Demographics and the long-horizon returns of dividend-yield strategies

The Quarterly Review of Economics and Finance 53 (2013) 202–218

Contents lists available at SciVerse ScienceDirect

The Quarterly Review of Economics and Finance journal homepage: www.elsevier.com/locate/qref

Demographics and the long-horizon returns of dividend-yield strategies King Fuei Lee ∗ Schroder Investment Management, 65 Chulia Street, #46-00 OCBC Centre, Singapore 049513, Singapore

a r t i c l e

i n f o

Article history: Received 2 April 2012 Received in revised form 31 December 2012 Accepted 5 February 2013 Available online 13 February 2013 JEL classification: G000 G350 C320

a b s t r a c t This paper investigates the relationship between demographic changes and the long-run returns of dividend-yield investment strategies. We hypothesise that in a world where components of wealth are mentally treated as being non-fungible, the preference for high dividend-paying stocks by older investors means that the excess returns of high dividend-yielding stocks, relative to other stocks, should be positively related to demographic clientele variation. In particular, we find that, consistent with the behavioural life-cycle hypothesis, long-run returns of dividend-yield investment strategies are positively driven by changes in the proportion of the older population. Our results are robust when controlled for the Fama–French factors, inflation rate, consumption growth rate, interest rates, tax clienteles, time trend and alternative definitions of both dividend-yield strategies and demographic variation. © 2013 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved.

Keywords: Dividend yield Demographics Investment style Investment strategy

1. Introduction The Modigliani and Brumberg (1954, 1980) life-cycle model of consumption and savings assumes that individuals try to smooth their consumption over their lifetimes. Because labour income flows are uneven over the course of life, therefore saving rates will also vary over the course of life. In particular, the model posits that individuals have low saving rates during their early adult years, but will save more with age as their incomes increase, before dissaving in their retirement as earnings fall. One of the implications of the life-cycle model is that financial asset prices are linked to changing demographics. The reasoning is that the middle-aged, who are at the peak of their earnings potential, tend to be heavily involved in the accumulation of net assets as they save for their retirement. Prices of financial assets such as stocks and bonds are therefore likely to rise due to the higher demand from a relative increase in the size of the middleaged cohort. However as this age group enters retirement, they will start to decumulate their wealth which will consequently cause a fall in financial asset prices. This thus forms the basis of the Asset Meltdown Hypothesis which posits that, just as the higher demand for financial assets by the baby boomers had led to the rise in US

∗ Tel.: +65 6535 3411; fax: +65 6535 3486. E-mail address: [email protected]

stock prices in the 1990s (Shiller, 2000; Sterling & Waite, 1998), the impending retirement of these baby boomers is also likely to cause a stock market meltdown around 2020. Given the theoretical importance of demographics to stock markets, there has been a wealth of empirical studies examining the relationship between them. While Poterba (2001, 2004) did not find any systematic relationship between demographic structure and returns on stocks, Jamal and Quayes (2004) analysed the impact of demographic structure on stock prices in the US and UK and showed that the proportion of population in the prime earning age has had a direct influence on stock prices. Bakshi and Chen (1994) similarly found that a rise in the average age in the US corresponded to a rise in the risk premium of an S&P500 portfolio, while Claude, Campbell, and Viskanta (1997) also discovered a positive relationship between the average age of the US population and the long-run returns on the S&P500. Davis and Li (2003) extended their analysis to seven countries in the OECD and found that generally an increase in the proportion of middle-aged people tends to boost financial asset prices. Arnott and Chaves (2012) adapted a polynomial curvefitting technique on a cross-section of 22 countries, and found that stocks perform best when the cohort of people aged 35–59 is large, and when the cohort of people aged 45–64 is fast-growing. Despite empirical evidence generally supporting the linkage between demographic structure and stock markets, there has been little work done investigating the effects of demographics on the performances of investment strategies within

1062-9769/$ – see front matter © 2013 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.qref.2013.02.001

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

stock markets, particularly on dividend-yield investment strategies. The effectiveness of dividend-yield strategies in enhancing portfolio returns is well-documented. Studies have generally identified a positive relationship between share price performance and dividend yield (Christie, 1990; Fama & French, 1988; Grant, 1995; Hodrick, 1992; Visscher & Filbeck, 2003). Several papers have tried to explain the effectiveness of dividend-yield investment strategies. Brennan (1970) proposed a tax-effect hypothesis that predicts that investors receive higher returns to compensate for the higher taxes on the dividend income of these stocks relative to capital gains. Naranjo, Nimalendran, and Ryngaert (1998) however determined that the positive correlation between dividend yield and return holds true even after adjusting for risk and tax effects. Gombola and Liu (1993a) attributed the effectiveness of dividendyield strategies to the stability of beta, while Gombola and Liu (1993b) tied it to the economic cycle. This paper explores the relationship between demographic changes and the long-run returns of dividend-yield investment strategies. We find that, consistent with the behavioural lifecycle hypothesis, long-run returns of dividend-yield strategies are positively driven by demographic clientele variation as represented by changes in the proportion of the older population.1 We show that a one percentage point increase in the proportion of seniors is related to a 221.7%, 199.9% and 172.3% increase in the ten-year excess returns of high dividend-yielding stocks versus low-yielding stocks excluding zero-dividend stocks, low-yielding stocks including zero-dividend stocks and the market respectively. Our robustness checks also show that our results hold when controlled for the Fama–French factors, inflation rate, consumption growth rate, interest rates, tax clienteles, time trend and alternative definitions of both dividend-yield strategies and demographic variation. There are several motivations for this work. Firstly, our research relates to the impending ageing population that will afflict the US over the next few decades. According to the US Census Bureau, the proportion of US population over the age of 65 is estimated to grow from the current 13% in 2012 to 21% in 2050. Given the importance of demographic forces and the looming powerful changes on the horizon, studies looking at demographic changes should attract more and more attention. Secondly, our findings have different implications for future stock market behaviour versus the Asset Market Meltdown Hypothesis. According to our hypothesis, the meltdown can be avoided because the old population will be switching within their portfolios into high dividend-yielding stocks instead of selling down their entire financial portfolios. Thirdly, our research also has important implications for corporate dividend policies. Since the buying of dividend-yielding stocks by the older population leads to better share price performances of these companies, managers of companies may be tempted to alter their dividend policies in order to cater to this demographic dividend clientele. This will have wide-ranging implications on the economy through its effects on investments and productivity. Fourthly, our paper adds to the understanding of the drivers of long-horizon returns of dividend-yield investment strategies. Specifically, practitioners may find the predictability of the returns of investment

1 We note that for the purpose of our hypothesis, the magnitude of the proportion of older population is not crucial. This is because according to the marginal opinion theory (Williams, 1938; Smith, 1967), in a market comprising of a number of interested parties who each possess an opinion as to the worth of the stock, the price of the stock is not set by the majority, regardless of how overwhelming it is, but by the last owner. Thus our hypothesis only requires that the marginal investor is represented by an older person.

203

strategies interesting. To our knowledge, there has not been any work done on this aspect. The rest of this paper is structured as follows: Section 2 discusses the behavioural life-cycle hypothesis, and introduces our hypothesis. Section 3 describes the data sample and the methodology applied. The empirical findings are reported in Section 4, while robustness tests are conducted in Section 5. Section 6 highlights potential further work that can be done in this area, while Section 7 concludes the paper. 2. Demographic clientele variations and dividend-yield investment strategies 2.1. Behavioural life-cycle theory and the dividend preferences of older investors According to the behavioural life-cycle theory (Thaler & Shefrin, 1988), households treat components of their wealth as nonfungible. Wealth is assumed to be broken into three mental accounts: current income, current assets and future income. The temptation to spend is greatest for current income and least for future income. The behavioural life-cycle theory therefore hypotheses that in the retirement stage of a household’s life cycle, the investor perception of the non-fungibility between dividends and capital gains should lead to a preference for high dividend-paying stocks by older investors for consumption purposes. Empirical evidence has generally been supportive. Graham and Kumar (2006) studied the stock holdings and trading behaviour of 77,995 households over the period of 1991–1996 and found that, compared to younger investors, older investors allocate a greater proportion of their equity portfolios to dividend paying stocks. Lee (2011) investigated the importance of demographic clienteles as a source of the time-varying demand for dividend payers, and found that the dividend premium is positively driven by changes in the proportion of the older population. 2.2. Hypothesis development We hypothesise that in a world where components of wealth are mentally treated as being non-fungible, the preference for high dividend-paying stocks by older investors means that the excess returns of high dividend-yielding stocks, relative to other stocks, should be positively related to changes in the proportion of the older population. The larger the increase in the proportion of old population, the greater the relative demand for high dividendpaying assets, and hence the stronger the relative performance of dividend-yield investment strategies. We formalise our hypothesis in a simple model that highlights the many strong assumptions that are needed for our conclusion. We start with the model of Poterba (2001) which expresses the relation between demographic structure and asset prices as p ∗ K = Ny ∗ s

(1)

where p is the relative price of assets in terms of the numeraire good, K is the fixed supply of durable assets, Ny is the proportion of young individuals in a world where they work when young (y) and retire when old, and s is the saving rate out of labour income for young workers. Thus according to the model of Poterba (2001), asset prices are positively related to the size of the working cohort. We can rewrite Eq. (1) with respect to high dividend-paying stocks (DY) to incorporate the stronger preferences by older investors for these assets: pDY,t ∗ KDY,t = No,t ∗ wo,t

