A Case Study to Compare Multivariate and Univariate Cointegration Analyses in Small Samples

In general, the small number of observations affects the validity of the estimation results. It is well known with the small sample the cointegration test can be subject to size and power bias. The purpose of this paper was to compare multivariate and univariate cointegration analyses in small samples. A simple inflation model is specified that includes liquidity (M2), government budget deficit, and official exchange rate, as well as the wheat support price as a macroeconomic approach. We report small sample results for the maximum-likelihood (ML) approach of Johansen and Juselius (0991), the Fully Modified OLS procedure, and Autoregressive Distributed Lag (ARDL) modeling that was originally used in simple model. Quantitative estimates, based on the time series annual data from 0991 to 2112, indicate that the long run estimated coefficients in ARDL approach the point of view of size element with the ML and FMOLS cointegration approaches are symmetrical

Keywords:

Small Sample , Multivariate Cointegration , Univariate Cointegration