How It Works
Two products, one goal: tell you when large price moves are likely across 28 FX pairs. Direction is unknown. Your job is to decide what to do with that information.
The Two Products
RV Forecast (Signal Board)
A machine learning model that predicts the probability of an above-average price move for each currency pair, across 27 time horizons (15 minutes to 24 hours).
- 28 pairs, 756 predictions per hour
- Calibrated probabilities (a 40% reading means it happens ~40% of the time)
- Trained on 98 features from price and positioning data
- Direction is unknown. This predicts size of move, not which way.
Positioning Events
Detected when the retail FX crowd reaches an extreme positioning level, measured by the Speculative Sentiment Index (SSI) crossing a statistical threshold.
- 91% of events produce a 15+ pip move within the measurement window
- Tracked for 8 or 12 hours from detection with live checkpoint updates
- Validated against 9,100+ historical patterns (2014-2026)
- Direction is unknown. Events detect volatility, not direction.
Reading the Signal Board
The Signal Board is a heatmap. Each row is a currency pair. Each column is a time horizon. Each cell shows the probability of an above-average move.
Pip estimates are approximate and based on historical move distributions at similar probability levels. The model updates hourly (sub-hourly for the 15m/30m/45m horizons).
Interpreting Positioning Events
When a positioning event fires, it means the retail crowd has reached a statistically extreme level of positioning on a currency pair. Here is everything you see on the event card and what it means.
When an Event Fires
Crowd Position: Shows the current retail positioning. “Crowd Long” means the majority of retail traders are long on this pair. “Crowd Short” means the majority are short.
Fade Direction: The contrarian direction. If the crowd is long, the fade direction is short (and vice versa). This is not a recommendation. It is the direction that historically follows crowd extremes, but our research (RP-004) shows the directional edge is statistically insignificant. The value of the event is the volatility detection, not the direction.
Measured Over: The fixed tracking window. 8 hours or 12 hours. Once the event is detected, we track price movement for this duration. At the end, the event resolves.
RV Score: If available, the current RV Forecast spike probability for this pair. An elevated RV score alongside a positioning event means both systems agree: something is likely to move.
During the Measurement Window
The event card updates with live checkpoint data every hour. The state label tells you two things: what price is doing relative to the fade direction, and what the crowd is doing.
Profitable + Crowd Holding: Price has moved in the fade direction and the crowd is still positioned at the extreme. The positioning pressure is still present.
Profitable + Crowd Exiting: Price moved in the fade direction and the crowd is unwinding their extreme position. The move may have further to run as the unwind continues, or it may be exhausting.
Underwater + Crowd Holding: Price has moved against the fade direction. The crowd's extreme positioning is being rewarded so far. This can reverse quickly or continue.
Underwater + Crowd Exiting: Price is against the fade direction and the crowd is unwinding. The thesis is failing on both dimensions. Price and positioning pressure are both going against.
Understanding MFE and MAE
MFE (Maximum Favorable Excursion): The best the price moved in the fade direction at any point during the measurement window. If MFE is 45 pips, the price was up 45 pips from entry at its peak. This tells you the opportunity that existed.
MAE (Maximum Adverse Excursion): The worst the price moved against the fade direction at any point. If MAE is 20 pips, you would have been down 20 pips at the worst point. This tells you the drawdown risk.
Historical average MFE across all events is approximately 50 pips. Average MAE is approximately 33 pips. These are the excursions you should expect. Events that produce large moves typically show the move within the first 4-6 hours of the window.
Historical Context on the Card
Each event card shows the vol hit rate: the percentage of similar historical events that produced a 15+ pip move. This is measured by max excursion (MFE or MAE), not by the final close-to-close P/L. A pattern can spike 50 pips, come back, and close flat. The 50-pip spike was real volatility. The vol hit rate captures that.
The directional fade rate is also shown: the percentage of events where the final P/L was in the fade direction. For most pairs, this is near 50-55%. This is why we frame these as volatility events, not directional signals.