(2)

204

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

where pDY,t is the relative price of high dividend-paying assets in terms of the numeraire good at time t, KDY,t is the fixed supply of high dividend-paying assets, No,t is the proportion of old individuals, and wo,t is the rate of equity ownership out of wealth for old individuals. The price return of high dividend-paying assets RDY,t is therefore expressed as RDY,t =

KDY,t-1 pDY,t No,t wo,t = ∗ ∗ pDY,t-1 No,t-1 wo,t-1 KDY,t

(3)

It can be seen from Eq. (3) that the relative price returns of high dividend-paying assets are a function of three factors: the change in the proportion of old individuals, the change in the rate of equity ownership of old workers, and the rate of change in the supply of high dividend-paying assets. In particular, our paper focuses on exploring the relationship between changes in the demographic structure, or demographic variation, and the price returns of high dividend-yielding assets i.e. dividend-yield investment strategies. While Eq. (3) sets the general framework of our hypothesised relationship, two points are worth noting. The first point is that we have chosen to focus on the long-horizon rather than shorterterm time frames. This is done for several reasons. Firstly from a practical perspective, investors typically re-position their portfolios gradually over long periods of time. Secondly, stock returns in the short horizon can be particularly susceptible to short-term noise and market vagaries which obscure the true long-run relationship. For example, when Yoo (1994) estimated the multivariate time-series regressions of annual U.S. stock, corporate and government bond returns to shares of total population for different age groups, he found that the statistical significance of his results increases dramatically when three- and five-year centred moving average data is used instead of annual data. Thirdly, as pointed out by Fama (1998), because “stock prices adjust slowly to information, one must examine returns over long horizons to get a full view of market inefficiency”. This is especially important when examining stock prices against slowly changing demographic variation because “long horizons provide a better test for low frequency population changes” (Arnott & Chaves, 2012). The second point to note is that the long-horizon returns of dividend-yielding investment strategies are matched against annual changes in the proportion of old population in our study here. Our thinking is that since retirement reflects a spot-in-time change in employment status (from being employed to being retired), and is known in advance, therefore the ageing population who is expecting to retire over the next few years is likely to gradually start re-positioning their portfolios today and continue to do so over time. In other words, the returns of the dividend-yield investment strategy over the ten-year period from year t-10 to t reflects the gradual portfolio re-positioning into dividend-yielding stocks by the additional proportion of ageing people who are anticipating to retire in year t when they reach 65 years old.2 We will discuss more in the later section.

2 According to the report Equity Ownership in America 2005 by the Investment Company Institute and the Securities Industry Association (available at http://www.ici.org/research/investors/equity owners), 56.9 m US households (i.e. half of all US households) own equities in 2004, with the median age of the investor being 51 years old and the median equity holdings being $65,000. Of these households, 33% are in the 50–64 years old age group and have median equity holdings of $87,500. This implies that almost half (i.e. 44.4%) of the retail money invested in US equities is controlled by the age group of 50–64 years old, and their investment patterns are likely to have a discernible effect on stocks. It also means that the total equity holdings of this age bracket stood at US$1.62tr in 2004. We note that the total market capitalisation of high dividend-yielding stocks then was US$1.20tr. It is therefore plausible that as these individuals gradually re-position their portfolios into dividend-yielding stocks in anticipation of their retirement, the weight of money is large enough to drive returns of high dividend-yielding stocks in a

3. Data sample and methodology This section briefly discusses the data sources and definitions of the variables used. Following Gwilym, Clare, Seaton, and Thomas (2009), we define a dividend-yield investment strategy as a zerocost long-short directionally neutral strategy comprising of equal positions in both a long and short portfolio of value-weighted stocks that are in the highest and lowest quintiles of stocks ranked by dividend yield respectively. The stock universe from which the portfolios are constructed includes all NYSE, AMEX and NASDAQ firms which have at least seven monthly returns from July of year t-1 to June for year t, and for which market equity data for June of year t is available. The portfolios are formed on the dividend yield at the end of each June using NYSE breakpoints, with the dividend yield that is used to form portfolios in June of year t being the total dividends paid from July of t-1 to June of t per dollar of equity in June of t. The annual performance of the portfolio is then measured from January to December of year t. Because of the observation of a “U-shaped” relationship between dividend yield and return in the US and UK by Keim (1985) and Morgan and Thomas (1998) respectively, Gwilym et al. (2009) highlighted the potential need to consider zero-dividend firms as a separate group rather than incorporating them into the lowest dividend quintile. We have therefore defined our longshort dividend-yield strategies to allow for both the inclusion and exclusion of zero-dividend stocks. We have also defined a third long-short dividend strategy consisting of long positions in high dividend-yielding stocks and short positions in the market portfolio. The returns from this strategy can be interpreted alternatively as being the excess market returns of a long-only portfolio of high dividend-yielding stocks. As highlighted before, this study is focused on the long-run horizon. Our definition of long-horizon is 10-years, a time frame that is also consistent with that of studies like Campbell and Shiller (1998) and Rapach and Wohar (2005). This long-run 10-year number is also close to the intuition of Arnott and Casscells (2003) who suggested that the effects of the demographic crisis in the U.S. which begins in earnest in 10 years are more likely to already have an impact on capital markets now. The long-run returns of our investment strategies are therefore calculated as the ten-year returns of our three dividend-yield investment strategies and denoted as variables RQ1–Q5 Excl zero-div , RQ1–Q5 Incl zero-div and RQ1–Rm . Data used for our calculations are downloaded from the website of Kenneth French.3 In this paper, we also focused on value-weighted portfolios when measuring the returns of the dividend-yielding investment strategies. This is done for several reasons. Firstly, as highlighted by Liew and Vassalou (2000), value-weighted portfolios result in more realistic returns in the presence of small capitalisation stocks. This is especially apt here because, as pointed out by Gwilym et al. (2009), dividend stocks are more susceptible to “the small firm issue (particularly with zero-dividend firms) [hence reporting value-weighted returns] is likely to offer a better representation of practical strategies”. Secondly, the use of value-weighted factor portfolios to represent investment strategies is also very common in academic research and widely adopted in various studies

meaningful manner. In addition, the proportion of seniors was the fastest growing demographic bracket over the last few decades (rising from 5.2% in 1928 to 13.1% in 2011) and is forecast to remain so going forward (reaching 20.2% in 2050). It is therefore also highly likely that seniors will be increasingly identified with the marginal investor. 3 The returns reported by Kenneth French are all multiplied by 100, and the data is available at http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ data library.html.

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

including Fama and French (1993), Bauer, Koedijk, and Otten (2005) and Asness (1997). Thirdly, intuition suggest that the older population, being more risk-averse (Harlow & Brown, 1990; Riley & Chow, 1992), are more likely to express their preference for high dividend-yielding stocks through the purchase of large capitalisation stocks versus small capitalisation stocks given the relative safety of large companies, particularly during recessionary times (Bernanke & Gertler, 1989; Gertler & Gilchrist, 1994; Perez-Quiros & Timmermann, 2000) Following Graham and Kumar (2006) and Poterba (2001), we use the old-to-population ratio Old/Population, defined as the proportion of population aged above 65 to the total population, as the variable representing the demographic structure. Demographic variation is therefore expressed as the annual change in the old-to-population ratio dOld/Population. We also employ an alternative measure of the demographic variation variable dOld/PSavers, defined as the annual change in the old-to-prime savers ratio, in our robustness test. The other alternative measure dOld/Population(t,t-10) represents the ten-year change in the old-to-population ratio. All demographic variables are expressed in percentage terms. The US population data used for the calculations of the measures of demographic variations is downloaded from the US Census Bureau4 website. In our robustness checks, we employ the Fama–French factors and measures of inflation rate, consumption growth rate, interest rates and tax clienteles. The Fama–French factors are constructed using the 6 value-weighted portfolios formed on market capitalisation and book-to-market. The small cap effect SMB is calculated as the average return on the three small portfolios minus the average return on the three big portfolios i.e. SMB = 1/3 (Small Value + Small Neutral + Small Growth) − 1/3 (Big Value + Big Neutral + Big Growth), while the value effect HML is calculated as the average return on the two value portfolios minus the average return on the two growth portfolios i.e. HML = 1/2 (Small Value + Big Value) − 1/2 (Small Growth + Big Growth). Rm –Rf is the excess return on the market and is calculated as the value-weighted return on all NYSE, AMEX, and NASDAQ stocks minus the one-month Treasury bill rate. Inflation rate is calculated as the percentage change in the Consumer Price Index CPI over the last ten years and is denoted by dCPI/dt, while the real consumption growth rate dC/dt represents the percentage change in the real per capita consumption over the last ten years. For interest rates, we define the short-term interest rate R as the one-year interest rate while the long-term interest rate RLong is the long government bond yield. The term structure or yield curve YC is therefore calculated as the difference between RLong and R, while the short-term interest rate movement dR/dt is the change in the one-year interest rate over the last ten years. The data used for these calculations is downloaded from the website of Robert Shiller.5 To account for the existence of tax clienteles, we calculate the control variable Tax which represents the relative marginal tax on dividend income versus capital gain for the highest tax bracket. The data for this calculation is obtained from the website of Citizens for Tax Justice.6 Following the methodology of Fama and French (1992), Fama and French (1998) and Poterba (2001), we employ multivariate Ordinary Least Squares (OLS) regression on overlapping data to estimate the relation. The regression is expressed as RQ1–Q5,t = ˛0 + ˛1 dOld/Populationt + ˛2 ControlVart + εt

4 5 6

Available from: http://www.census.gov. Available from: http://www.econ.yale.edu/∼shiller/data.htm. Available from: http://ctj.org/.

