This study analyzes the dynamic process of yield spread differences in 15 emerging countries’ bond markets by using alternately the state-space model and the threshold-GARCH (T-GARCH) model. Using the discrete Kalman filter technique, we decompose yield spread differences into fundamental and transient components in the adoption of the first order state-space model estimation. We then further decide whether spread changes coincide with both global and country-specific explanatory factors by adoption of T-GARCH estimation after extracting the fundamental values from the yield spread observations. This procedure can be taken by comparing the T-GARCH estimation results of the observed and the derived fundamental values from the state-space model. The existence of overreaction by investors’ psychological situation and also its magnitude can be checked in the emerging bond markets if the conditional volatility of observed spreads turned out to be significantly positive upon arrival of unanticipated negative shocks, whereas that of the fundamental one turned out not to be significant. In this analysis, EMBIGS (emerging bond market global index spread) from JP Morgan Chase are used as data for country yield spreads. For weekly data in use, the sample period covers the April 1998 to December 2008 period and the sample emerging countries cover 6 from Latin America, 4 from Europe and 5 from Asia. Among them, Korea and Thailand have data only up to Aril 2004 and March 2005, respectively, at which points they were excluded from the emerging market status. Major findings of this study are as follows. First, yield spread differences in emerging markets are mainly altered by their transient values rather than by their fundamental values, as the empirical results show that the fundamental values are almost deterministic in the long run and the transient values explain most of changes in spread. This is because three different crises such as the Russian moratorium in 1998, the Argentinean currency crisis in 2002 and the global financial crisis in late 2008 which all occurred during the sample period can be absorbed into the transient value process, whereas the fundamental values remained stable. This can be a meaningful evidence in that real spread changes reveal in the state of overreaction, especially in the bearish trend of the emerging bond markets during the sample period.. Second, in the procedure of T-GARCH estimation, clear evidence of overreaction in observed yield spread differences is found such that the impact on the conditional variance of the unexpected rise in observed spreads is significantly positive, whereas that for the unexpected rise in fundamental spreads does not show any significance. This evidence can be interpreted as that investors’ psychological attitude result from their on-going noise trading.. Third, in the estimation of multiple regressions in the T-GARCH model where two global factors, US T-bond rate and VIX, and two country-specific factors, domestic stock index return and US dollar based foreign exchange rate, are added as explanatory variables for the changes of spread, the same evidence of overreaction as in the above simple T-GARCH model is found. All the four estimators appear to coincide with the theoretical hypotheses in estimating the observed yield spreads. However, in the estimation of the fundamental yield spreads, most of the coefficients on the explanatory variables turn out to be insignificant. This may be partly because our model could not catch the structural break such as the October Crisis of 1987 in the US, using only a first order state-space model with a discrete Kalman filter technique. A more precise estimation technique such as the extended Kalman filter in the nonlinear estimation model need to be used to overcome this weakness, which is left for future research.
Read full abstract
Jan 22, 2012
I I Dedov 
Open Access
