Statistical Modeling in HFT
Out-of-sample R-squared is useful for avoiding overfitting but does not reliably determine profitability. Modern HFT firms often precompute model outputs into lookup tables, making execution latency less about real-time inference. Price mo…
1 sources - 4 claims
Out-of-sample R-squared is useful for avoiding overfitting but does not reliably determine profitability. Modern HFT firms often precompute model outputs into lookup tables, making execution latency less about real-time inference. Price movement features should be normalized as percentage returns rather than dollar moves. Overfitting is a primary modeling risk in trading because market data has a very low signal-to-noise ratio.