Evaluating Timing Models !!!
Stock market timing models are difficult to evaluate, but you do not have to completely understand the interworking of the models.
For any model you are considering investing with, you should at least know:
- What is the background of the model developer ?
- that is, is he/she qualified to develop a sophisticated timing model ?
- On average, how many times does the model trade every year ?
- some models issue about 12 trades/year and others over 100 times/year.
- Are the long-term performance results of the model outstanding ?
- for example, does the model beat buy-and-hold in both up and down markets ?
- to satisfy the above, your evaluation must include at least one market crash (as in 2008).
- Are the performance results Verified by a 3rd-party tracking company ? ***
- like, TimerTrac.com, ThetaResearch.com, etc.
*** Non-Verified Performance
If the hypothetical backtested (non-verified) performance of a model:
- comes from a company's first model, do not trust the results;
- comes from a company that has several older verified models that produced lousy performance results, do not trust the newer results;
- comes from a company that has several older verified models that produced very good performance, you can trust the results from the new models (because they usually retain 85-90% of the code/signals from the older models, together with 10-15% of new code that contains algorithms which produce a higher percentage of winning trades).
If the information that you have available for any model is not enough
to satisfy the conditions above, you should NOT use that model.
On the other hand, it you can evaluate a model as above,
you most likely do not need any other info !!!