Hit Rate is the simplest statistic on a research screen. Out of every observation that met your condition, what fraction went the way you wanted?
hit rate = positive outcomes / total observations
It appears in research contexts rather than trading ones. An event study reports the hit rate of forward returns after a signal fired. A factor scan reports the hit rate of months in which the long-short spread was positive. Win Rate is the closely related trading metric, computed over realised round trips with actual costs deducted, and the two are not interchangeable.
Why the number alone decides nothing
Two signals can have identical hit rates and opposite outcomes, because the metric throws away magnitude. Illustrative arithmetic on a five-day forward window:
| Signal | Hit rate | Average gain | Average loss | Expected outcome |
|---|---|---|---|---|
| A | 68% | 1.1% | 3.4% | 0.68 x 1.1 - 0.32 x 3.4 = -0.34% |
| B | 41% | 4.6% | 1.9% | 0.41 x 4.6 - 0.59 x 1.9 = +0.76% |
Signal A is right two times in three and loses money. Read the hit rate and the payoff together or do not read the hit rate.
The comparison that is always missing
A hit rate is a conditional figure and it means nothing without the unconditional one. If 62% of five-day windows after RELIANCE earnings were positive, the follow-up question is what share of all five-day windows in the same period were positive. In a sample that spans a long bull run, the unconditional figure might be 56%, and your apparent edge is six percentage points rather than sixty-two. That comparison is the base rate, and it is the first thing to look up when a hit rate looks impressive.
Caveats
Sample size governs how much you can read into the figure. Across 24 quarters of one company’s earnings you have 24 observations, and a 62% hit rate carries a standard error near 10 percentage points. It is consistent with anything from 42% to 82%.
Overlapping windows inflate the count without adding information. Twenty-day forward returns sampled daily give you roughly one independent observation per month, not one per day, so a hit rate computed over 2,000 daily rows rests on about 95 real data points.
High hit rates are behaviourally expensive. They feel like competence and they encourage cutting winners early to preserve the streak, which is the disposition effect operating directly on a research metric. The signals worth holding often have the least comfortable hit rates.