Essays on Market Microstructure and Financial Markets Stability

The present doctoral thesis consists of three main chapters. The chapters of the thesis can be considered independently. Each of the three chapters raises a research question, reviews the related literature, proposes a method for the analysis, and, finally, reports results and conclusions.

Chapter 1 is entitled Dark Trading and Financial Markets Stability and it is based on a working paper co-authored with Prof. Dr. Jorge Goncalves and Prof. Dr. Roman Kraussl. This paper examines how the implementation of a new dark order -- Midpoint Extended Life Order (M-ELO) on Nasdaq -- impacts financial markets stability in terms of occurrences of mini-flash crashes in individual securities. We use high-frequency order book data and apply panel regression analysis to estimate the effect of dark order trading activity on market stability and liquidity provision. The results suggest a predominance of a speed bump effect of M-ELO rather than a darkness effect. We find that the introduction of M-ELO increases market stability by reducing the average number of mini-flash crashes, but its impact on market quality is mixed.

Chapter 2 is entitled Dark Pools and Price Discovery in Limit Order Markets and it is a single-authored work. This paper examines how the introduction of a dark pool impacts price discovery, market quality, and aggregate welfare of traders. I use a four-period model where rational and risk-neutral agents choose the order type and the venue and obtain the equilibrium numerically. The comparative statics on the order submission probability suggests a U-shaped order migration to the dark pool. The overall effect of dark trading on market quality and aggregate welfare was found to be positive but limited in size and depended on market conditions. I find mixed results for the process of price discovery. Depending on the immediacy need of traders, price discovery may change due to the presence of the dark venue.

Chapter 3 is entitled Machine Learning and Market Microstructure Predictability and it is another single-authored piece of work. This paper illustrates the application of machine learning to market microstructure research. I outline the most insightful microstructure measures, that possess the highest predictive power and are useful for the out-of-sample predictions of such features of the market as liquidity volatility and general market stability. By comparing the models' performance during the normal time versus the crisis time, I come to the conclusion that financial markets remain efficient during both periods. Additionally, I find that high-frequency traders activity is not able to forecast accurately neither of the market features.