Bootstrapping methods and applications on Turkish macroeconomic variables

This paper mainly has two aims. In the first place, an extensive discussion on bootstrap methods used in statistical estimation and inference with cross section and time series data sets is presented. In particularly, dynamics through which bootstrap refines the statistics of interest and outperforms classical statistical approach are discussed in detail. In addition, a special attention is paid to the application of bootstrap with different data sets such as cross section and time series and how the procedure must be conducted are explained comprehensively. As the second aim of this study, three bootstrap applications with Turkish macroeconomic time series are performed. Because stationarity tests are quite common among empirical practitioners and because they have explicit implications in terms of long term cointegration relationships between time series, the first two applications of this study focus on bootstrapped unit root tests under the presence of small samples where bootstrap approach is known to be superior to conventional unit root tests suggested in literature. Besides, bootstrap unit root tests results and unit root tests allowing for single and multiple endogenous structural breaks are compared. The third application of the study is devoted to obtain bootstrap estimation of the Keynesian consumption function for the Turkish economy, by employing an auxiliary method used in determining the number of structural breaks occurring in the relationship between the series involved in the model. Results show that conventional and bootstrapped unit root test results show substantial differences on the presence of unit root. Furthermore, bootstrapped unit root test results are highly in line with the results suggested by the unit root tests allowing for single and multiple structural breaks. Furthermore, the Keynesian consumption function estimated by Ordinary Least Square and bootstrap method differ significantly in terms of the magnitude of the model coefficients estimated across three regimes determined endogenously.