In the short run, the stock market is driven mostly by demand, with

Transcription

1 Research Financial Analysts Journal A Publication of CFA Institute The Long-Run Drivers of Stock Returns: Total Payouts and the Real Economy Philip U. Straehl and Roger G. Ibbotson Philip U. Straehl is head of capital markets and asset allocation at Morningstar Investment Management LLC, Chicago. Roger G. Ibbotson is Professor in the Practice Emeritus of Finance at the Yale School of Management, New Haven, Connecticut, and chairman and chief investment officer at Zebra Capital Management LLC, Milford, Connecticut. We provide theoretical and empirical evidence over that total payouts (dividends plus buybacks) are the key drivers of long-run stock market returns. We show that total payouts per share (adjusted for the share decrease from buybacks) grew in line with economic productivity, whereas aggregate total payouts grew in line with GDP. We also show that a dividend discount model (DDM) based on current yields and historical growth rates underestimates expected returns relative to the total payout model. Finally, we demonstrate that the cyclically adjusted total yield (CATY) predicts changes in expected returns at least as well as the cyclically adjusted price-to-earnings ratio (CAPE). Disclosure: Morningstar Investment Management LLC uses a version of the methodology outlined in this article to manage multi-asset portfolios. CE Credits: 1 In the short run, the stock market is driven mostly by demand, with risk-on/risk-off environments or dynamic discount rates causing overand underreactions to underlying fundamentals. But in the long run, as we show in this article, the cash flows that corporations supply are the ultimate drivers of stock returns. We further show that these corporate cash flows participate in the underlying growth of the real economy. Ibbotson and Chen (2003) decomposed historical returns on the basis of such corporate fundamentals as dividends, earnings, and book value, finding that the majority of historical returns can be attributed to the supply of these components. In recent decades, a new source of stock market supply has emerged as companies increasingly use share buybacks instead of dividends to return cash to shareholders. Skinner (2008) found that in the United States, stock market buybacks are the dominant source of payouts, whereas Boudoukh, Michaely, Richardson, and Roberts (2007) provided evidence that the payout yield, which includes both dividends and buybacks, is more predictive of changes in expected returns than is the dividend yield. In addition, von Eije and Megginson (2008) documented a significant rise in buybacks by companies in the European Union, suggesting that buybacks are an increasingly important global phenomenon. Although a growing body of literature discusses the importance of buybacks as a form of payout, the impact of buybacks on stock returns has been largely ignored in practice because many practitioners continue to rely on traditional supply models that use dividends as the sole source of corporate payout. In these traditional models, payouts via buybacks lead to structural increases in per share growth rates (e.g., earnings per share), because the number of shares is decreased by buybacks even as per share growth exceeds the underlying corporate We thank Robert Arnott, William Bernstein, Andrew Carter, Peng Chen, CFA, Elroy Dimson, Thomas Idzorek, CFA, Antti Ilmanen, Paul Kaplan, CFA, and Deniz Yavuz for helpful comments and suggestions CFA Institute. All rights reserved. Third Quarter 2017

2 The Long-Run Drivers of Stock Returns cash flow growth. This structural change in per share growth not only complicates the attribution of fundamental supply components to payout and fundamental growth but also leads to potentially biased return forecasts when current and historical market data are combined. In this article, we develop models of stock returns on the basis of total payouts (i.e., dividends and buybacks), which provide not only a more consistent framework for examining the historical sources of returns, independent of changes in payout policy, but also a more robust returnforecasting model that can be related to the growth in the real economy. The total payout models that we propose in this article have a foundation in corporate finance theory. According to the theorem of Miller and Modigliani (1961), investors should be indifferent about whether they receive distributions via dividends or buybacks as well as how they participate in a buyback that is, by receiving cash from tendering their shares or by receiving an increased proportion in the company. 