portfolio-data-split

Splitting portfolio data for backtesting involves dividing datasets into training, validation, and walk-forward subsets. This requires input files like panel_ohlcv.parquet and returns_wide.parquet to generate the separated datasets needed for strategy evaluation.

Walk-forward data splitting in portfolio analysis is a technique that divides data into sequential subsets for testing trading strategies. It creates separate training, validation, and walk-forward sets to facilitate a comprehensive evaluation of strategy optimization over time.

You need parquet file formats for portfolio data splitting, specifically panel_ohlcv.parquet and returns_wide.parquet. These files provide the panel data and returns matrices required to generate the split training, validation, and walk-forward output datasets.

Yes, you can customize output paths when splitting portfolio datasets. The data splitting process supports specifying custom data paths and output directories, allowing you to direct the generated training, validation, and walk-forward subsets to your preferred locations.

Available split modes for portfolio strategy backtesting include full-only and train-validation-walk-forward. These modes allow you to control how the portfolio data is divided, accommodating different phases of the strategy optimization process.

Dividing portfolio data into training and validation sets prevents overfitting during strategy optimization. By separating the data, you can train your model on one subset and validate its performance on another before final walk-forward testing.