It works extremely well at 1 lot trades, but once you increase the volume profits reach maximum at some point once you do significant turnover compared to the rest. It's kinda obvious, you got adverse selected more and more.
What bothers me - maybe I simply quote inefficiently? I tried a lot of of approaches (imbalance, skewing, VWAP, hand crafted signals), and at some point they stopped improving my PnL.
Any way to estimate theoretical capability of the specific market? Is this market already exhausted and you just can't do anything else here.
Assuming that market indeed can be exhausted, it's not some US bonds/top stocks with infinite volumes.
And side question about backtesting. Is it even worth it? Now I filter past trades with new signals. Predictions of new trades suck for me, in real life competitors instantly react on my orders, so backtest overestimates PnL.