Debate surrounds the ability to predict stock returns , particularly, but not only arising from the dividend-price and earnings -price ratios. This lineage of research dates back to Campbell and Shiller (1988) and Fama and French (1988) and has seen recent interest (see, e.g. Campbell and Thompson 2008; Cochrane 2008, 2011; Kellard et al. 2010; McMillan and Wohar 2010, 2013; McMillan 2014, 2015). That predictability should occur is central to our understanding of asset price movement, i.e. such movement arises from changes to expected cash flow and risk premia, which in turn arises from changes in economic conditions. However, consistent empirical evidence in favour of predictability is lacking (see, e.g. Ang and Bekaert 2007; Goyal and Welch 2003; Welch and Goyal 2008; Park 2010). Nelson and Kim (1993) argue that relatively small samples used within this research agenda can lead to inconsistent results. One emergent line of research argues that the lack of consensus in the empirical literature arises from the potential for instability in the relation between prices and dividends . Paye and Timmermann (2006) suggest the potential for breaks in the coefficient values within the predictive regression, while Lettau and Van Nieuwerburgh (2008) consider the presence of shifts within the predictor variable. Moreover, Timmermann (2008) argues that predictability models generally perform poorly; however, there exist short-lived periods of time where predictability can be found. In a similar vein, Chen (2009) argues that predictability may switch between periods of returns predictability and dividend growth predictability over time (see, also McMillan and Wohar 2013), while Campbell and Thompson (2008) and Park (2010) argue that evidence in favour of predictability has declined over time.
Of course, returns predictability is one side of the coin on which cash-flow predictability lies on the reverse. In addition to the above-cited work of Chen (2009) and McMillan and Wohar (2013), there is an increasing body of work that supports such cash-flow predictability. Notably, Engsted and Pedersen (2010), Ang (2011) and Rangvid et al. (2014) all present evidence supporting cash-flow predictability across a wide range of international markets. An understanding of whether stock return or cash-flow predictability exists and the relative strength of any predictive effect is important in enhancing our knowledge of the key drivers of movements in asset prices. Equally, whether the nature of predictability changes over time, perhaps related to the business cycle, will provide useful information not only to investors but also to policy makers who could use such knowledge to understand the future course of the economy.
This book seeks to explore these themes by examining predictability for a range of international stock markets. The aims here are several. First, we seek to consider whether predictability exists in stock returns and/or cash-flow growth. In doing so, we consider a range of alternative methodologies that take advantage of not only the usual predictive regression approach, but also the underlying cointegrating relation that is hypothesised to exist within the stock price and dividend (earnings ) behaviour. Within these modelling frameworks, we also consider the potential for time-variation to exist and whether allowing such time-variation affects the nature of predictability and the balance between stock returns and cash flow. The presence of time-variation would seem a reasonable consideration, especially when viewed over a period of several decades. For example, market deregulation during the 1980s, the dot.com bubble and bust over the late 1990s and early 2000s, and the financial and sovereign debt crises of the late 2000s and early 2010s are likely to have an impact on the relation between stock prices and their fundamentals.
Second, while the set of results contained in Chap. 2 is based on in-sample behaviour within stock returns and cash flow , it is equally important to consider out-of-sample predictability. Through considering this type of forecast power, we can ensure that any results obtained within one sample of data can be generalised to a second sample. Indeed, within this context, it could be argued that the true test of a model’s predictive ability lies in its out-of-sample forecast performance. Moreover, by considering a range of international stock markets as well as out-of-sample behaviour with provide robust evidence on the nature of predictability. As with Chap. 2, in Chap. 3, we will continue to consider the role of time-variation within the econometric approach. In this case, we undertake that using fixed window, rolling and regressions. This approach, which drops old observations as it adds new ones, will allow the forecasts to take account of any breaks , regardless of the break dates. The influence of the break will be included and then excluded as the fixed window moves through the sample. This contrasts with a more static forecast approach where, if the break occurs in the out-of-sample period, the influence of the break will be ignored. Equally, a break during the in-sample period will condition the forecasts even when (if) the influence of the break declines. Notwithstanding this, it could be argued that the rolling window approach adopts a cliff-edge fall in terms of past information as it drops out of the window. Thus, we also consider a recursive, or expanding window, approach in which new information is added but older observations are not lost. Hence, a secondary interest in this chapter is whether the use of a rolling or recursive forecast approach is preferred.
As noted above, evidence suggests that both stock return and dividend growth predictability may both exist but over different periods of time, such that the nature of predictability switches. Chapter 4 will consider this issue more explicitly and examines the correlation between stock return and dividend growth predictability over a range of time horizons. Furthermore, by adopting a quantile regression approach within this chapter, we are also able to examine whether stock return and dividend growth predictability varies with the level of the dependent variable. These two approaches will add weight to the existing evidence concerning the nature of predictability and formalise any relation between stock return and dividend growth predictability.
While these three chapters consider stock return and cash-flow growth predictability through the dividend yield as expressed by the present value model. Chapter 5 will expand the set of predictor variables. Taking a general view of asset pricing, ultimately any variable that can proxy for economic cond...