Three essays on financial markets
- Autores: Julio Alberto Crego Cobelo
- Directores de la Tesis: Dante Amengual (dir. tes.), Enrique Sentana Iváñez (codir. tes.)
- Lectura: En la Universidad Internacional Menéndez Pelayo (UIMP) ( España ) en 2017
- Idioma: inglés
- Tribunal Calificador de la Tesis: Thomas Gehrig (presid.), Maria Teresa Gonzalez Perez (secret.), Mark Van Achter (voc.), Sabrina Buti (voc.), Roberto Pascual Gascó (voc.)
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
Recent events, such as the nancial crisis and the proliferation of ash crashes, have highlighted the important role of nancial markets in the economy. This thesis aims to better understand several features of these markets and the eects of dierent policies that have been implemented.
In the rst chapter, I study how liquidity changes around public information releases.
I restrict the empirical analysis to the publication of the Weekly Petroleum Status Report which has three features relevant to identify the causal eect of information release on liquidity. First, only a subset of rms are aected, thereby generating a treatment and control set-up. Second, its content is easily translated into a unique number, the change in inventories; third, it is released weekly which is substantially more frequent than other public announcements such as macro-news, or earnings announcements,among others.
Using transaction-level data, I estimate the eect of this report's publication on the bid-ask spread, volume, and midpoint returns via a dierence-in-dierence strategy by comparing the day of the release with the remaining days of the week. I nd that the mean bid-ask spread doubles immediately after the release and that volume increases by 32 percent. In contrast to the predictions of previous theory, this eect is independent of the report's content although prices react to this information.
To illustrate why the arrival of a public signal can yield an increase in adverse selection costs in nancial markets, I propose a dynamic model with a public signal and riskaverse informed investors. In this set-up, the public signal induces informed investors to participate in the market as it reduces uncertainty. At the same time, the market maker, which is the counterparty in the transaction, realizes that after the signal there are more informed investors in the market; thus, she widens the spread to maintain zero prots.
Additionally, since the noise traders do not change their behavior, volume spikes after the signal. All these movements in liquidity are a consequence of the resolution of uncertainty; therefore they do not depend on the signal's realization. Apart from the static eects, the model's dynamics deliver testable hypotheses about price and liquidity before and after the signal's release, which are also consistent with the data. In particular, the eect on spreads and volume is persistent whereas prices react just at the moment of the release.
This rst chapter concludes that in times of high uncertainty, there is an amplication mechanism aecting the release of information. In addition to the public announcement's content, which is directly incorporated into prices, the resolution of uncertainty lowers the barriers to entry for the informed traders, who enter the market and incorporate their information into prices through their trades. As a consequence, if the policy maker is willing to bear the increase in adverse selection, she should provide more information in times of turmoil. This measure might be complementary to others that regulators have taken in the past such as forbidding short sales, which is the focus of the second chapter of this thesis.
An extensive literature has analyzed the short selling bans that took place in 2008 and the main conclusion is that this policy hurts liquidity. Nevertheless, European regulators continued implementing restrictions on short selling in 2011 and 2012, which aimed to reduce aggressiveness in the market. Although this objective was also the main focus of the 2008 bans, the literature has remained silent about the behavior of this market characteristic before and after the ban. In the second chapter of this thesis, I propose a method to measure aggressiveness, which I dene as the intensity at which investors sell taking liquidity from the market. Next, using limit order book data from the main Spanish stock exchange, I document that indeed investors' aggressiveness just before the ban is twice as high as the one of previous weeks; meanwhile, it reverts to this prior level just after the ban's implementation.
Aside from assessing the impact of this policy on aggressiveness, I revisit its eect on liquidity using a novel identication strategy. While previous studies hinge on a treatmentcontrol comparison, in which rms not aected directly by the ban make up the control group; I use a design similar to regression discontinuity with time as my running variable.
Since the Spanish ban takes place in the middle of the trading day, I am able to identify the causal eect by comparing the seconds just before and just after the ban under the assumption that the exact implementation time is random. The estimates indicate that whereas the ban is detrimental for market quality, their magnitude is much lower than the one found by previous researchers.
The conclusion of this chapter is that regulators face a trade-o: on the one hand, when investor aggressiveness is extremely high the ban helps to reduce it; on the other hand, this policy undermines liquidity.
Besides the structure of the nancial market, the dependencies between dierent investment assets have played a key role in the recent crisis. The focus of the third chapter of this thesis is to study the dependence between the market basket, that we proxy using several indexes, and market neutral hedge funds (MNHFs). These funds' managers claim that their investment generates a positive return while maintaining independence with the market. Although there is a consensus that these two assets are not independent but their linear correlation is zero, the nature of this dependence is still unknown. Whereas most studies using non-parametric methods nd clear evidence of tail dependence, those who use parametric methods reach the opposite conclusion. The existence of this type of dependence is key to understand the contribution of hedge funds to market instability, as well to compute several risk measures such as VaR or Expected Shortfall.
Chapter 3, which is joint work with Julio Gálvez, reconciles these two strands of the literature and concludes that there is no tail dependence. Instead, we nd that the market and the MNHFs obtain low returns during bear periods and high returns during bull times which generate a dierent type of non-linear dependence. Additionally, the estimated linear correlation is dierent from zero in both states: positive in bull and negative in bear periods. However, the unconditional correlation coecient is insignicant.
Our results indicate that MNHFs do not move with the asset market in extreme episodes which suggest that these funds do not contribute to market instability. Moreover, as long as the state of the market is predictable to some extent, investors and managers can adjust their portfolios for hedging purposes.
To estimate tail dependence, we model the states as a latent Markovian binary variable and we rely on conditional copulas to characterize the complete joint distribution depending on the state. The resulting states correspond to sensible bear and bull periods; in particular, our states capture the 2008 global recession and the 2011 and 2012 European debt crises which are characterized by low average returns with high volatility.
As previous papers using parametric methods, we cannot reject that the parameter of the copula which generates tail dependence is zero. To reconcile these results with the non-parametric literature, we show that the methods used in those studies do not have the proper size if the data generating process is a two-state model. Specically, we simulate data from our estimated model with exactly zero tail dependence, and we conduct the non-parametric test proposed by the previous literature. We nd that their rejection rate is close to 100% with sensible sample sizes.
