Intraday Versus Daily Backtest Granularity

Before you can backtest anything you have to decide how finely to slice time. A daily backtest looks at one price per instrument per day — typically the close — and rebalances at most once a day. An intraday backtest steps through minutes, seconds, or individual ticks, letting a strategy see and react to price moves within the trading session. The choice sounds like a technical detail, but it determines which strategies you're even capable of testing honestly: a signal that decays over two hours simply cannot be evaluated on daily bars, no matter how clever the rest of the backtest is.

The tradeoff

Daily granularity is cheap to build, cheap to store, and fast to run — a decade of daily bars for a few thousand names fits comfortably in memory, and a full backtest takes seconds. But it hides everything that happens between opens and closes: a strategy that would have been stopped out at 11am and re-entered at 2pm just sees one number for the whole day. Intraday granularity captures that path, which matters enormously for anything with a stop-loss, an intraday signal, or execution costs that depend on when exactly you trade. The cost is data volume (tick data for liquid names runs into billions of rows a year) and a much harder simulation problem, since you now need to model the order in which events happen within a bar, not just the bar's summary statistics.

Worked example: a mean-reversion signal that only lives for hours

Suppose a signal says "buy when a stock drops more than 2% in the first hour of trading, and it tends to recover half that move by 2pm." A daily backtest only ever sees the open and the close. If the stock opened down 2%, dropped further to −4% by 11am, and finished the day at −1%, the daily bar records a −1% day — the signal's entire mechanism (drop, then partial recovery within the session) is invisible. An intraday backtest sampling hourly bars would see the −4% trough at 11am, the entry, and the recovery to −1% by the close, letting you measure the actual edge the strategy is trying to capture. Running this strategy on daily data wouldn't just be less precise — it would silently test a completely different, nonexistent strategy.

What this means in practice

The right granularity is set by the strategy's holding period, not by convenience: a rule of thumb is that your bar size should be several times shorter than the shortest thing the signal needs to see. A daily-holding-period portfolio strategy rarely needs tick data, and building tick infrastructure for it is wasted effort; a stat-arb strategy that unwinds within the day cannot be trusted on daily bars at all. Many research teams run daily backtests first to filter ideas cheaply, then re-test survivors on intraday data before committing capital — daily granularity is a fast, coarse filter, not a substitute for the finer test a short-horizon strategy actually needs.