I finally took a vacation. I was so relaxed that I didn’t even post about it.
Then, labor day weekend to tie it off.
Got to drive my new car around on some road trips. Got good and drunk a couple times.
Oh. Daily trader is going well:
Introducing a directional accuracy metric on historical forecasts across the full range of historical data made a huge, huge difference.
Unfortunately, even with the new metrics, one of the classic problems with a system like this is that even if you have good metrics to sort and filter by, you can overspecialize the parameters for those filters if you only look at one forecast set, so, winner selection might be great for that day, but you’ll get misses on the others unless you’re doing a range of dates. For 2000+ forecasts per day, that’s a little complex. So, it’s back to the drawing board a little to try to squeeze a little more out.
It’s been hard work, but, even in its current state now I think it’s almost ready to go live. I just need to get this last round of analyses performed to make sure I’m not cheating myself out of larger gains. 8.43% over 11 days is nothing to snuff at — it’s incredible actually — but, let’s see where it goes. I think it can do better.