205

where ControlVart represents the relevant control variables and εt is the random disturbance term. It is well known that while using overlapping data in regression helps achieve greater efficiency, it also induces a moving average process in the errors which invalidates the usual OLS standard errors. We therefore adjust for this by calculating the standard errors using the Newey and West (1987) heteroscedasticity and autocorrelation consistent variance matrix that is based on the Bartlett kernel. This thus provides asymptotically valid hypothesis tests when using data with overlapping observations. The time period employed in this study is from 1937 to 2011 which represents the period for which long-term data for the three dividend-yield strategies is available.

4. Empirical findings Fig. 1 shows the time series plots of long-run excess returns of the dividend-yield investment strategy (Q1–Rm ) and the annual change in old-to-population ratio. It is observed that while both variables are not perfectly synchronous, they are visibly positively related to each other. Indeed it can be seen from the correlation matrix in Table 2 that the contemporaneous correlation between the dividend-yield strategy and the demographic variation measure of annual change in older-to-population ratio is 0.427 at 5% significance level. Fig. 2 shows the time series plots of long-run excess returns of the dividend-yield investment strategy (Q1–Q5 excluding zerodividend stocks) and the annual change in old-to-population ratio, while Fig. 3 shows the plots of long-run excess returns of the dividend-yield investment strategy (Q1–Q5 including zero-dividend stocks). It can be seen that the returns of both dividend-yield investment strategies also appear to be positively related to the demographic variation variable. Table 2 shows that the correlations stand at 0.337 and 0.286 respectively, and are significant at 5% levels. Table 1 shows the descriptive statistics of the dividend-yield strategies, the demographic variation variables as well as the control variables, while Table 2 shows their unit root test statistics and the correlation matrix. The unit root test employed here is the Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test (Kwiatkowski, Phillips, Schmidt, & Shin, 1992) which uses the null hypotheses of linear stationarity and trend stationarity respectively. It can be seen that for the dividend-yield strategy variables, the demographic variation variables and the control variables, the unit root tests generally accept the null hypotheses of linear stationarity and trend stationarity. This supports our employment of OLS regressions in our empirical analysis, as consistent with Fama and French (1992), Fama and French (1998) and Poterba (2001). Fig. 5 shows the time series plots of the average firm sizes for the different dividend-yield portfolios. It can be seen that the average firm sizes for all the portfolios rose gradually during the first few decades of the sample time period before dramatically rising in the 1980s and 1990s. This was a period of sustained stock market booms which Bakshi and Chen (1994) ascribed to demographic changes and the financial asset accumulation by Baby Boomers. The Internet Bubble period 1998–2000, however, saw divergent paths in the average firm size of Q1 and Q5 portfolios, with the latter increasing very sharply while the former dwindled. This trend reversed in 2000 with the bursting of the bubble. The average firm sizes of all the portfolios declined in 2008 in the face of the Global Financial Crisis before staging a recovery in 2009 on the back of quantitative easing by the Federal Reserve. Fig. 6 shows the spreads in average firm sizes between the long and short portfolios of the different dividend-yield investment strategies. It can be

206

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

Fig. 1. Time series plots of long-run excess returns of dividend-yield investment strategy (Q1–Rm ) and annual change in old-to-population ratio, 1937–2011.

seen that the spread has remained mostly negative between 1939 and 1978, implying that smaller companies dominated the space of high dividend-yielding stocks during that period. The subsequent changes in the spreads reflect our earlier observations in Fig. 5. Figs. 7 and 8 show the average dividend-yields of the various portfolios and the spreads in average yields for the different investment strategies respectively. The average dividend yield of the Q1 portfolio experienced huge spikes in the years of the Great Depression and World War Two. This phenomenon is similarly witnessed by Cornell (2012) who believed them to be a reflection of the great uncertainty during those times which had impacted expectations of future long-run dividend growth. It can also be seen that outside

of those periods, the spread in average dividend yields between the long and short portfolios of the different investment strategies has remained mostly within the band of 2.5–7.5%. Column 1 of Table 3 shows the OLS regressions of the 10-year returns of the dividend-yield strategy (Q1–Q5 excluding zerodividend stocks) RQ1–Q5 Excl zero-div against the annual change in old-to-population ratio over the period of 1937–2011. It can be seen from our regression results that the annual change in the old-to-population ratio is a positive and statistically significant determinant of the long-run returns of the dividend-yield strategy. The returns of the dividend-yield strategy are high when the proportion of older population to total population increases, while

Fig. 2. Time series plots of long-run excess returns of dividend-yield investment strategy (Q1–Q5 excluding zero-dividend stocks) and annual change in old-to-population ratio, 1937–2011.

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

207

Fig. 3. Time series plots of long-run excess returns of dividend-yield investment strategy (Q1–Q5 including zero-dividend stocks) and annual change in old-to-population ratio, 1937–2011.

the returns are low when the proportion of older population to total population falls. Column 2 of Table 3 shows the regression of the 10-year returns of the dividend-yield strategy (Q1–Q5 including zero-dividend stocks) RQ1–Q5 Incl zero-div against the annual change in the old-topopulation ratio over the period of 1937–2011, while Column 3 of Table 3 shows the regression of the 10-year returns of the dividend yield strategy (Q1–Rm ) RQ1–Rm against the demographic variation variable. The relationships are positive and highly significant at 5% and 1% levels respectively, and show that both dividend-yield strategies outperform over the long-run when the proportion of older population increases, and underperform when the proportion of older population falls. Our results therefore confirm our hypothesis that demographic variations are important determinants of the long-run performance of dividend-yield investment strategies in the US.

5. Robustness checks and control variables 5.1. Fama–French factors Fama and French (1993) proposed that the excess returns on US portfolios can be predominantly captured by a three-factor model that uses the market portfolio and mimicking portfolios for the factors related to size and value-growth. In particular, they find that the expected return on a portfolio in excess of the risk-free rate is largely explained by the sensitivity of its return to the three factors: (i) the excess return on a broad market portfolio (Rm –Rf ); (ii) the difference between the return on a portfolio of small stocks and the return on a portfolio of large stocks (SMB, or small minus big); and (iii) the difference between the return on a portfolio of high-book-to-market (growth) stocks and the return on a portfolio of low-book-to-market (value) stocks (HML, or high minus low).

Table 1 Descriptive statistics, 1937–2011. Variable

Mean

Standard deviation

Maximum

Minimum

No. of observations

RQ1–Q5 Excl zero-div RQ1–Q5 Incl zero-div RQ1–Rm RQ1–Q5 Excl zero-div EW RQ1–Q5 Incl zero-div EW RQ1–Rm EW dOld/Population dOld/PSavers dOld/Population(t,t-10) Rm –Rf SMB HML R RLong dCPI/dt dC/dt YC dR Tax

19.704 14.384 23.479 6.476 −31.548 56.470 0.091 0.209 0.929 111.736 44.367 64.196 64.040 72.657 44.991 25.924 6.610 −0.147 19.186

40.601 43.938 23.626 29.699 43.881 50.093 0.062 0.525 0.515 105.124 59.381 46.620 49.443 45.657 33.866 7.941 8.018 3.677 23.838

124.918 163.375 78.287 91.673 170.379 146.410 0.193 1.082 1.523 388.266 215.548 238.947 169.347 172.197 129.577 42.066 19.622 12.199 66.000

−54.769 −69.668 −42.897 −47.777 −82.009 −38.624 −0.079 −1.103 −0.253 −39.523 −37.119 −37.921 7.171 25.691 −19.429 2.188 −9.322 −10.418 −26.000

75 75 75 75 75 75 75 75 75 75 75 75 75 75 74 73 75 75 75

208

Table 2 Unit root statistics and correlation matrix, 1937–2011. RQ1–Q5 Incl zero-div

RQ1–Rm

RQ1–Q5 Excl zero-div EW

RQ1–Q5 Incl zero-div EW

RQ1–Rm EW

dOld/ Population

dOld/ PSavers

dOld/ Population (t,t-10)

Rm –Rf

SMB

HML

R

RLong

dCPI/dt

dC/dt

YC

dR

Tax

Unit root test

KPSS: Trend stationarity

0.258 0.089

0.347*

0.410*

0.192

0.181

0.112

0.275

0.254

0.348*

0.162

0.193

0.095

0.303 0.126*

0.331

0.171

0.097

0.161

0.225

0.146

0.102

0.094

0.148**

0.147**

0.133*

0.08

0.076

0.083

0.091

0.060

0.063

0.102

0.086

0.071

0.079

0.09

0.933**

0.871**

0.851**

0.221

0.488**

0.337**

0.297**

0.339**

0.086

0.096

0.653** −0.272** −0.239*

0.163

0.240*

0.332** −0.086

−0.382**

1.000

0.131

−0.043

0.558** −0.177

0.111

Correlation matrix 1.000 RQ1–Q5 Excl zero-div 0.933** RQ1–Q5 Incl zero-div RQ1–Rm 0.871** 0.851** RQ1–Q5 Excl zero-div EW 0.221 RQ1–Q5 Incl zero-div EW RQ1–Rm EW 0.488** 0.337**

