Laurs Randbøll Leth defends his PhD thesis

Candidate: Laurs Randbøll Leth

Title: "Essays on High-Frequency Market Microstructure: Herding and Volume-Synchronized Probability of Informed Trading".

Time and venue
23 September 2019 at 14:30, University of Copenhagen, CSS, Øster Farimagsgade 5, 1353 Copenhagen K, building 4, room 4.2.26.

Evaluation committee
Professor Anders Rahbek, Department of Economics, University of Copenhagen, Denmark (chairman)
Professor Antonio Guarino, Depart. of Economics University College London, UK
Professor David Easley, Cornell University, USA

Abstract 
This thesis consists of three self-contained chapters within financial market microstructure. The focal point is the modeling and implications of information-based trading in financial markets affected by high-frequency traders.

Chapter 1 (“Rational Herding During a Stock Crash“) examines the extent of herd behavior in a stock market from 2005 to 2008. I consider an asymmetric information model with event uncertainty. This setting enables rational statistical herding to occur with positive probability. The model is fitted to the financial tick data of a NYSE traded stock, and findings reveal that the proportion of herd sellers increased during the Great Recession. This suggests that herd behavior may provide a part of the explanation of financial crises.

Chapter 2 (“Delta Hedging and the VPIN“) investigates how portfolio hedging can take advantage of a microstructure measure (VPIN) of toxic order flow. Suppose a portfolio manager with a short position in a European call option attempts to collect its volatility risk premium by engaging in dynamic delta hedging of the underlying asset. The results show that VPIN signals large intraday price movements leading to losses of the hedging portfolio. Conditional on high VPIN readings, the portfolio manager is advised to either expand her portfolio with VIX futures and/or adjust her daily risk exposure.

Finally, Chapter 3 (“Maximum Likelihood Estimation of VPIN: Toxic Order Flow and Warning Signals“) investigates if maximum likelihood estimation of the VPIN will improve its predictive power for short-term return volatility. This assessment is based on the classification of true and false positive events after high VPIN values were detected. Compared to the method of moment estimation, I show that maximum likelihood estimation is superior in terms of a lower false discovery rate.“