# Predictability of Stock Returns

**Disclaimer:** This dissertation has been submitted by a student. This is not an example of the work written by our professional dissertation writers. You can view samples of our professional work here.

Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of UK Essays.

INTRODUCTION

Many empirical studies have been done to show the predictability of stock returns and plenty of evidence supports that security returns are predictable.

## Purpose of this paper and summary of paper

DIVIDEND, DIVIDEND YIELD AND STOCK RETURNS

A company makes use of the profit or surplus it earns in two major ways: Firstly, It can reinvest the profits in the business which is normally termed as retained earnings where company has the advantage of investing its surplus in the positive Net Present Value (NPV) projects. Secondly, it can either distribute the excess profits to the shareholders in the form of dividends or make use of that profit in doing the stock repurchases. Most of the companies in the present scenario are retaining a fraction of their total earnings and are distributing the remainder to the shareholders in the form of dividend. Some terms which are used throughout the paper are discussed.

## DIVIDEND

Dividend is a taxable payment which is declared by the management board of a corporation for the class of its shareholders. It is paid to the shareholders out of the corporate profits usually quarterly earnings or retained earnings either in the form of cash, stock or property. A company usually announces quarterly and annually dividends based on a fixed schedule but sometimes announces the dividend at any time in the year and may refer it to special dividend. The allocation of a dividend is done as a fixed amount per shares called dividend per share, which results in shareholder receiving a dividend in regard to his/her shareholding pattern.

## DIVIDEND YIELD

Dividend yield is a rate of return an investor or a shareholder is getting from each buck invested in equity. Dividend yield ratio is a measure of calculating how much a corporation is spending on dividends each year comparative to its share price. It is calculated by dividing the company's annual dividend payments by its market capitalisation or by dividing dividend per share by the price per share.

## SECURITY

## REVIEW OF LITERATURE

A detailed literature review is carried out to focus on the existing research on the subject of predictability of stock returns. Articles published in the professional journals, books, reports and other sourced information from the internet, particularly from the main professional bodies in UK, US and other emerging economies are highlighted.

There is extensive literature available on this topic as various empirical studies have been done to show the predictability of stock returns. The time series behaviour of dividend yield has been observed by many researchers over the last twenty five years and extensive support for using the dividend yield as a prime ratio for measuring expected security returns is provided. Michael S. Rozeff (1984) shows evidence that equity risk premium is forecasted by dividend yield. Among various methods of calculating the equity risk premium, the method of realised market rates of return, Gordon- Shapiro constant growth model and spreads between different bonds classes are the popular ones. He used the constant growth model also known as Gordon growth model and gave it a new twist suggesting that the dividend yield on stock as an approximation of equity risk premium would be very much helpful to look at. Generally the model suggests that the expected rate of return on stock market is equal to a variable of dividend yield on the market plus the predicted growth rate of dividends. He wrote the model as:

RSTK = DYLD (1+GROW) + GROW (2)

Where,

RSTK = Expected rate of return (capital gains plus dividend yield) of the stock market

DYLD = Current dividend yield on the market

GROW = Nominal expected growth rate of the market dividends

Equation for measuring the equity risk premium (RPE) is observed by subtracting out the nominal riskless rate, RBILL (expected rate of return on treasury bills)

RPE = DYLD (1+GROW) + GROW – RBILL (3)

Some very useful characteristics were observed in this view as attention is being driven mainly on two major aspects of expected market return, the yield and growth, but the drawback in the model was that the predicted growth rate of dividend was unobservable due to which the risk premium could not be measured by using the model without estimating the predicted long-run growth rate of dividends. Although the model gives the suggestion that the variation in the expected stock returns should be captured by the dividend yields.

