Speakers & Abstracts

Belkacem was born in Mestegmer, Morocco. He studied Applied Mathematics at Université de Lille (MSc and Doctorate 3eme cycle), completing his PhD thesis in 1989 under the supervision of Professor Radu Theodorescu at Université Laval. After a brief time at Université de Moncton he joined Université de Québec à Trois-Rivières, where he rose to full professor. He then moved to the Department of Preventive and Social Medicine at Université Laval. Simultaneously, he was very involved with the Modelling and Simulation Laboratory in Morocco. In the fall of 2018 he was appointed as the Director of the National Institute of Statistics and Applied Economics (INSEA), Rabat, Morocco.

Throughout his academic career, Belkacem has carried out research in statistics, biostatistics and epidemiology. He has played an important role in many team research projects and is actively involved in the production of official statistics in Morocco, together with various national and international statistical activities and projects. Belkacem is a member of the Centre de recherches mathématiques, the Statistical Society of Canada, the Institute of Mathematical Statistics and the International Statistical Institute. His work has been supported directly by NSERC. Belkacem has supervised nine doctoral students and 14 MSc students. The excellence of his collaborative work was recognized with the Best Paper in The Canadian Journal of Statistics in 2004 with Ghoudi and Remillard. He has also done extensive consulting work for the Institut national de santé publique du Québec.

His current research is in the field of climate change, population health and official statistics. He played an essential role in developing tools to help government agencies, such as the Institut National de Santé Publique du Québec, monitor and support research on the relation between public health and climate changes. For example, in the case of adaptation to heat in urban settings, the large number of behaviours that can be adopted greatly complicates the monitoring of the evolution of preventive actions taken by individuals to counter the effects of the heat. Belkacem’s work regarding semiparametric estimators for a count regression function (e.g., the penalized MMLE and Kernel smoothing forthe non-parametric estimation of items) was essential for performing psychometric analyses and thus developing valid heat adaptation indices that are comparable over time and between countries. More specifically, he has successfully adapted these estimation methods to the field of climate change and population health. This has made it possible to develop valid climate change adaptation indices (heat, flooding, pollen, Lyme disease) as well as to reduce the size of a set of behavioural indicators without losing the underlying base information.

As chair of the local program committee and a member of the scientific program committee, Belkacem oversaw the local organization of the 61st World Statistics Congress in July 2017 in Marrakech. With respect to this he received an ISI Service award for outstanding and dedicated leadership and service as local program committee chair, ISI 2017, 61st WSC. Most recently he has co-chaired the High-level Group for Partnership, Coordination and Capacity-Building for statistics for the 2030 Agenda for Sustainable Development, United Nations StatisticsDivision.

Belkacem, with his wife Nadia, divides his time between Rabat, Morocco and Québec city. He has two adult children who have earned undergraduate and graduate degrees in accounting and computer science.

Title of the talk: Time scales concepts and smoothing techniques for discrete and mixed-type data

Summary: Time scales is a relatively recent and exciting mathematical theory introduced by the German mathematician Stefan Hilger in his dissertation [3], see also [4]. It provides a general unifying tool that builds the missing bridges between discrete and continuous analysis. Since then, time scales concepts have found applications in several theoretical and applied problems. For a comprehensive overview on time scales calculus see, for instance, Bohner and Peterson [2], Agarwal et al. [1]. In this talk, basic concepts of time scales calculus together with various applications will be presented. A particular emphasis will be put on nonparametric functional estimation. More precisely, we will show how time scales concepts might be used to cast continuous, discrete and mixed-type smoothing estimation problems under the same framework.

References

[1] Agarwal, R., Bohner, M., ORegan, D., and Peterson, A., 2002. Dynamic equations on time scales: a survey, Journal of Computational and Applied Mathematics 141 (2002) 126

[2] Bohner M. & Peterson, A. (2001). Dynamic Equations on Time Scales, Birkhäuser.

[3] Hilger, S., 1988. Ein Ma.kettenkalkül mit Anwendung auf Zentrumsmannigfaltigkeiten. (German). Universität Würzburgn, .

[4] Hilger, S., 1990. Analysis on measure chains: a unified approach to continuous and discrete calculus. Results in Mathematics, 18:1856.

