Independent intermarket research

Volatility is forecastable. Direction isn’t.

We tested conditioning across 5+ independent market states. The same result every time: the state moves volatility and correlation hard — a next-session volatility R² near 0.6 out of sample — and moves direction not at all. That is the whole reason intermarket signals keep failing. You have been asking one market to predict another's direction. Markets only tell you about its size.

The second moment bends to the state; the first moment doesn’t. Measured across volatility regime, session, quarter phase, dealer-gamma regime and inflation regime. Dealer-gamma conditioning alone hits a t-statistic near 9 on realised range. Direction stays null in all of them.

~0.6out-of-sample R² forecasting next-session volatility (HAR + implied)

nulldirection, conditioned on any of 5+ states

2,700+strategies & indicators audited the same way

The second-moment law, in one line

Condition on a market state — a high-vol regime, a session, a dealer-gamma setting — and you reshape the second moment: volatility, correlation, the dependence structure. You do not touch the first moment: which way price goes next. This is not our idea. It is one of the most durable results in the academic literature — the Dacco-Satchell and Guidolin-Timmermann finding that regimes help variance and allocation and fail for directional signals. We just confirmed it on our own data, five ways, and it explains the null result under the whole intermarket genre.

Every "when X is red, buy Y" rule is a direction bet dressed as an intermarket insight. That is the moment it dies. The relationship is real — it lives in the covariance — but it was never pointing at a candle's color.

Markets tell you how big the next move is. They do not tell you which way.

What conditioning actually moves

We ran the same test across five-plus states: volatility regime, trading session, quarter and calendar phase, dealer gamma (GEX), and inflation regime. In every one, the state governs realised volatility and cross-asset correlation with real statistical force. Dealer-gamma conditioning on realised range comes in near a t-statistic of 9. That is not a marginal effect you talk yourself into. It is a wall.

Then we pointed the identical machinery at direction. Nothing. A flat null in all five states, with nothing left to size down and salvage. The state that screams about volatility goes silent about direction. Same data, same window, opposite verdict.

What that buys you on the tradeable side

The second-moment law is not a dead end. It is a redirection. If the size of the move is what conditioning predicts, then size is what you build on.

Volatility is genuinely forecastable. A standard HAR model plus option-implied vol reaches an out-of-sample R² near 0.6 for next-session realised volatility. Direction models on the same inputs never clear their own noise.
The volatility-risk-premium is real. Implied vol sits above subsequently realised vol in about 85% of months. A matched-maturity short-vol carry runs a Sharpe near 0.74. This is a paid-for edge, not a chart pattern.
Crypto is a free, always-on volatility gauge. A 24/7 read on global risk. Useful where option-implied vol is missing — but for well-instrumented markets it is subsumed by the implied vol you already have.

The catch on the vol premium: read the tail before you sell it

A Sharpe near 0.74 looks clean until you see how it is made. Short-vol carry earns a trickle of premium most months and then hands back a crater in one. The return distribution has severe negative skew — it is tail-concentrated the wrong way. A single volatility spike can erase a year of carry in a session. The premium is genuine. The path is a landmine. It is tradeable only with hard tail-sizing, and it is the opposite of passive.

The myth we busted: "sell vol into the Fed"

Popular retail thesis: short the straddle into the Fed decision and pocket the volatility crush. The implied crush is real — implied vol does collapse after the announcement. The tradeable premium is not there. On our data, the Fed day is a below-average day to be short volatility: the event's realised move offsets the crush, so the Fed-day short-vol premium sits under the everyday premium by roughly 5 to 7 basis points of excess.

Read that again, because it inverts the pitch. You are taking extra risk on the one day everyone tells you to sell, for less premium than a random Tuesday. The vol-risk-premium is unconditional. Harvest it continuously, not around events. If you want the intermarket edge that did survive event-timing, it is a drift, not a crush — the pre-FOMC drift, covered separately.

How we test the second moment honestly

Volatility and correlation are easy to fool yourself with, because heteroskedasticity fakes a relationship where none exists. So the vol side runs through a three-part gate: a Forbes-Rigobon adjustment for the correlation bias volatility itself injects, GARCH-devolatilised residuals so we are not just measuring shared variance, and a scale-invariant dependence measure that survives when the simple correlation lies. Direction is benchmarked against a volatility-null so a signal has to beat noise, not just look busy. The vol figures are out-of-sample and walk-forward, 2012 to 2026, matched-maturity and cost-aware where it matters.

Underneath sits the same engine as the rest of the audit. Futures from 13 years of CME data, FX on tick-level real bid/ask, liquidity-aware stocks, crypto spot and perps. A fast model does the bulk work; the strongest model tries to break every apparent survivor, hunting look-ahead and impossible fills. Over 2,700 strategies and indicators have gone through it. Most do not come out the other side.

Research and education, not financial advice. No signals, no return promises. Independent, and not affiliated with TradingView.

The relationships, named — with the verdicts

This page gives you the law. What it does not give you is the ledger: which specific intermarket relationships and volatility rules we tested, which held, which broke, and the exact after-cost numbers behind each call. That is The No List — every strategy and indicator we audited, named, with its verdict and the reason it lived or died.

Get The No List → Not ready to buy? The method, the misses and the running results are free in our Discord ↗.

FAQ

Can intermarket analysis predict market direction?

Not in our testing. Across 5+ independent market states, conditioning had zero power over next-move direction — a flat null every time. What it does predict is volatility and correlation, the second moment, not the first.

If direction is a null, what is intermarket data good for?

Volatility and correlation. A HAR-plus-implied model reaches an out-of-sample R² near 0.6 forecasting next-session volatility, and the vol-risk-premium (implied above realised in ~85% of months) is a real carry edge — with a nasty tail.

Should I sell volatility into the Fed decision?

On our data, no. The Fed day is a below-average day to be short vol: the realised move offsets the implied crush, leaving the short-vol premium about 5 to 7 basis points below the everyday premium. Harvest the premium continuously, not around events.

Is the volatility-risk-premium safe to trade?

It is real — a matched-maturity short-vol carry runs a Sharpe near 0.74 — but the return distribution has severe negative skew. One volatility spike can erase a year of carry. Tradeable only with hard tail-sizing, and never passively.