Hi all,
Here’s the question: You build a model and made your best judgement in it’s constructions based on the backtest results and factor interactions. In the abstract you honestly feel like it’s picking performing well.
But then when you run it “for real” you might see a list of stocks, but maybe 1 or 2 out of 10 would be stocks that make you wonder “why is my model picking this dog?” And you look at the financials and wonder how did my model pick this? You look at the factor ranks but still don’t get it. You see a news story that the CEO even just resigned, sales are flat, and the stock price is about the same as it was 5 yrs ago. So you go in and start tweaking, trying to make sure your model picks stocks closer to your conception of what a good investment would be (by doing things that might disadvantage said “dog”). But no matter what you do, adjustments you make only reduce backtest results. By trying to exclude the “dogs” you end up hurting the performance of the model.
Do you conclude that the original model is better and just take what the model recommends? Or do you prefer the adjusted model with worse returns that tends to excludes companies you don’t like? I guess to what degree do you let these types of reactions to stock selection direct your modeling and investing process? Thoughts appreciated.
For me, I admit to finding it hard to hold my nose and buy a stock that I simply don’t understand - but I also admit that I might could learn to like them if I understood them better. I also see the case that having the stock selections be more in tune with my intuition of the types of companies I’d like to invest in is also desirable in many ways.
Thoughts appreciated.