Price forecast
Where will price be?
Future prices or price ranges for an instrument.
- Output
- A price level with its range
- For example
- The future price of gold.
- Scored with
- Mean absolute error, RMSE, median absolute error
01 of 16
What’s likely,and how sure.
What is likely to happen in financial markets, under what conditions, over what horizon, and with what degree of uncertainty. Every forecast answers all four, in advance, and is scored once it resolves.
XAUUSD
Gold · next 30 hours · example forecast
In the record as Forecast value
A single best estimate of where it ends up.
Sixteen kindsof forecast.
Each answers a different question about what comes next, and each is scored by the measure that suits it.
Every forecastis a record.
Instrument, horizon, range and probability are fixed when the forecast is made, so it can be judged fairly against what happens. Hover a field to see where it lives on the chart.
XAUUSD Range forecast · 24 hours
- Instrument
- XAUUSD
- Asset class
- Precious metals
- Forecast type
- Range
- Direction
- Higher
- Forecast value
- +0.1% from reference
- Reference price
- Price at creation
- Expected range
- −0.5% to +0.8% (80%)
- Probability
- 73% higher
- Confidence
- Defined separately from probability
- Time horizon
- 24 hours
- Created at
- Timestamped at publication
- Expiry
- Evaluated 24 hours later
- Methodology
- Hybrid ensemble
- Status
- Active
- Outcome
- Recorded at expiry
- Performance
- Scored at expiry
AI that reads, modelsand explains.
Machine learning and language models work inside the forecast, alongside econometrics and statistics, never instead of them. They are benchmarked like any other method, and what they explain stays grounded in the data.
Example: a central bank statement and a EURUSD forecast
Source · policy statement
The Governing Council kept interest rates unchanged but said further cuts are likely later in the year as inflation in the euro area continues to slow. Growth remains weak, while wage pressures are easing.
Named entities · event extraction · tone · matching to past events
- Event
- Policy decision: rates held
- Guidance
- Cuts signalled for later this year
- Inflation
- Slowing, wages easing
- Growth
- Weak
- Tone
- Dovish
- Exposed
- EURUSD, Bund yields, EURO STOXX 50
Learning
- Supervised and unsupervised
- Reinforcement learning
- Gradient boosting
- Random forests
- Support vector machines
- Ensembles
Deep learning
- Neural networks
- Transformers
- Time-series foundation models
- Representation learning
- Multimodal learning
Language
- Natural language processing
- Large language models
- Event and entity extraction
- Tone and topic classification
- Summarisation
Explainability
- Feature importance
- Sensitivity analysis
- Attribution
- Scenario comparison
- Historical analogues
- Horizon
- Hours ahead
- 80% range at the horizon
- −0.1% to +0.4%
Further out,wider.
From intraday to long-term, the range widens as the horizon grows. A method that works over hours is not assumed to work over months: each horizon is judged on its own record.
Classifiedfive ways.
By type, asset class, horizon, methodology and the market conditions it was made in. That is what lets every forecast be found, filtered, compared and evaluated.
A liquidity forecast for XAUUSD, short-term, from a hybrid ensemble, in high volatility.
Type
Asset class
Horizon
Methodology
Market condition
Scored againstwhat happened.
A forecast that says 70% should come true about 70% of the time. Probabilities are checked for calibration, ranges for coverage, prices for error, directions for hit rate.
- Misses stay visible in the record.
- Confidence and probability are never treated as the same thing.
- Every method is compared with simpler baselines.
When the method says 70%, it happens about 70% of the time. Points sit on the diagonal.
See what’s likely next.
Forecasts across every market, with the range, horizon and evidence stated.