I have a trend following strategy that has been optimized for the Dow30 dataset (WD) using Walk-forward and 10 years backtested with all and the out-of-sample data. It appears to be profitable and low risk. If the same strategy is just a regular to bad performer when applied to a different dataset such as SP500, does it mean it was overfitted?
Rename
Different DataSets do have different characteristics. A very obious one is volatility. Dow30 has mouch lower Volatility than Nasdaq100 for example.
If your strategy has components which depend on volatility (most do in one way or another) then you'll need to adjust some parameters for different data sets. Or - much better - you make a running volatility measurement (like an ATR indicator) part of your strategy and adjust such a parameter automatically.
There are quite some more characteristics of a dataset like tendency to bounce back,
Probability of gaps, tendency for trends and so on which are much harder to acount for automatically.
So the short answer: No.
If a Strategy works for one (large enough) data set only it is not necessarily a sign of over-optimization.
If your strategy has components which depend on volatility (most do in one way or another) then you'll need to adjust some parameters for different data sets. Or - much better - you make a running volatility measurement (like an ATR indicator) part of your strategy and adjust such a parameter automatically.
There are quite some more characteristics of a dataset like tendency to bounce back,
Probability of gaps, tendency for trends and so on which are much harder to acount for automatically.
So the short answer: No.
If a Strategy works for one (large enough) data set only it is not necessarily a sign of over-optimization.
Thank you Doc for your answer, it's always very good to read you.
One compensating option would be to place like stocks into like datasets. For example, you could place all small cap stocks in their own dataset. Or place all stocks that are sensitive to oil prices (or transportation costs) into the same dataset. But having your strategy correct for that leading indicator (as was suggested with volatility in Reply# 1) would be the most ideal option.
WL's IndexLab is designed to "aggregate leading indicators" and then publish them as indicators your strategy can trade against.
WL's IndexLab is designed to "aggregate leading indicators" and then publish them as indicators your strategy can trade against.
QUOTE:
One compensating option would be to place like stocks into like datasets. For example, you could place all small cap stocks in their own dataset. Or place all stocks that are sensitive to oil prices (or transportation costs) into the same dataset. But having your strategy correct for that leading indicator (as was suggested with volatility in Reply# 1) would be the most ideal option.
WL's IndexLab is designed to "aggregate leading indicators" and then publish them as indicators your strategy can trade against.
2
Excellent answer and an interest idea to work on. Thank you Superticker, I am hands on the task.
Your Response
Post
Edit Post
Login is required