Overfitting in investing

What overfitting is, why strategies tuned to past data fail in live trading, what a large study of community strategies found, and how to measure it.

What it is

Overfitting happens when a strategy's rules are tuned so closely to past data that they capture random noise rather than a lasting pattern. The backtest looks superb; the future does not cooperate.

How it happens

  • Adjusting thresholds until the backtest peaks (RSI 27 instead of 30).
  • Adding branches to dodge each past crash.
  • Choosing the best of many variants. See Signal mining and data snooping.
  • Picking the test period that looks best.

What the evidence shows

A study of thousands of public rules-based strategies on Composer, published by Composer Atlas, compared each strategy's backtest with its results after publication:

  • Only about 4 to 20% delivered their full backtested return afterwards.
  • Turnover was the strongest predictor of failure: the highest-turnover fifth gave up a median of 136 percentage points of return.
  • Longer backtests helped less than expected, and how concentrated returns were in a few days did not predict failure.

Measuring it

Compare the return a strategy earned on the data it was built on (the "fitted" period) with what it delivered afterwards:

Overfit score = 100 x (1 - delivered return / fitted return)

A score near 0 means it delivered what it promised; near 100 means the edge vanished. Always compare against similar strategies, because some decay is normal.

Avoiding it

  1. Fewer rules, each with a reason.
  2. Test nearby parameter values; results should not collapse.
  3. Keep data aside for Out-of-sample testing.
  4. Be suspicious of very smooth, very high backtests.

See also

Pages that link here: Frontrunner strategies, Leveraged strategies compared, Relative Strength Index (RSI), Sharpe ratio, Systematic investing, TQQQ For The Long Term (FTLT), VIX tier allocation, What is backtesting?

Last updated September 30, 2026. Education only, not investment advice.