Working papers
Outliers in Asset Pricing (Job Market Paper)
Abstract: This paper examines the effects of multivariate outliers on factor models and portfolio weights through expected returns estimates. To investigate these effects, we improve the machine learning method of Kozak et al. (2020) to detect and be robust against a given proportion of multivariate outliers. Theoretically, we show how outliers affect estimation and that the robust estimator takes into account model uncertainty. Empirically, we find that the robust methods outperform the non-robust method’s out-of-sample R^2 by at least 10-20 percentage points. Furthermore, multivariate outliers are shown to be prevalent in asset pricing data and we decompose the data into three pricing regimes corresponding to a low market volatility, high market volatility and extreme events/recession market state. Multivariate outliers affect coefficient selection, magnitude and sign but also the maximum Sharpe ratios of models. We create a robust unconditional factor model by pooling the coefficients from the regimes. Out of sample this model results in a Sharpe ratio improvement of 50% and reduced downside risk compared to the benchmark non-robust model.
Presented at: KU Leuven, IFABS Conference Oxford 2023 (Best PhD Paper nomination), University of Bristol, University of Nottingham, Imperial College
To be presented at:
To be presented at:
Abstract: We propose a set of dependence measures that are non-linear, local, invariant to a wide range of transformations on the marginals, can show tail and risk asymmetries, are always well-defined, are easy to estimate and can be used on any dataset. Additionally, in economics and finance applications the dependence measure can measure the degree of substitutability between two goods or risks. We propose a nonparametric estimator and prove its consistency and asymptotic normality. To show practical utility, we apply these measures to high-frequency stock return data around market distress events such as the 2010 Flash Crash and during the GFC. Contrary to ubiquitously used correlations we find that our measures clearly show tail asymmetry, non-linearity, lack of diversification and endogenous buildup of risks present during these distress events. Additionally, our measures anticipate large (joint) losses during the Flash Crash while also anticipating the bounce back and flagging the subsequent market fragility. Our findings have implications for risk management, portfolio construction and hedging at any frequency
Presented at: KU Leuven Finance Seminar, CFM-Imperial Market Microstructure Workshop (PhD poster), KU Leuven Statistics Seminar (Spring 2024) and the 18th Belgian Financial Research Forum 2024
I gratefully acknowledge financial support from Internal Funds KU Leuven (STG/20/033)
Making heads or tails of systemic risk measures
Abstract: This paper applies a result linking copulas to the CoVaR family of systemic risk measures to obtain results on relevant properties of these risk measures. The main property established is that the risk measures can be bounded by their value under independence and perfect positive dependence thereby providing model-free bounds on the risk. We establish other results concerning their sensitivity to power-law tails, outliers and their properties under aggregation. We show that the risk of model misspecification at high quantiles can lead to both significant quantitative and qualitative differences in estimates and their interpretation. To alleviate the problem we propose a nonparametric estimator and show its statistical behavior in simulations. We apply the new estimator to stock return data of US financial institutions and stock-based climate risk proxies. The empirical results show the merits of our estimation approach and the theoretical results.
Presented at: BSE Summer School on Banking 2022, CEBA Seminar St. Petersburg 2022 (online), IFABS Conference Naples 2022, the 17th Belgian Financial Research Forum 2023 and KBC Risk Model Development
I gratefully acknowledge financial support from Internal Funds KU Leuven (STG/20/033)
To be presented at: