STA 4753 Lecture 4: Time-series Analysis. Measure of Forecast Accuracy
30 Jan 2017
School
Department
Course
Professor
Document Summary
Forecast origin: time the forecast is made (time t). The further the forecast moves from the origin, the more difficult to predict. One-step-ahead forecast error: +=+ + h-step-ahead forecast +(cid:4666)(cid:4667): forecast for period (t+h, at time t h-step-ahead forecast error: +(cid:4666)(cid:4667)=+ + Fo r ecast ac cur acy m e asur em ent s. Percentage error measures must have y > 0: Mean absolute error/deviation: measures the absolute size of the errors in units. Mae and mape - analyzes/measures forecast variability (errors) Mean percentage error: indicates if bias is high ((actual forecast) / actual) x 100. Mpe= -3. 9 means actual results were 3. 9% less than the forecast. Mape: measures the size of errors in % terms. Want mase to be lower than the random walk. Mase < 1 = forecasts are better than random walk forecast. Mase > 1 = forecasts are worse than random walk forecast.
