Backtesting Data: Choose the Resolution You Need

Backtesting data: Choose observation resolution separately for the signal question and the fill question.

The data-spending discussion compared granularity, corporate actions, delisted instruments and quote histories. Vendor prices and capabilities in the comments were unverified. A more durable lesson is to specify the unresolved research question before deciding which additional observations would answer it.

The source discussion is a starting question, not verified strategy evidence. Accessible comments were reviewed incompletely; some replies remained unavailable. Author results are unverified. The exercise below is original educational reasoning with hypothetical inputs, not a reproduction of those results.

Backtesting data: backtest data resolution: separate signal and fill needs

Write the decision rule and execution assumption on different lines. A signal computed after a daily close may need daily history for its inputs, while an intraday limit-fill claim needs a different kind of evidence. Naming the timeframe alone does not establish that one dataset supports both tasks.

List each required field, its units, timestamp meaning and availability. Distinguish trades from quotes. Record instrument identifiers, session conventions, adjustment policy and universe membership. An attractive file with missing delisted names may answer a current-symbol exercise while remaining insufficient for a historical universe claim.

Worked example: one bar, two possible paths

Imagine a bar with open 100, high 105, low 95 and close 101. A hypothetical position has a stop at 96 and a target at 104. One intrabar path reaches the high first; another reaches the low first. Both fit the same OHLC values but imply different exits. These invented prices demonstrate an ambiguity rather than either actual path.

Write down the ambiguity before buying more history. Finer bars may narrow it without eliminating it. A trade sequence can help with event ordering, while quotes may be needed for spread assumptions. Even detailed historical observations do not automatically reveal the counterfactual queue position of an order that never existed.

Use a small sample to test the requirement

Build a minimal acceptance case using the fields the research question demands. Ask whether the file distinguishes the two possible outcomes in that case. Check that timestamps have a stated timezone and that missing rows are reported. Compare a few rows with an independent record where available, keeping disagreement visible.

Store the downloaded sample unchanged and record a hash and retrieval time. Document what the provider declares and what remains unknown. A manifest can preserve those declarations; it cannot transform them into verified point-in-time provenance. Make that distinction readable in the result report.

Avoid paying for unrelated precision

If the unresolved issue is symbol eligibility, more decimal places in prices will not answer it. If the issue is publication timing, additional trading volume may not help. Match the requested evidence to the failure case, then decide whether the proposed dataset can resolve that case.

A useful purchase criterion is a successful small reconstruction with documented limitations, rather than a vendor ranking or a promise of profitable research. Synthetic paths can test code behavior, but they cannot replace evidence of the historical conditions the strategy claims to exploit. Keep simulated exercises and historical observations clearly labeled throughout.

Reference and first useful step

Last-traded prices can differ from market-order execution prices. Investor.gov: Types of Orders.

Open Advisory Research to explore a local sample report and inspect its assumptions before comparing your own saved evidence. Choose a synthetic sample before importing offline JSON. Keep outputs advisory; this exercise does not place trades. For background, read backtesting fundamentals and paper trading.

Continue the research

Revised Price Data: Audit Backtest Snapshot Changes and Walk-Forward Testing: Log Every Holdout Access develop adjacent checks. Use a small synthetic fixture first; these educational diagnostics do not validate a trading edge or change account settings.