Forecasting

Forecasts are the core output of LTprophecy. This guide covers creating one-off and scheduled forecasts, interpreting results, and using scenario analysis.

Creating a Forecast

1. Navigate to Forecasts → New Forecast.
2. Select a Production model from the dropdown.
3. Set the Horizon — how many time steps ahead to forecast.
4. Choose the Confidence Level — 80%, 90%, or 95% prediction intervals.
5. Optionally attach Future Covariates (e.g., planned ad spend).
6. Click Run Forecast.

Scheduled Forecasts

Forecasts can be scheduled to run automatically using cron syntax. Navigate to Forecasts → Schedules and click Add Schedule.

Reading Forecast Results

The forecast results panel includes:

• Point Forecast — the model's single best estimate for each period.
• Prediction Interval — upper and lower bounds at the selected confidence level. Values outside this range are statistically unlikely.
• Historical Overlay — past actuals shown alongside the forecast for context.
• Component Decomposition — trend, seasonality, and residual breakdown (Prophet and ensemble models only).

Scenario Analysis

Scenario analysis lets you run "what-if" simulations by adjusting input assumptions without retraining the model.

1. Open a completed forecast and click Add Scenario.
2. Choose a scenario type:
   • Factor Drift — shift a feature column by a percentage
   • Volatility Multiplier — increase/decrease variance
   • External Override — paste in custom future values
3. Name the scenario (e.g., "Bear Case — 20% spend cut") and save.
4. Compare scenarios overlaid on the same chart.

Exporting Results

Forecast results can be exported as:

• CSV — point forecast + intervals per period
• JSON — full API response payload
• PDF Report — auto-generated executive summary (Enterprise)
• Webhook push — POST to a configured endpoint on completion

Forecast Accuracy Tracking

After the forecast horizon passes, LTprophecy automatically computes actual-vs-predicted accuracy metrics:

• MAPE (Mean Absolute Percentage Error)
• RMSE (Root Mean Squared Error)
• Coverage rate (% of actuals within prediction interval)

These are surfaced in Forecasts → Accuracy History and available via the evaluation API endpoint.