Vol-of-Vol Stock Screener for Unstable Volatility

Screen stocks by the instability of their own implied and realized volatility over 21 observations, using settled cross-sectional ranks rather than confusing the measure with VIX-of-VIX.

A vol-of-vol stock screener ranks how unstable each optionable equity's volatility has been. Alphanume's Vol-of-Vol Index dataset calculates the coefficient of variation for roughly 30-day implied volatility and 30-day realized volatility across the trailing 21 observations. It also serves the component mean and standard deviation plus daily cross-sectional percentile ranks and z-scores.

The Vol-of-Vol documentation defines the fields, window, filters, and finality flag. This is a cross-name equity measure. It is not the volatility of the VIX index, not VIX-of-VIX, and not a directional forecast for the underlying stock. Alphanume's published evidence is deliberately qualified: across 280,105 settled rows, full-distribution correlation with forward realized volatility is 0.05 and the deciles are non-monotonic. The supported result is concentrated in the extreme tail.

Understand the coefficient before ranking

The raw measure divides the trailing standard deviation of volatility by its trailing mean. That makes it dimensionless and more comparable across names than the standard deviation alone. It can still become large when the mean is low, so preserve the component values and avoid interpreting the ratio without its denominator.

The first 20 observations of a ticker's history are absent while the 21-observation window warms up. The twenty-first observation is the first eligible value. That missingness is mechanical, not evidence of stable volatility. Cross-sectional comparisons should report how many names were eligible on each date.

Retrieve final high-instability names

For a historical screen, request a completed date with only_final=true. The endpoint supports floors and ceilings on raw implied and realized vol-of-vol, plus percentile floors. A top-decile screen can use min_iv_vov_rank=0.9. Retain the raw measure, components, current IV and HV, and notional-volume context.

GET /v1/vol-of-vol
date=<completed trading date>
only_final=true
min_iv_vov_rank=0.90

retain ticker, iv, iv_vov, iv_mean_21, iv_std_21,
       hv, hv_vov, hv_mean_21, hv_std_21,
       n_obs_vov, notional_volume,
       iv_vov_cs_ranked, hv_vov_cs_ranked, is_final

Today's rows update intraday and carry is_final=0 until settlement after the close. An intraday ranking can change. Store last_updated and the exact retrieval time for live decisions, or require settled values for research intended to be reproducible from closing data.

Separate catalyst churn from persistent instability

A single earnings cycle, regulatory decision, or takeover rumor can push volatility through several observations and elevate the 21-day coefficient. Persistent instability should remain high across several completed screens. Build consecutive-day counts or rolling rank summaries, but use only previously settled observations at each decision time.

• Flag earnings and other scheduled catalysts separately.
• Compare implied and realized instability rather than using IV alone.
• Keep the 21-day means beside the coefficient ratios.
• Require contemporaneous option quotes before modeling execution.

A name with unstable IV but stable HV may reflect changing forward event pricing. A name with both measures unstable may be in a changing realized regime. Those are research descriptions, not buy or sell instructions, and neither predicts the direction of the next stock move.

Persistence should be measured on ranks and raw values separately. A ticker can remain in the top decile because its coefficient stays elevated, or because the rest of the cross-section becomes calmer. Record the daily universe size, coefficient, and percentile, then require a minimum number of qualifying days within a fixed window. This reveals whether the screen is finding sustained instability or only relative rank movement.

Cross-sectional z-scores provide another view of extremeness, but they can be dominated by a few outliers. Compare percentile and z-score selections and publish their overlap. If a winsorization or mean floor is applied, make it a research transformation outside the served fields and preserve the unmodified source values for audit.

Control denominator, liquidity, and finality failures

The main numerical failure is ranking a high coefficient caused by a very low mean as though it reflected a large absolute volatility change. Apply a prespecified mean floor or show both ratio and standard deviation. The main timing failure is mixing provisional current rows with final historical rows.

Underlying notional volume is a liquidity reference rather than an option-market quote. A trade simulation still needs the actual contract, bid, ask, displayed size, and fill convention. A vol-of-vol screen can prioritize names for that analysis but cannot establish executable vega exposure.

Audit one persistent-instability screen

As a concrete next action, retrieve twenty completed dates with final rows, select the daily top decile, and count how often each ticker qualifies. Export IV and HV coefficients, their means and standard deviations, ranks, eligibility count, and catalyst flags. Review the actual IV path for several persistent and one-day names. This separates repeated instability from a single event-driven burst without turning the ranking into a directional forecast.