Presentations - PhD students - EM - first year

3 October 2019
Versione stampabile

2 PM, Seminar Room, Department of Economics and Management, via Inama 5


Estimation of penalized partial correlation networks:
Methodological developments with an application in portfolio selection


The least absolute shrinkage and selection operator (lasso) proposed by Tibshirani [1] is now used in a wide variety of situations. Due to its peculiar position as a 1-norm in the p-norm penalties’ family, it is able to both shrink and select parameters of a specific model during their estimation process. This penalization is able to produce sparse models that can be interpreted more easily. Also it could potentially rule out some noise in the estimation procedure.
Initially used in penalized regressions, the lasso penalty has been employed more recently in the estimation of the precision matrix of multivariate normal random vectors [2][3][4]. Using the precision matrix it is possible to build a sparse conditional independence graph representing the dependence structure of the set of variables in these vectors. Since the normality assumption doesn’t always hold in reality, a more robust procedure combining a multivariate t-distribution and the lasso penalty has been proposed [5]. Using the t-distribution, the precision matrix can be used to build a sparse partial correlation graph since the absence of partial correlation doesn’t imply conditional independence anymore. This new procedure can be more suitable to reconstruct the partial correlation structure among financial assets using the times series of their returns and the estimate of the precision matrix can be used in a mean-variance portfolio optimization [6]. Recently it has been used also to reconstruct the banking network from publicly available data [7].
Firstly this paper wants to address some further developments of the framework used to estimate sparse partial correlation networks. A new procedure which uses the multivariate normal inverse Gaussian (MNIG) distribution [8] and lasso penalty is developed. The MNIG distribution could be able to fit data that show asymmetry and heavy-tails, two well-known stylized facts in finance [9]. Also the lasso penalty could be substituted by other penalties, like the elastic net [10]. The elastic net penalty has been introduced to address some limitations of the lasso penalty which arise, for example, when some variables are highly correlated. Ryali, Chen, Supekar and Menon [11] substituted the lasso penalty with the elastic net to better estimate the conditional independence graph representing the connections among different regions of the brain. In this paper two additional procedures are developed to combine the elastic net with multivariate t-distribution and MING distribution. Furthermore a study using simulations is proposed to compare their estimation performances for different network topologies and different data generating processes. Lastly these new procedures are tested in the context of mean-variance portfolio optimization. Using real datasets it is possible to compare the performances of different portfolios built using different procedures for the estimation of the precision matrix of the set of assets. In addition it is also analyzed the stability over time of the estimated partial correlation graphs. This stability could be extremely important because it could influence the frequency to which the portfolio’s weights should be rebalanced.

GUIDA Vittoria

Exploration-exploitation dilemma: Behavioural Insights and Individual Decision-Making


Exploration and exploitation are two activities that underpin a variety of decision-making fields such as innovation, strategy, technology and similar.
Since James March’s seminal paper in 1991, these activities have been central in one of the largest and most diversified research streams in the field of organization and management studies. Most scholars highlighted the importance of the individual level of analysis. The literature also suggests
that many heuristics and bias are at work in this type of scenario. However, there is no empirical study or evidence to support the emergence heuristics in exploration-exploitation choices and the individual level of analysis is just partially addressed.
This article investigates the effects of heuristics and biases in the individual decision-making context of exploration-exploitation. In particular, it focuses on their effects on the switch from exploitation to exploration.
The lack of normative theories on this type of decisions compels this study to investigate “induced” heuristics: the experimental design is manipulated so that the heuristics are elicitable. Perfect rationality of the participants is initially assumed and – subsequently – it is verified whether they are rational, or they systematically use some heuristics (e.g. sunk costs effect).
This paper contributes to the literature in several ways. First, it advances the knowledge about exploration-exploitation decision-making at the individual level. Second, it provides empirical evidence for the heuristics and biases at work in this decision scenario. Finally, it contributes to the investigation of the micro-foundations of strategy.
Managerial implications of this study are related to the nudging, organizational design, and strategic opportunities that firms can exploit given a systematic use of the investigated heuristics.


