Semtum

What does my backtest have to beat?

Work out the Sharpe ratio a strategy with no skill is expected to reach after the number of versions you tried. Free calculator, no signup, works offline.

Tool 1 of 5

Drag the slider to however many versions you tried. Watch the bar move.

Watch: what goes in
What it tells you

The score a strategy with no skill at all is expected to reach, given the same amount of searching you did. If your result is below this number, you have not found an edge. You have found the luckiest of however many things you tried. Most people discover their result sits below it, which is the normal and correct outcome.

Every setting, asset and date range you tested, including the ones you abandoned, and any the person who sold you the strategy tried before you.

The span you actually scored, not what you downloaded.

Do not have one? Tool 2 computes it from your data and sends it here.

If you scored against a static blend at your own average exposure, that bar carries a penalty: a rule with no information at all sits below the blend by (1 − √w) × the asset’s own Sharpe. w is the share of the time your rule held the asset, as a percent.

The asset’s own Sharpe: buy and hold over the same years, excess of cash. Tool 2 shows it as the benchmark when you paste a series.

What is a Sharpe ratio, exactly, and where do I find mine?

It is reward divided by bumpiness. Take your returns, subtract what cash would have paid you, average them, and divide by how much they jump around (their standard deviation). Then scale it to a year.

Sharpe = average(return − cash rate) ÷ standard deviation(return − cash rate) × √periods per year

Two strategies that both made 20% a year are not the same strategy. If one made it in a smooth line and the other made it by nearly dying twice, the first has a much higher Sharpe. That is the entire point of the number: it is the only common yardstick that does not reward you for simply taking more risk, which is why every bar on this page is expressed in it.

  • below 0Lost money against cash, which ends the question.
  • 0 – 0.5Indistinguishable from noise for most test lengths.
  • 0.5 – 1.0Roughly what buy-and-hold on a stock index gives you.
  • 1.0 – 2.0Genuinely good, if the test was clean.
  • above 2.0On a retail backtest, assume a bug until you have proved otherwise. Module 7 lists nineteen that produce exactly this.

Where it is already computed for you:

  • TradingView: Strategy Tester panel → Performance Summary tab → "Sharpe Ratio" row.
  • Freqtrade: printed in the backtesting summary table.
  • Portfolio-level tools (PortfolioVisualizer, QuantStats, pyfolio): report it as standard.
  • MT4 / MT5: does not report Sharpe. Export the balance history and use tool 2.
  • A broker statement or a spreadsheet: use tool 2.

One warning. A Sharpe someone else quotes at you is worth very little on its own, because they chose which number to show. The bar above is what makes it mean something.

Paper-trading bar
–
Real-money bar
–
Timing penalty
–

The lower bar is the average best-of-N luck result, so half of all pure-noise searches beat it: enough to justify watching something, not enough to fund it. The higher bar is beaten by one noise search in twenty.

Where this number comes from

The free first module works through the arithmetic behind this calculator: where the bar comes from, why looking twice raises it, and what to do when your result sits below it. Subscribe for one email every two weeks about what I tested.

Your address is used for this and nothing else. Every email has a one-click unsubscribe.

If a tool here told you something uncomfortable

Why your backtest lied to you

The free first module explains the arithmetic behind these calculators in full: where the bar comes from, why looking twice raises it, and what to do when your result sits below it. No signup, ~14 minutes.

Read it free

The full course is eight modules, nineteen data sources with the trap in each, twelve working download scripts and the searchable ledger of every approach I tested, one of which survived. See what is in it →

These match the Python toolkit. The noise-floor integration is the same calculation as the toolkit’s own noise floor; the real-money bar here uses the exact closed form where the Python simulates it, so it may differ in the second decimal. Everything runs in your browser; nothing you paste is sent anywhere, and the page works with no internet connection.

This is not financial advice and not a recommendation to buy or sell anything. A result that clears every bar on this page is a reason to collect forward data, not a reason to fund it. See Module 8.