How Traders Use the RV Forecast
The realized volatility forecast is a forward-looking filter for your existing workflow. It does not tell you what to trade. It tells you whether the volatility environment supports or undermines the trade you are already considering.
Think of it as a check engine light for your trading: it tells you when the environment is about to get dangerous, or when conditions are clear for your strategy to run.
How to Read the Numbers
Before the examples, two concepts that make everything below click.
AUC measures ranking ability, not hit rate.
An AUC of 0.75 means: take one hour where a spike did happen and one where it didn't — the model correctly assigns a higher score to the spike hour 75% of the time. It's about relative ordering, not absolute prediction. A model with 0.75 AUC reliably puts the dangerous hours above the quiet hours in its rankings.
Spike probability is calibrated, not a win rate.
A 30% score means there's a 30% chance the move exceeds the volatility threshold for that pair and horizon. The baseline probability is roughly 10% (varying by horizon). So a 30% reading is not “wrong 70% of the time” — it's the model correctly identifying that this hour has 3x the normal probability of a spike.
The value is in the ranking.
Every hour, the model ranks 308 forecasts (28 pairs × 11 horizons). In live forward-test scoring, the top-ranked signals (30–50% score) produce threshold-exceeding moves 32.5% of the time, versus 6.1% for the lowest-ranked. A trader using this knows which pairs and which hours are dangerous right now — and which are safe for range trades, tight stops, or full position sizing.
With that foundation, every use case below follows the same pattern: use the score to decide whether conditions support or undermine the trade you're already considering.
Breakout Traders
The Problem
Most breakout systems generate signals based on price crossing a level. They cannot distinguish between a breakout with momentum behind it and one that will fail and reverse within an hour. Lagging indicators like ATR measure what already happened. By the time ATR confirms elevated volatility, the breakout is already underway and the entry is late.
How the RV Forecast Helps
The spike probability tells you whether the market has enough predicted energy to sustain a breakout before you enter. A breakout signal during a high-probability window has structural support. A breakout signal during a quiet window is statistically more likely to fail.
Example
Your system flags a breakout on EURUSD as price crosses above a 4-hour resistance level. You check the dashboard.
The 4h spike probability is 7%. The confidence tier reads Quiet. The expected move is 0.09% (about 10 pips).
The model is saying: there is a 7% chance EURUSD moves more than 30 pips in the next 4 hours. You skip the trade.
Compare that to the same breakout signal when the 4h spike probability is 38% and the expected move is 0.32% (35 pips). Now the volatility forecast agrees with the price signal. You take the trade with higher conviction.
The Filter Rule
Take breakout signals only when the spike probability for your target horizon exceeds 25%. Below that threshold, the market lacks the predicted energy to sustain the move. This single filter can remove the majority of fakeout losses from a breakout system.
Post-News Continuation
A news spike has already moved USDJPY 100 pips. You missed the initial move. Should you chase it?
The 4h expected move is 0.55% (60 pips). The p90 stress scenario is 0.95% (104 pips). Price has already moved 100 pips — right at the p90 boundary. The statistical move is likely spent. You let it go.
Now imagine the same spike, but the 4h expected move is 1.2% (131 pips) and p90 is 1.8% (196 pips). There is statistical room for continuation. You enter with a stop at the p90 boundary.
Range and Mean-Reversion Traders
The Problem
Range-trading strategies profit when price stays within boundaries. They fail when a genuine breakout destroys the range. Mean-reversion entries face the same risk: entering a reversion trade just before a volatility spike turns a controlled pullback into a sustained trend move.
How the RV Forecast Helps
Low spike probability confirms that the range is likely to hold. High spike probability is the warning signal: the range may be about to break. The p90 stress scenario tells you the worst-case move if a spike does occur — use it to set your stop beyond the range boundary plus the predicted tail.
Example
GBPUSD has been trading in a 45-pip range for 8 hours. Price is at the upper boundary. Your system signals a short entry for a reversion to the range midpoint.
The 4h spike probability is 9%. The p90 stress scenario is 0.22% (24 pips). The worst-case tail is well within your stop. You take the mean-reversion trade.
Now imagine the same setup, but the 4h spike probability is 41% and p90 is 0.48% (52 pips). That 52-pip worst case exceeds your range boundary. You skip the trade.
The Filter Rule
Enter range trades when spike probability is below 15% for your hold time. Exit or reduce when spike probability crosses above 30%. Use p90 as the stress test for your stop placement.
Options Traders
The Problem
Options pricing depends on implied volatility. The opportunity is finding windows where realized volatility is likely to exceed implied volatility, meaning options are underpriced. Most traders estimate this using trailing realized vol (backward-looking) or gut feel. Neither is a calibrated forward-looking estimate.
How the RV Forecast Helps
The expected move output is a direct comparison point against implied volatility. When the model's expected move exceeds the implied move priced into options, the options are cheap. Conditions favor buying straddles or strangles. The calibration is what makes this actionable: when the model outputs a 30% spike probability, approximately 30% of such predictions historically produced large moves.
Example
USDJPY 4h options are priced with an implied move of 22 pips. You check the RV dashboard.
The 4h expected move is 0.33% (36 pips). The spike probability is 34%. The p75 scenario is 42 pips.
The model predicts realized vol will likely exceed the implied move by 60%. The straddle is underpriced. You buy the 4h straddle, sized for the 34% hit rate (roughly one in three such predictions produces a large move).
The Edge
This is not available from trailing ATR. ATR tells you what USDJPY did yesterday. The RV model uses 98 features — session timing, spread dynamics, positioning data, calendar effects, volume patterns — to estimate what USDJPY will do in the next 4 hours.
Systematic and Algorithmic Traders
The Problem
Automated strategies need to adapt to changing volatility regimes. A trend-following system optimized for normal conditions will get stopped out repeatedly during a volatility spike. Most bots use static parameters or lagging vol measures to switch between modes.
How the RV Forecast Helps
The API delivers 308 forecasts per hour (28 pairs across 11 horizons) with full move distributions. Your system can switch strategies, adjust position sizes, widen or tighten stops, or pause execution entirely based on a single API call per cycle.
Integration Examples
API Quick Start
import requests
resp = requests.get(
"https://api.fxengineer.com/rv/signals/latest",
headers={"X-API-Key": "fxe_your_key_here"}
)
signals = resp.json()["signals"]
for pair in signals:
score_4h = pair["horizons"]["4"]["score"]
p90_4h = pair["horizons"]["4"]["p90_move_pct"]
if score_4h > 0.35:
print(f"{pair['pair']}: HIGH VOL - pause bot")
elif score_4h < 0.10:
print(f"{pair['pair']}: QUIET - range strategy")Three lines of logic. One API call per hour. Your system now has a forward-looking volatility filter.
Risk Managers and Execution Desks
The Problem
Risk desks need to know when volatility is coming before it arrives. Most desks rely on implied vol from options markets, VIX-equivalents, or internal models that update daily. None of these provide hourly, pair-specific, calibrated spike probabilities across multiple time horizons. Execution desks scheduling large FX orders need to know whether the next few hours are likely to be calm or volatile.
How the RV Forecast Helps
The multi-horizon surface (15 minutes to 24 hours) provides a forward-looking volatility map at every relevant time scale. The p90 stress scenario at each horizon provides the number a risk manager needs for worst-case drawdown estimation — an hourly, pair-specific, forward-looking tail estimate.
Example: Hedging
Your desk holds EURJPY exposure. At the morning review, the dashboard shows:
1h spike probability is 11% (normal). 4h is 29% (approaching elevated). 12h is 35% (elevated). The 12h p90 is 0.52% (57 pips).
You increase your hedge ratio from 50% to 80% for the next 12 hours. The actual move was 43 pips. Your increased hedge saved approximately 9 pips of unhedged exposure.
Example: Execution Timing
You need to execute a $50M EURUSD order. The 1h spike probability is 5%. You begin TWAP execution immediately, expecting minimal market impact.
If instead the 1h spike probability were 38%, you delay execution until the spike probability drops below 15%. Executing during a predicted quiet window can reduce slippage by 2–5 pips on a large order — on $50M notional, that is $10,000–$25,000 in execution savings.
How It Compares
| FX Engineer RV Forecast | ATR / Bollinger Bands | Implied Volatility | |
|---|---|---|---|
| Type | Predictive (ML, 98 features) | Reactive (lagging average) | Market expectation |
| Update frequency | Every 15 minutes | Every candle close | Continuous (options hours) |
| Horizons | 27 (15m to 24h) | Single lookback window | Daily to monthly |
| Calibrated probability | Yes | No | Implied, not calibrated |
| Best for | Volatility filtering, timing, sizing | Visual charting | Options pricing |
| Limitation | Does not predict direction | Backward-looking only | Requires options access |
What It Costs to Not Know
A single fakeout breakout on a 2-lot EURUSD position with a 30-pip stop costs $600. Two of those per week is $4,800/month in avoidable losses.
A grid bot that runs through a volatility spike without warning can lose 3–5% of account equity in a single event.
An unhedged EURJPY position during a 50-pip spike on $10M notional is $50,000 of unexpected P&L.
The RV forecast costs less per month than any one of these individual events.
For systematic traders, the math works differently: if the volatility filter improves your execution by 2 pips per trade across 200 monthly trades, that is 400 pips of value — roughly 20x the subscription cost.
Get Started
The RV forecast is currently in early access. Dashboard access, API keys, and the full methodology paper are available on request.
Request Early AccessNo commitment. We respond within 48 hours.
FX Engineer provides volatility forecasts for informational purposes only. Nothing on this page constitutes investment advice. This product does not predict price direction. Past model performance does not guarantee future results. Trading foreign exchange involves substantial risk of loss.