Risk forecasting for professionals

Know how far the market can move over the next bars.

At every bar close, Sibyllium turns OHLC prices into VaR, Expected Shortfall and the probability that your stop, limit or barrier is touched.

Request a pilot See the results

Input: OHLC only
Bars: 5 min to daily
Coverage: Indices · equities · FX · crypto · commodities

DJI · 30 minClose 46,949.8Forecast at 2026-03-16 15:30Historical data

Horizon

Position

€

Probability of touching a level

Dist.

Price

Within 3 bars

Loss risk at 3 bars

VaR 97.5 %	1.74 %	174 €
VaR 99 %	2.27 %	227 €
ES 97.5 %	2.32 %	232 €
ES 99 %	2.84 %	284 €
Stop risk 99 %	2.64 %	264 €

Within 3 bars: 38 % chance of touching 46,642.5 (2 ATR below). Maximum probable loss (VaR 99 %): 2.27 %, i.e. 227 € on 10,000 €.

Forecast from a past bar. Historical data, not investment advice.

1.08 / 100exceedances of the 99 % VaR (target: 1)

−17 to −20 %VaR error vs historical simulation

0.66 vs 0.67forecast vs observed chance (per 100) of a 3-ATR stop being hit

+34 %exposure at equal risk budget vs historical simulation

Different asset classes, different timeframes, 516,000 bars, same settings everywhere.

At each bar

Four measures from the same forecast

Expected move

Full distribution of the move over 1, 3, 5 and 15 bars, with volatility regimes.

Loss risk

VaR and ES at 97.5 and 99 %, stop risk, amounts in your currency.

Touch probability

Chance that any price is hit within 1 to 5 bars, including at the session open.

Portfolio risk map

Portfolio VaR and ES, contribution of each position, hedges, diversification.

Platform

The Viewer, screen by screen

Decision at T

Risk, trend and your position at a chosen bar

A one-sentence summary, then risk and trend side by side. Enter a position to get its VaR and its probability of being stopped out.

Decision at T window: volatility and risk, trend by horizon

Probability of touching a level shown at the cursor

Probability at the cursor

Hover over a price to see its probability of being touched within 1, 3 and 5 bars.

Forecast vs realised

EUR/USD: 78.5 % inside the 80 % range over 2,398 periods.

VaR backtest with Basel traffic light and statistical tests

Backtest report

Basel traffic light, Kupiec and Christoffersen tests per instrument.

Trend reading and Market Radar

Trend direction and its measured edge

Short, medium and long-term paths with a predictability level, applied to a whole watchlist. The Radar Backtest scores every reading against a random direction.

51–56 %right direction at +5 bars when predictability is High, vs 50 % at random

5 / 5test periods above random at +5 bars (4 / 5 at +15 bars)

5-minute bars, 5 test periods from Oct 2025 to Jul 2026, same settings, before costs.

Decision mode chart with forecast paths for three horizons and VaR corridor

Use cases

What each team uses it for

Intraday desksStop placement and position sizing at each bar.
Risk managementInternal VaR and ES, or a challenger to historical simulation.
Model validationVaR backtests with Kupiec and Christoffersen tests, reproducible results.
Brokers, CFD, cryptoProbability of margin call or liquidation.
Prop-trading firmsChance of breaching a trader's daily loss limit.
Structured productsProbability of touching a barrier within 1 to 5 bars.
Corporate treasuryChance that a currency reaches the budget rate this week.
Wealth managersWhich positions drive the client's risk, and which hedge it.
Software vendorsRisk forecasts by file or API at each bar.

Evidence

Compared with standard risk models

Measure	Compared with	Reduction vs benchmark	Sibyllium
VaR, 1 bar	Historical simulation		−17 to −20 % error
VaR, 1 bar	GARCH-t, GJR-t, filtered HS		−4 to −5 % error
VaR, 3 and 5 bars	Historical simulation		−14 to −20 % error
Touch probability	FHS, GJR-t		−1 to −4.5 % Brier score
High–low range, 1 bar	HAR · GARCH		−7 % · −14 % error
Portfolio VaR	Multivariate FHS		−2.5 to −3.8 % error
Direction, High level, +5 bars	Random direction	other unit: hit-rate points	+1 to +6.5 pts hit rate

Method. Estimation on the first 60 % of each history, test on the last 40 %, over 516,000 bars and 1.85 million forecasts per model.

Scope. Risk measures validated from 1 to 5 bars. Tails beyond 1 % slightly thin; long and short VaR symmetric.

Pilot

Six to eight weeks on your data, compared with your current method

Success criteria are set in writing before any computation.

Week 0
NDA, instruments, criteria
Weeks 1–2
Out-of-sample backtest
Weeks 3–6
Live forecasts, timestamped
Weeks 7–8
Results on the agreed criteria

You provide OHLC bars. We deliver the forecasts, the reports and the evaluation code.

Contact

Request a pilot or a demo

Tell us which instruments and timeframes you work on, and what you want to measure.

contact@sibyllium.com