Return forecasts and optimal portfolio construction: a quantile regression approach

Abstract

In finance there is growing interest in quantile regression with the particular focus on value at risk and copula models. In this paper, we first present a general interpretation of quantile regression in the context of financial markets. We then explore the full distributional impact of factors on returns of securities and find that factor effects vary substantially across quantiles of returns. Utilizing distributional information from quantile regression models, we propose two general methods for return forecasting and portfolio construction. We show that under mild conditions these new methods provide more accurate forecasts and potentially higher value-added portfolios than the classical conditional mean method.return forecast, quantile regression, portfolio construction,

A Quantile Regression Approach to Equity Premium Prediction

“…As Koenker and Bassett () show, estimators of this type have high efficiency over a large class of distributions. A subset of the above specifications has been employed by Taylor () and Ma and Pohlman (), among others.…”

The predictive power of the yield spread for future economic expansions: Evidence from a new approach

“…Forecasting the whole distribution rather than merely its centre is better aligned with the original proposition by Granger (1969), who defines causality in terms of conditional distributions of variables, rather than its subsequent implementations which focus almost exclusively on conditional means. Predicting quantiles of the future distribution has been already shown to be largely successful for stock returns, and in particular, helping improve forecasts of the future centre of return distribution by utilising predictions of off-the-centre future returns (Cenesizoglou and Timmermann, 2008, Ma and Pohlman, 2008, Zhu, 2013, Meligkotsidou et al, 2014, Pedersen, 2015. Therefore, we expect that the ability to predict the shape of the future distribution of economic growth using yield spread values would improve our knowledge about the future growth, and potentially our estimates of future expected growth rates.…”

Stock return distribution and predictability: Evidence from over a century of daily data on the DJIA index

“…Yet another related branch of literature investigates the predictive power of observed economic variables for the entire future return distribution, as approximated by a set of different quantiles. Ma and Pohlman (2008) show that in-sample, different financial valuation factors can explain different quantiles of future return distribution. Pedersen (2015) also reports the in-sample and out-of sample predictive power of economic variables for stock and bond return distributions, finding that different variables predict different quantiles of future return distribution, most frequently in the tails and least strongly in the center.…”