Why does my PPO network always collapse to buy and hold?
retained text
I have a 10 year dataset across a range of companies. Unfortunately it's only 1m candles as I'm just doing initial exploration at the moment. My training loops always seem to collapse to buy and hold across the 9 year test data set. Why and what are typical ways to shake it up to prevent this one position learning? Or, do I stop early and fly on one wing and a prayer? Should I also experiment with A3C?
document fetched 2026-10-24 10:45 · canonical https://reddit.com/r/algotrading/comments/1xdlhqz/why_does_my_ppo_network_always_collapse_to_buy