What does the Sharpe ratio actually measure?
The Sharpe ratio is a strategy's excess return divided by the standard deviation of its returns — how much you were paid for the amount of bouncing around you had to tolerate.
The appeal is that it makes two very different strategies comparable. A rule returning 40% a year with wild swings and one returning 12% smoothly are hard to rank by return alone; dividing by volatility gives a common unit.
The limitation sits in that same divisor. Standard deviation treats an upside surprise exactly like a downside one, so a strategy that occasionally makes an enormous gain is penalised for it. That is why Sortino, which divides by downside deviation only, exists — and why quoting Sharpe alone is a choice worth being aware of.
It is also period-dependent in a way people forget. A Sharpe computed on daily returns and one computed on monthly returns are not the same number, and annualising involves multiplying by the square root of the number of periods — which quietly assumes the returns are independent. For a strategy with autocorrelated returns, that assumption inflates the annualised figure.
What Sharpe should you expect at retail scale?
Out of sample and after costs, a Sharpe around 1 is a genuinely good retail result, and the range from roughly 0.5 to 1.5 covers most strategies that survive honest validation.
Context matters more than the number. Large systematic funds operate at Sharpes that would look unremarkable on a retail forum, because they are running enormous capital where capacity, not edge, is the binding constraint. A retail trader has the opposite problem and the opposite freedom.
The figures that should trigger scepticism rather than excitement are the high ones. A backtested Sharpe of 4 usually means one of a handful of things: the test is in-sample, costs are missing or understated, the sample is too short, or the result is the best of many variants and the search was never counted.
None of those require dishonesty. They are the default behaviour of most testing setups, which is exactly why the high number is more common than the real edge it would represent.
How do you inflate a Sharpe ratio without meaning to?
Four mechanisms account for most inflated Sharpes, and all four are things a careful person can do by accident.
Reporting the in-sample figure. Parameters chosen on the same data that produced the number make the number a description of that data. This is the largest single effect and the easiest to fix.
Understating costs. Sharpe is computed on net returns, so every pip of unmodelled spread comes straight off the numerator. For a short-horizon strategy, realistic costs can halve it.
Too few periods. Sharpe estimated from a handful of months carries an enormous confidence interval. The estimate is not wrong so much as unstable — run the same strategy on a different year and it moves substantially.
Selection. The best Sharpe among fifty tested variants is a maximum, not a sample. There are deflated Sharpe methods that adjust for exactly this, and they require you to have counted the trials — which is the part nobody does by default.
Is a high Sharpe always better?
Not always, and the exception matters. A strategy can achieve a very high Sharpe by taking a small, steady profit while carrying a rare and enormous loss — the profile of selling options or of any short-volatility position.
Those strategies look magnificent right up until the event they are exposed to happens, and the Sharpe computed on the quiet period is not merely optimistic; it is measuring the wrong thing entirely, because the risk being taken does not show up as volatility until it shows up as a loss.
The practical defence is to look at the return distribution rather than only its summary. Skew, the worst single day, and the shape of the drawdowns tell you what kind of strategy you have. A high Sharpe with a very long left tail is a different animal from a high Sharpe with symmetric returns.
It is also worth asking how the strategy behaves in the periods that broke similar ideas. A rule that has never met a volatility spike has not been tested against the thing most likely to end it.
What should you report alongside Sharpe?
Report the sample size, the out-of-sample split, the costs assumed, and the number of variants tried — because Sharpe without those four is a number that cannot be evaluated by anyone including you.
Sample size sets the confidence interval. A Sharpe of 1.2 over 40 trades and one over 2,000 trades are not comparable claims, and quoting the figure without the count invites the reader to treat them as if they were.
The out-of-sample split is the difference between a measurement and a description. Where the split falls, and whether it was chosen before or after seeing results, decides how much the number means.
The variant count is the one almost nobody reports, and it is what turns a Sharpe into a maximum. If you tried forty parameter sets, say so; there are standard adjustments once the number is known, and none of them are available if it is not.