What a track record supports
Paste a list of closed trades and get each figure with the range its sample allows.
The question
- A track record arrives as a win rate and a profit factor. Neither figure says how many trades produced it, and on forty trades chance produces most of the pairs that get quoted.
- Twenty-four wins in forty is a rate of 60 percent. The range that sample supports runs from 44.6 to 73.7 percent, and a coin sits inside it.
How to read the table
| column | what it is |
|---|---|
| value | What these particular trades did. Not in dispute. |
| 95% interval | The range the true figure is compatible with, given how few trades there are. A short record has a wide range. |
| same figure in Python | The identical calculation, run beforehand by mesura and shipped with the package. |
| the sentence below each row | The same thing in words, including what it means when a coin or a break-even point sits inside the range. |
- The Python column is filled for the built-in examples only. A list you paste has no second column: Python does not run in a browser.
- A resampled bound agrees to the last bit. A Wilson bound agrees to about 1e-16, because the z value comes from Python's inverse normal on one side and from an approximation on the other. The page prints which of the two it is.
Where the intervals come from
| figure | method | why that one |
|---|---|---|
| win rate | Wilson score interval | The textbook normal interval puts the upper bound above 100 percent on a high rate and a short sample, which is where a record is usually quoted. |
| profit factor, payoff ratio, expectancy | percentile bootstrap, 20000 resamples | Ratios of a heavy-tailed sample with no usable closed form, so the range is measured by redrawing the same trades rather than assumed. |
| break-even win rate | 1 / (1 + payoff ratio) | An identity, not a measurement. It ignores costs: a record that breaks even before spread and commission loses money after them. |
What it cannot answer
- Whether the record beats chance. A coin flip has to be flipped against a market, and a list of closed trades does not carry one: the prices that were not traded are missing.
- Nothing you paste leaves the browser. There is no request, no upload and no storage, which the network tab will confirm.
Source
- LudwigJMarx/mesura-web: the page, the parser and the TypeScript port, with its tests.
- LudwigJMarx/mesura: the Python original the figures above are compared against.