Research
WORKING PAPER:
A semi-parametric study on dynamic linkages among international real interest rates
Abstract: Even in a perfectly efficient international interest rate market, equal real interest rates are thought not to be expected because of the existence of transaction costs and nonlinear approaches are necessary to evaluate the relationship between real rates. Through nonlinearity tests, we find strong evidence that nonlinearities substantially characterize real interest rate linkages. Then we consider the semi-parametric method to evaluate generalized additive models, and the generalized impulse response functions are calculated to investigate the dynamics of real rates after shocks of different sizes, different directions, and with different histories. All marginal effects from four models backed by different economic theories and generalized impulse responses from vector error correction models support the real interest rate parity hypothesis which indicates a high degree of integration in capital markets. Consequently, the conventional tests which rely on linearity assumptions may lead to misleading inferences.
Forecasting Chinese Inflation using SW Unobserved-Components Models
Abstract: This paper utilizes the nonparametric method to uncover the nonlinear relationship between money growth and inflation, finding the effect from money growth on inflation is significant only in the volatile money-growth period, which implies that using the inflation data-only model may provide reliable forecasts after 2015. We examine whether two unobserved-components (UC) models, which only use price data, can improve Chinese inflation forecasts. We estimate monthly Chinese inflation during the period December 2006--April 2019 based on these two UC models. Then we perform an out-of-sample inflation forecasting competition among UC models, including Bayesian vector autoregression models (BVAR) and autoregression models (AR) . The forecasting evaluation period was from January 2015 to April 2019. We find the multivariate UC model provides better forecasts compared with the univariate UC model according to the mean square error criterion, which means disaggregated inflation data can help improve inflation forecasts. Furthermore, both UC models performed better than BVAR in all horizons, which provides new empirical evidence about whether Chinese M2 growth rate or other macroeconomic variables are important for forecasting inflation.
A semi-parametric study on dynamic linkages among international real interest rates
Abstract: Even in a perfectly efficient international interest rate market, equal real interest rates are thought not to be expected because of the existence of transaction costs and nonlinear approaches are necessary to evaluate the relationship between real rates. Through nonlinearity tests, we find strong evidence that nonlinearities substantially characterize real interest rate linkages. Then we consider the semi-parametric method to evaluate generalized additive models, and the generalized impulse response functions are calculated to investigate the dynamics of real rates after shocks of different sizes, different directions, and with different histories. All marginal effects from four models backed by different economic theories and generalized impulse responses from vector error correction models support the real interest rate parity hypothesis which indicates a high degree of integration in capital markets. Consequently, the conventional tests which rely on linearity assumptions may lead to misleading inferences.
Forecasting Chinese Inflation using SW Unobserved-Components Models
Abstract: This paper utilizes the nonparametric method to uncover the nonlinear relationship between money growth and inflation, finding the effect from money growth on inflation is significant only in the volatile money-growth period, which implies that using the inflation data-only model may provide reliable forecasts after 2015. We examine whether two unobserved-components (UC) models, which only use price data, can improve Chinese inflation forecasts. We estimate monthly Chinese inflation during the period December 2006--April 2019 based on these two UC models. Then we perform an out-of-sample inflation forecasting competition among UC models, including Bayesian vector autoregression models (BVAR) and autoregression models (AR) . The forecasting evaluation period was from January 2015 to April 2019. We find the multivariate UC model provides better forecasts compared with the univariate UC model according to the mean square error criterion, which means disaggregated inflation data can help improve inflation forecasts. Furthermore, both UC models performed better than BVAR in all horizons, which provides new empirical evidence about whether Chinese M2 growth rate or other macroeconomic variables are important for forecasting inflation.