Batting average counts how often your portfolio finished a period ahead of its benchmark. Borrowed from baseball, where it answers the same question: not how far the ball travelled, but how often the bat connected.
Lemma Analysis computes it on monthly returns against SPY, over every month where both series have a figure. A tie is not a win, and a month the benchmark could not be measured in counts for neither side — it changes neither the numerator nor the denominator.
Over 40 months, a portfolio finished ahead of SPY in 26 of them: a batting average of 65%. Read as “in two months out of three, holding this portfolio beat holding the index.”
The number that makes it worth having: two portfolios can post the same cumulative return and the same alpha while one beat the index in 26 months of 40 and the other in a single explosive month. The curve cannot tell them apart. This can.
Read it against 50%, which is what a coin flip against the index looks like:
| Batting average | Reading |
|---|---|
| < 45% | usually behind the index; any outperformance rests on a handful of months |
| 45–55% | indistinguishable from the index on frequency |
| 55–65% | a genuine, repeated edge |
| > 65% | rare over long samples — check the losing months are not enormous |
Always pair it with a magnitude measure. A high batting average with negative alpha means many small wins funding a few catastrophic losses — the classic shape of a strategy that looks reliable right up until it does not.
Batting average matters most when you are deciding whether an edge is repeatable. Cumulative return and alpha collapse a whole history into one number, and one lucky quarter can carry both. Frequency cannot be carried by one quarter.
It is also the honest counterweight to hindsight: a portfolio that beat the market by picking one winner feels, in memory, like a portfolio that beats the market.
Completely blind to magnitude — beating the index by 0.01% counts exactly as much as beating it by 8%. Sensitive to the period length: the same portfolio measured daily, monthly and yearly produces three different batting averages, and only figures on the same frequency are comparable.
On short histories it is noisy in a way that reads as precision: twelve months quantizes it into steps of 8.3 pp, so 58% and 67% are one month apart.