1 This logic also applies to share issuance, whereby investors receive the same return whether they add a proportional amount of cash in the issuing company (while maintaining the same percentage ownership) or give up fractional ownership but benefit from the future payoff of the asset acquired with the issued shares. The total payout models that we derived in our study mimic the behavior of three hypothetical buyback/issuance investor types. The buy-andhold investor holds on to her shares in a buyback and receives a bigger fractional ownership in the company. The pro rata buyback investor tenders a proportional number of his shares and receives cash in return for participating in the buyback. The cap-weighted index investor participates proportionally in both the buyback and the issuance of shares. In this article, we demonstrate that the three investor types end up with the same overall return, even though the ways the return is generated differ. Importantly, from a return-forecasting perspective, the cash flows received by both the pro rata buyback investor and the cap-weighted index investor are the same, irrespective of the payout method (dividend versus buyback). The classification of buyback/issuance investor types is a key contribution of our study, which extends the literature in several directions: 1. We develop three total payout models of stock returns and show that US stock returns between 1871 and 2014 can be attributed almost entirely to the supply of both dividends and buybacks. 2. We relate the growth in total payouts to the real economy and show that total payouts per share (adjusted for the share decrease from buybacks) grow in line with economic productivity and that aggregate total payouts grow in line with aggregate GDP, suggesting that total payouts participate in the growth of the real economy. 3. We show that the dividend discount model (DDM) significantly underestimates the forward equity return when current market information (e.g., yields) is combined with historical data (e.g., historical growth). The dividend and cash buyback model and the dividend less net issuance model, which we derived in our study, apply universally across time periods owing to their independence from payout regimes. 4. Finally, we demonstrate that the cyclically adjusted total yield (CATY) is at least as predictive of changes in expected returns as the cyclically adjusted price-to-earnings ratio (CAPE). Overall, the total payout model represents a viable alternative to such traditional supply models of stock returns as the DDM, providing a framework for deriving macroconsistent forecasts of long-run stock returns as well as superior forecasts of short-term expected returns. Data We obtained data on monthly prices, earnings per share, dividends per share, and inflation for January 1871 December 2014 from Robert Shiller. 2 Consistent with Ibbotson and Chen (2003), we defined the risk-free rate as the income return of long-term US government bonds. For 1926 and beyond, we used income return data from the Ibbotson SBBI Classic Yearbook; for , we inferred returns from 10-year US Treasury bond yields (also from Shiller). At the beginning of each monthly holding period, we assumed a 10-year maturity, a bond price equal to par, and a coupon equal to one-twelfth of the beginning-of-period yield. Following Ibbotson and Chen (2003), we obtained the starting book-to-market value for 1871 from Vuolteenaho (1999). We obtained the book-to-market value for the S&P 500 Index at the end of 2014 from Morningstar and GDP and population data from the Maddison Project Database 3 (for ) and the US Bureau of Economic Analysis (for ). Ibbotson and Chen covered ; we provide a Volume 73 Number 3 cfapubs.org 33

3 Financial Analysts Journal A Publication of CFA Institute full update and backdate of the Ibbotson Chen return decompositions in Appendix A, covering For , motivated by Boudoukh et al. (2007), we calculated net issuance for the S&P Composite Index 4 based on constituent data from CRSP as the stock-level monthly change in shares outstanding times the share price summed across companies every month, divided by the monthly market capitalization. 