FORECASTING SINGAPORE'S GDP GROWTH : COINTEGRATION AND ERROR CORRECTION APPROACH

Abstract

The main objective of this exercise is to use cointegration theory and its associated error correction model (ECM) to construct a small scale model for forecasting Singapore's GDP growth. Based on the aggregate demand and aggregate supply framework, thirteen variables were selected initially. Subsequently the dimensions of the vector of variables was reduced to four variables, namely GDP, money supply(M2), unit labour cost(index) and total OECD GDP(volume index). As a pre-requisite for the analysis of cointegration, two testing methods were utilized to test for unit roots. One is the Augmented Dickey & Fuller test (ADF) which was applied to deseasonalized data while the other is the Hylleberg et al. (HEGY) procedure which used the original data. Both tests indicated the presence of a zero-frequency unit root in each series. The HEGY test, in addition, showed the presence of seasonal unit roots as well. For this reason, the analysis was carried out with data deseasonalized by the filter S(L)=1+L+L2 +L3 such that s(L)Xt = Xt + Xt-1 + Xt-2 + Xt-3. Based on Johansen' s approach in determining the dimensions of the co integrating vector, the study found that there are two cointegrating relationships among the four selected macroeconomic variables. Hence, it was inferred that the GDP and the other three data series are cointegrated. Therefore, was constructed within an error correction model (ECM) a vector autoregressive (VAR) framework. In this model, the OECD output was treated as exogenous and the other three variables as endogenous. The forecasts of the three endogenous variables were obtained by assuming a certain growth rate for the OECD output. The ECM model appears to forecast Singapore's GDP growth quite well. By improving upon this type of simple model, it would be possible to obtain seasonally accurate forecasts which are otherwise obtained through very large and expensive models.