Methodology

We grade our own signals. No one else does.

Most trading tools sell every metric as an edge. That's the business model — more signals, more subscriptions. We do the opposite: we read the research, grade each metric by how much the evidence actually supports it, and tell you when a popular one doesn't hold up. Honesty isn't a disclaimer here. It's the product.

The five grades

Every metric in GradientView — options desk, portfolio, or value — carries one of these badges. Tap it in the app and you get the paper, the finding, and the replication status in plain language.

Durable
Replicated out-of-sample, with a clear economic mechanism. These are the signals to build on. e.g. variance risk premium, IV term structure, options-implied borrow fees, box-spread funding.
Conditional
Real, but the signal depends on a model or an assumption you should be able to see and adjust. e.g. dealer gamma (GEX), vanna/charm — all rest on inferring the dealer's side, which we let you toggle.
Decayed
Genuinely worked in-sample, then faded after publication — often because everyone started trading it, or because a hidden factor explained it. e.g. the call-put IV spread, largely a borrow-fee artifact; the O/S ratio.
Folklore
Popular in trading culture, weak in the evidence. We still show these — traders expect them — but we label them clearly rather than dress them up. e.g. max-pain gravitation, "smart money" UOA alerts.
Framework
A discipline or lens rather than a backtested return signal — valuable for structuring thinking, not for mechanical prediction. e.g. Graham's defensive checklist, margin of safety, no-arbitrage surface fitting.

Interpretation, on every number

A number without meaning is just noise you pay for. Every metric in the product renders the same three ways — the standard we hold ourselves to.

Layer 1 · Read

What it means, now

One plain-language sentence about the current state — no jargon defining jargon.

"SPY premium is rich versus its own year — the seller's side is being paid."
Layer 2 · Context

Where it sits

The number's place in its own history, and the base rate of what has typically followed — as frequencies, never predictions.

"IV this rich resolved lower within 30 days in 78% of past instances."
Layer 3 · Evidence

How much to trust it

The badge — and behind it, the citation and replication status, in language you don't need a PhD to read.

"Durable · Bollerslev-Tauchen-Zhou 2009 · replicated internationally."

The principles behind the grades

01

Post-publication decay is real

Published market anomalies weaken about a quarter out-of-sample and more than half after publication, as the world trades against them. We assume any single-name "options predict stocks" edge is roughly half as strong now as its paper claimed — and grade accordingly. McLean & Pontiff (2016).

02

Beware the too-good backtest

Common look-ahead filters in options research inflate reported Sharpe ratios by an order of magnitude — from around 0.5 to above 5. A spectacular backtest is a reason for suspicion, not excitement. Nothing we surface leans on ex-post information a trader wouldn't have had. Duarte, Jones, Khorram, Mo & Wang (2025).

03

Mechanism over correlation

A signal with a clear economic reason to exist — a risk premium, a hedging flow, a funding cost — survives better than one that's merely a historical pattern. The durable grades all have mechanisms; we're skeptical of the ones that don't.

04

Inferred is not observed

Dealer positioning, the sign of options flow, "who's on the other side" — much of what popular tools present as fact is actually estimated. Where we infer, we say so, badge it Conditional, and give you the assumption to adjust.

05

Decision-support, never advice

GradientView computes; you decide. It issues no buy/sell signals, places no trades, and makes no recommendations. It shows you your risk and the market's structure honestly — the judgment stays yours.

The bottom line

A tool you can trust is one that tells you when not to.

That's the whole idea. Everything else — the desks, the risk lab, the screeners — is built on it.

See the desks →   Request access →