Extracting Multi-Day Intraday Features with Distributed Minute-Bar Processing

Summary

This BigQuant example addresses a limitation in which its high-frequency feature extraction module calculates within a single day, while a researcher may need features from minute data spanning multiple days. It shows a distributed workflow using FAI: read trade-level stock data, resample it into one-minute bars, calculate daily features by instrument, and combine results from parallel remote tasks.

The example computes return variance, skewness and kurtosis, average trade amounts, measures associated with rising and falling returns, and large-order flow and momentum features. It selects high-amount minute bars as a proxy for large orders. The post offers code as an implementation illustration, not empirical validation: it does not report predictive performance or compare the features against alternatives. Its results depend on the data fields, sampling, aggregation choices, and proxy definitions shown, and the example focuses on Chinese stock data.

Key ideas

• Resample trade-level observations into minute bars before calculating intraday features.
• Group calculations by date and instrument to produce daily feature records.
• Use parallel remote tasks to process instruments across a date range.
• The example includes return-distribution, average-trade-amount, large-order-flow, and momentum measures.
• Large-order activity is approximated by ranking minute bars by transaction amount.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.