I tried to test the accuracy with tick data. No luck yet. I think a mode flag or some parameters should be wrong. I don't understand why 10% of the trades are wrong even with tick data. There are 1 second bars with 200-300 price change and I run the backtest on 15sec bar period.