João Issler, da FGV-EPGE, vai apresentar o seu trabalho de investigação. 

A No-Arbitrage Approach to Asset Pricing using
Panel Data∗

We propose a no-arbitrage framework related to stochastic discount factors (or pricing kernels) that takes seriously the consequences of no-arbitrage in asset pricing. First, we establish identification of a valid stochastic discount factor (SDF) based on economic theory – no-arbitrage, the asset-pricing equation – and information contained in both cross-section and time-series of returns. Indeed, the cross-sectional dimension plays a pivotal role for the identification of a valid SDF. Second, based on the identification result, we propose a no-arbitrage estimator for a valid SDF. Third, we derive the asymptotic properties of this estimator, namely, its consistency and asymptotic normality when the number of assets and of time periods increase without bounds. The asymptotic character of this no-arbitrage SDF estimator is opposed to standard small-sample alternatives where it is hard to interpret empirical results since these often change when different groups of assets are used in estimation. In theory, asymptotic estimates are immune to this problem. Fourth, we derive a no-arbitrage one-factor model for the logarithm of asset returns, where the single factor is the logarithm of a valid SDF, containing all the pervasive elements of (log) asset returns. Based on the proposed approach, we first investigate which type of utility function best fits U.S. data among popular preference specifications in the lit-erature: the constant-relative-risk-aversion (CRRA) coefficient utility function; the external habit utility function; and the Kreps-Porteus specification. Sec-ond, based on these estimation results, we present a no-arbitrage simulation study assessing how close our consistent SDF estimator is to actual SDF for medium-sized panel-data samples. Third, we estimate a multi-effect linear re-gression model that allows for parameter heterogeneity in the intercept and in its slope that is consistent with our derived one-factor model. Using regression results, we assess how well this heterogeneous one-factor model fits the cross-section and time-series data of returns. We are also able to test directly our key identification assumption by means of a Wald test. In our final empirical study we ask if our approach can be used to price the cross-sectional distribution of asset returns for the U.S. economy using mid-sized data for N and T extracted from the Fama-French library. Results are surprisingly good.