Um estudo sobre a provisão de consórcio com séries temporais e machine learning

Abstract

The consortium sector of a particular company works by dividing its customers into groups that start and finish at the same time the payment of installments of a combined amount to obtain their own vehicle. The acquisition of this car is carried out by drawing lots that take place at any time between the payment of the installments. The provision is a capital that the company needs to maintain in the event that customers in this group withdraw from paying these installments. If, at the end of the group, the company is unable to obtain the amount it had determined, another sector of the same organization will supply the necessary with the loan that will be essential for the group. Therefore, the greater the group’s default, the greater the loan amount that the company must reimburse in the consortium. In order to find ideal models that can estimate the necessary forecast of the provision and the value of the loan over time, two of the usual time series models were used: ARIMA and Exponential Smoothing - Holt Winters, which more approached the predicted values; and a Booosting - Machine Learning application.