5 Our net issuance estimate is specific to the S&P Composite Index, which has a fixed number of constituents, and is different from net issuance estimated for the aggregate market. Although aggregate market net issuance includes net issuance owing to company entry (e.g., IPOs) and company exit (e.g., cash buyouts, cash mergers), net issuance specific to an index with a fixed number of constituents captures company-level net issuance (e.g., secondary offerings and buybacks). Sabbatucci (2015) showed that merger-and-acquisition (M&A) cash payouts are an important part of aggregate shareholder cash flows. Given our focus on the S&P Composite Index, we disregarded these aggregate cash flows in our study sample because if an index constituent is bought out, it is replaced with a similar stock, creating offsetting cash inflows and outflows. For the pre-1925 period, we estimated net issuance for the S&P Composite Index on the basis of aggregate market value and net issuance data assembled by Wright (2004). In particular, we backcast the market values over on the basis of the 1925 market value for the S&P Composite Index (from CRSP) and the relationships between price return, market value, and net issuance. We followed Wright (2004) in relating the share price (P) to the market value (MV) and net issuance (NI): MVt NIt =. MVt Solving for MV t 1 /MV t, we obtained an expression for the inverse of the change in market value, which we used to backcast market values for : MVt MVt NIt =. MVt MVt We then calculated aggregate net issuance for on the basis of the geometric difference between the change in market value and the change in the price index. 6 Using data from Compustat, we calculated the buyback yield of the S&P Composite Index for Consistent with Fama and French (2001), we estimated buybacks on the basis of the increase in a company s treasury stock, excluding repurchased shares earmarked for compensation, such as employee stock option programs, and payments-in-kind, such as acquisitions. 8 We assumed that buybacks were zero before We interpolated linearly between annual buyback data points to obtain a monthly buyback time series. The Rise of Buybacks A primary objective of our study was to shed light on the impact of buybacks on the return generation process. We started by documenting the rise of buybacks as a form of corporate payout relative to dividends. The drastic change in payout policy over the last few decades is highlighted in Figure 1, which plots dividend and buyback yields since Figure 1 shows that the gradual substitution of buybacks for dividends started in the early 1980s, with buybacks ultimately exceeding dividends over the more recent history. Buybacks surpassed dividends in 8 of the last 10 calendar years, supporting the claim that buybacks have become an important way for US companies to return cash to shareholders. Grullon and Michaely (2002) found evidence that the structural change in corporate payout policy, among other factors, coincided with the adoption of SEC Rule 10b-18 in 1982, which provided a safe harbor for companies to conduct share buybacks without being suspected of share price manipulation. Figure 2 depicts the total yield of US stocks from 1871 to The total yield series, which includes both dividends and buybacks, appears to be consistent over time, generally reverting around its longterm mean. Boudoukh et al. (2007) examined the time-series properties of dividend yields and total yields and found that total yields followed a stationary time-series process whereas dividend yields experienced a structural break in the early 1980s. Total payouts are thus a more stable measure of corporate payouts than are dividends. Buybacks have a fundamentally different impact on the return generation process than dividends do. Although payouts via dividends increase the income return, buybacks increase the price return (per share) because a buy-and-hold investor s share in a company is increased. Prior studies, including Ibbotson and 34 cfapubs.org Third Quarter 2017

4 The Long-Run Drivers of Stock Returns Figure 1. Dividend Yield and Buyback Yield, Yield (%) Dividend Yield Buyback Yield Figure 2. Total Yield, Yield (%) Total Yield Note: Total yield is the dividend yield plus the buyback yield depicted in Figure 1. Volume 73 Number 3 cfapubs.org 35

5 Financial Analysts Journal A Publication of CFA Institute Chen (2003; see Appendix A for updated Ibbotson Chen return decompositions), disregarded the fact that return components are sensitive to a company s payout method (i.e., dividends versus buybacks). Such traditional supply models as the dividend discount model (DDM), however, are not inaccurate per se. Figure A1 (in Appendix A) shows that traditional return models fully account for historical total return levels. Nevertheless, traditional return models are often wrongly applied in practice, given investors tendency to combine current dividend yields with historical per share growth rates when forecasting returns. Historical per share growth rates, measured over a time frame that includes the period before buybacks were prevalent, underestimate forward-looking per share growth by underestimating the impact of buybacks on ex ante growth rates. In other words, by relying on historical per share growth rates as a proxy for future growth, investors underestimate the fact that while any