from what I can see, you need much more than 20 rounds to evaluate TC performance. most of my V2 models with FN have slightly above zero TC over 50+ rounds, but if you look at individuals over 20 rounds then it does go up and down and hard to make a call. I also observe that V3 models seem to have higher TC volatility, but these also yield my best TC models.
personally, I like TC more than MMC - it does seem to favor FN models, and isn’t as correlate with Corr like MMC - relatively happy with lower value multipliers perhaps until the big -0.2 burns arrive!
I found my CORR and FNCV3 are good enough and start staking on TC, then gained very bad TC
Current TC can reward and punish us. I can accept penalty if my prediction is bad. But I cannot accept penalty for not adapting to the optimizer.
Most of high stake users on the leaderboard only use CORR, but their predictions made up current meta model. Obviously they worried about TC. Is there any way to make them less worried about TC？Reducing penalty for not adapting to the optimizer is my idea.
