Strategy correlation matrix

Correlation is the one thing a backtest tells you that still holds a year later. Across 322 strategies we measured how well year one predicted year two: pairwise correlation persisted at +0.707, while return persisted at −0.016 — indistinguishable from random. So this tool is built on the measurement that survived, not on the one that did not.

Pick your combination

Every strategy here is one of our free templates, backtested daily on the same rolling window. Tick a few and the numbers below recompute — including how the combination compares with running each one on its own account.

Loading…

The full 16 × 16 matrix

Daily returns, computed on the same window as the template cards. Red means the pair tends to win and lose together; green means they do not.

Loading…

What happens when one of them stops working

Those numbers all come from templates. The real question is: what if one of the EAs I add stops working?

From 322 non-template strategies we took those that were profitable in sample and lost money out of sample while staying uncorrelated with these six — 27 of them — and dropped in the two worst. That is deliberately picking the bad case, not a random one.

EA addedIn sampleIn-sample SharpeOut of sample
US30 · Ichimoku cloud breakout+22.9%1.34-32.1%
USDJPY · Williams %R + EMA200+19.1%1.15-27.6%
PortfolioLegsOOS returnMax drawdownSharpe
6 templates (baseline)6+151.34%7.33%3.88
+ 1 failed EA7+70.80%16.96%1.98
+ 2 failed EAs8+23.97%21.38%0.82
Random failed EAs addedMedian return25th pctMedian DDStill profitable
1+122.2%+101.9%12.17%100%
2+93.5%+70.8%16.25%100%

Two EAs that each lost about a third, dropped in, and the portfolio still returned +23.97%. The cost is real: down from +151.34%, with Sharpe falling from 3.88 to 0.82 — two bad legs out of eight is a quarter of the book subtracting.

Drawing the failed EAs at random instead of picking the worst, across 2,000 draws of two, every single one stayed profitable.

Forward test — frozen, tracked daily, win or lose

Everything above is a backtest, and every backtest on this page was built with hindsight. So on the date below we froze one combination — the six lowest-correlation templates, parameters locked — and from that day forward we record it once a day and publish whatever comes out. No re-selection, no restarts, no quiet edits. Screenshots and backtests are easy to fake; a timestamp is not.

Loading…
Method. Every figure comes from this site's engine re-running each template daily on a rolling 24-month window, with real spreads, excluding swap and commission. Portfolios are equal-weighted — the same risk percentage per trade on every leg — daily returns summed on calendar days, annualised with 365.

One thing that has to be said plainly: these 16 templates were selected using this same history, so the return figures carry selection bias and are not an expectation for the future. The correlation structure does not have that problem — in our out-of-sample test it persisted at +0.707 while return persisted at −0.016. Use this tool to judge diversification and risk, not to forecast returns.