0.854**

0.787**

0.423**

0.325**

0.286**

0.257*

0.448**

0.854**

1.000

0.740**

0.061

0.625**

0.427**

0.400**

0.421**

0.787**

0.740**

1.000

0.206

0.426**

0.265**

0.235*

0.075

0.056

0.423**

0.061

0.206

1.000

−0.453**

−0.291**

0.026

0.203

0.325**

0.625**

0.426**

−0.453**

0.286**

0.427**

0.265**

−0.246*

0.257* 0.448**

0.400** 0.421**

0.235*

−0.291**

0.075

0.026

−0.246*

−0.012

0.119

0.179

0.256* −0.119

−0.286**

0.046

0.279**

0.250*

−0.131

0.514** −0.272** −0.270** 0.565** −0.323** −0.305**

0.063

0.381**

0.267** −0.100

−0.263** −0.427**

−0.649**

0.152

0.369**

0.164

0.287**

−0.144

0.427**

1.000

0.643**

0.596**

0.336**

−0.451**

0.740**

0.468**

0.020

0.643**

1.000

0.893**

0.672**

−0.322**

0.491**

0.243* −0.029

0.596** 0.336**

0.893** 0.672**

1.000

0.622**

−0.213

0.622**

1.000

0.051

0.406** 0.388**

0.292** −0.106 0.246* 0.096

−0.451** 0.740**

−0.322** 0.491**

−0.213 0.406**

0.051

1.000

−0.355**

0.150

−0.388** −0.342** −0.111

−0.355**

1.000

0.055

−0.014

−0.045

0.055 −0.388** −0.014 ** −0.342 −0.045

1.000

0.121

0.093

0.121

1.000

0.974**

0.093

0.114

0.974** 0.680**

1.000

−0.111

0.597**

0.019

0.350** −0.019

−0.048

−0.066

−0.044

0.320**

0.289** −0.084

−0.131

0.259**

0.091

−0.265**

0.245* 0.149 0.379** −0.150

0.011

−0.258* 0.171

0.066 −0.114

−0.219

0.268** −0.067

−0.414**

0.519**

−0.094

0.225

−0.081

0.309**

0.042

−0.266**

0.058

0.096

−0.034

dOld/Population dOld/PSavers

0.297** 0.339**

0.048

0.152

dOld/Population(t,t10) Rm –Rf SMB

0.086

0.131

0.096

−0.043

R

0.653** −0.272**

RLong

−0.239*

HML

dCPI/dt

0.558**

−0.012 0.287** 0.514**

0.056 −0.131

0.203 −0.649**

0.388** 0.246*

−0.177

0.565** −0.323**

0.152

−0.272**

0.369**

0.020

−0.029

−0.106

0.096

−0.144

−0.270**

−0.305**

0.427**

−0.044

−0.131

−0.265** 0.245*

0.048

0.164

0.320**

0.381**

−0.048

0.289**

0.267**

−0.066

0.119

0.046

0.063

dC/dt

0.163 0.240*

0.179

0.279**

YC

0.332**

0.256*

0.250*

dR

−0.086

−0.119

Tax

−0.382**

−0.286**

0.011 −0.263**

0.468**

−0.100 −0.427**

−0.258* 0.066

−0.084 0.171 −0.114

0.243*

0.259** 0.091 −0.219 0.268** −0.067

0.292**

0.379**

0.149

−0.150

−0.414** 0.519**

−0.094

0.152

−0.081

0.225

0.150

−0.047 0.309** 0.042

0.058 0.096

−0.266** −0.034

Note: The KPSS test employs a Newey–West type variance estimator of the long-run variance of u(t), with truncation lag m = [c · n ] where c = 5, s = .25, n = 74. * Significance level: 10%. ** Significance level: 5%. s

−0.095 0.259** −0.177

0.627** −0.041

0.114

−0.047 0.019

0.597** 0.350** −0.095 0.259** −0.177 0.680** −0.019 −0.593** 0.048 0.427** ** ** 0.627 −0.041 −0.402 −0.133 0.295** 1.000 0.181 −0.438** 0.343** 0.030 1.000

−0.038

−0.593** −0.402** −0.438** −0.038 0.048 −0.133 0.343** 0.188 0.427** 0.295** 0.030 0.036

1.000

0.181

−0.610** −0.749**

0.188

0.036

−0.610** −0.749** 1.000 0.379** 0.379**

1.000

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

KPSS: Level stationarity

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

209

Fig. 4. Time series plots of long-run excess returns of dividend-yield investment strategy (D1–Rm ) and Fitted Model, 1937–2050.

We therefore employ the three Fama–French factors as control variables in our robustness test. This is particularly apt as high dividend-yield investment strategies are frequently also classified under the broader category of value strategies that includes low market-to-book strategies and low price-to-earnings strategies. It is therefore interesting to investigate whether the relationship between dividend-yield strategies and demographic variation

reflects a broader relationship between value strategies and demographics. Columns 4–6 in Table 3 show the multivariate regressions of the dividend-yield strategies against the three Fama–French factors. It can be seen that the demographic variation variable remains a statistically significant positive determinant of RQ1–Q5 Excl zero-div and RQ1–Rm at 5% levels, and of RQ1–Q5 Incl zero-div at 1% level. The

Table 3 Regressions of long-run returns of dividend-yield investment strategies against annual change in old-to-population ratio and control variables, 1937–2011. Multivariate regressions of various long-run returns of dividend-yield investment strategies against demographic variation and the control variables, RQ1–Q5,t = ˛0 + ˛1 dOld/Populationt + ˛2 ControlVart + εt , The long-run returns of the dividend-yield investment strategy RQ1–Q5 is the ten-year cumulative return of a zero-cost long-short directionally neutral portfolio comprising of long and short positions in the highest (Q1) and lowest (Q5) quintiles of stocks ranked by dividend yield respectively. The demographic variation measure is given by dOld/Population which represents the annual change in old-to-population ratio. ControlVar represents the relevant control variable that is used in the robustness checks.

Dependent variable

Explanatory variables dOld/Population Fama–French: Rm –Rf Fama–French: SMB Fama–French: HML dCPI/dt Constant R-squared No. of observations

(1) (2) (3) Long-short directionally neutral

(4)

Ten-year cumulative excess returns

Ten-year cumulative excess returns

RQ1–Q5

RQ1–Q5

RQ1–Rm

Excl zero-div

Incl zero-div

222.370 (2.740)*** – – – – – – – – −0.641 (−0.117) 0.114 75

204.156 (2.134)** – – – – – – – – −4.295 (−0.455) 0.082 75

163.766 (3.145)*** – – – – – – – – 8.495 (2.246)** 0.183 75

(5)

(6)

Incl zero-div

149.358 (2.611)** 0.024 (0.561) −0.018 (0.886) 0.514 (4.824)*** – – −28.783 (−4.217)*** 0.464 75

201.912 (2.783)*** 0.036 (0.694) −0.132 (−0.887) 0.458 (3.226)*** – – −31.669 (−2.852)*** 0.372 75

(8)

(9)

Ten-year cumulative excess returns RQ1–Rm

Excl zero-div

(7)

102.444 (2.333)** 0.015 (0.541) 0.062 (0.632) 0.218 (2.593)** – – −4.263 (−0.972) 0.379 75

RQ1–Q5

RQ1–Rm

Excl zero-div

Incl zero-div

209.446 (2.535)** – – – – – – 0.098 (0.437) −3.810 (−0.449) 0.119 74

206.707 (2.233)** – – – – – – 0.057 (0.232) −6.525 (−0.5738) 0.092 74

182.102 (3.540)*** – – – – – – −0.054 (−0.507) 9.826 (2.186)** 0.214 74

Note: T-statistics are shown in parentheses and are based on the Newey–West (1987) heteroscedasticity and autocorrelation consistent variance matrix. Lag lengths (l) used to evaluate the serial correlation for the Newey–West correction follows the recommendation by Newey and West (1994) and is computed as l = 4(T/100)0.25 where T is the number of observations. * Significance level: 10%. ** Significance level: 5%. *** Significance level: 1%.

210

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

Fig. 5. Time series plots of average firm size of various dividend-yield portfolios, 1928–2011.

Fig. 8. Time series plots of spreads in average dividend yield of dividend-yield investment strategies, 1928–2011.

returns over the period 1926–1991 and the sub-period 1941–1991. We have therefore shown that our hypothesis of demographic variation as an important determinant of the long-run returns of dividend-yield investment strategies is robust to the inclusion of the Fama–French factors. 5.2. Inflation rate

Fig. 6. Time series plots of spreads in average firm size of dividend-yield investment strategies, 1928–2011.

