VaR Performance Report Summary: 2/1/2025

VecViz’s analytic performance reports document the behavior and performance of our analytics. This post focuses on the “VecViz Value At Risk (VaR) Performance Report” as of February 1, 2025, Here we will summarizes the key findings, and explain how to use the report to investigate the performance of specific tickers. You can find this report on the “Reports” section of the VecViz website, along with updated reports over time.

VaR is the maximum amount you could lose by being long a ticker as of a specified date in the future at a specified level of probability.  VaR is likely of interest to risk adverse/ underinvested investors, as a tool to help them imagine the downside they are avoiding. VaR should be of interest to risk exposed investors, and especially those subject to margin. VaR is also relevant to investors in the options market, although less so than our Option Fair Value Estimates (they too have an associated Performance Report).

Performance Metrics Utilized in the VaR Report

The VaR Performance Report compares the behavior and performance of 95% and 99% VaR calculated by VecViz’s Vector Model to the same as calculated by VecViz’s implementation of the Sigma Model. The primary performance metric is the Breakage Rate, and the secondary metric is the Return on VaR Based Capital (ROVBC).¹

VaR Breakage Rate

VaR breakage occurs when actual losses exceed the VaR estimate. The primary goal of a VaR estimate is a breakage rate consistent with stated intentions: 5% for 95% VaR and 1% for 99% VaR. Ideally, breakage would be randomly distributed across model dates and tickers., and so we also evaluate the consistency of the breakage rate accordingly.

ROVBC

For a given average breakage rate, a VaR estimate that is more conservative for tickers with below-average returns and less conservative for tickers with above-average returns is preferable. Such an estimate would yield attractive ROVBC (see our FAQ or the report for a definition). Maximizing ROVBC is an important, though secondary, criterion for evaluating VaR performance.

The VaR report considers the “alpha” of the Vector Model relative to Sigma² as, detailed in the tables below, to help determine whether differentials in ROVBC performance are due to VaR differentials across tickers or relative changes in VaR over time.

95% & 99% VaR Summary Tables

We summarize how well the Vector Model fits these criteria relative to Sigma in the table below, for each of the six time horizons (denominated in trading days, where 21d ~= 1 month, 252d ~= 1 year).  Where the Vector Model better fits the criteria we enter a “V”, and where Sigma is superior, we enter an “S”.  For sake of brevity, we present here results only for the entire out of sample performance record (“ALL TMD” = all ticker model dates, but the complete report also specifically addresses the trailing 1 year, 3 month, and 1 month time lookback windows.

95% VaR, ALL TMD (all ticker-model dates and associated forward horizon performance to the extent passage of time allows from 1/31/2022 – 1/30/2025)	1d	10d	21d	63d	126d	252d	1d V / S	252d V /S	page(s)
Average VaR Estimated Loss Severity	V	V	S	S	S	S	-5.4 / -4.0	-47.3 / -56.2	16
Average Breakage Proximity to 5.00%	V	V	V	V	V	V	4.78 / 4.06	5.52 / 2.38	27
Breakage Consistency Across Model Dates	V	V	S	S	S	S	na	na	35, 37, 39, 41, 43, 45
Breakage Consistency Across Tickers	S	S	S	S	S	S	na	na	47-52
Average ROVBC	S	S	S	S	V	V	0.03 / 0.04	19.80 / 15.08	28
VM ROVBC Alpha Across All TMD’S	-0.01	-0.13	-0.28	-0.95	-1.60	-3.01	na	na	28
Average VM ROVBC Alpha By Ticker, Across All Model Dates	0.00	0.11	0.24	0.72	1.55	2.31	na	na	28

