Free lesson · HFT & microstructure

Volume clocks, trade imbalance and VPIN

Open interactive lessonPractice calculationsExplore labs

First distinguish a quote from a trade. A quote is resting willingness to transact; a trade is executed volume. Buy-initiated volume means a buyer crossed to a seller’s quote, not that the trade had no seller. Every trade has both sides.

Choose a bucket size using only past information. Accumulate executed volume until the bucket is full, splitting a trade across boundaries if necessary. Label the completed bucket at its closing timestamp. A bucket completed in the future cannot be attached to its earlier start for a historical decision.

Given estimated buy and sell volumes in equal-sized completed buckets, calculate the absolute imbalance in each and average it relative to bucket volume. This is the volume-imbalance aggregation used in VPIN-style measurement. Taking the absolute value removes direction. High imbalance can be a stress feature without identifying whether you should buy or sell.

Trade signing is itself an estimation problem. Venue aggressor flags, quote tests and bulk-volume classification can disagree. Compare signed imbalance, activity, spread and volatility controls, and evaluate incremental information about subsequent markouts at a fixed horizon. The VPIN name does not make its score a calibrated probability that a particular trader is informed.

Apply it in a strategy

• Build equal-volume buckets causally and audit their closing times and volume conservation.
• Compare trade-signing conventions and bucket sizes chosen only on training data.
• Test incremental future markout prediction controlling for trade intensity, spread and volatility; keep inventory limits independent.

Research deliverable

Hand-fill several buckets, show leftover volume crossing a boundary, and compare signed versus absolute imbalance.

Sources & evidence · reviewed 12 September 2026

Reviewed 12 September 2026: README and repository overview only; code was not executed or reproduced. The README identifies Python 2.7.1, Wind data from January 2015–October 2018 and CSI-300 examples. This is implementation inspiration rather than validation evidence. The lesson implements only the completed-bucket aggregation with modern self-contained Python.

Further reading: yt-feng/VPIN · supplied implementation repository ↗

Primary empirical critique, publisher abstract reviewed: challenges VPIN as a toxicity or early-warning measure. Full empirical tables were not reviewed here; no independent replication is claimed. Compare classification, activity and volatility controls rather than assuming the acronym establishes predictive validity.

Further reading: Andersen & Bondarenko · Assessing Measures of Order Flow Toxicity and Early Warning Signals for Market Turbulence ↗

The existing checkpoint reviews a September 2026 simulated order-book study. It motivates stress tests, not new validation of VPIN.

Further reading: Recent HFT regime-stress research ↗

from math import isfinite, isclose

def vpin_completed(buy,sell,bucket_volume):
    if not buy or len(buy)!=len(sell) or not isfinite(bucket_volume) or bucket_volume<=0: raise ValueError("Completed equal-volume buckets required")
    for b,s in zip(buy,sell):
        if not all(isfinite(x) and x>=0 for x in [b,s]) or not isclose(b+s,bucket_volume,rel_tol=1e-9,abs_tol=1e-9): raise ValueError("Invalid or incomplete bucket")
    return sum(abs(b-s) for b,s in zip(buy,sell))/(len(buy)*bucket_volume)

print(vpin_completed([70,40],[30,60],100))