What the Sharpe ratio hides

The Sharpe ratio is the average excess return divided by the variability of that excess return. William F. Sharpe described the historic version as the average differential return per unit of variability of that differential return, in The Sharpe Ratio (Journal of Portfolio Management, Fall 1994). A higher ratio means more average excess return for each unit of that variability. It does not describe the order of gains and losses.

What goes into the number

Sharpe's historic calculation starts from a differential return: the result minus a benchmark. When the benchmark is a riskless rate, the differential return is excess return. The ex post ratio is the average of those differentials divided by their standard deviation.

Sharpe ratio = average excess return / standard deviation of excess return

Sharpe notes two limits that matter before any trading screen. Mean and variance ignore other features of the outcome distribution, so two paths with the same average and the same variability can still differ in ways an investor cares about. The ratio also depends on the length of the period you measure. Annualizing short periods is common so that a monthly series and a weekly series can be compared, and that step assumes a simple relationship across time that real returns may not have. The ratio also says nothing about correlation with your other positions.

Tradee computes one specific version for backtests and for execution analytics. Bar-to-bar returns come from the equity series. The risk-free rate is zero, so excess return is the bar return itself. Variability is the sample standard deviation. The result is annualized by multiplying by the square root of the number of bars per year. A zero standard deviation produces no ratio. Changing that convention would change both surfaces together, which is why it lives in one function.

Same returns, two drawdowns

The figures in this article's images are a worked example, not a recorded backtest. Start at $10,000. The sample is twelve monthly returns: ten gains of 4% and two losses of 8%. The risk-free rate in the example is zero, matching Tradee's setting. Annualizing uses 12 periods per year because the step is one month.

The average return is 2% per month. The sample standard deviation is about 0.0467. Dividing and annualizing gives a Sharpe ratio of 1.48 when rounded to two decimal places. Every input above is shared by both orders, so the ratio matches.

The second path is down 15.36% before it recovers. The first path's deepest fall is 8.0%, later in the year, from a higher peak. A ranking that sorts only on the Sharpe ratio ties them. A ranking that also sorts on maximum drawdown does not. Why win rate is not enough makes the same point about hit rate: one summary can hide the payoff shape. Here the hidden shape is the path.

You can rebuild both endings from the return list. Compound $10,000 by each monthly return in order. The drawdown at each step is the drop from the highest equity so far, divided by that high. The spread path's 8.0% drop is the -8% month that follows the peak near $11,698.59. The clustered path's 15.36% drop is the two opening losses, and it needs no later peak to explain it.

What else the ratio leaves out

A short sample moves the ratio a lot when one month changes. Twelve points are enough to show the order effect. They are not enough to treat 1.48 as a stable property of a strategy. Sharpe also points out that the historic ratio is tied to the t-statistic of the mean: with the same number of observations, a higher ratio is a larger t-statistic, not a different kind of proof.

Costs, funding, and skipped trades change the return series before the ratio is computed. Two bots with the same signal and different size rules can post different ratios because the equity path changed. Size the loss before the trade is about that input. The ratio will not tell you which sizing rule you used.

Tradee's zero risk-free rate is a convention, not a claim that cash returns nothing. If you compare a Tradee figure with a published fund ratio that subtracts a Treasury yield, the benchmarks differ. Compare the inputs, then the numbers.

How to read one on a result

1. Record the bar size and the window. The annualization factor has to match the bars.
2. Read maximum drawdown and the length of that drawdown beside the ratio.
3. Read trade count. A smooth equity line with very few trades can look calm for mechanical reasons.
4. Check whether the worst losses are clustered. The example is the template: identical inputs, different pain.
5. Keep the strategy version fixed. A ratio from the best of many trials is a research result, not a forecast.

On the Backtests hub, Sharpe is one of the filters next to profit and drawdown. Use more than one. An execution result uses the same Sharpe calculation on its equity series. It is still a summary of modeled or simulated equity, not a promise about the next window.

FAQ about Sharpe ratio

Is a higher Sharpe ratio always better?

Inside one convention, a higher ratio means more average excess return per unit of variability. It can still hide a deep clustered loss, a tiny sample, or a benchmark that differs from the ratio you are comparing it with. Read it with the path.

Why can two equal ratios have different drawdowns?

The average and the standard deviation do not include order. Rearranging the same returns keeps the Sharpe ratio and can change the worst peak-to-trough fall. The twelve-month example is that rearrangement.

Methodology and sources

The definition follows Sharpe's 1994 historic ratio: average differential return divided by the standard deviation of that differential return. The worked series uses a zero risk-free rate and annualizes monthly returns with the square root of 12, which is the same structure as Tradee's bar-return calculation. Drawdowns are peak equity minus later equity, divided by the peak. Panel figures are -8.0%, -15.4%, and 1.48. The unrounded clustered drawdown is 15.36%. This article is educational and does not provide personalized investment or trading advice.