Quantitative Research

Test 2 — Is the significance inflated by non-independent events?

Several companies often file on the same calendar day, sharing the same market conditions — treating each as an independent observation can overstate a t-statistic. Re-aggregating one observation per calendar day, rather than per filing, removes that inflation.

The trade journal is barely profitable at a token $200 stake per trade. Before concluding the effect is too small to matter, three ways it could be a statistical illusion were tested directly against the raw data — a crash-clustering artifact, inflated significance from correlated same-day events, and pure bid-ask bounce in illiquid names. All three were built to make the finding disappear. None of them did.

Barely moves. Collapsing tens of thousands of events into a few thousand independent trading days leaves the t-statistic almost unchanged — the original count was not doing the work.