Multivariate unobserved components models in a globalised world

Student thesis: Doctoral Thesis


This thesis is a multi-country study and takes the literature on the country-specific unobserved components (UC) model as its starting point. In three increasingly flexible essays, we gradually extend the country-specific UC model to consider the affects of globalisation. Additionally, we propose to incorporate heterogeneities across countries in a data based fashion. The contribution is provided in three essays. In the first essay (Chapter 2) we estimate a country-specific unobserved components model with time-varying parameters and stochastic volatility. We consider 34 countries (23 advanced economies and 11 emerging market economies) and independent assumptions across countries are imposed. To consider the affects of globalisation, the model incorporates two observed global factors (oil price and global output) which account for global determinants of inflation. We find that inflation dynamics are explained by the combination of domestic factors (lagged domestic inflation and domestic output) and observed global factors (global output and oil price). Effects of these variables are constant over time. The Phillips curves are generally flat for the period under consideration (1995-2018), and different from zero. The global demand seems to matter more in emerging market economies than in advanced economies. Our results also point to the dominant role played by oil price as a key factor behind inflation dynamics over time. In the second essay (Chapter 3) we relax the assumption that errors across countries are independent, and allow for cross-country linkages in the error covariance matrix. First, we use the factor stochastic volatility specification to allow for cross-country linkages in the error covariance matrix. This method assumes that all countries' errors are driven by latent factors. However, one practical problem is that heterogeneities are very likely to exist in errors, since we include both advanced economies and emerging market economies. For that reason, we allow for stochastic volatility in all errors and propose a method to remove stochastic volatility in a data based fashion. We rewrite the process of log-volatility using the non-centered parameterization and impose the Horseshoe prior on the coefficient that controls time-variation in the log-volatility. We apply these methods to the data in Chapter 2. We find evidence that there are global factors driving all countries' inflation (output). The estimates under this model are in line with previous studies and, for certain countries, the estimates indicate that they are influenced by both domestic factors and global factors. Allowing for cross-country linkages in the error covariance matrix will decrease the persistence and flatten the Phillips curve. We also find that this model provides a superior in-sample fit and accurate density forecasts compared to existing models in the literature, especially if the period of uncertainty is the period being forecasted. Finally, in the third essay (Chapter 4) we further relax the assumption that the conditional mean depends on domestic factors, and allow for cross-country linkages both in the error covariance matrix and in the conditional mean. We name this model a panel unobserved components model. It extends Chapter 3 in three ways. First, it takes dynamic interdependencies into account by allowing for cross-country linkages in the conditional mean, more specifically, in coefficient matrices associated with lagged variables. Second, it takes static interdependencies into account. This is done through two blocks. One block is allowing for cross-country linkages in the error covariance matrix. Another block is allowing for cross-country linkages in the conditional mean, more specifically, in the coefficient matrix associated with the Phillips curve. Ther
Date of Award10 Dec 2021
Original languageEnglish
Awarding Institution
  • University Of Strathclyde
SponsorsUniversity of Strathclyde
SupervisorGary Koop (Supervisor) & Julia Darby (Supervisor)

Cite this