Overview
Time series forecasting model selection is driven by two questions most practitioners skip: is the series stationary, and what drives its variation? A non-stationary series fed into an ARIMA model without differencing produces forecasts that drift toward the historical mean. A series with strong external drivers modeled without those drivers produces forecasts that are systematically wrong whenever the drivers change.
The right forecasting model is determined by the series' temporal structure: trend type (linear, exponential, none), seasonality (single, multiple, none), stationarity, the forecast horizon relative to the series length, and whether external regressors are available and predictable. Each combination points to a different model family.
The Time Series Forecasting Model Prompt generates a complete forecasting specification: temporal structure decomposition, model selection by structure, stationarity handling, external regressor integration, and a forecast evaluation framework that tests performance at the actual forecast horizon.
What you get: - Temporal structure decomposition (trend/seasonality/stationarity) - Model selection matrix by structure and horizon - Stationarity testing and transformation - External regressor integration specification - Horizon-appropriate forecast evaluation
Built for: data scientists and analysts building production forecasting systems for demand, sales, traffic, and any temporally ordered business metric.