Facoltà di scienze economiche

An econometric analysis of time-varying risk premia in large cross-sectional equity datasets

Ossola, Elisa ; Gagliardini, Patrick (Dir.)

Thèse de doctorat : Università della Svizzera italiana, 2013 ; 2013ECO001.

In this thesis, we develop a new econometric methodology to estimate the time-varying risk premia implied by conditional linear asset pricing models. In contrast to the classical approach, we estimate risk premia from a large dataset of returns of individual stocks instead of portfolios. The aim is to avoid the potential bias and loss of information implied by sorting and grouping stocks into... Plus

Ajouter à la liste personnelle
    Summary
    In this thesis, we develop a new econometric methodology to estimate the time-varying risk premia implied by conditional linear asset pricing models. In contrast to the classical approach, we estimate risk premia from a large dataset of returns of individual stocks instead of portfolios. The aim is to avoid the potential bias and loss of information implied by sorting and grouping stocks into portfolios. When working with individual stock returns we face several econometric challenges. First, our datasets are characterized by large cross-sectional and time series dimensions, and this fact complicates the numerical implementation and the study of the statistical properties of the estimators. Second, in available datasets, we do not observe asset returns for all firms at all dates, i.e., the panel of stock returns is unbalanced. Third, data feature cross-sectional dependence because of the correlation structure in error terms. To address these challenges, we propose a new estimator that uses simple weighted two-pass regressions. Our estimation methodology accounts for the unbalanced characteristic of large panel data. We study the large sample properties of our estimators in a double asymptotics scheme that reflects the large dimensions of the dataset. In this setting, we test the asset pricing restrictions induced by the no-arbitrage assumption in large economies and we address consistent estimation of the large-dimensional variance-covariance matrix of the errors by sparsity methods. We apply our methodology to a dataset of about ten thousands US stocks with monthly returns from July 1964 to December 2009. The conditional risk premia estimates are large and volatile in crisis periods, and do not match risk premia estimates on standard sets of portfolios.