Forecast Signal

Switchgear, Connectors, and Controls Import Customs Value: Six-Month Forecast

$3.85B projected by Feb 2027 (−20.6% from $4.86B in Aug 2026), 80% band $3.45B to $4.26B. A six-month statistical projection for switchgear, connectors, and controls import customs value, shown with a prediction band and historical backtest error.

The six-month projection

  • Projected for Feb 2027: $3.85B, −20.6% versus $4.86B in Aug 2026.
  • Prediction band at the six-month horizon: 80% band $3.45B to $4.26B.
  • Direction: Falling, from $4.86B in Aug 2026.
  • Out-of-sample backtest error (MAPE): 15.2%, one step ahead on months the model had not seen.

Statistical extrapolation, not a market or policy forecast

  • It knows nothing about the next Fed decision, OPEC action, or policy change. Read the prediction band and the backtest error, not just the central line.
  • This series publishes monthly, so the model forecasts directly on it and is anchored on August 2026, the newest month the source has released. Any gap to today is the agency's publication lag, not a stale model.

How it is built

  • The series is collapsed to one point per month, then projected six months with Holt-Winters seasonal exponential smoothing with a damped trend and a twelve-month seasonal pattern. The smoothing parameters are grid-searched by walk-forward validation on an early tuning segment, and the reported error is the mean absolute percentage error on a final evaluation segment never used for selection or band calibration. Deterministic: the same data always yields the same forecast.
  • The 80% band is an empirical residual band: at each horizon its width is the 80% quantile of this series' own fixed-parameter walk-forward residuals (13 residuals at the six-month horizon).

Download and query it

  • This forecast, its band, and its backtest are free to download (CC BY 4.0) and available as an open, keyless JSON API and an MCP tool.
  • Source: USITC DataWeb (monthly import customs value). Underlying series is U.S. government open data, public domain.

Last reviewed 2026-10-08.