Automatic threshold selection for extreme value regression models of tail risks

EI-ERIM-OR seminar
Dollar sign
Speaker
Dr. Julien Hambuckers
Date
Thursday 1 Dec 2022, 12:00 - 13:00
Type
Seminar
Spoken Language
English
Room
C1-3
Building
Theil Building
Add to calendar
Dollar sign

We introduce a robust method to simultaneously estimate the tail and the threshold parameters of an extreme value regression model. This model finds its use in finance to assess the effect of market variables on extreme loss distributions of assets or investment vehicles, such as hedge funds.

(joint with A. Usseglio-Carleve and M. Kratz)

Splicing regression model

A major limitation of these models, though, is the need to select ex ante a threshold below which data are discarded, which leads to estimation inefficiency.

To solve this issue, our approach relies on a splicing regression model that extends the extreme value regression model below the threshold, and uses an artificial censoring mechanism of the likelihood contributions in the bulk of the data at the estimation stage to decrease specification issues.

Censoring strategies

We discuss several censoring strategies (both theoretical and data-driven) of the proposed estimator. We illustrate its superiority for tail inference over classical peaks-over-threshold methods in an extensive simulation study. We then investigate the determinants of hedge funds' tail risks over time, using an unbalanced panel of 1,400 long/short equity hedge funds over the period 1995-2021. We find a significant link between hedge funds tail risks and factors such as equity momentum, financial stability indicators and credit spreads.

     Julien Hambuckers

    About Julien Hambucker

    Julien Hambuckers is an associate professor of finance at University of Liège (Belgium), and a faculty member of  HEC Liège (the business school of the university).  From 2016 to 2018 I was a researcher at University of Göttingen (Germany), Chair of Statistics (Business and Economics faculty).

    My research areas are statistics applied to finance, financial econometrics and Extreme Value Theory. I also work on distributional regression models in the time series context, and on solving applied statistical questions (e.g. model selection or endogeneity) in financial economics.

    Organisers

    Maria Grith an Michal Mankowski

    More information

    Secretariat Econometrics
    Phone: +31 (0)10 408 12 59/ 12 64
    Email: eb-secr@ese.eur.nl

    See also

    Inference on derivatives of high dimensional regression function with deep neural network

    Weining Wang (University of Groningen)
    Students standing

    Asymmetric Violations of the Spanning Hypothesis

    Paul Riva, Northwestern University
    Paula Riva smiling at the camera

    A New Branching Rule for Range Minimization Problems

    Bart van Rossum (Erasmus University Rotterdam)
    Grote Europese subsidie voor duurzame campus

    Conformal prediction for enhanced decision-making of PV Power in electricity markets

    Tarek Alskaif (Wageningen University)
    Campus faculty flags

    Why Has the Number of Billionaires Increased So Much?

    Coen N. Teulings and Simon J. Toussaint
    A women wearing fashionable sunglasses whilst holding up multiple high end shopping bags

    Unearthing Financial Statement Fraud: Insights from News Coverage Analysis

    Jianqing Fan, Princeton University
    Jianqing Fan smiling at the camera

    Extending the Scope of Inference About Predictive Ability to Machine Learning Methods

    Juan Carlos Escanciano (Universidad Carlos III de Madrid)
    Polak Building and autumn trees

    Beyond Arbitrage: Deviations from Risk-Return

    Benjamin Holcblat, (University of Luxembourg)

    On changepoint detection in functional data using empirical energy distance

    Lorenzo Trapani, (University of Leicester)

    Saddlepoint techniques for the statistical analysis of time series

    Davide La Vecchia (University of Geneva)
    Spring in Rotterdam

    Compare @count study programme

    • @title

      • Duration: @duration
    Compare study programmes