Quantitative Methods in Empirical Finance: Insights into Economic Forecasting, Designing Financial Market Surveillance Systems and Modeling Extreme Returns

Abstract

The volume of high-frequency economic and financial data that is currently available facilitates the opportunity for a comprehensive analysis of important financial phenomena, but also necessitates the design and application of innovative statistical methods. Accordingly, this thesis presents three empirical studies at the intersection of the fields of financial market microstructure, financial econometrics and statistical learning.

The first study investigates how economist forecasts are related to trading activity in the over-the-counter treasury bond market. While economic forecasting is ubiquitous within the industry, its role in the trading process has received little attention. Consistent with models of heterogeneous opinions, we show that the forecasting economist’s employing institution places a disproportionately large but asymmetric reliance on the forecast. That is, only forecasts which imply a fall in future treasury bond prices are associated with an abnormal trading reaction. Empirical evidence shows that reference dependence and loss aversion may offer one explanation for this asymmetric trading response.

The second study designs a market surveillance model to detect illegal insider trading. Effective market surveillance underpins the integrity of financial markets. Now, more than ever, there is substantial pressure on financial institutions to develop more comprehensive compliance and regulatory technology. Our surveillance model combines nearest neighbour dynamic time warping, a pattern recognition algorithm for time series data, with contemporary methods for analyzing extreme value statistics to identify illegal trading patterns. Importantly this approach does not require access to any confirmed illegal transactions in the training phase. Using a high-frequency dataset provided by an investment bank, we demonstrate that our surveillance model achieves substantial improvements in the identification of illegal insider trading over alternative approaches.

The third study explores how exogenous information can be incorporated into a financial risk forecasting model and whether or not this additional information improves risk forecasts. Accurate risk forecasting is essential for maintaining the stability of financial markets. Yet, the majority of existing financial risk forecasting models ignore the wealth of economic and financial data which may provide additional information about the size and likelihood of future losses. We hypothesize that the optimal model is sparse and that the statistical importance of covariates is time-varying. Accordingly, Lasso regularization is integrated into our risk model. This regularized risk model dramatically improves financial risk forecasts during periods of elevated volatility and financial market stress, when accurate forecasts are needed the most. By studying the selection frequency and predictive duration of economic and financial covariates we provide a detailed insight into how and when exogenous information is related to financial risk.