How Traders Use This Data
We do not recommend specific trades. The data supports multiple approaches:
- Volatility filter: Use events to identify when a pair is likely to move, then apply your own directional analysis.
- Fade at detection: Enter in the fade direction when the event fires. Accept the ~52% directional hit rate and manage risk with the historical MAE data.
- Wait for confirmation: Watch the first checkpoint (1-2 hours). If the state is “Profitable + Crowd Holding,” the move has started and the crowd is still trapped.
- Limit orders: Use historical MFE data to set limit entries at better prices if you expect a pullback before the main move.
- Options / straddles: Use the volatility forecast (not direction) to buy options when events fire, profiting from movement in either direction.
The common thread: the event tells you something is likely to happen. What you do with that is your decision.
The Data Behind It
SSI data covers 28 FX pairs from December 2002 to present (4.4 million+ observations). The detection model uses dual-timeframe z-score thresholds calibrated per pair. All historical statistics are computed from bid/ask M1 price data.
For the full methodology, see our research papers: RP-001 (data provenance), RP-002 (SSI-price causality), RP-003 (session volatility structure), RP-004 (directional edge analysis), RP-101 (volatility predictability surface).
How We Got Here
FXE is not the product we set out to build. We started with a hypothesis — that fading retail crowd positioning at extremes would produce a directional edge. Five research papers later, the data told a different story.
Established the SSI dataset: 4.4 million observations across 28 pairs since 2002. Validated data provenance and quality.
Tested whether SSI causes price movement. Result: null. Price leads SSI, not the other way around. The crowd reacts to price, not vice versa.
Mapped session volatility structure. Found stable, exploitable patterns: London AM consistently produces the largest moves. This structure does not decay.
Tested the directional edge of fading retail extremes. Result: the corrected favorable rate is 50.3% — a coin flip. Our original analysis showed 81%. The correction was published with the same prominence as every other finding.
Built the volatility predictability surface. Found that SSI extremes reliably detect large price moves — 91% produce 15+ pip excursions. Direction is unknown, but the move itself is predictable. This became the product.
When the data contradicts our hypotheses, we say so. The 81% to 50.3% correction is published in RP-004. The year-by-year performance table shows every year, including the ones that lost money. We do not cherry-pick results.
Why This Exists
The founder behind FXE ran a CFTC-registered CTA using SSI-based contrarian strategies. The directional edge was real: 60% win rate from 2014 to 2020. Then it decayed. 53% in 2021. 49% in 2022. 46% in 2023. Rather than keep trading a dying edge, we built a research program to understand what happened.
Four published papers and 9,100+ tested events later, the answer was clear: SSI extremes predict volatility, not direction. The directional edge decayed from 60% to 43%. The volatility detection signal, 91% of events produce a 15+ pip move, never decayed at all. That finding is why this product exists.
Why FXE
| FX Engineer | Typical Sentiment Product | |
|---|---|---|
| Publishes null results | Yes — RP-002, RP-004 | No |
| Shows year-by-year decay | Full table, including losing years | Cherry-picked periods |
| Sample size | 9,100+ validated events | Unspecified |
| Methodology | Published research papers | Black box |
| Track record | Public, auditable, exportable | Screenshots or none |
| Directional claims | None — volatility only | “Fade the crowd” |
| Data corrected when wrong | 81% → 50.3% published | Original numbers kept |
Who This Is For
Discretionary FX Traders
Use positioning events and RV forecasts as a volatility filter alongside your own technical or fundamental analysis.
Systematic / Algo Traders
Integrate via API. 756 predictions per hour, 28 pairs, 27 horizons. Machine-readable JSON. Build volatility-aware entry/exit logic.
Options Traders
Directionally agnostic volatility forecasts. Buy options when spikes are predicted. Use positioning events to time straddle entries.
Risk Managers / Portfolio Overlay
Monitor when the crowd reaches extremes on your exposure pairs. Anticipate volatility before it hits. Adjust hedges accordingly.
Questions? Want a walkthrough?