Skip to content

Feature: Forecast Confidence Diagnostics Across Forecast Horizon #442

Description

@coderleeon

Is your feature request related to a problem?

TimesFM provides strong zero-shot forecasts, but it can be difficult to determine how confidence degrades as the forecast horizon increases.

In practice, users often trust short-horizon predictions more than long-horizon predictions, yet there is limited visibility into where forecast reliability begins to break down.

Describe the solution you'd like

Add an optional forecast diagnostics report that evaluates prediction confidence across different forecast horizons.

Example:

  • Horizon 1–7: High confidence
  • Horizon 8–30: Moderate confidence
  • Horizon 31–90: Low confidence

Potential metrics:

  • Prediction interval width
  • Forecast variance growth
  • Calibration error by horizon
  • Historical backtest performance

Describe alternatives you've considered

Users can manually run rolling backtests, but this requires additional experimentation and implementation effort.

Additional context

This would make TimesFM easier to evaluate in production settings where forecast reliability is often more important than a single aggregate accuracy metric.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions