Alternative data
Alternative Data: More Features, More Ways to Leak the Future
Novel datasets can add information, but their publication history is often harder to reconstruct than their values.
Working definition
Alternative-data leakage occurs when a backtest uses a value, mapping, revision, or coverage decision that was not actually available at the simulated decision time.
01
The timestamp you received is not always the timestamp you need
A dataset may contain an event time, ingestion time, vendor publication time, revision time, and the time the file became available to the researcher. Backtests often keep only one of them. If the chosen field is earlier than the information could have been consumed, the model is given an artificial head start.
The safe approach is to define an explicit availability timestamp and apply a conservative lag. When the vendor cannot reconstruct historical delivery, that uncertainty belongs in the limitations rather than being replaced with a precise but unsupported assumption.
02
Entity mapping can leak too
Company identifiers, locations, product categories, and event labels are revised as businesses change. A present-day mapping table can quietly connect old observations using knowledge learned later. Point-in-time joins need versioned identifiers and effective dates, not only clean keys.
- Retain raw vendor delivery files and ingestion timestamps.
- Version entity mappings and classification rules.
- Measure historical coverage instead of filling it with current coverage.
- Test longer publication lags as an adverse scenario.
03
Novel does not mean independent
Alternative features may be proxies for familiar exposures such as size, geography, seasonality, or sector momentum. Validation should compare the signal with simpler public baselines and report incremental value after neutralization. Otherwise expensive data can reproduce an existing factor with a more complicated provenance problem.
Practical takeaways
- Model availability time, not only event time.
- Make historical entity mapping point-in-time.
- Stress publication lags and incomplete coverage.
- Compare novel features with simple factor baselines.
Want an independent read?
Test the claim, the data, and the implementation together.
Validraft scopes the hypothesis, checks feasibility, and delivers a descriptive validation report with visible evidence and limitations. Research and simulation only; never investment advice.