Fama and French (1988) examined the ability of dividend yields to predict stock returns. The dividend / price ratios i.e. dividend yield is used to project the returns on the equal weighted and value weighted portfolios consisting stocks of New York Stock Exchange (NYSE) and having the holding period i.e. return horizons from one month to four years. Their result supports existing evidence that the predictable component of return is a small fraction of short horizon return variances. By using overlapping multiple-year horizon returns, very strong evidence is provided by them in support of the dividend yield effect. They offered evidence that predictability power increases with the return horizon. Less than 5% of monthly or quarterly return variances are explained by performing regression of returns on yields. The discount-rate effect has remained the focus point for the explanation in their studies which suggest that the counter adjustments of security prices are set off by the shock to the discount rates and returns expected. Fama and French (1988) found out that there is an increase in the explanatory power of dividend yield in the time horizon of the returns. They also noticed a massive high of 64% in the R2 over the horizon of 4 years.

Studies over the span of past twenty years have shown that some econometric difficulty exists in the test involving long holding period (horizons) return and hence resulted in the biasness towards the rejection of null hypothesis. Stambaugh (1986) pointed out another problem that the explanatory variable or the independent variable “Dividend yield” contains a level of price that already appear in regressand and therefore it may not be properly a externally originating variable. The error-in-variable problem pointed out by Fama and French (1988) is due to the fact that the information about the prediction of future returns and dividend growth is already present in the yield. It might actually result in the bias down towards the regression coefficient in the regression of dividend yield.

Some issues arising out of these researches were addressed by Hodrick (1992), where he explored the statistical properties of three methodologies in monte calro experiments and used them to carry out conclusion and measurement in long run forecasting experiment along with an application to dividend yields as predictor variable for stock returns. He uses Ordinary Least Square (OLS) regression which is used by Fama and French (1988b) as his prime methodology along with Vector autoregression (VAR) as second methodology which is used in Campbell and Shiller (1988), Kandel and Stambaugh (1988) and Campbell (1991).

By establishing a link between long and short run predictability of return, he demonstrated that consistency exist only between a reasonable large amount of long horizon predictability and small amount of short horizon predictability. For the sample of 1952-1987, the strong support has been provided by the VAR test for the explanatory power of one month ahead returns. The significant constant changes in expected stock returns are forecasted by the changes dividend yield and this conclusion was supported by estimates and results of Monte Carlo experiment.

Simple autoregressive approach is used by Mei (1993) where the ability of multi factor model is examined in explaining the mean return. The capability of model in capturing the dividend yield effect is evidenced by Mei.

A small sample bias in imitation of a VAR system for yields and returns, under null of zero predictability of returns is analysed by Nelson and Kim (1993). They focused on the possibility that there could be an important role played by a bias occurred from a small sample in judging the inference of stock return predictability from the usage of financial fundamentals like dividend yields. It is also reported by Nelson and Kim that t-values could misled as they are prone to two small sample biases:

If the predictor is endogenous then the regression coefficient is biased.

In the case of overlapping periods, the standard error is biased.

Both these biases work in the similar direction and indicate return to be more predictable then they actually are. Using annually sampled return from the U.S. market, they accounted that there is an upward displacing in the simulated distributions of t-statistics, but still traced certain signs of predictability at standard significance levels.

Biased coefficient estimates is result of the regression run on lagged dependent variables is the flaw that ushered from the research of Dickey and Fuller (1976) and Stambaugh(1986). This flaw was addressed by Goetzmann and Jorion(1993), as they used the bootstrapping methodology in modelling the distribution of regression statistics formed under null hypothesis that returns on stocks are identically and independently distributed, without having any relation with the past dividends. By using the bootstrapping method the historical values of dividends are fixed to the past dividends and return series is bootstrapped which resulted in the reconstruction of pseudo-dividend yield, thereby giving only minor evidence of predictability. They fail to reject the null hypothesis that returns follow a random walk and are unrelated to the past dividend yields.

The methodology employed by Fama and French (1988) is again put into the research by Cochran, Define and Mills (1993) to perform further research and came up with two main results that in many of the major stock markets across the globe, the dividend yield has the predictive power to measure the stock returns, and as the length of the return horizon rises from one month to forty eight month, the predictability power also rises thereby giving signs of positive correlations between predictability and the length of return horizon.