Fama–French factors of market premium Rm –Rf and small cap effect SMB are not significant. However, the Fama–French factor of the value effect HML is significant at 1% level for RQ1–Q5 Excl zero-div and RQ1–Q5 Incl zero-div , and at 5% significance level for RQ1–Rm , a finding that is consistent with Kothari and Shanken (1997) who found reliable evidence that both dividend yield and book-tomarket track time-series variation in expected real one-year stock

Conventional wisdom holds that stocks, particularly dividendpaying stocks, are good hedges against inflation. For example, Carrel (2010) advised that “one of the best ways to keep inflation from taking a bite out of your investment earnings is to invest in dividend-paying stocks. The big advantage dividends hold over other income generating investments is they have the potential to keep pace with inflation. As prices rise, profits also tend to rise, and companies can afford to raise their dividend payments”. His view is similarly echoed by Arnott (2003), who pointed out that “the importance of dividends for providing wealth to investors is self-evident [because] dividends [. . .] dwarf inflation”. Empirical evidence has however been mixed. While earlier studies (Fama & Schwert, 1977; Geske & Roll, 1983) have generally found a negative relationship between short-horizon stock returns and inflation, recent studies (Boudoukh & Richardson, 1993; Kolari & Anari, 2001) have found that stocks can serve as effective longterm inflation hedges. In particular, Basse (2009) and Basse and Reddemann (2011) have found positive cointegrating relationships between inflation and dividends in Australia and US respectively. We therefore include inflation as a control variable in our study. Columns 7–9 in Table 3 show the results of our multivariate regressions. The annual change in old-to-population ratio remains a statistically significant determinant of the long-run returns of all three dividend-yield strategies. It can also be seen that inflation rate is not a significant explanatory variable, thus supporting the empirical findings of Geske and Roll (1983) and Fama and Schwert (1977) while casting doubts on the conventional belief of high dividendyielding stocks as inflation hedges. Our results also show that our earlier findings hold even with the inclusion of inflation rate considerations. 5.3. Consumption

Fig. 7. Time series plots of average dividend yields of various dividend-yield portfolios, 1928–2011.

The mental accounting theory (Thaler & Shefrin, 1988; Thaler, 1980) posits that households do not view dividends and capital gains as fungible, and have a higher propensity to consume out of the mental account for dividends than for capital gains. Empirical

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

evidence has generally been supportive of this concept of mental accounting by investors. For example, when Baker et al. (2006) examined the micro data sets from the Consumer Expenditure Survey and a large discount brokerage, they found strong evidence that the marginal propensity to consume out of dividend income is much higher than that of capital gains income. It is worth noting here that the prediction by mental accounting theory, that households prefer to consumer out of dividends, applies to the general household and is not unique to specific age groups. While the behavioural life-cycle hypothesis extends on the prediction to imply a stronger preference for dividend-paying stocks by the older population compared to the rest of the population, it is still within the theoretical framework for the general population to also have a stronger preference for dividend-paying stocks if they are planning to increase their overall consumption. It is therefore conceivable that the price performance of high dividend-yielding stocks simply reflects the greater purchase of these stocks by investors in general to fund their increased consumption. We therefore include the consumption growth rate as a control variable in our robustness check. Columns 10–12 in Table 3 show the results of the multivariate regressions. The demographic variation variable remains a statistically significant positive determinant of the long-run returns of the dividend-yield strategies RQ1–Q5 Excl zero-div , RQ1–Q5 Incl zero-div and RQ1–Rm . The consumption growth rate variable is not significant for RQ1–Q5 Excl zero-div and RQ1–Q5 Incl zero-div although it is significant at 10% level for RQ1–Rm . Our hypothesis that the long-run returns of dividend-yield strategies are driven by demographic changes is therefore robust to the inclusion of consumption growth considerations.

5.4. Interest rates Because the yield of a dividend-paying stock is a significant component of its total return, high dividend-yielding stocks are often compared against other income-generating options such as putting the money into savings or money market accounts, or buying bonds or certificates of deposits (Carrel, 2010). Mladjenovic (2009) thus highlighted that “income stocks can be sensitive to rising interest rates. When interest rates go up, other investments (such as corporate bonds, U.S. treasury securities, and bank certificates of deposit) are more attractive. [. . .] As more and more investors sell their low-yield stock, the prices for those stocks fall”. Given the potential sensitivity of dividend-yielding stocks to interest rates, we therefore include measures of interest rate as control variables in our robustness checks. We use four different measures of interest rates in our tests. They are the short-term one-year interest rate, the long bond yield, the term structure or yield curve, and the change in short-term interest rates. Columns 13–15, 16–18 and 19–21 in Table 4 and Columns 22–24 in Table 5 show the results of the multivariate regressions of the long-run returns of the three dividend-yield strategies to both the demographic variation variable as well as the short-term interest rate, long bond yield, yield curve and annual change in short-term interest rate respectively. In all the regressions, the annual change in old-to-population ratio remains an important determinant of the long-run returns of the dividend-yield strategies, with the significance levels at either 1% or 5%. It can also be seen that the various interest rate measures are generally not significant except for the yield curve which is a positive and statistically significant determinant of long-run returns of dividend-yield strategies at 5% levels. Our earlier hypothesis is therefore robust to the inclusion of interest rate considerations.

211

5.5. Tax clienteles The theory on dividend tax clienteles was first proposed by Miller and Modigliani (1961) who conjectured that investors are likely to sort themselves into clienteles in which low-tax investors collect dividends and high-tax investors realise capital gains so as to reduce their overall tax bills. Recent empirical studies have been supportive. For example, Allen, Bernardo, and Welch (2000) and Dahlquist et al. (2006) conducted studies on the US and Swedish markets respectively and found the existence of dividend tax clienteles in both markets. To investigate whether the demographic clientele effect that we report here is proxying for tax clienteles, we include the relative marginal tax rates of dividends versus capital gains as a control variable. Columns 25–27 of Table 5 show the results of our multivariate regressions. It can be seen the demographic variation variable remains positively related to the ten-year returns RQ1–Q5 Excl zero-div , RQ1–Q5 Incl zero-div and RQ1–Rm at 5%, 10% and 1% significance levels respectively. The sign of the tax control variable is also negative as theorised for all three regressions. However it is only significant at 10% level for RQ1–Q5 Excl zero-div and not for the rest. Our results are therefore robust to the inclusion of tax clienteles as control variable. 5.6. Time trend While a visual observation of the graphs and the results of our earlier unit root tests will suggest that the demographic variation variable and the three measures of dividend-yield strategies are trend-stationary, it is nevertheless prudent to include the time trend as a control variable to test the robustness of our results. Column 28–30 of Table 5 shows the results of the multivariate regressions. Even with the inclusion of the time trend as a control variable, the annual change in the proportion of old population continues to be positively related to the long-run returns of RQ1,Q5,Excl zero-div and RQ1–Rm . The explanatory power of the demographic variation variable is however not significant for the dividend-yield strategy that includes zero-dividend stocks, RQ1–Q5 Incl zero-div . Our results therefore largely support our earlier conclusions even when controlled for the time trend. 5.7. Alternative definition of demographic variation variable While our measure of demographic structure is intuitive from our hypothesis, we acknowledge that there are alternative definitions to the demographic structure used in other research. In their analysis of the effects of demographic structure on asset prices in Asia, Eskesen, Lueth, and Syed (2008), for example, have defined the demographic structure as the ratio of prime consumers (aged 65+) to prime savers (aged 40–65). We have therefore adopted the definition of Eskesen et al. (2008) as an alternative definition of the demographic structure and calculated the equivalent demographic variation variable as the annual change in the old-to-prime savers ratio. Table 2 shows that the correlations between the annual change in old-to-prime savers and the three measures of longrun dividend-yield strategy returns are all positive and statistically significant at 5% level. The results of the univariate regressions are shown in Columns 31–33 of Table 5. It can be seen that the alternative definition of demographic variation remains an important determinant of the ten-year returns of the RQ1–Q5 Excl zero-div , RQ1–Q5 Incl zero-div and RQ1–Rm at 5%, 10% and 1% significance levels respectively. We chose to compare ten-year returns to annual changes in the demographic variable for the reasons highlighted in Section

212

(10) (11) Long-short directionally neutral

Dependent variable

Explanatory variables dOld/Population dC/dt R RLong YC Constant R-squared No. of observations

(12)

(13)

(14)

10-year cumulative excess returns

10-year cumulative excess returns

RQ1–Q5

RQ1–Q5

RQ1–Rm

Excl zero-div

Incl zero-div

211.050 (2.867)*** 1.093 (1.201) – – – – – – −27.831 (−1.240) 0.158 73

208.242 (2.286)** 0.849 (1.021) – – – – – – −25.669 (1.211) 0.116 73

169.492 (3.938)*** 0.702 (1.878)* – – – – – – −9.203 (−0.943) 0.276 73

(15)

RQ1–Rm

Excl zero-div

Incl zero-div

217.417 (2.965)*** – – −0.216 (−1.479) – – – – 13.635 (1.306) 0.183 75

200.717 (2.191)** – – −0.150 (−0.822) – – – – 5.616 (0.445) 0.110 75

160.913 (3.415)*** – – −0.124 (−1.732)* – – – – 16.720 (2.937)*** 0.250 75

(16)

(17)

(18)

(19)

(20)

10-year cumulative excess returns

10-year cumulative excess returns

RQ1–Q5

RQ1–Q5

RQ1–Rm

Excl zero-div

Incl zero-div

205.278 (2.922)*** – – – – −0.177 (−1.103) – – 13.765 (1.035) 0.153 75

194.069 (2.158)** – – – – −0.104 (−0.499) – – 4.206 (0.261) 0.094 75

152.877 (3.336)*** – – – – −0.113 (−1.424) – – 17.673 (2.514)** 0.229 75

(21)

RQ1–Rm

Excl zero-div

Incl zero-div

283.990 (3.087)*** – – – – – – 2.160 (2.527)** −20.557 (−2.201)** 0.287 75

256.443 (2.445)** – – – – – – 1.833 (2.136)** −21.195 (−1.759)* 0.189 75

194.148 (3.389)*** – – – – – – 1.065 (2.445)** −1.324 (−0.230) 0.307 75

Note: T-statistics are shown in parentheses and are based on the Newey–West (1987) heteroscedasticity and autocorrelation consistent variance matrix. Lag lengths (l) used to evaluate the serial correlation for the Newey–West correction follows the recommendation by Newey and West (1994) and is computed as l = 4(T/100)0.25 where T is the number of observations. * Significance level: 10%. ** Significance level: 5%. *** Significance level: 1%.