Title of the talk: Smooth conditional distribution estimators using Bernstein polynomials

Summary: In a variety of statistical problems, estimation of the conditional distribution function remains a challenge. To this end, a two-stage Bernstein estimator for conditional distribution functions is introduced. The method consists in smoothing a first-stage Nadaraya–Watson or local linear estimator by constructing its Bernstein polynomial. Some asymptotic properties of the proposed estimator are derived, such as its asymptotic bias, variance and mean squared error. The asymptotic normality of the estimator is also established under appropriate conditions of regularity. Lastly, the performance of the proposed estimator is briefly studied through a few examples.

Title of the talk: Stock returns in the extreme

Summary: In this paper, we present a measure of systemic risk that includes the downside risk and the upside potential. The measure is based on the Conditional quantiles risk measure and the copulas. In order to capture the observed asymmetric behavior in both the right and the left tail, a split copula model is suggested. Our results, show that the dependence behaviour of returns is not symmetric. The extreme potential risk and extreme potential gain spillover is different across industries, countries and time. For example, the effect of the US financial market on other countries and US industries is different when it is in its extreme bad state compared to the case when it is booming.

Title of the talk: Nonparametric measures of local causality and tests of local non-causality in time series

Summary: The study of the causal relationships in a process (Y_t, Z_t)_t∈Z is a subject of a particular interest in finance and economy. A widely-used approach is to consider the notion of Granger causality, which in the case of first order Markovian processes is based on the joint distribution function of (Y_t, Z_t−1) given Y_t−1. The Granger causality measures proposed so far are global in the sense that if the relationship between Y_t and Z_t−1 changes with the value taken by Y_t−1, this will not be captured. To circumvent this limitation, this paper proposes local causality measures based on the conditional copula of (Y_t, Z_t−1) given Y_t−1 = x. Exploiting the asymptotic behavior of two kernel- based conditional copula estimators for α-mixing processes, the asymptotic normality of nonparametric estimators of these local measures is deduced and confidence intervals are built. Tests of local non-causality are developed as well. The efficiency of the proposed methods is investigated via simulations and their usefulness is illustrated on the bivariate time series of Standard & Poor’s 500 prices and trading volumes.

Title of the talk: A unified approach to backtesting expected shortfall and other co-tail systemic risk models

Summary: After the 2007-2008 market turmoils, a key proposal of the Basel Committee on Banking Supervision (BCBS) was to move the metric for risk assessment from value-at-risk (VaR) to Expected Shortfall (ES). However, in contrast to VaR models, there is a lack of formal backtesting methods to assess the accuracy of the ES models. This paper fills this gap by developing a two-step methodology to evaluate the validity of the ES models. The proposed methodology can be used to backtest internal ES models without a priori knowledge about their specifications. Our inferential approach combines the conditional moment (CM) test procedure and the generalized method of moments (GMM) method. It also provides a quite general framework to backtest the models with systemic co-tail risk measures, such as the marginal expected shortfall and the CoVaR.

Title of the talk: Testing the Eigenvalue Structure of Integrated Covariance with Applications (joint with A. Taamouti and J. Williams).

Summary: Decomposing a covariance matrix of a high dimensional dataset into collections of eigenvalues and eigenvectors has numerous applications, notably for Principal Component Analysis and Factor Analysis. In this paper and for Itô semimartingale dynamics, we derive the asymptotic distribution of a likelihood-ratio-type test statistic for the purpose of identifying the structure of eigenvalues of integrated covariance matrix estimated using high-frequency data. Unlike the existing approaches where the cross-section dimension grows to infinity, our test does not require large cross-section and thus it opens the door to a wide variety of applications. We also show how our test can be used to test for some specific factor decompositions of the integrated covariance matrix that are very useful for the empirical analysis of financial data. Furthermore, a test for ‘unexplained’ quadratic variation is proposed to investigate whether a given set of factors ‘explains’ at least a given proportion of integrated variance in the continuous part of the underlying process. These tests, however, are based on non-standard asymptotic distributions with many nuisance parameters. Another contribution of this paper consists in proposing a wide bootstrap method to approximate the asymptotic distribution. While standard bootstrap methods focus on sampling point-wise returns, the proposed method replicates features of the asymptotic approximation of the statistics of interest that guarantee its validity. A Monte Carlo simulation study shows that the bootstrap based test controls size and has a very good power even in samples with moderate size. Finally, we consider two empirical applications. The first one pertains to optimal hedging to crude oil futures while the second one concerns the extraction and usefulness of pricing factors from S&P 500 high-frequency stock prices.

