Publications
Publications
We develop an innovative application of Kelly et al’s 2018 instrumented principal component analysis model, wherein regression-based exposures (betas) to risk factors are used as characteristics. We show that this new type of model, which hybridizes elements from cross-sectional, statistical and time series models, has many advantages. It inherits the high precision and depth of analysis typically found in cross-sectional models, while dramatically reducing their data requirements. In addition, it is precise, allows the inclusion of many characteristics while remaining numerically stable, and greatly simplifies the construction of multiregional models. Finally, although calibrated using a universe of funds, this model has excellent precision and low bias when used to analyze and optimize portfolios of stocks.
[Paper]
Many investors form views about the evolution of the macroeconomic environment. Yet the available academic research on the impact of macroeconomic shifts on equity returns provides mixed results. In this paper we adopt a recent approach that consists in measuring how equity returns react during periods of strong macroeconomic shifts. Extending previous works on expected returns, we provide new insights on how macroeconomic shifts affect both risk and extreme risk. We also introduce a new measure of stability for expected returns that focuses on the direction of the performance rather than its actual expected value. We also whos that while most macroeconomic indicators affect the returns of equities in-sample, only a few do so reliably out-of-sample. Based on our observations, we finally provide some practical insight into the behaviour of common equity strategies during macroeconomic shifts.
[Paper upon request]
We examine how active risk- and holdings-based diversification of equity portfolios affect performance and vulnerability to large losses. Conducting a comprehensive empirical study of US-based funds, we find that risk-based and sector-based diversification significantly reduce active tail risk and the likelihood of extreme losses, without substantially diminishing portfolio performance. These effects are nonlinear and decreasing, suggesting that investors need not minimizing the concentration of their portfolios. We also examine these relationships on an unprecedented large sample of portfolios using a novel methodology that allows the production of portfolios with similar levels of risk, and find that they are robust to several definitions of extreme risk. Our results highlight the practical value of diversification in managing portfolio risk while maintaining competitive performance.
[Paper]
A precise understanding of the risks driving portfolio performance is a key component of asset management. We show that classical methods used to estimate risk factor contributions can lead to important “misattributions” of performance. These misattributions are not well documented, in spite of their potentially significant impact on the understanding of investment strategies. We investigate the drivers of misattribution episodes, including model uncertainty, time-varying factor parameters and extreme data points, and propose analytical approaches to mitigate their effects. We also show that introducing more model complexity does not necessarily lead to better estimates. By addressing misattribution drivers, our research contributes to better-informed decisions and more effective risk management in dynamic market environments.
[Paper]
The latest development in empirical Asset Pricing is the use of Machine Learning methods to address the problem of the factor zoo. These techniques offer great flexibility and prediction accuracy but require special care as they strongly depart from traditional Econometrics. We review and critically assess the most recent and relevant contributions in the literature grouping them into five categories defined by the Machine Learning (ML) approach employed: regularization, dimension reduction, regression trees/random forest (RF), neural networks (NNs), and comparative analysis. We summarize the empirical findings with special attention to their economic interpretation providing hints for future developments.
[Paper]
Selected Working Papers
We propose a hybrid generative framework to simulate realistic equity risk factor returns. Our approach combines an EGARCH-like volatility model with a Conditional Restricted Boltzmann Machine, a neural network used to learn complex multivariate probability distributions, including their serial correlation. Compared to other generative techniques that require heavy data and intensive training, this combination of econometric and generative models (E-CRBM) reproduces many static and dynamic features of returns’ distributions while maintaining the simplicity and computational efficiency of the original model. We show the advantages of this hybrid approach in the context of extreme risk management and macroeconomic analysis.
We introduce a new methodology that generates random equity portfolios with a fixed level of risk and weight concentration. It uses elements of group theory to define a rotation with random angles which, when applied on a portfolio, transforms its weights but conserves its risk and concentration. The implementation of this approach is straightforward to implement and only requires commonly available price data. The ability to generate random portfolios with fixed levels of risk is very desirable in many areas of research. We illustrate its potential by applying it for the study of active risk diversification and the creation of new benchmarking paradigms.
Investors are inherently exposed to changes in market liquidity conditions. However, comparatively little research has been directed towards the optimal way to invest given a view on market liquidity. The paper shows how to construct derivatives written on the market liquidity of the underlying asset, and identifies their potential buyers and sellers. A simple valuation model shows that liquidity can be hedged by purchasing volatility in amounts based on the trading volume of the underlying asset. The model calibration delivers risk-neutral expectations of market liquidity. Equilibrium analysis shows that derivatives on market liquidity can contribute to financial stability.
(Previous title: Liquidity Derivatives) Presented at the 2023 Global Finance Conferece, the 2023 Workshop Banks and Financial Markets, the 2022 Credit Conference, the 2022 DGF Conference, 2022 Derivative Market Conference, the 2022 Financial Markets and Corporate Governance Conference, the 2022 GSB Research Symposium, the 4th International Conference on European Studies, the ISB Summer Research Conference 2022, the 15th edition of the IRMC, the Leibniz Institute for Financial Research SAFE, the Nova School of Business and Economics, the 2022 Southern Finance Association Meeting, and the 11th Annual Stern/Salomon Microstructure Meeting
Cross-predictability denotes the fact that some assets can predict other assets' returns. I propose a novel performance-based measure that disentangles the economic value of cross-predictability into two components: the predictive power of one asset's signal for other assets' returns (cross-predictive signals) and the amount of an asset's return explained by other assets' signals (cross-predicted returns). Empirically, the latter component dominates the former in the overall cross-prediction effects. In the crosssection, cross-predictability gravitates towards small firms that are strongly mispriced and difficult to arbitrage, while it becomes more difficult to cross-predict returns when market capitalization and book-to-market ratio rise.