JerryBoyle |
07-03-2018 06:37 PM |
I bet exclusively through signals generated via a model, so luckily, the human issues that show themselves in a slump and cause decision making that only makes things worse aren't as much of an issue.
That said, my experience over the past 2 yrs of doing this has been:
1. Improve model and backtest to see that it generates a positive roi historically
2. Go live with model
3. Make pretty good money for a few weeks leading to day dreams of retirement at an early age
4. Model starts to fail, I let it run until only a small amount of the profit earned in step 3 remains
5. Rinse & repeat
I've gone through this process 3 or 4 times now, and step 4 is my slump. Makes me think smart CRW teams have a way to work against me once I start, but different discussion for different day. So far, my only busting out move has been to improve the model, and retest over the period where it failed. However, my current effort is trying to find some signal or value about the model fit or the racing season or types of races that can indicate ahead of time if my model is starting to degrade. Surprisingly (maybe not), the pseduo r2 thrown around here often actually is the best I've got. Currently, I'm in a slump with the model turned off, but it looks like r2 is starting to creep back up, so I will hopefully begin betting again soon...
|