Essays on Risk in the Financial Market

The first of two papers in my dissertation investigates to what extent the time-variation in portfolio allocation of U.S. investors can be accounted for by time-varying risk. The topic of international portfolio allocation has received significant attention in the literature in the context of portfolio home bias. The focus of the literature has been mainly on the cross-sectional characteristics of portfolio holdings, considering the allocation of portfolios across countries at a point in time. Not much work has been done on the time series dimension of international portfolio allocation, largely as a result of data limitations. This has changed due to the recent availability of carefully constructed monthly data on equity positions of U.S. investors across over 40 countries. This paper investigates to what extent the time-variation in portfolio allocation of U.S. investors can be accounted for by time-varying risk. This involves both changes in the variance of U.S. and foreign equity returns and their covariance. This time-varying risk is estimated through a multivariate GARCH model. The portfolio allocation model adopts a simple mean-variance framework, modified to include information frictions that generate the observed portfolio home bias. We find that time-varying risk can account for 40% of the variance of the portfolio share invested abroad by U.S. investors based on monthly data. The model can account for most of the large drop in the share invested abroad during the 2008 financial crisis.;The second paper explores the linkages between realized volatility of US equity returns with certain financial variables. Linear regressions of assets' returns on sets of putative explanatory factors typically yield low R-squareds. Even less successful are attempts to infer from such linear models how the factors account for returns' "realized" volatility. We show that the problem is caused by instability of estimated factor exposures and propose a direct and more successful way of attributing volatility. The method, employing high-frequency data, involves a nonlinear regression of returns' realized volatility on the realized covolatilities of factors. We find that about 40% of the realized volatility of US equity returns can be explained by a set of plausible factors for which such high-frequency data are available.