Testing the predictive power of term spread in the Euro area

Abstract

This thesis tests the predictive power of term spread in predicting the Euro area's real economic activities. The objectives of this study are to test the predictive power of term spread in the negative interest rate period in the Euro area, to examine the joint predictive power of term spread and EPU, and to reveal the Granger causality of the variables.

Term spread and GDP growth rate are significant variables; however, the term spread model is augmented with EPU. Term spread is derived from the three-month interest rate and triple ‘A’ rated ten-year government bond. The sample of this thesis ranges from 1999Q1 to 2019Q4. The in-sample model fit is tested with the full sample data, and the out-of-sample prediction is tested using the data before the negative interest rate period in the Euro area. The vector autoregression method is used in this study; furthermore, a linear model is estimated using some dummy variables such as the financial crisis 2008-9, high uncertainty period, and negative interest rate period.

The following are the five most significant findings of this thesis. First, the predictive power of term spread is low, but it has slightly increased during the negative interest rate period. Although term spread's predictive power is increasing, the estimate coefficients of term spread are not statistically significant yet. Such a low predictive power of the term spread is found in Germany, Italy, Spain, Belgium, Ireland, and Finland. Only in France, term spread has significant predictive power. Second, the relatively low predictive power of term spread is observed particularly during the recession caused by the European sovereign debt crisis and during the high uncertainty period. Third, the lags of GDP growth rate have better predicting power than the term spread has. The model’s adjusted R2 decreases by only 0.01 when term spread is removed from the independent variables, but the adjusted R2 drops from 0.93 to 0.61 as the lags of GDP are removed from the independent variables, indicating that the real economic activities in the Euro area can be better predicted by GDP growth rate’s lags than by term spread. Fourth, the estimate coefficients for EPU are almost zero and it cannot increase the model's predictive power either. Last, term spread Granger causes GDP growth in lower lags, optimally at lag two. A fragile form of bidirectional Granger causality between term spread and GDP growth rate is observed, while EPU does not Granger cause the GDP growth rate at all.