Variational inference for a large dimensional Markov switching model
[DRAFT AVAILABLE ON REQUEST]
Abstract: The multivariate Markov switching model identifies bull and bear markets from disaggregated indices and allows for asset allocation decisions. However, the Monte Carlo Markov chain algorithm's computational cost increases significantly with its dimension and quickly reaches a prohibited computational time. I propose a new variational inference (VI) algorithm to estimate a large-dimensional Markov switching model fast and accurately in the bull and bear markets. While taking substantially less time to compute, this method achieves comparable in-sample and out-of-sample results to its MCMC counterpart. In addition, this inference allows for the inclusion of important restrictions to identify hidden market states. The forward filtering backward smoothing algorithm of my novel VI is also common in economic literature. My simulation studies emphasize the accuracy and timely benefit of the new technique in, for example, identifying the bull and bear states, detecting regime switching, and providing forecasts for investment strategies. I investigate the empirical applications of three sets of stock returns that are listed in the S&P 500 and one set of industry portfolios and find similar insights.
The dynamic Dirichlet process mixture model: An application to the banking dynamics in the long run
[DRAFT AVAILABLE ON REQUEST]
(with Yong Song, Mei Dong, and Silvio Contessi)
Abstract: This paper proposes a new dynamic Bayesian nonparametric model to capture time varying distributions. Our model is built on Gutierrez et al. (2016) and Mena and Ruggiero (2016). We improve their algorithm through incorporating a break indicator and a hierarchical prior structure governing the parameters of all components in the mixture. Using the Australian banking statistics from 1925 to 2019, we apply the model to estimate the time varying bank size/growth distributions. Our results suggest that the skewness of the weighted bank growth distribution is procyclical to the business/financial cycle. Different quantiles of the weighted bank growth distribution exhibit different correlations with financial cycles.
Variational inference for dynamic Dirichlet process mixture model (with Yong Song)
[DRAFT AVAILABLE ON REQUEST]
Heterogeneity in the Banking System in the Long Run: Evidence from New Australian Data Spanning a Century (with Silvio Contessi, Mei Dong, and Ainura Tursulunieva)