Next-Day Movers vs a Simple Implied-Volatility Screen

Benchmark Next-Day Movers against a same-date high-IV/HV baseline using absolute next-session movement, equal universe rules, and fully settled observations.

Next-Day Movers and a simple implied-volatility screen answer different versions of the movement-selection problem. Alphanume's Next-Day Movers dataset provides a fixed daily list from a maintained volatility model and later records signed return plus absolute next-session movement. A high movement candidate means elevated expected magnitude, not expected direction.

The comparison baseline can use the top daily decile from the IV/HV Premium dataset. Its field reference defines ratio ranks and finality. Alphanume's proof material gives broader evaluation context. The benchmark must use the same dates, eligible universe, observation finality, and outcome definition or it will compare data construction rather than models.

Define the two selection rules

The maintained model supplies its selected tickers for each list date. The baseline selects tickers whose settled iv_hv_ratio_ranked is at least 0.90 on that same date. Decide whether to match the number of baseline names to the model list or retain the full decile. Report the choice and overlap.

Use absolute movement as the primary outcome because both screens concern magnitude. Signed return can be reported descriptively but should not be interpreted as a direction forecast. Define the return close convention and corporate-action adjustment consistently for both groups.

Join selections on completed dates

Retrieve a completed historical range from Next-Day Movers and the same dates from IV/HV Premium with final rows only. The newest mover outcome remains null until the following session completes, so exclude it from realized statistics without removing the selection record.

for each completed date:
  movers = GET /v1/next-day-movers?date=<date>
  iv = GET /v1/iv-hv-premium?date=<date>&only_final=true&min_ratio_rank=0.90

  model_set = set(movers.ticker)
  baseline_set = set(iv.ticker)
  overlap = intersection(model_set, baseline_set)

  evaluate both sets on identical next-session absolute returns
  retain null outcomes and exclusion reasons

Do not use the Next-Day Movers absolute_move for one group and a differently sourced or timed outcome for the baseline. Build one outcome table if possible, then join both selection flags. This prevents subtle discrepancies in sessions, adjustments, or missing tickers.

Compare coverage, overlap, and tails

Report mean and median absolute move, hit rates above prespecified thresholds, and tail quantiles. Show event counts and unique ticker counts. Because the same name can recur, use issuer-clustered or date-clustered uncertainty rather than treating every row as independent.

• Publish the daily list size and baseline size.
• Show overlap as its own cohort.
• Report missing outcomes instead of dropping them silently.
• Apply identical liquidity and price-coverage rules.

The baseline can be rationally rich because of known event risk. Stratify earnings and other scheduled catalysts rather than calling high IV a naive mistake. The maintained model may also select event names, so the event-control rule must apply to both sides.

Avoid leakage and post-selection tuning

The main failure mode is using settled end-of-day IV to simulate a trade before that value was final. Choose a next-session execution rule for both screens. Another failure is tuning the IV percentile after observing which threshold makes the model look strongest. Predefine the top decile or publish a complete sensitivity grid.

Do not treat the newest null outcome as zero. Historical mover selections are fixed, while their outcome fields populate after the next session. Freeze an evaluation cutoff and report how many selections are awaiting outcomes.

Universe alignment requires more than matching the date. The IV/HV cross-section covers liquid optionable equities, while the maintained model may apply its own eligibility rules. Define the comparison universe as the intersection when testing ranking skill, then also show coverage outside the intersection. Otherwise a difference in average movement can arise because one method sees securities the other cannot select.

List-size matching is another choice with real consequences. Comparing a short model list with an entire top decile gives different breadth and weighting. A clean primary test can take the top ratio-ranked names equal to the model's daily list size, with the full decile as sensitivity. Break ties using a documented raw measure and never use realized movement to decide the cutoff.

Daily aggregation should weight dates equally as well as events. A few sessions with unusually broad baseline coverage can dominate an event-weighted mean. Publish event-weighted and date-weighted results, plus the distribution of daily differences. This reveals whether any apparent advantage is persistent or concentrated in a small number of volatile market sessions.

Run one matched benchmark table

As a concrete next action, retrieve sixty completed dates, construct the top settled IV/HV decile for each, and build one combined ticker-date outcome table. Add model-selected, baseline-selected, and overlap flags. Report equal-weight absolute-move statistics, tail hit rates, missing outcomes, and event counts under one timing convention. That matched table tests whether the maintained model adds information beyond the obvious volatility baseline.