AR1 Prediction Model
Gradient boosting trees can handle NaN values natively, while most other models require filtering NaN rows before training. The forecasting model predicts future log return as a linear function of the current log return. The Model class re…
2 sources - 9 claims
Gradient boosting trees can handle NaN values natively, while most other models require filtering NaN rows before training. The forecasting model predicts future log return as a linear function of the current log return. The Model class receives weights and bias from offline research and can handle scalar and vector inputs with the same predict method. The AR(1) setup assumes there may be serial predictability in price series. A vectorized drop_nan function should use a NumPy boolean mask instead of a Python loop. The feature matrix consists of lagged log returns in row-major order after NaN rows are dropped. The research notebook downloads historical OHLCV data, fits the model with a time-series-aware train/test split, and reports performance statistics. An AR(1) model predicts the next log return from the previous log return.