Overfitting / curve fitting
Picking parameters that explain the past and not the future. The central error of Part XIV.
In the chapter
A system is not a signal: it is seven decisions taken beforehand
From the same chapter
- R multipleThe risk of one trade. The unit that lets you compare systems.
- ExpectancyAverage R per trade. If it is negative, nothing else matters.
- Profit factorGross won divided by gross lost.
- DrawdownWorst fall from a peak. The one you lived is a draw, not the distribution.
- Maximum Adverse ExcursionThe furthest against a trade went before closing.
- In-sample · out-of-sampleThe data you chose the parameters on, and the data it never saw while being chosen.
- Look-ahead biasUsing information in a backtest that did not exist live yet.
- Risk of ruinProbability of going broke even with an edge, through position size.