Financial Data Structures: Comparing Time, Volume, and Dollar Bars

Summary

This article compares time, tick, volume, and dollar bars as ways to organize market data for machine learning. Time bars use fixed intervals; tick and volume bars use trade counts or traded quantity; dollar bars use traded value. The proposed rationale for activity-based bars is that they may better reflect changes in market activity than fixed clock intervals. Dollar bars also keep the value exchanged more consistent when asset prices change, though the appropriate threshold is not settled and may depend on the instrument and intended trading frequency.

The article analyzes S&P 500 E-mini futures data and compares return distributions using Jarque-Bera and Shapiro-Wilk statistics, along with a visual distribution check. In the reported sample, volume bars score best on the Jarque-Bera comparison, while both volume and dollar bars improve on time bars under the Shapiro-Wilk comparison. The authors choose dollar bars for their subsequent work because of their intuitive interpretation, not because they outperform on every test. These distribution checks do not establish predictive value or trading profitability, and findings may depend on the data and bar thresholds.

Key ideas

• Time, tick, volume, and dollar bars define observations using different measures of market activity.
• Dollar bars group trades by exchanged value, which adjusts the quantity traded as asset prices change.
• The article argues that activity-based sampling may reduce distortions from quiet periods in clock-time bars.
• In the reported distribution tests, volume bars perform best on one normality statistic, while volume and dollar bars improve on another compared with time bars.
• The choice of bar threshold remains uncertain, and distributional properties alone do not prove trading usefulness.

Tags

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