TORQ Gamma Exposure (GEX): What It Measures

TORQ compresses a large options surface into one comparable reading. That sentence is simple. The boundary behind it is not.

Some parts of TORQ come from dated market data. Other parts must be estimated because public option-chain data does not reveal every dealer's complete position, economic owner, or hedge. The calculation is deterministic—it is not an AI prediction—and uses a consistent set of implementation rules.

This article explains what enters the instrument, what remains modeled, when the data would have been available, how the historical series was built, and what the research does and does not support. It is a methodology overview rather than a line-by-line technical specification, so some detailed calculation rules are not displayed.

What does gamma exposure measure?

Gamma exposure estimates how quickly aggregate option delta may change when SPX moves. It translates position-level gamma into a market-level sensitivity estimate for a defined options universe.

Delta describes how an option's value responds to a change in the underlying. Gamma describes how quickly that delta changes. When a delta-hedged position's delta changes, the holder may need to rebalance its hedge.

The familiar market intuition is:

The words modeled and may matter. Gamma exposure describes a possible hedging mechanism. It does not prove that this mechanism will dominate news, futures flow, volatility repricing, liquidity conditions, or positions outside the model.

What market data goes into TORQ?

TORQ begins with a dated snapshot of the short-dated SPX options market. The inputs describe what is listed, how the options are positioned in the chain, and how sensitive they are around the observation time.

At a public level, the input categories are:

The complete calculation contains additional inclusion, data-quality, transformation, and normalization decisions. Those low-level implementation details are not displayed in this overview.

The useful transparency boundary is therefore not “every line of the formula” versus “trust us.” It is knowing which inputs are observed, which conclusions are modeled, when the snapshot existed, and which claims the result can support.

Is dealer long-versus-short gamma directly observed?

No. Public open interest reports how many contracts remain open; it does not disclose the complete economic owner and direction of every position. TORQ therefore estimates dealer positioning through a consistent model.

That distinction limits every GEX-style measure built without a complete dealer ledger. The same open contract can sit inside a larger position with other strikes, expirations, SPY options, futures, or over-the-counter exposure. Market makers may also hedge options with other options rather than only with futures.

TORQ does not claim to reconstruct all of those books. It applies a repeatable set of positioning and aggregation rules to a defined market-data surface. This overview shows the main components without displaying every sign, weighting, transformation, and exception-handling rule.

This also explains why two gamma providers can publish different values from the same trading day. GEX is not an exchange-published fact. Scope, snapshot time, positioning assumptions, data cleaning, normalization, and aggregation all change the result.

Is TORQ trained by artificial intelligence?

No. TORQ is a deterministic calculation, not a machine-learning model generating a forecast from strategy outcomes.

The same valid inputs and the same version of the calculation produce the same reading. No daily profit and loss, winning-trade label, or later market outcome enters that morning's TORQ value.

There is still a separate research issue: Cashflow Engine evaluated multiple gamma-index constructions historically before selecting the production instrument. That is model development, even though it is not machine learning. Evidence measured on overlapping historical data must therefore be described as development-sample evidence rather than fully independent confirmation.

That is why the validation process matters as much as the formula. A deterministic instrument can still be over-selected if too many constructions and thresholds are compared on the same outcomes.

Are historical TORQ readings strictly point-in-time?

The inputs are dated to the premarket information set for each session, but the oldest readings were reconstructed later rather than recorded live on those historical mornings. Those are related but different claims.

The daily design combines a previous-session options sensitivity surface with the open-interest information available before the new session opens. It does not use the new session's realized trading outcome.

For dates before the production collector existed, Cashflow Engine rebuilt TORQ from historical snapshots carrying those dates. For current sessions, the production process captures the reading contemporaneously and stores timing and provenance with it.

“Point-in-time inputs” should therefore not be expanded into the stronger claim that every historical number was archived live at the time. Cashflow Engine keeps reconstructed and contemporaneous provenance separate precisely because the distinction matters.

Why does TORQ use one premarket reading per session?

A fixed daily observation favors comparability across sessions. It answers a different question from a continuously updating intraday gamma estimate.

An intraday surface changes as SPX moves, new trades arrive, implied volatility changes, and same-day options approach expiration. A live estimate may better describe the current minute, but it is not directly interchangeable with a consistently timed premarket series.

