Mixed data sampling (MIDAS) method: Theory and application

As known, the most important requirement for traditional time series regression models is that all dependent and independent variables in the model must be at the same frequency. However, this requirement is not always ensured because economic and financial variables are released at different frequencies. The traditional solution of this aforementioned requirement is to perform aggregation method in applied literature. However, it is possible that the useful and necessary information in the high frequency variable will be likely lost as a result of aggregation. Gyhsels et al. (2004) developed a method in which variables with different frequencies can be used in the same model in order to eliminate this problem in the literature. This method is called Mixed Data Sampling (MIDAS). The MIDAS method enables high frequency variables to be included in multivariate models without being subjected to aggregation. With the MIDAS method, the importance of using high frequency information in forecasting of the economic growth rates of countries has increased in the relevant literature. The aim of this study is the real-time application of nowcast of Turkey's economy quarterly frequency of Gross Domestic Product (GDP) growth rates during a given time interval by benefiting monthly frequency variables under the MIDAS method. In the analyses, the growth rate of GDP between the 1st quarter of 2015 and the 2nd quarter of 2020 was nowcasted in real time by providing data sets accessible in June 2020 for monthly and quarterly frequency variables. In the study, the nowcasting performance of the MIDAS model was compared with the forecasting performance of the conventional model with aggregated variables. As a result of this comparison, it is determined that more accurate forecast is generally obtained with the MIDAS model.