shareholder benefits from the underlying growth in corporate cash flows, buybacks increase a buy-andhold investor s proportion in the company over time. The advent of share buybacks as the dominant form of payout has created a need for a new set of return models that are independent of the payout method. In the remainder of this article, we describe our development of total payout models of stock returns that provide a more consistent framework for examining the sources of stock returns. Total Payout Models of Stock Returns In this section, we derive three models of stock returns on the basis of total payouts (dividends plus buybacks) that are distinguished by how they are affected by share buybacks and share issuance. The dividend and cash buyback model and the dividend less net issuance model are independent of a company s choice of payout method, providing a more consistent way to analyze historical returns. We relate each model to a hypothetical investor type differentiated by how each investor type participates in share buybacks or share issuance. The one-period total return R of a stock over period t 1 to t is given by Rt Dt D P = + 1 = t + t 1, (1) where D is the dividend per share and P is the share price. The first right-hand term is the income return, and the second term is the price return. Total payouts refers to the payouts that investors receive from both dividends and buybacks. The advantage of measuring corporate performance on the basis of payouts instead of such accounting measures as earnings is that payouts are unaffected by changes in accounting standards or by such transitory factors as special items, providing a better measure of structural drivers of the supply of stock returns. We can write the price return in Equation 1 as a function of the change in the price-to-totalpayout ratio and the change in the total-payoutper-share ratio (TP), where the latter is the sum of dividend per share and buyback per share: Rt Dt T TP = + t T 1. (2) T We can simplify Equation 2 and write geometric average returns as follows: 10 R = DY + gtp + gp/ TP + CPI+ Interaction, (3) where DY is the dividend yield, g TP is the growth in real total payout per share, g P/ TP is the change in price to total payout, CPI is inflation, and Interaction captures the reinvestment return and geometric interaction among the components. Appendix B provides a derivation of moving from arithmetic single-period returns to multi-period geometric returns on the basis of the log return, eliminating the need for an interaction term. Note that buybacks do not affect the dividend yield but do increase growth in total payout per share ( g TP ), because the investor s proportion in the company increases. The buyback impact captured by this model is akin to that of a hypothetical buy-andhold investor who holds on to her shares as other investors tender theirs, thereby receiving a larger fractional ownership in the company because corporate cash is used to decrease share count. Similar to the traditional supply models discussed earlier, the choice of payout method (i.e., dividends versus buybacks) matters in this model because dividends are captured as a cash payout whereas buybacks affect the growth term. Although this 36 cfapubs.org Third Quarter 2017

6 The Long-Run Drivers of Stock Returns model is not independent of the payout method, we include it as a reference point for the other models. We call this model the dividend per share model. To obtain a measure that is independent of a company s payout method, we need to normalize Equation 2 to account for the effect of buybacks on payouts and growth in total payout per share. We use both buyback yield and net issuance to adjust the model for the impact of buybacks. Buyback yield is defined as PB 1 + Buyback yield t = 1 + t t B = PS t t t + S S t t, (4) where S is the number of shares outstanding, B is the number of shares repurchased, and P is the share price. Conversely, net issuance is defined as P 1 + Net issuance = 1 + t( St St ) t = PS t. (5) t t St Note that Equation 5 gives the change in share count. It implicitly consists of the shares issued less share buybacks. Equations 4 and 5 allow us to derive two different return models on the basis of total payouts. We can rewrite Equation 2 as a function of Equation 4, obtaining Rt Dt Bt + St T St TP = + t St T Bt + St 1. (6) T We can simplify Equation 6 and write geometric average returns as follows: R = TY + gtpexb + gp/ TP + CPI+ Interaction, (7) where the total yield (TY) is the dividend yield plus the buyback yield, g TPexB is the growth in real total payout per share adjusted for the share decrease from buybacks, 11 g P/ TP is the change in price to total payout, CPI is inflation, and Interaction