Help us double down on what's working, instead of guessing. Takes 5 seconds, totally optional.
alternativedata · 8 min read
Dealer Positioning as Alternative Data: An Evaluation Guide for PMs
Dealer positioning - gamma, delta, vanna and charm exposure computed from the options market - is an alternative data family with a mechanical causal story, daily-to-minute frequency, and a published methodology. This guide evaluates it the way an allocator evaluates any alt-data set: signal content, orthogonality, capacity, decay, point-in-time discipline, and integration cost.
Search for alternative data for hedge funds and you will find satellite imagery, card panels, and web-scraped exhaust. What you will rarely find catalogued is dealer positioning data - the gamma, delta, vanna and charm exposure of options market makers - despite it having the property most alt data lacks: a mechanical reason to matter.
The one-paragraph thesis: options dealers must hedge what they carry. Their aggregate exposure, computable from open interest and greeks, determines whether their re-hedging flow dampens moves (long gamma: sell rallies, buy dips) or amplifies them (short gamma: sell dips, buy rallies). That flow is not a statistical artifact - it is an obligated participant whose direction and size can be estimated from public options data, continuously.
4
Signal family: gamma, delta, vanna, charm exposure
1-min
Native frequency, live and in historical replay
8 yrs
Backtested exposure series with VIX-controlled results published
100%
Methodology published - no black-box scores
What the dataset actually is
Four exposure series per underlying, each derived from the options chain with a stated dealer-side convention:
Gamma exposure (GEX) - the flagship: how much delta dealers must trade per 1% move. Sign sets the regime (dampening vs amplifying); the strike profile locates walls and the flip. Primer: what is GEX.
Delta exposure (DEX) - the directional inventory dealers carry; context for how forced their hedging is.
Vanna and charm - how dealer deltas shift with vol and time; the second-order flows behind pinning, expiration drift and vol-spike unwinds. Guide: vanna and charm.
Two methodology choices separate research-grade from decorative versions of this data, and both are published here: how open interest is signed (see the effective-OI methodology, which infers positioning from live flow rather than assuming every contract is dealer-long), and whether the series distinguishes settled overnight exposure from intraday flow-adjusted exposure. Ask any vendor of positioning data those two questions first.
Evaluating it like an allocator
Criterion
Assessment
Causal story
Mechanical (hedging obligation), not correlational. The main failure mode is estimation error in who holds what, not spurious correlation.
Orthogonality
Derived from options positioning, not price/volume history - low overlap with momentum, value, quality factor libraries. Overlaps partially with vol-regime signals (VIX state), so test incrementally against those.
Frequency
Minute-level live; daily bars for slower books. Event-driven spikes around expirations are part of the signal, not noise.
History
Replayable point-in-time from 2017 (SPY) across the covered universe - enough for regime-conditioned tests, short of a multi-cycle academic panel. Coverage details: the archive contents.
Capacity
Signals concentrate in the most liquid underlyings (index complex, mega-caps), where capacity is deepest. Positioning effects are strongest exactly where you can trade size.
Decay
Published honestly: which of our own endpoints erode and which persist. Structural (mechanical) effects have proven more durable than statistical anomalies.
Point-in-time testability
The historical API replays the exposure series as computed at the time, minute by minute - the test in the next section runs without lookahead. See backtest integrity.
The published evidence
We maintain an open backtest of the exposure family rather than asking anyone to trust adjectives: GEX, DEX, VEX and CHEX tested over eight years of SPY with VIX-regime controls, with artifacts downloadable. The honest headline from that work is that exposure signals are regime-conditioning tools more than standalone alpha: negative-gamma states change the distribution of forward returns and realized vol materially, which is exactly the shape of information a PM folds into sizing, hedging timing, and entry conditioning rather than into a naive long-short. The GEX-conditioned VRP study shows the compounding version: positioning as a filter on a premium-harvesting strategy.
Integration: from CSV to production
# One panel: daily exposure state, point-in-time, since 2019
import requests, pandas as pd
rows = []
for d in pd.bdate_range("2019-01-02", "2026-08-01"):
r = requests.get(
f"https://historical.flashalpha.com/v1/exposure/summary/SPY",
params={"at": f"{d.date()}T15:59:00"},
headers={"X-Api-Key": KEY},
).json()
rows.append({"date": d, "net_gex": r["exposures"]["net_gex"],
"regime": r["regime"], "flip": r["gamma_flip"]})
panel = pd.DataFrame(rows) # join to your returns and test incrementally
The live and historical endpoints share response schemas, so the panel your researcher builds is the same shape production consumes - one schema from research to production. For firms wanting the exposure state pushed rather than polled, sub-second delivery is the streaming tier; entitlement and firm-wide licensing questions go through the institutional desk.
Frequently asked questions
Is dealer positioning data actually alternative data?
It fits every accepted definition: non-traditional source (derived from options market structure rather than fundamentals or price history), differentiated signal content, and limited penetration among traditional factor shops. It is unusual within alt data in having a mechanical rather than correlational link to returns.
How is this different from just watching the VIX?
VIX prices expected volatility; positioning measures obligated hedging flow and locates it by strike. Two days with identical VIX can carry opposite gamma regimes and radically different wall structure. The signals correlate in stress but decouple exactly when the positioning detail matters - test incremental value, not headline correlation. Context: are dealers long or short gamma.
What is the capacity of positioning-based signals?
The effects concentrate in the index complex and mega-caps - the deepest markets in the world - and the signal degrades gracefully: it conditions regimes rather than picking illiquid names. Capacity constraints bind later than for most alt-data families, though front-running crowded expiration mechanics is a real crowding channel to monitor.
Can I test it without engineering work?
Yes - the exposure summary endpoint returns the full state in one call, the Python SDK wraps it, and the MCP connector lets an LLM research assistant pull the panel directly. A single Alpha seat covers the historical replay needed for a proper point-in-time evaluation.
Dealer positioning earns a place on the alt-data shortlist for one reason: the flows it measures have to happen. Evaluate it with the same rigor as any dataset - incremental, point-in-time, capacity-aware - using the published methodology and the 8-year open backtest as the starting evidence base. Research access is self-serve on Alpha; firm-wide and streaming delivery run through the institutional desk and streaming. Related: the buy-side options data guide, the quantitative approach to dealer positioning, and build vs buy.
