Essays on high-frequency financial econometrics
My dissertation consists of three essays which contribute new theoretical and em- pirical results to Volatility Estimation and Market Microstructure theory as well as Risk Management. Chapter 2 extends the ACD-ICV method proposed by Tse and Yang (2012) for the estimation of intraday volatility of stocks to estimate monthly volatility. We compare the ACD-ICV estimates against the realized volatility (RV) and the generalized autoregressive conditional heteroskedasticity (GARCH) estimates. Our Monte Carlo experiments and empirical results on stock data of the New York Stock Exchange show that the ACD-ICV method performs very well against the other two methods. As a 30-day volatility predictor, the Chicago Board Options Exchange volatility index (VIX) predicts the ACD-ICV volatility estimates better than the RV estimates. While the RV method appears to dominate the literature, the GARCH method based on aggregating daily conditional variance over a month performs well against the RV method. Chapter 3 propose to model the aggregate trade volume of stocks in a quote- driven (specialist) market using a compound Poisson distribution. Trades are as- sumed to be initiated by either informed or uninformed traders. Our model treats trade volume endogenously and calibrates two measures of informed trading: rela- tive frequency of informed trading and relative volume of informed trading. Empiri- cal analysis of daily volatility estimates of 50 NYSE stocks shows that trade volume initiated by informed traders increase volatility, while trade volume initiated by un- informed traders reduce volatility. However, for both informed and uninformed traders, the disaggregated effect of trade frequency is to increase volatility. Our results also confirm that trade frequency dominates trade volume and trade size in affecting volatility. Yet trade volume and trade size have incremental information for volatility beyond that exhibited in trade frequency. Chapter 4 propose to estimate the intraday Value at Risk (IVaR) for stocks using real-time transaction data. Transaction data filtered by price durations are modeled employing a two-state asymmetric autoregressive conditional duration (AACD) model, and the IVaR is computed using Monte Carlo simulation. Empirical analysis of New York Stock Exchange (NYSE) stocks show that IVaR estimated using the AACD ap- proach track closely to those using the Dionne, Duchesne and Pacurar (2009) and Giot (2005) methods. Backtesting results show that our method performs the best among other methods.