macrocephalopod

Are you saying, dont try to find simple tendencies like this? I think maybe it's worth exploring, bc if the asset remains stable you can make money in theory

1. Don’t look for the thing you’re testing 2. Don’t overfit parameters based on the thing you’re targeting 3. Don’t simply select the best parameter value from one historical test 4. Understand max theoretical sharpe ratio for a given strategy (# indie bets)

Never tried 4, how do you calculate that? For 2&3, isnt it true that you have to pick parameters at a certain pt to run the strat? If so, may as well optimize w history, avg bets betw top 3 parameterVals, check sensitivity to param not too high. For 1, what should we do instd?

Is number 4 related to Grinold & Kahn IR = IC x sqrt (breadth)

Yup

So this is what not to do, instead do we need to start from an economic thesis and try to linearly model some future excess returns based on particular input variables? Is there truly no pt in testing the consistency of certain simple/naive tendencies/patterns 1/2

In thry, maybe we observe a particular asset tends to have two regimes, small mean reversion around trend, and messy chop, if we have a good model at identifying these regimes, when we trade the reversion one, couldnt this be a way to design an implementation? Sorry for all theQs

1. How sure are you that there are really two regimes? Couldn't you just be seeing patterns where there are none? 2. Regime-based models need to be right twice -- first about which regime you are in, second your signal needs to work in that regime.