Forecasting models: an application to home insurance

Abstract

Forecasting in time series is one of the main purposes for applying time series models. The choice of the forecasting model depends on data structure and the objectives of the study. This study presents a comparison of Box Jenkins SARIMA and Holt-Winters exponential smoothing approaches to time series forecasting to increase the likelihood of capturing different patterns in the data (in this specific case, home insurance data) and thus improve forecasting performance. These methods are chosen due to their ability to model seasonal fluctuations present in insurance data. The forecasting performance is demonstrated by a case study of home insurance monthly time series: total and frequency rate time series. In order to assess the predictive and forecasting performance of the two methodologies adopted, several evaluation measures are used, namely MSE, RMSE, MAPE, and Theil's U-statistics. A comparison is made and discussed, and the results obtained demonstrate the superiority of the SARIMA model over the other forecasting approach. Holt-Winters also produces accurate forecasts, so it is considered a viable alternative to SARIMA.