Abstracts

Belkacem ABDOUS, The National Institute of Statistics and Applied Economics (INSEA), Rabat, Morocco

TITLE OF THE TRAINING COURSE: SMOOTHING TECHNIQUES, REGULARIZATION, RIDGE AND LASSO REGRESSION

Chafik BOUHADDIOUI, United Arab Emirates University, UAE

TITLE OF THE TALK: TESTS OF CAUSALITY AT MULTIPLE HORIZON BETWEEN TWO INFINITE-ORDER COINTEGRATED VECTOR AUTOREGRESSIVE SERIES

Summary: We propose a generalized test for non-causality at various horizons for infinite-order cointegrated vector autoregressive. We first introduce multiple horizon infinite-order cointegrated VAR which can be approximated by a multiple horizon finite-order vector autoregressions. The order is assumed to increase with the sample size. Under some regularity conditions, we study the estimation of parameters obtained from the approximation of the infinite-order autoregression by a finite-order autoregression. The test can be implemented through linear regression methods. The asymptotic distribution of the new test statistic is derived under the hypothesis of non-causality at various horizons. Bootstrap procedures are also considered. The methods are applied to a cointegrated VAR models from economic data.

Sophie DABO-NIANG, University of Lille, France

TITLE OF THE TALK: STATISTICS AND ECONOMETRICS OF SPATIAL DATA

Summary: Spatial Statistics/Econometrics includes any(statistical/econometrics) techniques which study phenomenons observed on spatial sets. Such phenomenons appear in a variety of fields such as economics, epidemiology, environmental sciences, physics, image processing and many others. Modeling spatial data is among the most interesting research subjects in dependent data analysis. This is motivated by the increasing number of situations coming from different fields for which the data are of spatial nature. This is the case for instance in finance where hedonic pricing are usually estimated using data from the housing or labor markets that may be strongly spatially dependent. In fact, housing prices depend on location while markets vary by city or region in a given country. Modeling these type of spatial dependent data requires using statistical techniques (spatial auto-regression or prediction/kriging models) that incorporate an appropriate spatial dependency.

Abdelaati DAOUIA, Toulouse School of Economics, France

TITLE OF THE TALK: EXTREMILES: A NEW PERSPECTIVE ON ASYMMETRIC LEAST SQUARES (with Irène Gijbels and Gilles Stupfler)

Summary: Quantiles and expectiles of a distribution are found to be useful descriptors of its tail in the same way as the median and mean are related to its central behavior. This paper considers a valuable alternative class to expectiles, called extremiles, which parallels the class of quantiles and includes the family of expected minima and expected maxima. The new class is motivated via several angles, which reveals its specific merits and strengths. Extremiles suggest better capability of fitting both location and spread in data points and provide an appropriate theory that better displays the interesting features of long-tailed distributions. We discuss their estimation in the range of the data and beyond the sample maximum. Some motivating examples are given to illustrate the utility of estimated extremiles in modeling noncentral behavior. There is in particular an interesting connection with coherent measures of risk protection.

Mohamed DOUKALI, CIREQ Montréal, Canada

TITLE OF THE TALK: JACKKNIFE LIML ESTIMATOR WITH MANY INSTRUMENTS USING REGULARIZATION TECHNIQUES (with Marine Carrasco)

Summary: We consider instrumental variables regression in a setting where the number of instruments is large. However, in finite samples, the inclusion of an excessive number of moments may be harmful. Such a situation can arise when there are many weak instruments. We propose a Jackknife Limited Information Maximum Likelihood (LIML) based on three different regularizations methods:Tikhonov (T), Landweber-Fridman (LF), and Principal Components (PC). We show that our proposed regularized Jackknife estimators (RJLIML) are consistent and asymptotically normally distributed under heteroskedastic error. We derive the rate of the approximate mean square error and propose a data driven method for selecting the tuning parameter.

Simulation results demonstrate that the leading regularized estimators (LF and T of RJLIML) perform very well (are nearly median unbiased) even in the case of relatively weak instruments. An empirical application illustrates the relevance of our estimators.

Ahmed EL GHINI, Mohammed V University in Rabat, Morocco

TITLE OF THE TRAINING COURSE: MULTIVARIATE GARCH MODELING WITH APPLICATIONS IN ECONOMICS AND FINANCE

Guy MELARD - ECARES, Université Libre de Bruxelles, Belgium

TITLE OF THE TRAINING COURSE: FORECASTING HIGHT-FREQUENCY TIME SERIES BY STATISTICAL METHODS

Summary: Les données temporelles en haute fréquence (semaine, jour, heure, etc.) sont de plus en plus souvent disponibles alors que les méthodes classiques de prévision traitent essentiellement de données trimestrielles ou mensuelles. En conséquence, les séries temporelles sont plus longues. A première vue, c'est excellent du point de vue statistique excepté que le nombre de paramètres des modèles ARIMA ou en espace d'état peut être accru. En réalité, plus de paramètres ne sont pas nécessairement requis mais l'utilisati on de données hebdomadaires, journalières ou en haute fréquence implique des difficultés. Dans une première partie de l'exposé, on présentera quelques-unes de ces méthodes classiques (décomposition saisonnière, lissage exponentiel, régression linéaire multiple) pour passer aux modèles ARIMA.