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

Table 4 Regressions of long-run returns of dividend-yield investment strategies against annual change in old-to-population ratio and control variables, 1937–2011. Multivariate regressions of various long-run returns of dividend-yield investment strategies against demographic variation and the control variables, RQ1–Q5,t = ˛0 + ˛1 dOld/Populationt + ˛2 ControlVart + εt , The long-run returns of the dividend-yield investment strategy RQ1–Q5 is the ten-year cumulative return of a zero-cost long-short directionally neutral portfolio comprising of long and short positions in the highest (Q1) and lowest (Q5) quintiles of stocks ranked by dividend yield respectively. The demographic variation measure is given by dOld/Population which represents the annual change in old-to-population ratio. ControlVar represents the relevant control variable that is used in the robustness checks.

(22) (23) (24) Long-short directionally neutral

Dependent variable

Explanatory variables dOld/Population dOld/PSavers dR Tax t Constant R-squared No. of observations

(25)

(26)

10-year cumulative excess returns

10-year cumulative excess returns

RQ1–Q5

RQ1–Q5

RQ1–Rm

Excl zero-div

Incl zero-div

255.758 (2.975)*** – – −2.099 (−1.650)* – – – – −3.999 (−0.701) 0.147 75

244.574 (2.384)** – – −2.524 (−1.506) – – – – −8.364 (−0.955) 0.123 75

175.267 (3.054)*** – – −0.718 (−0.996) – – – – 7.338 (1.808)* 0.194 75

Excl zero-div

Incl zero-div

206.276 (2.410)** – – – – −0.615 (−1.858)* – – 12.635 (1.236) 0.244 75

191.153 (1.874)* – – – – −0.495 (−1.489) – – 6.394 (0.491) 0.154 75

(27)

(28)

10-year cumulative excess returns

10-year cumulative excess returns

RQ1–Rm

RQ1–Q5

RQ1–Q5

157.629 (2.923)*** – – – – −0.233 (−1.502) – – 13.537 (2.329)** 0.238 75

(29)

(30)

RQ1–Rm

Excl zero-div

Incl zero-div

170.585 (2.384)** – – – – – – −0.328 (−0.900) 650.836 (0.899) 0.139 75

106.116 (1.239) – – – – – – −0.620 (−1.788)* 1229.085 (1.786)* 0.158 75

99.705 (2.106)** – – – – – – −0.405 (−2.398)** 814.409 (2.422)** 0.294 75

(31)

(32)

(33)

RQ1–Rm

Excl zero-div

Incl zero-div

– – 22.950 (2.120)** – – – – – – 14.914 (2.317) 0.088 75

– – 21.503 (1.642)* – – – – – – 9.895 (1.200) 0.066 75

– – 18.007 (3.176)*** – – – – – – 19.720 (5.447)*** 0.160 75

Note: T-statistics are shown in parentheses and are based on the Newey–West (1987) heteroscedasticity and autocorrelation consistent variance matrix. Lag lengths (l) used to evaluate the serial correlation for the Newey–West correction follows the recommendation by Newey and West (1994) and is computed as l = 4(T/100)0.25 where T is the number of observations. * Significance level: 10%. ** Significance level: 5%. *** Significance level: 1%.

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

Table 5 Regressions of long-run returns of dividend-yield investment strategies against annual change in old-to-population ratio and control variables, 1937–2011. Multivariate regressions of various long-run returns of dividend-yield investment strategies against demographic variation and the control variables, RQ1–Q5,t = ˛0 + ˛1 dOld/Populationt + ˛2 ControlVart + εt , The long-run returns of the dividend-yield investment strategy RQ1–Q5 is the ten-year cumulative return of a zero-cost long-short directionally neutral portfolio comprising of long and short positions in the highest (Q1) and lowest (Q5) quintiles of stocks ranked by dividend yield respectively. The demographic variation measure is given by dOld/Population which represents the annual change in old-to-population ratio. ControlVar represents the relevant control variable that is used in the robustness checks.

213

R-squared No. of observations

Constant

dOld/Population(t,t-10)

Note: T-statistics are shown in parentheses and are based on the Newey–West (1987) heteroscedasticity and autocorrelation consistent variance matrix. Lag lengths (l) used to evaluate the serial correlation for the Newey–West correction follows the recommendation by Newey and West (1994) and is computed as l = 4(T/100)0.25 where T is the number of observations. * Significance level: 10%. ** Significance level: 5%. *** Significance level: 1%.

527.446 (6.198)*** – – 8.367 (0.769) 0.421 75 129.359 (1.839)* – – −5.321 (−1.010) 0.072 75 – – 26.726 (2.108)** −5.128 (−0.535) 0.115 75

Incl zero-div Excl zero-div

– – 38.248 (3.171)*** −21.154 (−3.104)*** 0.201 75

– – 19.303 (2.492)** 5.544 (0.840) 0.177 75

493.960 (4.523)*** – – −27.267 (−3.754)*** 0.236 75

Incl zero-div Excl zero-div

366.111 (3.324)*** – – −26.574 (−2.276)** 0.155 75

306.424 (5.887)*** – – −10.142 (−1.921)* 0.389 75

156.620 (1.855)* – – −2.933 (−0.359) 0.066 75

180.374 (2.182)** – – −7.380 (−0.918) 0.078 75

83.057 (1.398) – – 3.957 (0.646) 0.046 75

−175.723 (−2.2245)** – – −15.522 (−1.376) 0.061 75

Incl zero-div Excl zero-div Incl zero-div Excl zero-div

RQ1–Q5

5-year cumulative excess returns

RD1–Rm RD1–D5

10-year cumulative excess returns

RQ1–Rm 10-year cumulative excess returns

RQ1–Q5 Dependent variable

Explanatory variables dOld/Population

RQ1–Q5 EW RQ1–Rm

10-year cumulative excess returns

(45) (44) (43) (42) (41) (40) (39) (38) (37)

We have also varied the definition of the dividend-yield investment strategies in our robustness checks. This is done in three ways: Firstly, we vary the dividend yield threshold cut-off points used in the construction of the long-short portfolios to capture the top/bottom deciles of stocks rather than the top/bottom quintiles. Secondly, we shorten the time horizon of the strategies to five-years instead of ten-years. Thirdly, we adopt equal-weighted portfolios instead of value-weighted portfolios. Columns 37–39 in Table 6 shows the multivariate regressions where the long-short directionally neutral portfolios are calculated as long and short positions in the highest (D1) and lowest (D10) deciles of stocks ranked by dividend yield respectively. It can be seen that the demographic variation variable is a highly significant determinant of the long-run returns of the alternatively defined dividend-yield strategies, RD1–D10 Excl zero-div , RD1–D10 Incl zer-div and RD1–Rm , at 1% levels. In fact it appears that the R-squared of the regressions are also higher when the definition of high dividend-yielding stocks is tightened. This supports our hypothesis that the preference for high dividend-yielding stocks by the old population leads to the long-run outperformance of these stocks as their prices are being bidded up by the higher demand over time. Columns 40–42 in Table 6 show the multivariate regressions when the definition of the long-run time frame is shortened to fiveyears. It can be seen that the annual change in old-to-population ratio continues to be a positive determinant of the dividend-yield strategies RQ1–Q5 Excl zero-div and RQ1–Q5 Incl zero-div at 10% and 5% significance levels respectively, although it is not significant for RQ1–Rm . Our results therefore largely support our hypothesis even when alternative time horizons are employed. We however note that the R-squared of the regressions also drop when the time horizon is reduced, thus supporting our choice, as well as the recommendations of Fama (1998) and Arnott and Chaves (2012), of using a longer time horizon in the analysis of the effects of demographics on dividend-yield strategies. As explained earlier, this paper focuses on value-weighted portfolios when defining the dividend-yield investment strategies. While we do intuitively also expect the demographic clientele effect to be more effectively reflected in value-weighted portfolios versus equal-weighted portfolios, we want to check if our hypothesis continues to hold in the face of the small firm issue (Gwilym et al., 2009). Columns 43–45 of Table 6 show the results of our univariate regressions of the equal-weighted strategies RQ1–Q5 Excl zero-div EW , RQ1–Q5 Incl zero-div EW and RQ1–Rm EW . It can be seen that the demographic variation variable remains an important determinant at 5%, 10% and 1% significance levels respectively. Our findings are also robust to alternative definitions of both the dividend-yield strategies.

(36)

5.8. Alternative definition of dividend-yield strategies

(34) (35) Long-short directionally neutral

2.2. However, we acknowledge that our simple model represented in Eq. (3) reflects more the relationship between ten-year returns versus ten-year demographic variation. We therefore also conduct regressions of the ten-year returns of the dividend-yielding strategies against the ten-year change in the proportion of old population dOld/Population(t,t-10) as a robustness check. Table 6 Columns 34–36 shows the results of our regressions. It can be seen that the ten-year demographic variation variable continues to be positively related to the ten-year returns RQ1–Q5 Excl zero-div , RQ1–Q5 Incl zero-div and RQ1–Rm at 5%, 1% and 5% significance levels respectively. Our hypothesis is therefore robust to alternative definitions of the demographic variation variable.

