This thesis, through three empirical applications, provides an analysis of extreme events in financial markets. Robust growth in financial markets has greatly increased the ability of economic agents to share risk according to their preferences or tastes. Despite this, many markets have demonstrated extreme instability at times. These events have the potential to shake the confidence of investors and this fear can lead to inefficient outcomes with respect to risk sharing and resource allocation. By investigating the dynamics of securities during extreme events one can gain intuition as to their root causes and a better understanding of the inherent risk.

The first chapter analyzes how international equity markets interact during extreme events. Using a novel set of high-frequency data on exchange traded funds (ETFs), designed to track international equity markets, I examine the dynamics of intra-day returns between 11 countries. Using non-parametric tests designed to identify jumps in the price process I examine the dynamics across markets during jumps, as well as continuous movements. Contrary to other literature that uses coarser data, I find a high-degree of commonality in the jump components. Specifically, there are many instances when different markets co-jump and returns are significantly more correlated on jump days. I also find substantial evidence of self and cross excitation across markets and that international markets respond to US macroeconomic news announcements. These findings suggest that international financial markets are heavily intertwined and that shocks propagate across markets. This information is valuable from a modeling perspective as it provides evidence of channels through which economies are linked that must be accounted for. Further, it provides valuable information to investors into the benefits and risks associated with international diversification that allows them to take a more proactive, rather than a reactionary, approach to risk management.

The second chapter, based on Gualtieri and Sizova (2015), investigates the joint dynamics of portfolios considered to represent priced risk in asset markets. Specifically, it considers the joint modeling of the market return, and two zero net cost portfolios that are used as proxies for systematic risk factors: Value and Momentum. As in the case of chapter 1, we allow for a separation between continuous and jump dynamics. We find a number of interesting relationships between factor dynamics that have implications for risk-based explanations of factor risk premia as well as factor investing. Specifically, we find that although volatilities are highly correlated, the orthogonal (to the Market volatility) component of Momentum volatility contains information about the Market's dynamics. With respect to extreme events, we find that volatility co-jumps are present in both-return pairs (Market-Momentum and Market-Value). We find that Value does not jump independent of the Market, whereas Momentum does. We also find that a number of the Momentum jumps occur in bear markets, which is consistent with documented Momentum crashes (see for example Daniel and Moskowitz (2013). We also use the model output to investigate the merits of factor investing. We estimate a variety of metrics on jump days to analyze the benefits of diversifying away from the market and into additional stylized portfolios. We find that the a combination position in the Market and Value significantly improves performance during extreme events in terms of average loss, volatility and value-at-risk.

Aside from the empirical analysis we also provide a generalization of the univariate stochastic volatility conditional jump (SVCJ) model of Eraker et al. (2003) to the multivariate case. We provide a detailed appendix documenting the sampling scheme that can be used to investigate joint dynamics in extreme events.

The third chapter, based on Bada et al. (2015), examines whether algorithmic trading (AT) has a time varying effect on measures of liquidity such as bid-ask spreads and volatility. Specifically, in the context of a panel model with individual and time fixed effects we allow for structural breaks in the slope parameters at an unknown number of times and automatically detect the break points. The model is free from any ad-hoc identification of break points or restrictions on the number of breaks imposed by the econometrician a priori. The study is the first to use this estimator (in any context) and the results show clear evidence of breaks in the relationship between AT and liquidity during the financial crisis. These results are in contrast to prior literature that demonstrates a clear positive relationship between AT and market liquidity. The timing of the breaks is important as the merits of added liquidity during relatively stable periods versus its withdrawal during periods when it is in high demand are somewhat ambiguous and may possibly present a net welfare loss to society. The results indicate the presence of a state contingent relationship between AT and liquidity.