Nonlinear Combination of Time-Series Forecasting Models, a Case Study with ARIMA and Prophet

Analyzing and predicting time-series data is a fundamental and widely used technique in economics. For instance, predicting monthly product sales figures can be influenced by various external factors, but analyzing historical data can uncover significant patterns such as trends and seasonality. To achieve precise predictions, advanced models have been proposed in the literature. Moreover, combining different models can enhance the accuracy of predictions. In this paper, we propose a nonlinear combination of models to create a more accurate time-series model. While we focus on two primary categories of models, namely the Prophet model and the ARIMA model, our proposed method is not limited to these two models.