Thank you! Definitely a good idea to not assume persistence but to go looking for it in the data.
One suggestion: rather than backtesting selection criteria, think about how you can look more directly at those selection criteria. Simple data analysis tools like scatterplots and factor plots will get you a long way (but sometimes you need to get creative).
Reason being, backtesting is quite an indirect way to quantify an effect... lots of path dependency and generally an inefficient use of data. And data being so limited, we should really extract as much information as possible from what we have.
I'd be really happy to do a piece on this idea if people are interested.
Thank you! Definitely a good idea to not assume persistence but to go looking for it in the data.
One suggestion: rather than backtesting selection criteria, think about how you can look more directly at those selection criteria. Simple data analysis tools like scatterplots and factor plots will get you a long way (but sometimes you need to get creative).
Reason being, backtesting is quite an indirect way to quantify an effect... lots of path dependency and generally an inefficient use of data. And data being so limited, we should really extract as much information as possible from what we have.
I'd be really happy to do a piece on this idea if people are interested.