A comparative study of univariate time-series methods for sales forecasting

Firms use time-series forecasting methods to predict sales. However, it is still a question which method forecaster best, if only single forecast needed. This study investigates and evaluates different sales methods: multiplicative Holt-Winters (HW), additive HW, seasonal auto regressive integrated moving average (SARIMA) [a variant of (ARIMA)], long short-term memory (LSTM) recurrent neural networks the Prophet by Facebook on 32 univariate time-series. The data used taken from Time Series Data Library (TSDL). With respect root mean square error (RMSE) evaluation metric, we find that with SARIMA offers best performance, average, relative other compared methods. To support findings, both mathematical economic drivers observed performance are provided.