Overcoming Optimization Pitfalls
When optimizing a strategy, you adjust the inputs of a strategy to work well with a specific set of historical data. It is possible for the inputs to become 'fitted' so tightly to the historical data set that the entire strategy will not work well on any new data. This is called 'over-optimization' or 'curve-fitting'.
Some ideas to help you avoid over-optimization:
- Develop or use trading strategies that are based on solid market theories and trading ideas. Use logic that will capture the intended market activity.
- The values closest to the optimal value should also be profitable (that is, if the optimum value is 10, the values of 8, 9, 11, and 12 should also profitable).
- Once you optimize, perform the same historical back and forward testing on different time periods. For example, optimize the input values using data from 1996-1998. Then, using the optimized values, test the entire strategy on data from 1985-1989. Optimize the trading strategy on data from 1985-1989, then use these optimized values to test the trading strategy on data from 1994-1998. Results of the different tests will vary; the key is that the strategy should hold up under all these conditions.
Related Topics
Optimizing Strategies on a Chart