macrocephalopod

Discretising continuous variables can also lead to jumps in your forecast if the underlying variable meanders around the threshold over time, which lead to undesirable turnover in your strategy (which costs you money)

Ah I understand now!! How about output? Is it generally to try to model a continuous variable (futureRet), or to a logit regression to model odds of a category (TopDecileRet)? If Y is cont, confused on how to trade a model, do you adjust your exposure as expctd y changs?

I exclusively try to predict continuous variables, generally either (excess) return over some period or (excess) return scaled by a volatility forecast

Following this, another example is categorical to predict as a way to get around non-linear dynamics. E.g. if you have a factor implicated in extreme conditions, it may show up as a skew in returns (24hRet < 0) but not show up properly in linear regression.

So in this case simply checking for the categorical (a pretty naive parameter threshold of 0, not overfitting) will give you quicker insight into the value of the relationship

I just try to make money every month call me old fashioned

The real answer is that a binary variable is likely a shitty continuous variable you tried to flatten to make your backtest look better. Example: rising google trends binary works better than continuous bc it smooths news spikes. There are better ways to normalize (web vs search)

Gotcha, but how do you set your exposure, do you use scaling or draw lines in the sand? Also arent there some signals that mean something diff if you scale it. For example, I believe a breakout from a range (True/False) has diff meaning than dist from range top