Restricted (Penn State Only)
Ni, Jun
Graduate Program:
Doctor of Philosophy
Document Type:
Date of Defense:
April 05, 2018
Committee Members:
  • Jingzhi Huang, Dissertation Advisor
  • Jingzhi Huang, Committee Chair
  • Anna L Mazzucato, Committee Member
  • Xiantao Li, Committee Member
  • Runze Li, Outside Member
  • VIX modeling
  • Bond premia prediction
  • Asset allocation
This dissertation contains three essays. The first part studies the continuous-time dynamics of VIX with stochastic volatility and jumps in VIX and volatility. Built on the general parametric affine model with stochastic volatility and jumps in the logarithm of VIX, we derive a linear relationship between the stochastic volatility factor and the VVIX index. We detect the existence of a co-jump of VIX and VVIX and put forward a double-jump stochastic volatility model for VIX through its joint property with VVIX. Using the VVIX index as a proxy for stochastic volatility, we use the MCMC method to estimate the dynamics of VIX. Comparing nested models of VIX, we show that the jump in VIX and the volatility factor are statistically significant. The jump intensity is also stochastic. We analyze the impact of the jump factor on VIX dynamics. The second part establishes a forecast framework for the bond excess return based on macroeconomics fundamentals. Empirical evidence has suggested that excess bond returns are forecastable with macroeconomics fundamentals. In our study, we build new links to tie the forecastable variation in excess bond returns to underlying macroeconomic series. Based on two types of models, the linear model and additive model, and utilizing different combinations of screening methods, nonlinearization techniques and regularization techniques, we extract different factor combinations from 131 macroeconomic series, including employment, housing, financial, and inflation factors. This approach results in stronger forecast power for the excess bond returns compared with existing macro-based return predictors. The nonlinear effect of the macroeconomic predictors on the excess bond returns is recovered if we incorporate nonlinearized macro data in the analysis. A horse race comparing different variable selection approaches allows us to propose a robust model that generates highly accurate predictions of bond risk premia. Finally, we perform a comprehensive analysis of risk premia with an ETF dataset. The third part of this dissertation is a summary of traditional asset allocation methods performance on Chinese market. Since traditional asset allocation methods are well analyzed in US capital market, similarly, we want to conduct a comprehensive analysis of asset allocation techniques on Chinese market. Based on a horserace comparison among the trading performance by different asset allocation approaches with investment universe of Chinese capital market indices and the associated ETFs, we achieve a clear understanding on the relative ranking of different methods, finding the link between trading performance with different parameter estimation time windows and different investment universe as well. To explain the difference in the trading performance of several methods, we perform a simulation study and attribute bad performance as the inaccuracy of return estimation.