Daily Backtests: Put the Signal Before the Fill

Consider an illustrative rule that buys a stock when its final closing price exceeds a moving average. The simulator computes the condition from today's completed bar and books an entry at today's close. The arithmetic is easy to reproduce, but the decision uses a price that was not yet final when the order needed to be submitted. This is a hypothetical timing example, not a result from a Validraft client strategy.

A review should separate the availability of the last required input, the completion of the signal calculation, order submission, and execution. Keep the exchange calendar, timezone, and session alongside those times. If a daily dataset cannot establish the required ordering, the report should identify the unresolved assumption before interpreting performance.

A closing-auction execution can be a legitimate model when the order is decided using information available before the applicable submission deadline. NYSE's auction documentation sets out different entry and cancellation rules for different order types. Its research on closing orders also distinguishes an order being marketable from its exact execution probability. A generic instruction to fill at the close leaves these mechanics unspecified.

Our practical implication is straightforward: retain the inputs that existed when the order was committed. A rule based on a pre-close snapshot and a rule based on the completed daily bar are different candidates, even if they often select the same stocks. The snapshot may be missing the final price, volume, or cross-sectional ranking that appears in the completed file.

For a rule that genuinely requires the completed daily bar, a next-session execution is a candidate assumption to assess. It still needs an explicit price convention, costs, and treatment of delayed openings or missing prices. Merely shifting a signal column does not prove that the rest of the simulator follows the same clock.

In particular, a position first acquired at tomorrow's open should not earn the preceding close-to-open move. Test the bookkeeping with a small artificial price path containing a large overnight gap: inspect when cash changes, when exposure begins, and which interval contributes to P&L. The expected result can be established by hand without judging whether a real strategy is profitable.

Before comparing results, document which execution schedule represents the intended process and which schedules are sensitivity checks. A pre-close decision, a next-open entry, and an entry later in the next session answer different implementation questions. Choosing the schedule with the best chart after seeing all three turns the comparison into another selection step.

The delivery should show the decision clock, price convention, exposure interval, and evidence supporting each assumption. If the economic conclusion changes materially when the chronology is corrected, preserve that change in the report. A smaller result with an executable sequence is more useful than a precise estimate whose first fill could never have happened as modeled.