Aggregating Intraday Price and Volume Factors into Daily Features

Summary

The document discusses converting one-minute stock data into daily observations for high-frequency price and volume factors. The example query constructs features such as volume shares across intraday intervals, price and volume correlations, correlations with current and neighboring minute returns, interval returns, and return variance, skewness, and kurtosis. It then uses a row number within each day to retain one row, raising the question of whether to keep the first or last row and whether the window ordering should be reversed.

The central issue is that the shown window expressions are ordered cumulatively by timestamp, so many values can differ from minute to minute; selecting the first row may not represent completed daily or interval statistics. The document reports this discrepancy from inspecting query output but does not resolve the correct aggregation. A researcher should distinguish cumulative windows from full-partition calculations and verify frame semantics and the intended observation point before using these features. The code is an implementation discussion, not evidence that the proposed factors predict returns.

Key ideas

• The query derives intraday volume distribution, price-volume relationships, returns, and return-distribution statistics from minute bars.
• Window calculations ordered by time can produce evolving values within a trading day.
• Choosing the first or last row changes which point in those cumulative calculations is retained.
• The document raises the aggregation question but does not establish a definitive solution.
• Window frame definitions and the intended daily measurement point should be checked before using the resulting factors.

Tags

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