Do The Virtues of Volume Extend to VecViz?

2/2/2026

Our thanks to Jessie Li for her assistance with key elements of the analysis discussed below.

TL/DR: They do, despite the fact that volume is not a direct input to any VecViz model. More explicit consideration of volume leads to improvements in baseline VecViz analytic performance, particularly for the V-Score.

Volume is a well-regarded quant model input.

Trading volume has long been cited as a reliable proxy for information flow. The merits of weighting ticker returns by volume when modeling volatility have been known since the 1970’s^{1 2}. In 2012 the “Volume Clock” approach to collecting intraday price data was introduced as a way to better detect “toxic” market making conditions³. Numerous studies in the years between⁴ and since^{5 6} have continued to confirm the “virtue” of volume as a quant model input.

Nevertheless, volume is not a direct input to any VecViz model.

The VecViz conceptual framework is about the influence of the bullish and bearish “vectors” within a ticker’s fundamental narrative upon its price chart.  Thus, though volume has many virtues, it is exogenous to VecViz. 

But, price chart Tops and Bottoms are at the core of VecViz, …

VecViz views major price Tops and Bottoms as turning points in the net bullish / bearish vector balance driving price. We use their dates as the basis for embedding VecEvent narrative elements on our dashboard charts. They are also key inputs to our Vector Model of price probability and V-Score ranking of forward expected price return:

• Several Vector Model and V-Score feature inputs are simple transformations of Top and Bottom prices or dates (see the blue circled features in the V-Score spider chart below).
• The angles of the “VectorSet” price channels we construct from the Top and Bottom dates serve as key inputs to the Vector Model and V-Score (see grey circled features in the V-Score spider chart below)

… and they tend to occur amidst high trading volume.

We recently studied the ratio of average volume for Top & Bottom (TB) dates to volume on non-TB dates, which we will refer to as the “Volume Ratio”. We included the 141 active tickers in our coverage universe for which volume data was readily available, and included market data from 1/31/2015 through 1/27/2026.

The median ticker studied had a Volume Ratio of 1.41. Mann-Whitney testing⁷ found the Volume Ratio was significantly > 1.00 for

• 78% of all tickers studied with 90% confidence.
• 63% of all tickers with 95% confidence.

Do the “virtues” of volume permeate the machine learning ensemble residing between the Tops and Bottoms and VecViz’s published analytics?

If they do, then VecViz metric performance should be stronger in tickers for which the Volume Ratio is significant. We will examine Vector Model price probability analytics first, and then the V-Score, to determine whether or not that is the case.

We have several metrics that are variations of Vector Model price probability percentile forecasts:

• Value at Risk (VaR) and Opportunity at Risk (OaR), measured at both the 95th percentile and 99th percentile.
• Expected Up Body (EUB) and Expected Down Body (EDB)⁸
• Option Fair Value estimates

Of these we tested VaR and OaR on the basis of their breakage rate accuracy relative to targets (5% for 95% VaR and OaR, 1% for 99% VaR and OaR). On this basis we found no impact, but on a “relative to Sigma basis” we did find something.

Higher Volume Ratio tickers tend to have better 95% VaR 1d performance vs. “Sigma”.

Volatility tends to be correlated with volume⁹. Therefore, tickers experiencing especially higher volume on their Top and Bottom dates could arguably be expected to have higher VaR breakage rate differentials vs. target.

The bell curve based “Sigma” model of price probability¹⁰, which we compare Vector Model output to in many performance reports, does not incorporate trading volume at all – neither explicitly nor implicitly. Given the potential confounding influence of the volume – volatility relationship on Vector Model VaR breakage rate performance, “Sigma” VaR breakage relative to target is a helpful control when assessing the impact of volume on Vector Model VaR performance.

On a “relative to Sigma” basis, we did find a nearly significant¹¹ improvement in 95% VaR actual breakage rate vs. target performance for the 1d forward horizon of 0.14% in tickers with significant Volume Ratios over those with insignificant Volume Ratios.

VecViz’s V-Score has a significant high vs. low Volume Ratio ticker performance differential.

We found significant Spearman (i.e. rank¹²) correlation between ticker level Volume Ratio and ticker level differential between positive and negative V-Score forward price return for the 10d horizon (p-value of 0.035). Correlation at the 21d, and 63d horizons were also significant or nearly significant (p-values of 0.137 and 0.054, respectively), depending on your confidence standards.

To assess what this meant for performance in terms of price returns, we reran our V-Score performance report excluding the 31 tickers for which the Volume Ratio was not significant¹³. We found that positive V-Scores outperformed negative V-Scores for those horizons by an incremental ~250bps, annualized¹⁴. The improvement by V-Score category and time horizon is detailed further below.

Conclusion

We put Tops and Bottoms at the very core of VecViz analytics because they are the link between a ticker’s narrative and its price chart. In so doing, we inadvertently tapped into one of the most robust phenomena in quant finance: the signal strength of price associated with elevated trading volume.

Testing of the relationship between VecViz metric performance and TB trading volume confirms that this “virtue” of volume filters through somewhat to VaR, and to a much greater extent, the V-Score.

This conclusion helps explain why the performance of VecViz analytics over their four year out of sample period has held up so well, despite zero retraining, and offers a new route to enhanced metric performance going forward.

1. Clark, P. K. (1973). “A Subordinated Stochastic Process Model with Finite Variance for Speculative Prices.” Econometrica ↩︎
2. Epps, T. W., & Epps, M. L. (1976). “The Stochastic Dependence of Security Price Changes and Transaction Volumes.” Econometrica ↩︎
3. Easley, D., López de Prado, M., & O’Hara, M. (2012). “Flow Toxicity and Liquidity in a High-Frequency World.” The Review of Financial Studies ↩︎
4. Karpoff, J. M. (1987). “The Relation Between Price Changes and Trading Volume: A Survey.” Journal of Financial and Quantitative Analysis.
   ↩︎
5. Padungsaksawasdi, Chaiyuth & Daigler, Robert. (2018). Volume weighted volatility: Empirical evidence for a new realised volatility measure. International Journal of Banking, Accounting and Finance. ↩︎
6. NBER WORKING PAPER SERIES HOW AND WHEN ARE HIGH-FREQUENCY STOCK RETURNS PREDICTABLE? (2022) Yacine Aït-Sahalia Jianqing Fan Lirong Xue ↩︎
7. Mann-Whitney is a non-parametric test, which is more appropriate than a t-test given the highly skewed distribution of daily trading volume by ticker. ↩︎
8. EUB and EDB are probability weighted average prices between the model date price and 95% OaR, and between the model date price and 95% VaR, respectively ↩︎
9. volume and price volatility are driven upward by the arrival of new information, as per Clark and Epps, above. ↩︎
10. Sigma is the standard deviation of returns—the foundational volatility concept used in the Black-Scholes option pricing formula and introductory finance courses. VecViz calculates Sigma using daily log returns over a two-year lookback period with a 6-month half-life decay in the weighting accorded to each observation. ↩︎
11. the Mann-Whitney test statistic p-value was 0.12 ↩︎
12. given the highly skewed distribution of the Volume Ratios across tickers Spearman (rank) correlation is more appropriate than Pearson correlation (which is heavily influenced by outliers). ↩︎
13. with 90% confidence ↩︎
14. See the “PosNegDiff” row in the tables below and multiply by 26 for the 10d (2 week) horizon, 12 for the 21d (1 month) horizon, 4 for the 63d (3 month) horizon ↩︎