Overfitting
Also called: curve fitting, curve-fitting, data mining
Tuning rules so closely to past data that they capture random noise rather than a lasting pattern, so they look excellent in tests and fail afterwards.
The more conditions and settings a strategy has, and the more versions were tried, the more likely its past results reflect luck. Overfitted rules often have oddly precise settings and no clear reason why they should work.
Example
A rule that buys only when a 13-day average crosses a 47-day average on a Tuesday may fit the past perfectly and mean nothing for the future.
Learn more in these lessons
- Misleading backtests and cherry-picked resultsDoes a good backtest prove that a trading strategy works?
- Overfitting, sample size and out-of-sample testingHow do you tell whether a backtested strategy is real or just fitted to the past?
- The limitations of technical analysisDoes technical analysis work?
- Quantitative analysisHow do quantitative analysts build and test models, and where do they go wrong?
Related terms
Spotted a mistake or something unclear? Report it by email. Here is how corrections work.