forex-scalping.com

R squared of the cumulative R line: two journals that both end at +10R

Two traders each took 20 trades and ended at +10R. Their cumulative R after each trade:

Trader A: 1, 0, 2, 3, 2, 3, 4, 3, 5, 6, 5, 6, 7, 8, 7, 8, 7, 8, 9, 10
Trader B: 3, 2, 1, 0, 2, 1, 5, 4, 3, 2, 5, 4, 3, 2, 7, 6, 5, 4, 7, 10

Fit a straight line through each, with trade number on the x axis. The R squared of that fit is a measure of how straight the climb is. For A it is 0.94 and for B 0.55. In a spreadsheet: =RSQ(cumulative_R_range, trade_number_range).

The largest drawdown from a peak is 1R for A and 3R for B.

Same result, different experience. B was at +3R after the first trade, back at 0 by trade 4, and made the last 3R in the final trade. Judge B on 19 trades and the total is +7R. A has no single trade doing the work: its biggest gain is +2R. Which of the two would you rather be sizing up after 20 trades?

Nice illustration. Both traders made the same money, but Trader A's smoother path is much easier to stick with. R squared is a simple way to measure that, and a useful complement to drawdown when comparing strategies.

R squared is not a complete measure. A line that is flat, with R squared of 0.95 because it barely moves, is not better than a rising one at 0.8. Always read it with the slope, which is the R per trade: 0.46 for A. A trader who wants a figure combining both can divide the slope by its standard error, or use the K ratio family of measures, but for a journal the two numbers side by side are enough. A rising line with a high R squared and a positive slope is what steady edge looks like. A high value with a slope near zero is an account going nowhere in a very orderly way.

With only 20 trades the number is noisy. Trader B's R squared is 0.55 with the final +3R and 0.48 without it, and 0.59 if you drop the first trade instead. Those are moves of 0.07 to 0.11 from removing one trade out of 20. Do the same on your own journal by recalculating with the last trade left out, then the first. If the figure moves by 0.2 or more from a single trade, it is not telling you about the process. With 100 trades the pull of any one trade falls a lot, so use the statistic on bigger samples and judge a 20 trade version cautiously.

Use the line for something practical: a comparison of the live account with the backtest. Say, with made up figures, the backtest line had an R squared of 0.97 over 200 trades and a slope of 0.25R per trade, and live over 60 trades the slope is 0.12R and the R squared is 0.71. Both differences are expected somewhat, since live trades have costs and noise that the test missed, but they give you two concrete numbers to review monthly. A live slope of half the test slope for 60 trades might still be inside the noise, which you can check by seeing how much the slope changes between the first 30 and the last 30 trades.

Plot it, since the numbers do not replace the picture. A chart of cumulative R against trade number, with the fitted line added as a trendline and the equation displayed, takes a minute in a spreadsheet. The vertical distance between the curve and the line is where the account was ahead of or behind its average pace. Long stretches below the line followed by a rush above it are the pattern of a lumpy system, and the plan for lumpy results is different from the plan for a smooth one, for example in how long you give a new setup before judging it.