I named a model after myself. A 1993 model beat it.
Research note #1 · The Dimopoulos Chain, v1 · July 2026
Black-Scholes prices everything with one constant volatility. Markets don’t work that way: volatility arrives in regimes, and one structural driver of those regimes is options dealer positioning — when dealers are long gamma their hedging pins the market, and when they are short gamma the same hedging amplifies every move. Our pipeline had already validated that signal (it predicts next-day trading range beyond the VIX, t = 7.4). So we promoted it to a model and gave it a name: the Dimopoulos Chain — a Markov chain over nine market states (dealer-gamma tercile × realized-vol tercile) whose state-conditional distributions forecast tomorrow’s volatility.
House rule: a name is earned Sharpe-style, by surviving gates — or the model gets an exhibit here under its own name. Out-of-sample, 2019–2026, 1,794 days, all forecasts refit monthly with no look-ahead. The ladder, strictest last:
| Chain states add information beyond… | Newey-West t | gate ≥ 2.5 |
|---|---|---|
| VIX (the market’s implied forecast) | 4.34 | PASS |
| GARCH(1,1) | 3.19 | PASS |
| HAR (realized-vol workhorse) | 2.03 | below gate |
| HAR + VIX + GJR-GARCH combined | 1.02 | FAIL |
GJR-GARCH — Glosten, Jagannathan and Runkle’s 1993 asymmetric volatility model — turned out to be the single best forecaster in the stack, and once it enters the baseline, the Chain’s states add nothing significant. The likely reason is the interesting part: the leverage effect GJR captures (down moves raise volatility more than up moves) is plausibly a reduced form of the dealer mechanism itself. Markets fall → investors buy puts → dealers go short gamma → volatility amplifies. A 33-year-old econometric model was already pricing the shadow of the structure we measured directly.
What survived
Two things, and they matter more than the scoreboard:
1. Every decent forecast in the stack — GJR, HAR, the Chain, combinations — beats the implied forecast by a wide, significant margin. Options are traded against implied, not against your best econometric model; that gap is the variance risk premium, and its timing is the live research program.
2. The state map cleanly locates the danger. Measuring implied-minus-realized variance by state: the premium is positive in eight of nine states (hit rates up to 90% when dealers are pinned) and negative in exactly one — dealers exposed and volatility already high. One state out of nine is where short-volatility strategies die, and it is identifiable a day in advance.