Data engineering
Bars, ticks, and choosing data resolution
Match the data resolution to the strategy's decisions so the backtest does not invent precision the live system cannot use.
Resolution changes the question
A bar dataset can answer questions about completed intervals, but it may not identify the intrabar sequence of a stop and target. Tick or quote data can provide more detail, yet it also introduces larger storage, cleaning, and modelling requirements.
Use the lowest resolution needed for the decision you are testing, then validate important results at a more realistic resolution when possible.
Intrabar ambiguity matters
If a candle touches both a stop and a target, the order of those touches matters. A backtest must use a defined rule or richer data rather than silently choosing the favourable path.
- Document the intrabar fill convention.
- Use bid and ask when the strategy is spread-sensitive.
- Check whether tick history is complete for the entire sample.
- Avoid mixing resolutions without a clear reason.
More data is not automatically better
High-resolution data can contain feed-specific noise, duplicated quotes, and changing liquidity conditions. It increases fidelity only when the cleaning and execution model are also credible.
The goal is not maximum data volume. The goal is a defensible connection between the test input and the live decision.
Discuss this article
Ask a setup question, share a backtest, or compare notes with other algorithmic traders. Use a display name; your email remains private.
Create an account or sign in above to join the conversation.
Loading discussion…