99% VaR, ALL TMD  (all ticker-model dates and associated forward horizon performance to the extent passage of time allows from 1/31/2022 – 1/30/2025)	1d	10d	21d	63d	126d	252d	1d V / S	252d V /S	page(s)
Average VaR Estimated Loss Severity	V	V	V	V	V	S	-11.3 / -5.7	-67.1 / -72.9	111
Average Breakage Proximity to 1.00%	V	V	S	S	S	S	1.18 / 1.46	2.33 / 1.23	122
Breakage Consistency Across Model Dates	V	V	V	S	S	S	na	na	130, 132, 134, 136, 138, 140
Breakage Consistency Across Tickers	S	S	S	S	S	S	na	na	142-147
Average ROVBC	S	S	S	S	S	V	0.03 / 0.04	17.09 / 15.08	123
VM ROVBC Alpha Across All TMD’s	-0.01	-0.10	-0.22	-0.66	-1.23	-1.18	na	na	123
Average VM ROVBC Alpha By Ticker, Across All Model Dates	0.01	0.09	0.22	0.72	1.47	1.29	na	na	123

Summary Conclusions:

Over the period studied, which includes both the bear market of 2022 and the bull market that followed up through most of January 2025, both 95% and 99% Vector Model VaR levels:

1. were typically more conservative than Sigma VaR for shorter time horizons, and less conservative for longer time horizons;
2. had breakage rates closer to target for every horizon at the 95% level, and for the shorter time horizons at the 99% level. Note that if the Failed Banks³ are excluded Vector Model 99% VaR breakage was also closer to target than Sigma for all horizons;
3. exhibited breakage rate consistency across tickers inferior to Sigma;
4. exhibited greater breakage rate consistency across model dates than Sigma;
5. had higher ROVBC than Sigma for longer term horizons and lower ROVBC for shorter term horizons. Note that if the Failed Banks are excluded Vector Model ROVBC was greater than or equal to Sigma for all horizons;
6. had positive ROVBC alpha by ticker across model dates on average, but negative ROVBC alpha when measured simultaneously across model dates and tickers, suggesting poor calibration in Vector Model VaR for certain tickers (consistent with comments above about the impact of the failed banks on breakage and ROVBC results)

Appendix: Ticker Level VaR & ROVBC Performance Report Inquiry

When we press “Ctrl-F” on the VaR report pdf file and enter “NVDA”, for example, we learn that it appears 171 times in the document. Clicking through the search results, we learn that NVDA was (for most or all horizons and lookback windows unless otherwise noted) a

• Top 30 Ticker with regard to: Sigma 95% and 99% VaR Breakage, 95% and 99% Vector Model and Sigma ROVBC
• Bottom 30 Ticker with regard to: Vector Model VaR Breakage, Vector Model vs Sigma VaR Differential for the 1d thru 63d time horizons (i.e., the Vector Model forecasted steeper potential losses), Vector Model vs Sigma VaR Breakage (Vector Model had lower VaR Breakage), Vector Model vs Sigma ROVBC for most time horizons (Vector Model had lower ROVBC).

For further ticker level inquiry visit the Dashboards page. There you can:

1. view a chart of 95 and 99% VaR for the Vector Model and Sigma for the entire out of sample period, for each of the six time horizons;
2. view a table that provides the 95% and 99% VaR breakage and ROVBC for any ticker (or group of tickers) for an evenly spaced sampling of 20 model dates across the entire out of sample back testing / live daily production period;
3. see how VaR has varied with the issuer’s V-Score over those same 20 evenly spaced model dates spanning the entire out of sample back testing/ live daily production period;

Note that the dashboards are best viewed on a desktop or laptop, that there are 8 of them, they load progressively as you scroll down, with a few of them taking 15-30 seconds to initially load. After the initial load they load much quicker as you toggle between tickers, horizons, etc.

1. See our FAQ page for definitions of the Vector Model, Sigma, VaR, ROVBC, and many other terms. ↩︎
2. We ascribe underlying ticker-model date forward horizon price returns to Sigma, so the Alpha and Beta can also be said to be relative to the underlyting ticker-model date forward horizon price returns as well. ↩︎
3. Three major regional banks that failed in 2023 (SIVBQ, SBNY, and FRCB) are included in the results presented. All three banks saw their shares suddenly drop to nearly zero in 2023 due, in part, to the acceleration of depositor flight by social media, a dynamic that was arguably beyond the scope of the Vector Model’s training data. VecViz’s training data for the model results presented in this report ends from a model date basis in January 2021, and from a forward horizon basis in January 2022. See the Appendix of the report for detailed results excluding the Failed Banks, and several other ticker groupings. ↩︎