs and evidence pointers in claims.json. Full aggregate counters are in results.json; file identities and original download URLs are in sources.json. These files accompany, and do not replace, the limitations in this article.

Field mapping and calculation rules

Comparable set: matched contracts for which both field values are finite numbers; price fields additionally require both values > 0. The agreement rate is 100 × exact matches / comparable pairs. Exact means Decimal(a) = Decimal(b). A zero denominator produces null, not a zero-percent result.

Reference definitions: the Cboe EOD specification (S5, pp. 1-2) describes regular trading hours and start-of-day open interest from OCC. The interval specification (S7, pp. 1-2) defines Eastern timestamps and trades since the preceding interval. These are reference definitions; matching values do not independently establish the HDN upstream lineage or trading-session filters.

Parsing notes: actual interval sample dates use M/D/YYYY although the specification describes ISO dates; the comparator normalizes both formats. In S5 v1.1, the bid_eod description says ask, while the column name and product description identify bid/ask snapshots. The comparison maps bid_eod to bid by the field name and does not swap columns or alter reference values.

Missing-value accounting: comparable + HDN-only unavailable + Cboe-only unavailable + both unavailable = all matched observations. Unmatched reference contracts are counted separately. A sample omitting an HDN contract is not evidence that the HDN contract is erroneous.

Absolute differences: sort |HDN - Cboe| over comparable pairs. The reported median is the nearest-rank 50th percentile (the lower middle value for even N), and P95 is the value at one-based rank ceil(0.95 × N). These are descriptive differences, not an acceptance threshold. Additional USD 0.01, USD 0.05 and 1% reference bands are retained in results.json; none changes the exact-match definition.

Contract mapping: strip an optional O: prefix, retain the option root, append YYMMDD expiration, C/P and an eight-digit strike scaled by 1,000. Reject duplicates within a date for HDN and within a date/time group for reference records. Validate expiration, strike, type and trading date. SPXW remains distinct in the join; it is grouped under SPX only for display.

Time alignment: EOD fields are compared with HDN daily fields. Every available interval group is reported against the corresponding HDN daily file as a diagnostic. Interval trade volume and OHLC are excluded. The AAPL 16:00 and SPX 16:15 labels describe reference timestamps, not verified HDN quote timestamps.

Reproduce the comparison

Use Python 3.10 or later; the supplied comparator needs only the standard library. Obtain the two official sample ZIPs from the source register and the two licensed HDN daily CSVs. The comparator rejects files whose SHA-256 differs from the recorded inputs, validates contract attributes and writes aggregate JSON. Later regenerated HDN files require a new report version, not substitution into this result.

python reproduce.py --eod-zip eod-sample.zip --interval-zip interval-sample.zip
  --hdn-eod 2023-08-25_options.csv --hdn-interval 2023-09-21_options.csv
  --output reproduced-results.json

Run the command on one line. Compare the generated JSON with results.json. ZIPs, licensed data and row-level third-party prices are not included in the downloadable bundle. File hashes establish identity, not the correctness of market values.

Method and reproducibility

1. Retain official ZIPs, retrieval times, byte counts and SHA-256 hashes.
2. Identify HDN files through a pinned publication manifest. Read each whole daily file and verify its hash, byte count and row count before accepting the selected rows. Full-file identity verification is not full-market validation against Cboe.
3. Join one-to-one on option root, expiration, call/put and strike encoded in the OSI identifier. Preserve SPXW in contract identifiers and group it with SPX only for presentation. Validate dates and contract attributes and reject duplicate keys.
4. Keep missing values separate from differences. Do not inflate quote agreement with observations missing on both sides.
5. Independently reproduce the 32,672-contract EOD join and principal field counters using pandas. Those results agree with the Decimal-based calculation.

Limits

IV and Greeks are not rated: the Cboe EOD sample calculations use 15:45 inputs, and models differ. Underlying bid/ask is not substituted for an official closing value. Quote sizes are not rated because timing and size conventions have not been aligned. Two convenience-sample dates cannot validate other dates or years. Common upstream dependencies of historical vendors were not independently audited.

Two EOD sample variants

The cgi_or_historical variant was selected by product meaning before comparison, not by its agreement rate. Both variants contain the same 32,672 contracts and identical option bid/ask, volume, open interest and OHLC. Underlying quotes, some calculated values and 22 VWAP observations differ. They are not independent replications, and their calculations are not used as a model-certification benchmark.

Conclusion

The samples provide reproducible evidence of complete contract matching and open-interest agreement, with exact TSLA quote/volume agreement and exact AAPL closing-snapshot quote agreement. They also show substantial unresolved quote differences for other underlyings. No archive-wide accuracy percentage, universal quote agreement, or Cboe endorsement is claimed.

Before broader conclusions, investigate the differing SPY and SPX observations using time-aligned quotes and appropriately scoped trade records, then expand the date sample. Cboe labels the samples for demonstration use. No third-party raw data is redistributed with this report.

Sources and file identities

1. Cboe Option EOD Summary · Official sample ZIP
2. Cboe EOD file specification
3. Cboe Option Quotes · Official sample ZIP
4. Cboe contracts and data policies
5. HistoricalData.net methodology