Title of the talk: Spillover Effects among European, the US and Moroccan Stock Markets before and after the Global Financial Crisis

Summary: This paper assesses return and volatility spillovers among stock markets in Morocco, the US, UK, France and Germany represented respectively by MASI, S&P 500, FTSE 100, CAC 40 and DAX 30 indices, both before and after the global financial crisis (GFC) of 2008. The daily frequency data cover the period from January 2nd, 2002 to June 30th, 2016. Using the Diebold and Yilmaz approach, the results show varying financial connectedness between the Moroccan and the above mentioned developed stock markets. In fact, the significant increase of spillover index during the post-financial crisis period demonstrates that the US and European stock markets were the most affected. On the other hand, despite a relative increase of spillover effects coming from the US and German equity markets, our results show decline in the total net spillovers experienced by the Moroccan market after the recent financial crisis. These findings may provide some useful information to support decision-making and trading strategies for international investors. (Joint work with Karim Belcaid.)

Title of the talk: Clustering, Neural Networks and Data Science

Résumé : La 4ème révolution industrielle transforme l’ensemble des activités des sociétés. En fait, individus, familles, entreprises, organismes à but non lucratif, universités, gouvernements locaux, régionaux et nationaux, organismes transnationaux génèrent des données qui peuvent être structurées ou non, de sources diverses et plus ou moins complexes, et souvent massives. Le but est d’extraire de l’information utile, soit de la connaissance, à des fins de modélisation, d’analyse, de prédiction, de prise de décision… C’est le premier objectif de la science des données, une science qui s’appuie sur des méthodes issues de la statistique, des mathématiques, de l’informatique, de l’ingénierie, de l’apprentissage automatique et de la visualisation des données. La statistique y joue un rôle très important.

Cette présentation s’inscrit en science des données et vise à mettre dans le même cadre conceptuel les méthodes classiques de classification et les méthodes neuronales. Nous présenterons les différents aspects formels de chacune de ces méthodes et nous les confronterons sur un jeu de données réel afin d’identifier les forces et les faiblesses de chacune d’entre elle.

Des références à la littérature et des indications d'implémentation logicielle seront données.

Title of the talk: An alternative common shock model and its applications to construct new copula families

Summary: The so-called trivariate reduction method is a popular approach widely used to construct multivariate distributions. It is well known that this method has two major drawbacks. On one hand, it can only model dependence positive; on the other hand, it cannot always span the full range of positive correlation. To remedy these drawbacks, Genest, Mesfioui and Schulz (2018) introduced an alternative method which, contrary to the original, spans all possible degrees of dependence. This presentation will show that this novel idea can be used to construct a new class of copulas having an interesting stochastic representation. In particular, an extension of the Marshall--Olkin family of copulas will be presented. Some properties of this new family of copulas will be discussed.

Title of the talk: Managerial Preferences for Gambling On Expected Loan Loss Recognition

Summary: We study the incentives of bank managers to keep low loan loss reserves (LLR) relative to nonperforming loans and the consequences of this activity on firm’s insolvency. The central premise is that LLR inadequacy results from a class of speculative decisions that essentially trade off current costs implied by timely credit risk recognition with potentially magnified, but uncertain costs implied by delayed recognition. Because there is a straightforward analogy with insurance avoidance, we hypothesize that a greater mismatch between LLR and impaired loans can be related to general incentives that underlie the demand of hedging and insurance. According with the speculative nature of this activity, we also address the predictive ability of uncovered impaired loans on bank-specific measures of distress. We analyze a broad sample of publicly traded U.S. banks in the period 2001-2017. The results from our analysis show that, all else equal, the propensity to uncover impaired loans is greater (1) in banks with a corporate risk culture characterized by stronger preferences for gambling; (2) under distress, when building up LLR is more costly; (3) in banks with systemic characteristics, as moral hazard increases the propensity for risk shifting, and (4) in banks with greater discretion in loss provisioning. Our analysis also supports the pervasive consequences of this activity, showing that greater propensity to gamble with loss recognition increases the likelihood of future losses, earnings volatility, market volatility, and deteriorates solvency in horizons of three and five years.

Title of the talk: Value-at-Risk under Market Microstructure Noise

Summary: In this paper, we propose a robust estimation technique for Value-at-Risk that takes into account the effect of market microstructure noise. We show that the noise poses serious problems for estimating risk. In particular, it can cause an underestimation of risk, which makes the existing methods inconsistent in the presence of contaminated prices. Using Fourier transform and a deconvolution kernel estimator for the probability distribution function of the unobserved returns (prices), we derive a numerical approximation for Value-at-Risk in the presence of microstructure noise. Monte Carlo simulation and real data analysis illustrate satisfactory performance of the proposed method.