captures the reinvestment return and geometric interaction among the components. Note that both TY and g TPexB are independent of the payout method, because both dividends and buybacks are captured in the total yield term and the growth term is adjusted for the share decrease from buybacks. The buyback impact S captured by this model is analogous to that of a hypothetical pro rata buyback investor who tenders a proportional number of his shares and receives cash in return for participating in the buyback. Whether a company performs a payout via dividends or buybacks, this investor receives cash. We call this model the dividend and cash buyback model. Conversely, we can rewrite Equation 2 as a function of Equation 5, obtaining Rt Dt St T St TP = + t St T St 1. (8) T Equation 8 can be simplified to the following geometric average return components: R = NTY + gtpagg + gp/ TP + CPI + Interaction, (9) where net total yield (NTY) is the dividend yield plus the inverse of net issuance (arithmetically, dividend yield less net issuance), g TPAgg is the real aggregate total payout growth, g P/ TP is the change in price to total payout, CPI is inflation, and Interaction captures the reinvestment return and geometric interaction among the components. NTY implicitly consists of dividends and buybacks less shares issued. Equation 9 is similar to the stock return forecasting model proposed by Grinold, Kroner, and Siegel (2011). The return impact captured by this model can be related to that of a hypothetical cap-weighted index investor who participates proportionally in both the buyback and the issuance of shares. Similar to the previous case, independent of whether a company performs the payout via dividends or buybacks, this hypothetical investor receives cash. Moreover, this investor adds new capital to an issuing company and thus benefits from the aggregate growth of the company. We call this model the dividend less net issuance model. Together, the dividend per share model (Equation 3), the dividend and cash buyback model (Equation 7), and the dividend less net issuance model (Equation 9) provide the basis for the analysis of the historical sources of returns and stock return forecasts discussed in the remainder of this article. In the dividend per share model, buybacks increase the growth term, consistent with the traditional supply models discussed earlier. In both the dividend and cash buyback model and the dividend less net Volume 73 Number 3 cfapubs.org 37

7 Financial Analysts Journal A Publication of CFA Institute issuance model, buybacks affect the payout term and are thus independent of the payout method. The key difference between the dividend and cash buyback model and the dividend less net issuance model is the treatment of issuance. In the dividend and cash buyback model, issuance is captured by the growth term ( g TPexB ); all changes in share count (buybacks and issuance) are captured by the net total yield term (NTY) in the dividend less net issuance model. In the dividend and cash buyback model, buybacks are accounted for in the payout term (TY). Historical Return Decompositions In this section, we examine the return decompositions of the three supply models based on total payouts. Both the dividend and cash buyback model and the dividend less net issuance model allow for an examination of the return drivers independent of the payout method, providing a consistent framework for studying the historical supply components of stock returns over time. The results of the historical return decompositions for three periods ( , , and ) are reported in Table The central insight from Table 1 that applies to all three models is that the long-run supply of total payouts almost entirely explains realized returns over and The change in price to total payout which is the return component unrelated to the supply of total payouts common to all three models accounts for only 0.20% of realized returns over , suggesting that more than 97% of realized returns are related to the supply of total payouts. The total payout model thus provides a better description of long-term historical returns than does the earnings model in Figure A1, which shows an annualized growth in P/E ( g P/ E ) of 0.41%. Overall, this finding suggests that the realized level of returns is almost entirely attributable to the supply of total payouts. For the shorter period, the change in price to total payout is 0.53%. The results in Table 1 for the dividend per share model and the dividend and cash buyback model show that Table 1. Historical Return Decompositions Dividend per share model (buy-and-hold investor) Dividend yield DY 4.50% 4.29% 3.03% Total payout growth g TP Change in price to total payout g P/TP Inflation CPI Interaction Dividend and cash buyback model (pro rata buyback investor) Total yield TY 4.89% 4.78% 4.26% Total payout growth (adjusted for share decrease) g TPexB Change in price to total payout g P/TP Inflation CPI Interaction Dividend less net issuance model (cap-weighted index investor) Net total yield NTY 3.03% 1.80% Aggregate total payout growth g TPAgg Change in price to total payout g P/TP Inflation CPI Interaction Total return 9.05% 9.64% 10.41% 38 cfapubs.org Third Quarter 2017

