Definition. Fitting a method's rules to historical data so closely that they describe random noise instead of lasting relationships. Such a method usually fails on new data.
How it works
The more parameters and conditions you add to improve the historical result, the greater the risk that the method has “learned” random cases. The backtest looks great, live results look poor.
Example
A method with a dozen filters achieves an excellent result on one year's data and loses money the following year.
What to watch out for
What helps: simple rules, testing on data that wasn't used to build the method, and a reasonable number of observations.
You'll find the full explanation with graphics, examples and exercises in the 📖 Order flow encyclopedia space, and its use in the live market in the 🔴 Live Room. Both are available to Club members (in Polish).
