Résumé en anglais :
In the last ten years two new strands of research in finance have emerged. The first takes its root in the availability of highfrequency data, leading to the possibility of better measures and model free of financial variables that are essential to the conduct of portfolio allocation or risk management. Volatility is such a key variable, which is now very accurately measured non-parametrically as the sum of squared intra-day returns (realized volatility). At the other end of the spectrum, financial economists have stressed the importance of long-run risks to solve resilient asset pricing puzzles such as the level of the equity premium and the large cross-sectional differences in average returns across equity portfolios such as value and growth portfolios.
The long-run risks in question are the fluctuations in the long-run growth prospects of the economy and the time-varying level of economic uncertainty. In other words, the quantities investors care about are the level of consumption or output and its volatility. These quantities are measured at low frequencies, typically at monthly or quarterly levels. How to reconcile these long-run economic concerns with the high-frequency and noisy movements in asset returns? The common thread of this research proposal will be to show the importance of mixing frequencies to provide more informative tests of asset pricing models or to build more efficient risk management tools.
The first part will focus on testing asset pricing models whereas the second part will propose and study methods of inference for realized statistics with good finite sample properties. The first part of the research agenda deals with asset pricing tests and contains three projects. The main goal of the first one is to explain the short and long-run stylized facts underlying the risk-return trade-off by an equilibrium consumption-based asset pricing model that involves long-run risks and generalized disappointment aversion preferences.
The second project will revisit the empirical evidence of the capital asset pricing model (CAPM), which is empirically rejected. We will study the relation between returns and betas (coefficients of regression in the CAPM) averaged over long horizons and provide a theoretical model explaining the aggregation results. We will also consider the long run properties of the conditional CAPM, where conditioning variables such as realized volatility computed over long horizons will be used to introduce time variation in betas.
The general goal of the third project is to study the high frequency dynamics and the forecasting power for excess stock returns of a measure of idiosyncratic risk based on the cross sectional dispersion of stock returns recently introduced in the literature by a member of the team. We intend to explore the usefulness of this measure for risk management applications at high frequencies. The second part of the research agenda deals with statistical inference based on high frequency data and contains three important topics for risk management purposes.
The main goal of the first and second projects is to improve the finite sample properties of statistics and confidence intervals of measures developed recently in the literature, under the potential presence of market structure noise. The first project will propose the bootstrap for jump tests. In contrast, by using Edgeworth expansions, the second project proposes non-linear transform of quantities based on multivariate realized measure like the beta in the CAPM in order to improve the quality of their finite sample properties. The third project will test the distributional assumptions implied by continuous time models on high frequency data contaminated by market microstructure noise.
Project date: 01/12/2011 au 30/11/2014
Contact in TSE: Nour MEDDAHI