RQ1–Rm EW

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218 Table 6 Regressions of long-run returns of dividend-yield investment strategies against annual change in old-to-population ratio and control variables, 1937–2011. Multivariate regressions of various long-run returns of dividend-yield investment strategies against demographic variation and the control variables, RQ1–Q5,t = ˛0 + ˛1 dOld/Populationt + ˛2 ControlVart + εt , The long-run returns of the dividend-yield investment strategy RQ1–Q5 is the ten-year cumulative return of a zero-cost long-short directionally neutral portfolio comprising of long and short positions in the highest (Q1) and lowest (Q5) quintiles of stocks ranked by dividend yield respectively. The demographic variation measure is given by dOld/Population which represents the annual change in old-to-population ratio. ControlVar represents the relevant control variable that is used in the robustness checks.

214

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

215

Fig. 9. Time series plots of long-run excess returns of dividend-yield investment strategy (Q1–Q5 excluding zero-dividend stocks) and annual change in old-to-population ratio for international markets.

6. Further research

demographic structure starts to negatively impact returns of the strategy.

6.1. Forecasting model 6.2. Extension to international markets While this paper only seeks to investigate the relation between demographic variation and the long-horizon returns of dividend-yield investment strategies, the next step of work can be focused on constructing forecasting models for predicting these long-run returns. Although not within the scope of this paper, we made a whimsical attempt at forecasting the long-run returns of RD1–Rm in Fig. 4. Our fitted model, based on the regression coefficients in Column 39 of Table 6 and the demographic projections from the US Census Bureau, is forecasting an increase in the long-run returns of the dividend-yield investment strategy from now until 2025 before the shift in the

Another area of further research will be to extend the empirical study to other international markets. We do this here by examining the relationships graphically, as well as performing panel regressions. The data for international markets are obtained from Kenneth French’s website and from the US Census Bureau. Fig. 9 shows the individual time series plots of the long-run excess returns of the dividend-yield investment strategy of Q1–Q5 excluding zero-dividend stocks (represented by the black line in each country plot) and the annual change in old-to-population ratio (represented by the grey line in each country plot) for the

216

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

Table 7 Regressions of long-run returns of dividend-yield investment strategy variables against annual change in old-to-population ratio in US and international markets. Multivariate regressions of various long-run returns of dividend-yield investment strategies against demographic variation and the control variables, Rt = ˛0 + ˛1 dOld/Populationt + εt , The long-run returns of the dividend-yield investment strategy R is the ten-year cumulative return of a zero-cost long-short directionally neutral portfolio comprising of long and short positions in the specified high and low quintiles of stocks ranked by dividend yield respectively. Rank IC is the rank correlation coefficient between the long-run returns of the dividend-yield investment strategies and their respective quintiles. The demographic variation measure is given by dOld/Population which represents the annual change in old-to-population ratio.

Dependent variable

Explanatory variables dOld/Population Constant R-squared No. of observations

(46) (47) Long-short directionally neutral 10-year cumulative excess returns

(48)

OLS: US

Panel Regression: International markets

RQ2–Q4

167.180 (2.119)** 8.324 (1.223) 0.130 75

Rank IC

2.605 (1.669) 0.148 (0.787) 0.099 75

(49)

RQ1–Q5

(50)

RQ1–Rm

Excl zero-div

Incl zero-div

248.897 (1.818)* 33.413 (1.501) 0.062 357

131.225 (1.651)* 38.876 (2.063)** 0.031 356

26.701 (0.573) 35.365 (3.091)*** 0.004 363

Note: T-statistics are shown in parentheses and are based on the Newey–West (1987) heteroscedasticity and autocorrelation consistent variance matrix. Lag lengths (l) used to evaluate the serial correlation for the Newey–West correction follows the recommendation by Newey and West (1994) and is computed as l = 4(T/100)0.25 where T is the number of observations. * Significance level: 10%. ** Significance level: 5%. *** Significance level: 1%.

nineteen international markets of Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Hong Kong, Italy, Japan, Netherlands, New Zealand, Norway, Singapore, Spain, Sweden, Switzerland and United Kingdom. The charts for each country are plotted for the fullest time period for which data of both the demographic variation variable and the excess return variable are available. It can be graphically seen that while the relationship appears to be inverse for Austria and Norway, the relation appears to be at least moderately positive for most other countries. In particular for the countries of Hong Kong, Italy, Netherlands, Singapore and Sweden, the relationship between the annual change in the proportion of old population and the long-run excess returns of the dividend-yield investment strategy looks to be very strong. In Table 7 Columns 48–50, the results of the random-effects7 unbalanced panel regressions are presented. It can be seen that the annual change in old-to-population ratio is a positive determinant of the dividend-yield strategies RQ1–Q5 Excl zero-div and RQ1–Q5 Incl zero-div at 10% significance level, although it is not significant for RQ1–Rm . Our results therefore provide certain evidence that our hypothesised relationship also holds in international markets. Further work in this area should account for various other considerations such as the differences in the forms of local pension schemes and the differences in regulatory/legislative frameworks for a more complete analysis. 6.3. Other control variables The control variables employed in our robustness checks are by no means exhaustive, as we are constrained by the availability of long-dated data. Future work can therefore also examine the relationship of demographic variation and dividend-yield strategies in the presence of changes in equity ownership as well as the supply of dividend-paying assets as highlighted in Eq. (3).

7 Our choice of random-effects model over fixed-effects model is determined by the results of the Hausman test which we conducted on the two models. The results showed that the null hypothesis of no correlation between the unique errors and the regressors is not rejected at 5% significance level.

In addition, an implicit assumption adopted in this paper is that of constant risk aversion by the older population. Malmendier and Nagel (2011a, 2011b), for example, found that the risk appetites of individuals are often affected by their macroeconomic experiences. The implication therefore is that the preference for high dividend-yielding stocks can also be more pronounced for certain generations than others. Further investigation allowing for differences in risk appetite will be desirable. 6.4. Share issuance as an alternative test It has been shrewdly pointed out that our hypothesis about the preference for dividends over capital gains for consumption purposes by the old population can be alternatively tested by examining the long-run excess returns of companies engaging in stock purchases. Pontiff and Woodgate (2008), for instance, conducted Fama–Macbeth cross-sectional regressions and found that between the period 1970–2003, companies that repurchased stock also enjoyed a return premium compared to other companies. Jain (2007) went further by investigating US investor preference for dividends versus share repurchases over the period of 1989–1996 and found that despite the associated tax penalty, individual investors generally prefer cash dividend payouts to share repurchases. Further research into the strength of these relationships to demographic changes will be very interesting. We believe that there are at least three areas that can be explored. The first is to investigate the returns of a long-short strategy of stocks doing repurchases (versus companies issuing shares) against changes in the proportion of old population. The second is to calculate the relative returns of the long-short share-repurchase strategy versus those of a long-short dividend-yield strategy, and investigate how the relative returns compares against demographic changes. The third is to explore the returns of a long-short strategy comprising of a long portfolio of companies engaging in stock repurchases versus a short portfolio of high dividend-yielding stocks against demographic variation.

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

6.5. Distribution of dividend-yielding stocks Our paper has thus far focused on the extreme quintiles of the dividend yield spectrum. It will also be interesting to see if our hypothesised relation holds true across the distribution of stocks. We investigate this in two ways. In our first approach, we calculate the ten-year returns of a dividend-yield investment strategy RQ2–Q4 calculated using Q2 and Q4, and then we investigate its relation with the demographic variation variable. In our second approach, we estimate another variable Rank IC which measures the monthly Pearson correlation coefficient between the average dividend yields of the quintiles and their respective long-run returns. This is essentially a measure of the strength of the dividend yield effect throughout the whole distribution of stocks. We then regress the Rank IC variable against the demographic variation variable. Our results are shown in Columns 46–47 of Table 7. It can be seen that the demographic variation variable is an important positive determinant of RQ2–Q4 at 5% significance level. This appears to suggest that our hypothesised relationship holds true across the distribution of stocks. The demographic variation variable is however not an important determinant of Rank IC at 10% significance levels. We however note that p-value of the regression coefficient is 0.124. 7. Conclusion According to the behavioural life-cycle theory, households treat components of their wealth as nonfungible. As such, in the later stage of a household’s life cycle when they reach retirement and begin to dis-save, the investor perception of the non-fungibility between dividends and capital gains should lead to a preference for high dividend-paying stocks by older investors for consumption purposes. We hypothesise that this stronger preference for high dividendpaying stocks by older investors should mean that the excess returns of high dividend-yielding stocks, relative to other stocks, are positively related to changes in the proportion of the older population. In particular, we find empirical evidence that, as consistent with the behavioural life-cycle hypothesis, the long-run returns of dividend-yield investment strategies are positively driven by demographic clientele variation that is represented as changes in the proportion of the older population. Our results are robust when controlled for the Fama–French factors, inflation rate, consumption growth rate, interest rates, tax clienteles, time trend and alternative definitions of both dividend-yield strategies as well as demographic variation. Acknowledgements The author gratefully acknowledges the very insightful comments and helpful suggestions from Massimo Guidolin (the editor) and the two anonymous reviewers. Also, special thanks to Jade Foo who has painstakingly copy-edited this paper. All errors and omissions are solely the author’s responsibility. References Allen, F., Bernardo, A., & Welch, I. (2000). A theory of dividends based on tax clienteles. Journal of Finance, 55, 2499–2536. Arnott, R. (2003). Dividends and the three dwarfs. Financial Analysts Journal, 59(2), 4–6. Arnott, R., & Casscells, A. (2003). Demographics and capital market returns. Financial Analysts Journal, 59(2), 20–29. Arnott, R. D., & Chaves, D. B. (2012). Demographic changes, financial markets, and the economy. Financial Analyst Journal, 68(1), 23–46. Asness, C. (1997). The interaction of value and momentum strtategies. Financials Analysts Journal, 53(2), 29–36.

