A t-statistic answers one question: "could this result just be luck?" It measures how large an average result is relative to how noisy the individual results were. On our track record it tests our signals against the assumption that they have no edge at all.
Suppose a set of trading signals finished, on average, slightly profitable. Is that a real edge, or did we just get lucky? Flip a fair coin twenty times and you will often see twelve heads. Twelve out of twenty is "above average" and means absolutely nothing. Any track record faces the same problem: some positive average results are real, and some are the coin landing well.
The t-statistic is the standard tool for telling those apart. It does not ask "did we make money?" — it asks "is this result big enough, and consistent enough, across enough trades, that luck is an uncomfortable explanation?"
The t-statistic on our track record page is computed as: mean(R) divided by (standard deviation of R, divided by the square root of n). Here R is the R-multiple of each closed signal — its profit or loss measured in units of the risk taken — and n is the number of closed signals.
Read the formula as a ratio of signal to noise. The top is the size of the average result. The bottom is how much the individual results scattered around that average, shrunk by the square root of how many trades you have. So the t-statistic grows when results are larger, when they are more consistent, and when there are more of them. All three make luck a worse explanation.
That square root matters, and it is the part people underestimate. To double your t-statistic without improving the strategy at all, you need four times as many trades. Statistical confidence is expensive to buy with sample size alone.
H0 (pronounced "H-nought") is the null hypothesis — the boring explanation you are trying to rule out. Here it is: "these signals have no edge; the average R is really zero, and anything above zero is noise." The t-statistic measures how awkward the observed results are for that boring explanation.
Rough rules of thumb, using the conventions statisticians use in practice: a t-statistic near 0 means the results are entirely consistent with having no edge. Around 2 is the traditional threshold where a result is called "statistically significant" — loosely, less than about a 1-in-20 chance of seeing something this strong if there were truly no edge. Above 3 is a genuinely strong signal. Below 2, the honest reading is: not proven.
This is why our track record says so bluntly, when the sample is small, that the edge is statistically unproven. That is not false modesty — it is what the number says.
Example (illustrative — invented numbers chosen to show the mechanics, not our results). Say 25 closed signals averaged +0.20R, with a standard deviation of 1.0R across the individual trades. Then t = 0.20 / (1.0 / √25) = 0.20 / 0.20 = 1.0. That is a positive average and a t-statistic that proves nothing: entirely consistent with luck.
Now keep the exact same average and the same consistency, but run 100 signals instead of 25: t = 0.20 / (1.0 / √100) = 0.20 / 0.10 = 2.0. The strategy did not improve by one basis point. The evidence did. This is the single most useful intuition on this page — and it is also why a strategy showing you a handful of winning trades has told you nothing at all.
The big one: a t-statistic computed on a backtest is much weaker evidence than the same number computed on live, forward results. If you try a hundred strategy variants on history and report the best one, a t-statistic of 2 is roughly what you would expect to find by chance — that is overfitting, and the t-statistic cannot see it. It only scores the sample you hand it; it has no idea how many samples you discarded first.
Second: significance is not size. With enough trades, a genuinely real but economically useless edge — one that fees and slippage eat entirely — will produce an impressive t-statistic. "Statistically significant" and "worth trading" are different claims, and the t-statistic only speaks to the first.
Third: it assumes the trades are independent and the world is not changing. A batch of signals that all fired in the same week, in the same market conditions, on correlated stocks, is not really the number of independent bets it appears to be — and the formula will happily count them as though it were.
A measure of how hard it is to explain a result by luck. It compares the average outcome to how much the individual outcomes scattered, adjusted for how many of them there were.
Roughly: below 2 is unproven, around 2 is the traditional "statistically significant" threshold, and above 3 is strong. Below 2 the honest reading is that the result is consistent with having no edge at all.
H0 is the null hypothesis — the assumption being tested, which here is "these signals have no edge and the average result is really zero." The t-statistic measures how poorly that assumption explains what actually happened.
No. It says the past results are unlikely to be pure luck. It says nothing about the future, and it cannot detect overfitting — if many strategy variants were tried and only the best reported, a high t-statistic is exactly what chance would produce.
Browse all S&P 500 tickers to see this metric applied to individual companies.
Educational research only — not investment advice.