The regime
model.
A regularized logistic model estimates the probability that QQQ will have a positive return 45 calendar days ahead. The probability is translated into a score from −100 to +100.
Analysis date
October 1, 2026
Regime score & QQQ close
Gold line: trailing five-observation average · Jul 06–Aug 07: frozen-model backfill · Aug 10–Oct 01: production history
Four factors enter the logit.
Each current value is standardized against the frozen training sample, multiplied by its coefficient, and added to the intercept.
How to read the scores and calculationsExpand
Current value is the latest observed input. Standardized value (σ) is (current − training mean) ÷ training standard deviation; positive is above the frozen training average and negative is below it.
Logit contribution, shown at the right of each bar, is standardized value × frozen coefficient. A positive contribution raises the model’s estimated probability; a negative contribution lowers it. Magnitude shows influence on the current calculation, not causal importance.
High-yield OAS
Lowers current probability · -0.247The option-adjusted spread on below-investment-grade debt. It measures the extra yield demanded over comparable Treasuries and serves here as a credit-stress input.
VIX term structure
Raises current probability · +0.005The three-month VIX level minus spot VIX. It describes the shape of the implied-volatility term structure rather than the absolute level of volatility.
RSP / SPY breadth
Lowers current probability · -0.031The 20-session change in equal-weight S&P 500 performance relative to the capitalization-weighted S&P 500. It measures whether participation is broadening or narrowing.
QQQ 200-day slope
Raises current probability · +0.720The annualized rate of change of QQQ’s 200-session moving average. It represents the direction and pace of the long-horizon price trend.
From factor contributions to today’s regime score
Logit = 1.7437 − 0.247 + 0.005 − 0.031 + 0.720 = 2.1909
Probability = 1 ÷ (1 + e−logit) = 89.94%
Regime score = 200 × 0.8994 − 100 = +79.89
Probabilities are bounded from 10% to 90%, so the score ranges from −80 to +80. Display bands are: bearish below −50, moderately bearish from −50 to −9, neutral from −9 to +9, moderately bullish from +9 to +50, and bullish above +50. These labels summarize model output; they do not prescribe an action.
Regime score versus realized 45-day QQQ return
Each point pairs a historical score on its origin date with QQQ’s subsequently realized 45-calendar-day return. Only observations with complete outcomes are included.
Historical descriptive sample: Mar 26, 2026–Jun 23, 2026 origin dates · Pearson r = 0.90 · latest completed endpoint 2026-08-07
At the current score, the fitted historical average subsequent 45-day QQQ return is +15.98%. Historical relationship only—not a forecast or investment recommendation.
The best model wins ticker by ticker.
For each supported ticker, purged walk-forward Brier score chooses among three candidates. A direct refit controls production only when it beats both simpler benchmarks out of sample.
The direct four-factor score is always retained for audit—even when a benchmark is selected.
Production universe
139 published tickers
What the model does—and does not do.
Frozen specification
Daily scoring never refits the model. Retraining requires an explicit research action.
Walk-forward evidence
Validation is purged and out of sample; the production QQQ Brier score is 0.1709 across 378 predictions.
Probability, not prophecy
The output estimates the chance of a positive 45-day QQQ return. It is not a short-term trend forecast.