8 The Long-Run Drivers of Stock Returns the payout terms (i.e., dividend yield and total yield) account for the majority of historical returns. Total yield, which includes both dividends and buybacks, accounts for more than two-thirds of the real return for Despite the fact that share buybacks were prevalent only over the last three decades of our sample, explicitly accounting for buybacks as a payout increases the return attributed to payouts from 4.50% (dividend yield) to 4.89% (total yield). Notably, for , including share buybacks along with dividends in the payout terms dramatically raises the payout yield from 3.03% (dividend yield) to 4.26% (total yield), highlighting the important role buybacks have played in returning cash to shareholders in recent decades. In contrast to the dividend per share model and the dividend and cash buyback model, the net total yield payout term (NTY) in the dividend less net issuance model is net of share issuance, thus lowering the payout term. Over , the return attributable to NTY is 3.03%. NTY can be further decomposed into a dividend yield of 4.29%, a buyback yield of 0.47%, and a negative contribution from issuance of 1.70%. Over , NTY is only 1.80%, driven by a high issuance yield of 2.41%; dividends and buybacks are 3.03% and 1.20%, respectively. Figure 3 plots the issuance and buyback yields since 1901, highlighting that saw an increase in both buybacks and issuance, with the issuance yield peaking in the late 1990s/early 2000s and buybacks exceeding issuance in 7 of the last 10 calendar years (ending in 2014). The growth term is another key differentiator among the three models. In general, the per share impact of buybacks and issuance on the growth term is opposite to the impact on the payout term, because the three models mechanically add up to the same total return, g TPexB, which excludes the impact of buybacks on the share decrease but includes issuance; this growth term ( g TPexB ) amounted to 1.67% over Aggregate growth ( g TPAgg ) excludes buybacks and issuance, contributing 3.27% to the return over In contrast, growth in total payout per share ( g TP ), which amounted to 2.05% over , is net of issuance but includes the impact of buybacks on the decrease in share count. Because g TP includes buybacks, the structural shift from dividends to buybacks, which started in the 1980s, also led to a structural change in total payout per share growth. Therefore, g TPexB and g TPAgg allow for a more consistent analysis of historical growth trends over time, which is a key contribution of the dividend and cash buyback model and the dividend less net issuance model that we introduce in this article. Figure 4 depicts the impact of adjusting the per share growth rate by the share decrease from buybacks. Prior researchers have measured historical growth rates on the basis of per share statistics (e.g., earnings per share) without adjusting for the impact of buybacks on share count (see, e.g., Figure A1). Figure 4 plots the growth in real total payout per share ( g TP ) and the growth in real total payout per share adjusted for the share decrease from buybacks ( g TPexB ). The annualized growth rate of the former was 4.70%, whereas the growth rate of the latter was only 3.16%. Figure 3. Issuance Yield and Buyback Yield, Yield (%) Issuance Yield 1 0 Buyback Yield Note: Issuance yield is net issuance plus buybacks divided by market capitalization. Volume 73 Number 3 cfapubs.org 39

9 Financial Analysts Journal A Publication of CFA Institute Figure 4. Buybacks Increase per Share Growth, Year-End 1979 = g TP 2 g TPexB The difference is attributable to the buyback yield over the period, suggesting that a significant portion (32% over ) of growth measured by per share statistics is due to buybacks, not to growth of the underlying cash flows of the businesses. This example suggests that the adjustment of per share growth rates for the share decrease from buybacks has important implications for the measurement of fundamental corporate cash flow growth. The return decompositions in Table 1 highlight the long-term drivers of stock returns, but long-run averages mask the variations in the return components over shorter periods. To illustrate how the different return