The Use of Principal Component Analysis (PCA) In Modeling Yield Curve Dynamics of The Romanian Government Bond Market

Based on previous research addressing the use of Principal Components Analysis (PCA) in modeling the dynamics of sovereign yield curves, we investigate certain characteristic of the Romanian bond market and try to build yield curve scenarios that are historically plausible and of plausible shape and magnitude. The empirical analysis performed on weekly data revealed that changes at slope level occur according to changes implied by the first principal component. Therefore, when a market movement of large amplitude is expected, predicting the yield curve based on the first principal component appears to be the most historically plausible approach. An interesting finding in our research was that the first principal component went from explaining about 68 per cent of the yield curve changes in February 2020, to explaining roughly 93 percent in April 2021, suggesting that the main pattern of government yield changes throughout that year was that of a more parallel appearance. Also, the principle components coefficients (factor loadings) at 2 years were lower than those at 10 years, suggesting that in the case of market rallies, yields at 10 years will lower more than those at 2 years, leading to a flattening pattern of the yield curve. The analysis performed on the Romanian government bond market at different points in time -before the pandemic and one year after its start- showed that the first three principal components (depicting changes in level, slope and curvature) explained more than 93 per cent of the yield curve changes in all cases. However, one year after the pandemic they explained up to 98 per cent of the term structure dynamics.