Level-adjusted exponential smoothing for modeling planned discontinuities

Based on contextual information, forecasters often make judgmental adjustments to the forecasts of exponential smoothing models. This approach can produce sound initial forecasts, but it can result in diminished accuracy for forecast updates as well as misleading signaling effects. A proposed technique lets the forecaster include judgmental adjustments within an exponential smoothing model. For 20 real data series representing Virginia Medicaid expenses, forecasts are developed using the proposed technique and several alternatives, and they are updated through various simulated level shifts. The study shows that the proposed technique is more accurate than the alternatives in updating forecasts when a shift in level occurs approximately as planned. Other advantages of the proposed technique include automation through final reporting when the shift occurs as planned, better detection capability when the planned shift does not occur, and ease of adjusting the forecast when new assumptions about the planned change are warranted.