Modelling Time‐Variation in the Stock Return‐Dividend Yield Predictive Equation

• Published date: 01 December 2014
• DOI: http://doi.org/10.1111/fmii.12021
• Author: David G. McMillan
• Date: 01 December 2014

Modelling Time-Variation in the Stock

Return-Dividend Yield Predictive Equation

BYDAVI D G. MCMILLAN

Using data for forty markets, this paper examines the nature and possible causes of time-

variation within the stock return-dividend yield predictive regression. The results in this

paper show that there is significant time-variationin the predictive equation for returns and

that such variation is linked to economic and market factors. Furthermore, the strength and

nature of those links are themselves time-varying. The inclusion of this time-variation in

the predictive equation increases the predictive power compared to the standard constant

parameter predictive model. Evidence is also reported for time-varying dividend growth

predictability.Long-horizon predictability is also examined with evidence reported that the

nature of the factors affecting time-varyingpredictability changes with horizon. The results

here, while directly contributing to the returns predictability debate, in particular regarding

its existence and source, may also inform the discussion that links time-varying expected

returns (and risk premium) to economic factors.

Keywords: Stock Returns, Predictability, Time-Variation, Dividend Growth, Panel.

JEL Classification: C22,C23, G12.

1. INTRODUCTION

The existence of predictive power of the dividend yield for stock returns remains

one of the most contested questions in empirical finance with evidence both

in favour for and against such a predictive relationship. While examination of

predictive powerfor stock returns in itself may be of interest to market participants,

the greater interest lies in the economic content of this regression. That is, the

nature of this relationship and the equivalent relationship with dividend growth is

important for our understanding of the drivers of variationin equity valuation. That

is, the predictive equation can be used to revealwhether movements in the dividend

yield and equity valuation are primarily driven by cash flow considerations or

variation in the risk premium (see, in particular, Cochrane, 2011).

This paper seeks to contribute to the debate by examining the presence of

predictability for returns (and dividend growth) for a wide range of markets.

Further, our primary belief is that while such predictability exists,it is time-varying

in nature and thus there exist periods of time where movements in valuation are

driven more by expected cash flow and other periods where valuations are driven

more by considerations of expected returns. Hence, periods where stock return

predictability is absent and periods where it is present. Furthermore, we believe

that this time-variation has systematic features that, in turn, can be linked to the

properties of the predictor variable, the state of the market and the state of the

economy (such as the level of inflation or the rate of output growth). It is hoped

Correspondingauthor: David McMillan University of Stirling Phone: +44(0)1786-467309; Fax: +44(0)1786-467308

E-mail: david.mcmillan@stir.ac.uk

C2014 New YorkUniversity Salomon Center and Wiley Periodicals, Inc.

274 David G. McMillan

the results of this paper will further our understanding of stock price movements.

In particular, where Cochrane (2011) argues that the dividend yield proxies for

changes in expected return (and ultimately the risk premium), the results here

may help identify the links between such variables and market and economic

variables. Notwithstanding that, evidence of cash flow predictability has recently

been provided by Ang (2011). Thus, this paper will consider both these sources

of predictability.

The debate surrounding stock return predictability largely goes back to

Campbell and Shiller (1988a,b) and Fama and French (1988) who argued that

the dividend yield could be used as a predictor for stock returns. Subsequent to

this, a vigorous debate has been ongoing as to whether such predictability is robust.

For example, Campbell, Lo and MacKinlay (1997), Campbell and Shiller (2001),

Campbell and Yogo (2006), Campbell and Thompson (2008) and Kellard et al.

(2010) have provided further supportive evidence. While several authors have ar-

gued against such predictability due to econometric issues relating to persistence

in the regressor or small sample bias (see, for example, Nelson and Kim, 1993;

Stambaugh, 1999; Wolf, 2000; Lanne, 2002; Valkanov, 2003; Ang and Bekaert,

2007; Welch and Goyal, 2008). Of particular note, Campbell and Yogo (2006)

argued that the over rejection of the null of no predictability can arise due to

persistence within the regressor variable. While Cochrane (2008) has retorted that

that the dividend yield must havepredictive power for returns (or dividend growth)

otherwise, in the context of the present value model, the dividend yield would be

a constant. Furthermore, this latter point considers the economic content of the

predictive regression as opposed to just the nature of the statistical relationship.

That is, the present value model for stock prices states that the dividend yield must

be related to either, or both, of changes in expected returns or changes in expected

dividend growth. The absence of predictability for both ofthese is evidence against

the present value model.

Most recently,attention has turned to whether there is evidence of time-variation

with predictability. Chen (2009) has reported evidencethat the dividend yield may

predict either returns or dividend growthbut across different time periods. Building

upon the work of Campbell and Yogo (2006), Park (2010) argues that in a sub-

sample of US data that includes the 1990s the predictive power of the dividend

yield disappears. Park argues that this is related to the non-stationary behaviour

of the dividend yield over this timeframe. Related, in an earlier study, Goyal and

Welch (2003) argued that increased persistence in the dividend yield had led to

decreased predictive power evenprior to the 1990s. Kellard et al. (2010) using the

same methodology however, findsome predictive power for the UK. Engsted and

Pedersen (2010), employing long-term annual data for the US, Sweden, Denmark

and the UK, report evidence of time-variation in the strength of predictive power

for returns and dividend growth. They report that for the US post-war period

there is strong evidence of predictability for long-horizon real returns, with the

correct coefficient sign and for real dividend growth but with the wrong sign. In

contrast, they report no returns predictability but dividend growth predictability