Methodology & risk disclosure
How our forecasts and signals are built, how we measure them, and the risks you should understand before using them.
Data
We use market data from exchanges and data providers, on-chain data, news and macroeconomic series. Data is stored point-in-time: each record reflects what was known at that moment, preventing look-ahead bias. Delisted assets remain in historical universes to avoid survivorship bias.
Models
Our forecasting ensembles combine time-series transformers, gradient-boosted trees, recurrent networks and regime-switching models. Outputs are probability distributions summarised as P10, P50 and P90. Every production model has a model card describing training data, features, retraining schedule and known limitations.
Evaluation
All published performance figures are out-of-sample, produced by walk-forward testing: models are trained on a historical window and scored on the following unseen window, repeated across the full history. We report:
- Error — mean absolute percentage error (MAPE) of the median forecast.
- Directional hit rate — share of periods where the forecast direction matched the realised move.
- Calibration — how often realised prices fall inside the stated P10–P90 range (target: 80%).
- Strategy metrics — Sharpe ratio, maximum drawdown and turnover, after estimated costs.
Every metric is shown with its evaluation period, asset universe and horizon.
Volatility cone on this website
The live chart on our home page shows a volatility cone computed from the realised volatility of the displayed price history, assuming no drift. It is a statistical illustration of a plausible range, not a model forecast. Live prices are provided by the CoinGecko public API and may be delayed.
Limitations
Markets change. Models trained on the past can fail in new regimes, during liquidity shocks, or when data is incomplete or manipulated. Cryptoasset markets are especially volatile and can move far outside any forecast range.
Risk disclosure
Capital at risk. Investing and trading in financial instruments and cryptoassets carries a high level of risk and may not be suitable for everyone. You could lose all the money you invest.
Cryptoassets. Cryptoassets are high-risk and largely unregulated in the UK. You are unlikely to be protected if something goes wrong, including by the Financial Ombudsman Service or the Financial Services Compensation Scheme.
No advice. Chakra AI Ltd provides research software, data and educational content only. We are not authorised or regulated by the Financial Conduct Authority and do not provide investment advice, arrange trades or manage money. Nothing on this website is a recommendation or invitation to buy, sell or hold any investment or cryptoasset. Cryptoassets are high-risk and largely unregulated in the UK; you could lose all the money you invest and are unlikely to be protected if something goes wrong. Forecasts and model outputs are uncertain, and past or simulated performance is not a reliable indicator of future results.
Past performance and simulated or backtested results are not reliable indicators of future results.