Abstract (EN):
The paper describes the procedure developed to derive short-term forecasts of the daily sales of hypermarkets. The underlying model includes 6 smoothing parameters, accounting for level, trend, 3 seasonal components (weekly, monthly and annual) and holidays. A sequential least square procedure was devised to specify the values of these parameters. The detection of significant autocorrelations of the one-step ahead forecast errors led to the posterior modification of the original forecasts by adding to them a proportion of the original one-week lagged errors.
Language:
Portuguese
Type (Professor's evaluation):
Scientific