components vary over shorter-term market cycles, Figure 5 depicts the return decomposition based on the dividend and cash buyback model for rolling 10-year real returns. Total yield (TY) is the most stable component of real returns, generally fluctuating between approximately 4% and 6%. The growth term ( g TPexB ) is significantly more volatile than total yield, with the 10-year growth rate ranging between 8.96% and 6.56% over the sample period. The volatility in growth rates can be attributed to the fact that corporate payouts are sensitive to the business cycle, with companies paying out more of their earnings during good economic times, when profits are high, and cutting back payouts during recessions. Note that although g TPexB excludes the impact of buybacks on share count, buybacks are still a component of aggregate total payout growth. For instance, Skinner (2008) found that buybacks are more sensitive than dividends to variations in a company s earnings, because buybacks offer greater flexibility than dividends in making short-term adjustments, suggesting that buybacks have been a key driver of the short-term volatility in the growth term in recent history. Finally, growth in the price-tototal-payout ratio ( g P/ TP ) varies most significantly in the short run. Although the contribution of g P/ TP to long-run returns in Table 1 is only 0.20%, g P/ TP varies materially in the short term. Classical financial theory assumes a constant equity premium, but a growing body of research suggests that expected returns vary over time in ways that are predictable (see, e.g., Cochrane 2011). Time-varying expected returns are a potential explanation for the observed variation in changes in valuation in the short term. A variance decomposition of 10-year real log returns, reported in Table 2, confirms that the change in valuation g P/ TP is the most variable return component in the short run. Changes in valuation explain more than half the variance in real returns, whereas changes in real growth rates explain almost a third of the variance. Conversely, only 3.8% of the variance is explained by changes in total yield. Therefore, although total yield and the growth in total payouts 40 cfapubs.org Third Quarter 2017

10 The Long-Run Drivers of Stock Returns Figure 5. Decompositions of Rolling 10-Year Real Returns, Return (%) 9 Total Yield g TPexB g P/TP Notes: Total yield is the dividend yield plus the buyback yield, g TPexB is the growth in real total payout per share adjusted for the share decrease from buybacks, and g P/ TP is the change in price to total payout. The three components sum to the real total return over each 10-year period. explain the majority of the return level in Table 1, the analysis of 10-year returns suggests that changes in valuation account for a significant portion of the variance of returns in the short run. The historical return decompositions analyzed in this section, based on the total payout models that we Table 2. Variance Decompositions of Rolling 10-Year Real Log Returns, Dividend and Cash Buyback Model Contribution to Variance Total yield TY 3.80% Total payout growth (adjusted for share decrease) g TPexB Change in price to total payout g P/TP Interaction 7.56 developed, show that (1) long-term historical stock returns can be almost fully attributed to the supply of total payouts and (2) total payouts have been more stable than dividends over time. Total Payouts and the Real Economy Following prior research on long-term drivers of stock returns, we now examine the relationship between growth in total payouts and growth in the real economy. Unlike in previous studies, however, the growth rates in our models are adjusted for changes in payout policy, allowing for an applesto-apples comparison of growth trends over time. Diermeier, Ibbotson, and Siegel (1984) related the return of aggregate financial assets to the performance of the real economy and stressed the importance of capital market forecasts being macroconsistent : Long-run growth expectations for financial assets need to be anchored in reasonable growth expectations for the overall economy. After all, financial assets cannot outperform the economy Volume 73 Number 3 cfapubs.org 41