TORQ's design asks: What did the modeled short-dated gamma backdrop look like before this session began, measured through the same production process?

It does not ask: What is the complete dealer position at this instant?

This distinction also helps explain why a premarket association could be stronger for morning entries than for afternoon entries. By later in the session, the market state may have changed materially from its starting snapshot. That is a hypothesis to test, not a guarantee.

What did Cashflow Engine's historical research find?

The historical research found a directional relationship between TORQ and strategy outcomes across more than a dozen tested strategy and portfolio series, but the size and risk impact varied. Stronger controls made several early claims smaller.

The durable public conclusions are deliberately limited:

• the historical relationship was not universal across portfolios or entry times;
• portfolios already optimized for consistency tended to show less incremental value;
• TORQ did not reliably identify the worst loss sessions;
• apparent drawdown improvements can be mechanical when a rule simply uses less exposure;
• same-exposure controls, era normalization, and influential-session checks materially changed the interpretation; and
• the studied relationship remains context, not proof of causation.

Internal research evaluated the relationship across more than a dozen strategy and portfolio series and two de-identified live books. Several first-pass effect claims were reduced or retracted after stronger controls.

The live-book comparisons support external relevance, but they were not pristine, large-sample prospective trials. Cashflow Engine therefore records current research decisions before the outcome is known and treats the accumulating prospective sample as the cleaner test.

Multiple instrument constructions and hypotheses were examined during development. Results can be affected by selection bias, market-regime change, transaction assumptions, and influential sessions.

How should the reported t-statistics be interpreted?

A t-statistic describes an estimated difference relative to its estimated uncertainty. It does not measure the dollar size or practical importance of the difference.

In the community analysis, a negative t-statistic means the tested low-TORQ cohort had a lower estimated outcome than the comparison cohort. An absolute t-statistic below roughly two generally provides limited evidence against a zero-difference model when viewed as a single conventional test.

Three cautions matter:

1. Not significant does not mean equivalent. A small t-statistic does not prove the true effect is zero or economically negligible. That requires a confidence interval and a pre-defined smallest effect that matters.
2. The table contains many overlapping tests. Entry windows, TORQ thresholds, and strategy selections share observations. An isolated cell is weaker evidence than a pre-specified pattern that survives an appropriate multiple-testing control.
3. The outcome is strategy-specific. These tests ask whether the analyzed strategy results differed across TORQ cohorts. They do not establish that TORQ predicts every market outcome.

The responsible summary is that the shared tables showed more consistent historical separation in morning entries and little incremental separation in the afternoon and last hour. They do not justify the statement “the later entries are proven unaffected, therefore trade them.”

What can TORQ tell you—and what can it not tell you?

TORQ can provide a consistent lens on the estimated premarket gamma backdrop. It cannot remove uncertainty from the trading day.

TORQ does not replace defined loss, buying-power limits, execution controls, portfolio diversification, or Monte Carlo simulation. Some severe sessions looked ordinary in advance. Risk architecture still carries the tail.

How much of the TORQ calculation is displayed?

Cashflow Engine displays the main data inputs, timing, modeled assumptions, output unit, validation approach, and limitations. The detailed production calculation is not shown line by line.

Public documentation includes:

• the instrument's purpose, unit, and premarket cadence;
• the main categories of observed market inputs;
• the modeled nature of dealer positioning;
• the distinction between historical reconstruction and live capture;
• the high-level validation design and its weaknesses; and
• the conclusions TORQ does and does not support.

This overview does not display:

• the final formula or calculation sequence;
• complete contract-inclusion rules;
• sign conventions and positioning logic;
• relative weights, transformations, or normalization;
• internal data-quality and fallback rules;
• thresholds, percentile mappings, or regime labels; or
• sizing and decision logic.

This level of detail makes the reading understandable and the claims testable while keeping the article focused on interpretation rather than implementation.

Related reading and sources

• Explore TORQ in the Workbench
• Cboe: Evaluating the Market Impact of SPX 0DTE Options
• Cboe: 0DTE Positioning, Trends, and Market Impact
• American Statistical Association statement on p-values
• SPX vs. SPY options
• SPX settlement explained
• How automated options trading works