Dans une seconde partie, on montrera quelques exemples (vitesse du vent au sommet d'une éolienne, trafic dans une cellule d'un réseau de téléphonie mobile, ventes journalières de produits en magasin, soldes de trésorerie dans des organisations, consommation d'énergie par heure dans des bâtiments de bureaux) et les problèmes qu’ils posent. Dans une troisième partie, on discutera quelques solutions comme les effets de calendrier (jours fériés, promotions, longueur de mois), la flexibilité de l'horaire de travail, l’introduction de variables explicatives, le traitement automatique des données aberrantes (« outliers »), le traitement des données manquantes. Une autre implication des séries en haute fréquence est que les variables de comptage prennent souvent des petites valeurs entières positives alors que la plupart des techniques de modélisation supposent des variables continues.

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

Jose OLMO - University of Southampton, UK

TITLE OF THE TALK: A RE-EXAMINATION OF THE SIZE EFFECT (with Richard McGee from University College Dublin)

Summary: We investigate the monotonic relationship between firms' market capitalization and stock returns in a predictive setting. This relationship is modelled using a conditional logit model within a stochastic utility framework. This parametric choice to predict the ranking of cross-sections of risky assets allows us to propose a joint and a sequential test of monotonicity. Rejection of the tests provides valuable information about the point in the cross-section at which the monotonicity hypothesis is rejected. We apply these tools to analysing the ranking of U.S. stocks for 20 industries over the period January 1970 to December 2015. Market capitalization has significantly better predictive ability for ranking the cross-section of stocks during episodes of industry positive returns than during negative episodes. We also observe a major role of the winning stocks in determining such monotonic relationship and on the profitability of top-minus-bottom trading strategies based on size. In contrast, the predictive power of the size effect for obtaining the factor loadings of asset pricing models is robust to the specific choice of the size factor portfolio.

Jeroen ROMBOUTS - ESSEC Business School, Paris, France

TITLE OF THE TALK: DYNAMICS OF VARIANCE RISK PREMIA: A NEW MODEL FOR DISENTANGLING THE PRICE OF RISK

Summary: This paper formulates a new dynamic model for the variance risk premium based on a state space representation of a bivariate system for the observable ex-post realized variance and the ex-ante option implied variance expectation. A regime switching structure accommodates for periods of unusually high volatility, heterogenous dynamics and changes in the dependence between the latent states. The model al- lows separating the continuous component of the variance risk premium from the impact of jumps on option implied variance expectations. Using options and high frequency returns for the S&P500 index, we address a puzzle in the existing literature related to return predictability by disentangling the part of the variance risk premium associated with normal sized price fluctuations from that associated with tail events. The latter component predicts to a significant extent, and asymmetri- cally with respect to their sign, future market return variations.

Antonio RUBIA - University of Alicante, Spain

TITLE OF THE TALK: MULTIVARIATE TESTING FOR FRACTIONAL INTEGRATION UNDER CONDITIONAL HETEROSKEDASTICITY

Summary: We introduce a new testing approach to detect fractional integration in a multivariate context. This setting extends the testing procedure in Breitung and Hassler (2002) under fairly general assumptions that permit weakly-depedent errors on a MDS basis, such as VAR-type and conditional time-varying second-order moments. Our test can readily be implemented in a regression-based context building on feasible generalized least squares (FGLS) estimation. The asymptotic null distribution is a standard Chi-square distribution, independently of the true order of fractional integration, and the test exhibits non-trivial power against sequences of local alternatives. Using this procedure, we address joint long-run dynamics in trading volume and different volatility measures in the stocks integrated in the Dow Jones Industrial index.

Abderrahim TAAMOUTI - Business School, Durham University, UK

TITLE OF THE TALK: MEASURING NONLINEAR GRANGER CAUSALITY IN QUANTILES

Summary: We introduce new measures of granger causality in quantiles, which detect and quantify nonlinear causal effects between random variables. The measures are based on nonparametric quantile regressions and defined as logarithmic functions of restricted and unrestricted expectations of quantile check loss functions. They can easily and consistently be estimated by replacing the unknown expectations of check loss functions by their nonparametric kernel estimates. We derive a bahadur-type representation for the nonparametric estimator of the measures. We establish the asymptotic distribution of this estimator, which can be used to build tests for the statistical significance of the measures. We also examine the properties of the latter under local alternatives. Thereafter, we show the validity of a smoothed local bootstrap that can be used in finite-sample settings to perform statistical tests. a monte carlo simulation study reveals that the bootstrap- based test has a good finite-sample size and power properties for a variety of data generating processes and different sample sizes. Finally, we provide an empirical application to illustrate the usefulness of measuring granger causality in quantiles. We quantify the degree of nonlinear predictability of the quantiles of equity risk premium using the variance risk premium, unemployment rate, inflation, and the effective federal funds rate. The empirical results show that the variance risk premium and effective federal funds rate have a strong predictive power for predicting the quantiles of the risk premium when compared to that of the predictive power of the other two macro variables. In particular, the variance risk premium is able to predict the center, lower and upper quantiles of the distribution of the risk premium; however, the effective federal funds rate predicts only the lower and upper quantiles. Nevertheless, unemployment and inflation rates have no effect on the quantiles of the risk premium.

Francesco VIOLANTE - ENSAE, France

TITLE OF THE TALK: DYNAMIC PROPERTIES AND CORRELATION STRUCTURE OF LARGE PANEL CRYPTO CURRENCIES

Summary: The behaviour of a large portfolio of highly valued and most actively traded cryptocurrencies is studied. Unlike more traditional financial assets, the dynamic behaviour of cryptocurrencies returns is characterised by a particularly high level of volatility, by abnormally large variations, and is affected by extreme shocks to liquidity. We aim at investigating the dynamic properties of cryptocurrencies and particularly the correlation structure linking them to identify whether and to what extent there exist diversification opportunities in these markets.