By using a long annual UK data series and monthly US data from 1871, covering 121 years Goetzmann and Jorion (1995) extended their previous analysis of dividend yield regressions, and little evidence has been shown throughout the sample period in support for the long horizon return predictability by the use of dividend yield. Dividend yield test is performed using the two new historical series. They used monthly series of dividend yields and returns for New York Stock Exchange (NYSE) and annual UK stock exchange return and yield series beginning in 1872. To explain the regression biases they employed empirical marginal significance level using two main procedures of “Fixed Dividends” and vector autoregression (VAR) with stochastic dividends and prices. The argument highlighted in the research says that the survivorship simulations may affect the tests over longer periods and can show that in case of drawing sample solely from surviving markets, the regression statistics can be critically biased towards the predictability finding.

Kothari and Shanken (1997) tried to evaluate the ability of Book to Market (B/M) ratio in order to track the variation occurring in the time series of expected market index returns and to compare the forecasting ability of B/M ratio to the forecasting ability of dividend yield. For the purpose of evaluating the statistical significance of regression evidence showing the predictability of expected stock returns, they employed vector autoregressive (VAR) framework. They created a null hypothesis of no predictability and tested it by using the bootstrap simulation procedure. The degree and statistical significance level of the relation between expected return and B/M along with dividend yield is also assessed by doing the bootstrap simulation.

They extend the conventional procedure of bootstrap by developing a Bayesian-bootstrap simulation which is then used to estimate the probability of various slope values on the estimates of historical ordinary least squares (OLS). The predictive power of dividend yield is also estimated by using the univariate least square technique. Results obtained from both the conventional and the bootstrap test gives the confirmation that dividend yield has the predictability for the value and equally weighted return.

Claessens, Dasgupta, and Glen (1998) performed a study on the data collected from nineteen emerging markets and examined the effect on the return of an asset by the several different risk factors in addition to the major risk factor . They found out that apart from , size and trading volume lays down significant explanatory power in most of the markets. Further, their results confirmed the already published evidence in regard to the developed markets with throwing lights on some opposite findings. The variable dividend yield seemed to be an important factor playing explanatory role in seven out of nineteen countries studies.

It is evidenced by many researchers that dividend yield has some degree of predictive power to forecast the return component associated with the stock or an asset. This predictive power of dividend yield is re-examined by Ang and Bekaert (2001) in order to forecast the excess returns, cash flows, and interest rates. Their findings were the end result of work they did on two different sets of data, a long data set for US, UK and Germany and a shorter data set for a sample of four countries namely United States, United Kingdom, France and Germany. Dividend yield and earning yield were constructed using the whole lot from past year earnings and dividend. Due to the seasonal component, monthly and quarterly figures were not included. Growth figures for earnings and dividends were constructed from the ratios, and the rates of annual dividend or earning growth over the period of month or quarter were produced.

The results suggest that dividend yield has the power to predict the excess stock return only at short horizons and that too with the short rate. It does not possess any predictability power for the long horizons excess returns in any of the four countries examined. The predictability power of dividend yield to forecast the future dividend growth is not strong enough across sample period and countries studied. Present value model employed by Ann and Bekaert shows that there is a large role played by the discount rate and short rate movements in providing explanation supporting the variation occurring in dividend yields.

Lewellen (2002) focused mainly in short-term horizon test of regression returns and in order to avoid the complication arising from overlapping in returns, monthly returns were regressed on lagged dividend yield. Weak evidence has been provided by all the previous studies that dividend yield forecast stock returns. In order to find stronger evidence, he estimated an Ordinary Least Square (OLS) regression on equal and value weighted returns on log lagged dividends based on New York Stock Exchange (NYSE) by using the same model employed by Stambaugh (1999) and Nelson and Kim (1993). His results provided robust evidence of predictability power for the data period ranging from 1946-2000 and for a range of sub sample.

Another research was carried out by Goyal and Welch (2003) where a simple recursive residuals (out of sample) graphical approach is suggested in order to evaluate the predictability power or of dividend ratio. Based on the US stock market, Returns with lag dividend ratio were regressed for the period from 1926 to 2002. By using the graphical approach they concluded that there is no forecasting power shown by dividend yield for the returns of a year ahead as well as years prior to 1990s.

### Cite This Dissertation

To export a reference to this article please select a referencing stye below: