How to stress test a Buy Rule

Hi

This links to my previous topic, but from a different angle.

I have a buy rule, which is based on a sound principle but there are different ways to implement it.

I've come up with a rule through using "Optimizer Studies" (which I hadn't realised was there until someone pointed it out so thank you so much!).

However, the rule is so basic that I don't trust it. I prefer something more complicated - but that gives worse results on a 20 year backtest.

What are the ways one would stress test my simple rule to validate it, or break it?
e.g. I've compared annual performance of my simple rule and it wins in 16 out of 20 years.

Thanks!

Simple is good

Haha yes completely agree. Simple is brilliant. This is just too simple, and works better than my slightly less simple rules. So want to stress test it on historical data, as well as then watching it operate live for a few months.

If you have to use a buy rule (ranking systems are more robust!) I recommend testing the rule using the mod(stockid)-method:

This way you can test the rule in many subset of your full universe. If it's a good rule it should work in most of them.

So want to stress test it on historical data, as well as then watching it operate live for a few months.

If your rules are designed for U.S. stocks, then you might try testing your rules in non-U.S. universes (ADR, Canada, Europe) to see if they are better than no rules at all.

Live test gives little information because only the most serious problem would be revealed.

2 Likes