ML Forecast
Machine learning models analyze historical price patterns to generate research forecasts across multiple time horizons. Up to eight models are trained per symbol: RandomForest, ExtraTrees, Ridge, KNeighbors, LSTM, XGBoost, LightGBM, and CatBoost. Forecasts are generated nightly for all tracked stocks.
Kronos · Candle Path Research
A pretrained candlestick model samples the next 20 daily candles. The median and 5th–95th sample percentiles describe model dispersion; they are not a calibrated probability range. This experimental model is evaluated separately from the existing ML consensus.
Load a recorded forecast to inspect the sampled range.
Enter a cached symbol, choose Auto history, and select Generate forecast. View recorded forecast displays a saved result without starting a new job. At least 40 valid completed daily candles and supported asset metadata are required. Each symbol reports its own data errors.
Daily requests are queued at 18:30 Eastern on weekdays. On-demand requests use completed candles; a stale-data message means the daily data update must finish first.
History: grey. Pointwise median: blue. Sample 5th and 95th percentiles: dashed. “Samples above origin” is the fraction of model samples, not an observed chance of a price increase. Accuracy is measured only after recorded forward forecasts mature. Daily research context does not establish intraday entry timing.
Cached provider prices have not been verified for split or dividend adjustments. Equity horizons count trading sessions; crypto horizons count calendar days. Targets are counted from the last completed candle, so check the generation time and target date when comparing forecasts.