217

Baker, M., Nagel S., & Wurgler, J. (2006). The effect of dividends on consumption. National Bureau of Economic Research Working Paper 12288. Bakshi, G., & Chen, Z. (1994). Baby boom, population aging, and capital markets. Journal of Business, 67(2), 165–202. Basse, T. (2009). Dividend policy and inflation in Australia: Results from cointegration tests. International Journal of Business and Management, 4(6), 13–16. Basse, T., & Reddemann, S. (2011). Inflation and the dividend policy of US firms. Managerial Finance, 37(1), 34–46. Bauer, R., Koedijk, K., & Otten, R. (2005). International evidence on ethical mutual fund performance and investment style. Journal of Banking and Finance, 29(7), 1751–1767. Bernanke, B., & Gertler, M. (1989). Agency costs, net worth, and business fluctuations. American Economic Review, 79, 14–31. Boudoukh, J., & Richardson, M. (1993). Stock returns and inflation: A long-horizon perspective. American Economic Review, 83, 1346–1355. Brennan, M. (1970). Taxes, market valuation & corporate financial policy. National Tax Journal, 23, 417–427. Campbell, J., & Shiller, R. (1998). Valuation ratios and the long-run stock market outlook: An update. National Bureau of Economic Research Working Paper 8221. Carrel, L. (2010). Dividend stocks for dummies. New York, NY: John Wiley & Sons. Christie, W. (1990). Dividend yield and expected returns: The zero-dividend puzzle. Journal of Financial Economics, 28, 95–126. Claude, B., Campbell, R., & Viskanta, T. (1997). Demographics and international investments. Financial Analysts Journal, 1997(July/August), 14–28. Cornell, B. (2012). Dividend-price ratios and stock returns: Another look at the history. Available from SSRN: http://ssrn.com/abstract=2071014 or http://dx.doi.org/10.2139/ssrn.2071014 Dahlquist, M., Robertsson, G., & Rydqvist, K. (2006). Direct evidence of dividend tax clienteles. Swedish Institute for Financial Research Working Paper. Davis, E. P., & Li, C. (2003). Demographics and financial asset prices in the major industrial economies. Brunel University Working Paper March 2003. Eskesen, L., Lueth, E., & Syed, M. (2008). The graying of Asia: Demographics, capital flows, and financial markets. IMF Regional Economic Outlook: Asia and Pacific, 2008(November), 41–57. Fama, E. (1998). Market efficiency, long-term returns and behavioral finance. Journal of Financial Economics, 49, 283–306. Fama, E., & French, K. (1988). Dividend yields and expected stock returns. Journal of Financial Economics, 22, 3–25. Fama, E., & French, K. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics, 33(1), 3–56. Fama, E., & French, K. (1992). The cross-section of expected stock returns. Journal of Finance, 4, 427–466. Fama, E., & French, K. (1998). Value versus growth: The international evidence. Journal of Finance, 53(6), 1975–1999. Fama, E., & Schwert, G. (1977). Asset returns and inflation. Journal of Financial Economics, 5, 115–146. Gertler, M., & Gilchrist, S. (1994). Monetary policy, business cycles, and the behaviour of small manufacturing firms. Quarterly Journal of Economics, 109, 309–340. Geske, R., & Roll, R. (1983). The fiscal and monetary linkage between stock returns and inflation. Journal of Finance, 38(1), 1–33. Gombola, M., & Liu, F. (1993a). Considering dividend stability in the relation between dividend yields and stock returns. Journal of Financial Research, 16(2), 139–150. Gombola, M., & Liu, F. (1993b). Dividend yields and stock returns: Evidence of time variation between bull and bear markets. Financial Review, 28(3), 303–327. Graham, J., & Kumar, A. (2006). Do dividend clienteles exist? Evidence on dividend preferences of retail investors. Journal of Finance, 61, 1305–1336. Grant, J. (1995). A yield effect in common stock returns. Journal of Portfolio Management, 21(2), 35–40. Gwilym, O., Clare, A., Seaton, J., & Thomas, S. (2009). Dividends and momentum. Journal of Investing, 18, 42–49. Harlow, V., & Brown, K. (1990). The role of risk tolerance in the asset allocation process: A new perspective. Charlottesville, VA: The Research Foundation of the Institute of Chartered Financial Analysts. Hodrick, R. (1992). Dividend yields and expected stock returns: Alternative procedures for inference and measurement. Review of Financial Studies, 5(3), 357–386. Jain, R. (2007). Institutional and individual investor preferences for dividends and share repurchases. Journal of Economics and Business, 59, 406–429. Jamal, A. M. M., & Quayes, S. (2004). Demographic structure and stock prices. Economic Letters, 84, 211–215. Keim, D. (1985). Dividend yields and stock returns: Implications of abnormal January returns. Journal of Financial Economics, 14(3), 473–489. Kolari, J., & Anari, A. (2001). Stock prices and inflation. Journal of Financial Research, 24, 587–602. Kothari, S., & Shanken, J. (1997). Book-to-market, dividend yield, and expected market returns: A time-series analysis. Journal of Financial Economics, 44(2), 169–203. Kwiatkowski, D., Phillips, P. C. B., Schmidt, P., & Shin, Y. (1992). Testing the null of stationarity against the alternative of a unit root. Journal of Econometrics, 54, 159–178. Lee, K. F. (2011). Demographics, dividend clienteles and the dividend premium. Quarterly Review of Economics and Finance, 51(4), 368–375. Liew, J., & Vassalou, M. (2000). Can book-to-market, size and momentum be risk factors that predict economic growth? Journal of Financial Economics, 57, 221–245. Malmendier, U., & Nagel, S. (2011). Depression babies: Do macroeconomic experiences affect risk-taking? Quarterly Journal of Economics, 126(1), 373–416.

218

K.F. Lee / The Quarterly Review of Economics and Finance 53 (2013) 202–218

Malmendier, U., & Nagel, S. (2011b). Learning from inflation experiences. Working Paper. Available from: http://emlab.berkeley.edu/∼ulrike/Papers/InflExp 4.pdf Miller, M., & Modigliani, F. (1961). Dividend policy, growth and the valuation of shares. Journal of Business, 4, 411–433. Mladjenovic, P. (2009). Stock investing for dummies. New York, NY: John Wiley & Sons. Modigliani, F., & Brumberg, R. (1954). Utility analysis and the consumption function: An interpretation of cross-section data. In K. K. Kurihara (Ed.), Post-Keynesian economics (pp. 388–436). New Brunswick, NJ: Rutgers University Press. Modigliani, F., & Brumberg, R. (1980). Utility analysis and aggregate consumption functions: An attempt at integration. In A. Abel (Ed.), The collected papers of Franco Modigliani: Volume 2. The life cycle hypothesis of saving (pp. 128–197). Cambridge, MA: The MIT Press. Morgan, G., & Thomas, S. H. (1998). Taxes, dividend yields and returns in the UK equity market. Journal of Banking and Finance, 22, 405–423. Naranjo, A., Nimalendran, M., & Ryngaert, M. (1998). Stock returns, dividends yields and taxes. Journal of Finance, 53(6), 2029–2057. Newey, W., & West, K. (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55, 703–708. Newey, W., & West, K. (1994). Automatic lag selection in covariance matrix estimation. Review of Economic Studies, 61(4), 631–654. Perez-Quiros, G., & Timmermann, A. (2000). Firm size and cyclical variations in stock returns. Journal of Finance, 55(3), 1229–1262. Pontiff, J., & Woodgate, A. (2008). Share issuance and cross-sectional returns. Journal of Finance, 63(2), 921–945.

Poterba, J. M. (2001). Demographic structure and asset returns. Review of Economics and Statistics, 83(4), 565–584. Poterba, J. M. (2004). The impact of population aging on financial markets. National Bureau of Economic Research Working Paper 10851. Rapach, D., & Wohar, M. (2005). Valuation ratios and long-horizon stock price predictability. Journal of Applied Econometrics, 20(3), 327–344. Riley, W., & Chow, V. (1992). Asset allocation and individual risk aversion. Financial Analysts Journal, 48(6), 32–37. Shiller, J. R. (2000). Irrational exuberance. Princeton, NJ: Princeton University Press. Smith, R. G. E. (1967). The ‘marginal opinion’ theory of stock price. Financial Analysts Journal, 23, 127–132. Sterling, W., & Waite, S. (1998). Boomernomics: The future of your money in the upcoming generational warfare. New York, NY: Ballantine Publishing Group. Thaler, R. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior and Organization, 1, 39–60. Thaler, R., & Shefrin, H. (1988). The behavioral life-cycle hypothesis. Economic Inquiry, 26, 609–643. Visscher, S., & Filbeck, G. (2003). Dividend-yield strategies in the Canadian stock market. Financial Analyst Journal, 59(1), 99–106. Williams, J. (1938). The theory of investment value. Cambridge, MA: Harvard University Press. Yoo, P. S. (1994). Age distribution and returns of financial assets. Federal Reserve Bank of St. Louis Working Paper 94-002B.