Three essays on market impact and systemic risk

This thesis is divided into two parts. The first part consists of two chapters, while the second part consists of one chapter. Chapter 2, starts by considering the relationship between trades and midpoint changes, typi- cally modeled using constant parameter Structural Vector AutoRegressive (SVAR) models as in Hasbrouck (1991a). A parameter called instantaneous impact represents the contemporary rela- tionship between trades and prices. It is a liquidity measure and quantifies a trade’s simultaneous impact on price. In general, the market or price impact corresponds to the average response of prices to trades, i.e., the effect of the trading activity on prices. According to the law of supply and demand, if a trader buys shares, the price is pushed, on average, upward and vice versa downward. When a buy/sell occurs, the effect on prices does not vanish immediately, but it dissi- pates after a while and converges to a level called permanent impact. In Hasbrouck’s framework, the latter quantity is named the information content of stock trades. The main idea behind our modeling approach is that the instantaneous impact is not constant during the day and depends on the state, or the history of prices and trades of the market at a given time. The interest of the study in the intraday variation of the instantaneous and permanent impact is motivated by the fact that these financial quantities are fundamental for traders for efficient transaction cost analysis and optimal execution strategies. In Chapter 2, we jointly model the midprice return and the sign of trades by introducing several significant modifications to the Hasbrouck (1991a) model. In particular, we use the heterogeneous aggregation technique of Corsi (2009) to exploit the long memory property of the order flow parsimoniously. Then, instead of directly modeling the trade sign, we decided to model the probability that a trade occurs, introducing a logistic specification for the trade sign equation. Our choice avoids complicated dependencies in the residuals of the trade equation and allows one to compute the likelihood of the model. Taking into account the intraday variation of the liquidity and volatility of the market, we let the instantaneous impact parameter and the volatility vary over time. For the time-variation of the parameters, we rely on the Dynamic Conditional Score Driven (DCS) framework of Creal et al. (2008) and Harvey (2013). The instantaneous impact shows a huge intraday variation and a decreasing pattern. Our analysis reveals that the information content of stock trades depends positively on the instantaneous impact and negatively on the state of the market. This chapter brings another important contribution to the transaction cost analysis: we derive a closed-form expression for the time-varying permanent impact coefficient. Our approach provides an efficient real-time estimation of the permanent impact. In Chapter 3, we study the execution of metaorders, which are large trading orders split into small orders executed incrementally. Chapter 3 continues the studies of Chapter 2. The main difference is that in Chapter 3, we analyze the effect of (signed) volume on prices of a metaorder instead of the sign of a single transaction. We deepen the analysis and introduce the concepts of realized and temporary impacts. The realized impact is the average price paid during the execution of the metaorder. In contrast, the temporary impact is the short-time price fluctuations caused by the execution of the metaorder. In Chapter 3, we face the nontrivial problem of estimating the permanent and temporary impact coefficients without proprietary metaorder data. We propose a general framework for assessing the impact of metaorders based on a novel definition of a nonstandard-Impulse Response Function capable of reproducing the incremental execution of shares. The proposed methodology allows us to derive the permanent and the temporary impact coefficients in a closed-form expression, relying solely on the market data. The knowledge of these coefficients is important for practical reasons. Indeed, the permanent and the temporary impact coefficients are fundamental for traders for efficient transaction cost analysis and optimal execution of orders. For the application, we consider the Hasbrouck (1991a) model and the Transient Impact Model of Bouchaud et al. (2004). Then, we modify them by aggregating regressors in a similar spirit of Corsi (2009). Empirical results show that our methodology is easy to apply, does not require proprietary data, considers the dynamic response of the order book, and produces appropriate estimates of the permanent and the temporary impact coefficients. In Chapter 4, we focus on a different topic. We address the issue of modeling negative interest rates as a proxy for economic (in)-stabilities. Specifically, we study the financial time series of bond yields and their importance for the well-being of a given macroeconomic area. After the 2011 sovereign debt crisis and the Quantitative Easing (QE) launched in 2015 by the European Central Bank (ECB) as a response to it, two main empirical facts emerged: the possibility of negative interest rates and common shocks affecting similarly common countries. In Chapter 4, we consider the model of Recchioni and Tedeschi (2017) for stochastic interest rates with common volatility, which is suitable for reproducing the characteristics of the data. The maini dea behind the model is to take into account a system of n stochastic differential equations, representing the interest rates of the different countries in the eurozone, and a stochastic differential equation of type Heston (1993), representing the common source of stochasticity affecting the yields. After carefully analyzing the dataset, we calibrated the model on a set of European yields with different maturities. Empirical results indicate that the parameter’s values contain information on the investigated area’s economic conditions, which also depends on the maturity of the yields. The result of the estimates allows us to develop an early warning indicator incorporating macroeconomic factors. We show that such an indicator is sensitive in the event of changes in market conditions. An OLS analysis reveals that the indicator, which is a function of the model’s estimated coefficients, is correlated with macroeconomic variables. The results indicate that the early warning indicator is positively and statistically correlated with changes in GDP and return on yields.