High Frequency Trading and Market Quality

Since the last decade, technological advances have been creating many dramatic changes in every aspect of our lives. In the meantime, the financial industry has always been striving to match with the ever-changing and developing environment. A significant breakthrough of technology in finance is high-frequency trading (HFT hereafter) which keeps adopting any technological progress in ultra-fast data transferring due to its competitive nature on latency. High-frequency traders (HFTr) trade financial instruments with the ultra-high speed in an electronic exchange where all orders placed in the computer system to achieve legitimate reward on their speed advantage.
This research aims to study HFT activities and their impact on market quality. The focus is at first on categorizing HFT activities in different groups of strategies using the machine learning process and second studying impact of each strategy on the other participants of the market. Since HFT participations are diverse from market making to manipulation and taking effect of all activity in pool might produce an unfounded result. This is a problematic issue that has not been considered in extant literature comprehensibly. However, few scholars investigated HFT clustered in passive and aggressive, yet the identification of each type within aggressive or passive group of strategies and their impact of each on market quality has been neglected.
The study benefits from a featured database containing all messages sent to the electronic stock exchange at the level of micro-second (1x10-6 second). A functional approach is taken to account for assessment and identification of HFT activities since the focus of the study is on the HFT activities taken by either entities registered as HFT or other participants.
The study contributes to the literature from different aspects. Developing methods of identification of HFT activities and their type of strategies contributes methodologically and paves the way for investigating each strategy for market regulators and researchers. The findings based on separated strategies can reconcile conflicts.

KORCA Blerita

The impact of regulation on Corporate Social Disclosure in Europe

Corporate Social Responsibility (CSR) is becoming increasingly important in Europe, from an institutional, empirical and theoretical perspective. Within this broad stream of research, there is a focus of studies on the communication tool which enables companies to share their responsible
actions toward environment and society, that is the Corporate Social Disclosure (CSD). The importance of CSR in the European context is stressed by the fact that the European Union (EU) has issued a regulation which requires from large companies to publish a report disclosing their social
and environmental impacts. The Directive 2014/95/EU (Directive) requires from European companies with more than 500 employees to disclose in their annual report (or in a stand-alone report) information regarding their social and environmental actions. As a result, organizations had to comply
with this law starting from 2018, regarding the financial year of 2017.
This study aims to examine the impact of regulation on companies’ reporting process both at the European and at the company level. To do so, mixed methods will be employed. First, a comprehensive literature review will be conducted to understand how the Directive is transposed in each EU member state. This contribution is relevant for the reason that the EU gave broad discretion to member states in terms of transposing the law into their national legislation. Consequently, states could impose some changes to the law or transpose it in the original version. Second, a multiple case study will be conducted employing both the Italian and the UK context. The objective is to analyze the process and the changes that occurred inside the organization because of the Directive. Many companies across Europe have been disclosing social and environmental information voluntarily.
When the new law applied, some of these companies felt into its scope and had to comply with its requirements. Thus, it would be of particular importance to analyze the shift from voluntary to mandatory disclosure and analyze the changes in terms of reporting but also inside the organization
actions. Finally, many studies argue that some of the motives that companies have to disclose such information are related to reputation, legitimacy and financial performance. The aim instead of the third paper is to analyze the relationship between CSD and Corporate Financial Performance (CFP).
Additionally, it will be examined how the Directive as a moderator impacts the interaction between CSD and CFP.
The overall contribution of this project is to shed light on the ongoing debate if the disclosure of social and environmental information should be mandatory or not. Furthermore, the longitudinal analysis before and after the legislation allows presenting the findings regarding the effectiveness of the directive as a new law. The last but not the least, with the analysis of the relationship between CSD and CFP, this study will offer insights on the potential benefits one organization can have in terms of financial performance if disclosing social and environmental information.