The impact of time-varying risk on stock returns: an experiment of cubic piecewise polynomial function model and the Fourier Flexible Form model

Abstract

<abstract> <p>With fast evolving econometric techniques being adopted in asset pricing, traditional linear asset pricing models have been criticized by their limited function on capturing the time-varying nature of data and risk, especially the absence of data smoothing is of concern. In this paper, the impact of data smoothing is explored by applying two asset pricing models with non-linear feature: cubic piecewise polynomial function (CPPF) model and the Fourier Flexible Form (FFF) model are performed on US stock returns as an experiment. The traditional beta coefficient is treated asymmetrically as downside beta and upside beta in order to capture corresponding risk, and further, to explore the risk premia attached in a cross-sectional context. It is found that both models show better goodness of fit comparing to classic linear asset pricing model cross-sectionally. When appropriate knots and orders are determined by Akaike Information Criteria (AIC), the goodness of fit is further improved, and the model with both CPPF and FFF betas employed showed the best fit among other models. The findings fill the gap in literature, specifically on both investigating and pricing the time variation and asymmetric nature of systematic risk. The methods and models proposed in this paper embed advanced mathematical techniques of data smoothing and widen the options of asset pricing models. The application of proposed models is proven to superiorly provide high degree of explanatory power to capture and price time-varying risk in stock market.</p> </abstract>