Previsão de vendas de peças para máquinas pesadas por meio de séries temporais

Abstract

This monograph aims to forecast, for 12 months, the amount of sales of spare parts for heavy machinery, based on its own sales history and potential correlated variables. Thus, univariate models were estimated, in which the regression variable is the time-lagged variable itself, and multivariate models, in which exogenous variables were added along with the introduction of the dependent variable lagged in time, as a regression variable. ARIMA and SARIMA models were applied for the univariate models, and dynamic regression technique for the multivariate models, which is the combination of a regression model with an additional model to estimate the regression residuals, since, in time series, the residuals are autocorrelated. In this monograph, it was decided to estimate the ARIMAX models for the multivariate models, which use a transfer function to introduce covariates or exogenous variables to the model. The sales forecast for spare parts is important so that it is possible to balance the stock at satisfactory levels, keep costs at adequate levels and serve the customer in a timely manner.