SCRO / Journées de l'optimisation

HEC Montréal, 29-31 mai 2023

CORS-JOPT2023

HEC Montréal, 29 — 31 mai 2023

Horaire Auteurs Mon horaire

NNQF New Numerical Techniques in Quantitative Finance

29 mai 2023 10h30 – 12h10

Salle: Procter & Gamble (vert)

Présidée par Lars Stentoft

4 présentations

  • 10h30 - 10h55

    A Novel Approach to Forecasting Equity Option Implied Volatility Surface

    • Letourneau Pascal , prés., University of Wisconsin-Whitewater

    This paper presents a novel approach for forecasting the implied volatility surface of equity options. Our method involves transforming the domain of the implied volatility surface and proposing a set of simple functions to model the entire surface. We show how to select a parsimonious set of these functions, which allows for a manageable set of dependent variables for estimating a forecasting model. To test the effectiveness of our approach, we apply it to a large sample of equity and compare our forecasting results to benchmark methods. Additionally, we demonstrate the practical application of our approach through a trading exercise. Our results indicate that our proposed method provides a superior forecasting performance compared to existing approaches.

  • 10h55 - 11h20

    Capital Structure Modeling Under Alternative Processes

    • Hatem Ben Ameur, GERAD, HEC Montréal
    • François-Michel Boire, prés., HEC Montréal
    • Mark Reesor, University of Western Ontario
    • Lars Stentoft, HEC Montréal

    In a contingent claim approach to capital structure modeling, the relationship between credit and equity risks crucially depends on the dynamics of the firm’s asset value. However, asset values are unobservable, posing important challenges for estimation. Viewing levered equity as an installment call on assets, the quasi-maximum likelihood (QML) approach of Ben-Abdellatif et al. (2021) provides the basis for estimation from equity data, and lends itself to a variety of asset models. Specifically, we develop procedures to estimate and solve capital structure models under Constant Elasticity of Variance (CEV) and Normal Inverse Gaussian (NIG) Lévy processes. This generalizes the log-normal assumption found in Merton (1975), Leland (1994), and Duan (1994), highlighting the significance of volatility clustering and the non-normality of asset dynamics for equity valuation.

  • 11h20 - 11h45

    Commodity option return predictability

    • Constant Aka, prés., Université Laval
    • Gabriel Power, Université Laval
    • Marie-Hélène Gagnon, Université Laval

    The paper investigates the predictability of commodity option returns using
    option-based characteristics, futures-based predictors, and macroeconomic
    variables. We use a set of 103 predictors and various linear and nonlinear
    machine learning models to analyze the predictability of returns on options on
    seven commodities. The out-of-sample R-squared is used as a statistical
    criterion to assess the performance of models, and the economic gains from a
    strategy that trades on models’ forecasts are used as an economic criterion. We
    find that Random Forest is the most effective model for predicting commodity
    option returns, and grains and meats option returns are the most predictable.
    Also, nonlinear models outperform linear models, and low-frequency option
    returns are the most predictable for energy commodities, while higher-frequency
    option returns are the most predictable for food commodities. The implied
    volatility is found to be the most important variable to predict commodity
    option returns and the inclusion of macroeconomic variables increases predictive
    power when examined with option-based characteristics. We demonstrate that a
    strategy that trades on nonlinear machine learning forecasts can yield
    substantial profits, with food commodity options being more profitable than
    other types of commodity options. The study also finds that there is good
    predictive information in past returns to predict future option returns, but a
    model based on past returns alone is less profitable than the full model.
    Finally, updating the hyperparameters more frequently improves the performance
    of machine learning models considerably.
    Keywords: Commodity options, option returns, machine learning, forecasts,
    predictability

  • 11h45 - 12h10

    Calibrating American Option Pricing Models to Large Panels

    • Stentoft Lars, prés., University of Western Ontario
    • Letourneau Pascal , University of Wisconsin-Whitewater

    In this paper we demonstrate how recently proposed efficient simulation and regression based option pricing models can be used to calibrate realistic models with time-varying volatility and non-normality to large panels of American options. We demonstrate that the trick of calibrating to synthetic European options, created from the American options, leads to significantly different parameter estimates suggesting that this approach may not perform well.

Retour