A new decomposition-based model for forecasting health expenditures

Increasing access to and demand for healthcare services day by day also causes healthcare expenditures to increase. Therefore, planning, tracking, and forecasting health expenditures; is very important to maintain health care policies at the highest quality. In this thesis, a new forecasting model based on time series decomposition called Trend-Residual (T-A) model is proposed to model and forecast Türkiye's total health expenditure. The proposed T-A model has a cascade structure and tries to model and estimate the two subcomponents of the health expenditure time series (trend and residual) separately. The T-A model first determines the trend component of the total health expenditure time series with the polynomial regression model, which has many advantages. Then, a new residual model is developed with the linear parameters optimized by the Least Squares Estimation method and the non-linear parameters by the Neural Network Algorithm, and the trend removal time series, that is, its residual component, is modeled. Türkiye's total health expenditures between 1999 and 2021 were used as the data set. Modeling and prediction performance of the proposed T-A model; compared to gray models, regression models, exponential smoothing models, and ARIMA models. Comparison results show that the modeling and forecasting performance of the proposed T-A model is better than other models. As a result, according to the forecast results obtained with the T-A model, it is predicted that the total amount of health expenditure will reach 2.2 trillion TL in 2030 and increase approximately five times from 2022 to 2030.