11 Financial Analysts Journal A Publication of CFA Institute indefinitely because the assets would ultimately become the economy itself! Applying the supply model to stocks, Ibbotson and Chen (2003) showed that growth in earnings per share for US stocks is in line with growth in US GDP per capita. To think about growth in both the real economy and stock market payouts in a consistent manner, we start with a simple model for the aggregate economy and then derive growth rates that are consistent with our definition of growth in our equity forecasting model. Rather than derive exact macroeconomic equivalents of the stock market and the real economy, our goal is to relate the growth rates in our total payout models to widely used measures of the performance of the real economy, such as GDP. For more elaborate production-based asset pricing models, see, for example, Cochrane (1991). We assume that the aggregate output (Y) of the economy follows a standard Cobb Douglas production function: 1 ββ Y = AK L, (10) where A is the total factor productivity (the amount not explained by labor and capital and often attributed to improvements in technology), K is the capital stock, L is the labor hours worked, and β is the output elasticity, which is assumed to be <1 and constant over time. Taking logs (denoted in small letters) and the first difference (i.e., the difference between t + 1 and t) in Equation 10, we obtain an expression for the drivers of aggregate output growth (i.e., GDP): ( ) + y = a + 1 β k β l. (11) Equation 11 suggests that GDP growth (Dy) is a function of the change in total factor productivity (Da), the change in capital stock (Dk), and the change in labor hours worked (Dl), with β determining the sensitivity of aggregate growth to changes in the factor inputs (i.e., changes in the labor and capital inputs). Relating the macroeconomic output in Equation 11 to the stock payouts, we can see that a stock represents a claim on the residual product of the economic process that is available to owners of capital after all other claims have been satisfied. In theory, owners of capital choose factor inputs (e.g., labor and capital) to maximize their share in the economic process that is, expected profits, which are ultimately returned to owners of capital via dividends and buybacks. Therefore, because the choice of the factor inputs (i.e., Dl and Dk) is driven by a company s expected payouts and the company also benefits from improvements in productivity (Da), it is reasonable to expect aggregate total payouts ( g TPAgg ) to grow in line with the overall economic process (Dy), barring changes in factor share. A portion of the overall economic output (Dy) is financed with new capital, represented by the change in the capital stock (Dk) in Equation 10. For example, the owner of the company may issue more stock to buy a new machine. In this way, the growth available to owners of capital is the growth per unit of capital invested. To calculate income per unit of capital, we divide Equation 9 by K: Y K 1 ββ = AK L A L K = β K. (12) As in Equation 11, taking logs and the first difference, we obtain growth rates: ( ) y k = a + β l k. (13) Equation 13 shows that the growth in income per unit of invested capital is the growth in overall output (Dy) minus new capital investment (Dk), which equals the growth in productivity (Da) plus the relative growth of labor versus capital. If the labor-to-capital ratio is constant (i.e., Dk = Dl), the last term in Equation 13 goes to zero and GDP per unit of capital reduces to the growth in total factor productivity (Da). From the perspective of a stock market index, such as the S&P Composite examined here, new capital investments take the form of new share issuance. Because the number of constituents in the S&P Composite is fixed (i.e., for every stock added to the index, another stock leaves the index), company-level issuance, such as secondary offerings, by existing companies is the key driver of new issuance in our dataset. For stock market indexes that cover the universe of stocks, such as the Wilshire 5000 Total Market Index, 13 IPOs are an additional source of new issuance. New issuance increases the number of shares outstanding, which, for a given level of earnings, leads to lower earnings per share. Thus, per share growth, such as the growth of earnings per share (adjusted for the share decrease from buybacks), is the natural stock market equivalent of the growth 42 cfapubs.org Third Quarter 2017

12 The Long-Run Drivers of Stock Returns in income per unit of invested capital (Equation 12). Buybacks, however, constitute a payout similar to a dividend with the money that investors receive from tendering their shares to the repurchasing company flowing back into the economy and thus must be excluded from our stock market measure of invested capital. Prior studies on the supply of equity returns have made no adjustment to growth terms for the share decrease from buybacks. Having identified the macroeconomic equivalents of the growth rates in our total payout models from a theoretical perspective, we now examine empirically the relationship between economic growth and stocks total payout growth. In our study, we compared aggregate real total payout growth ( g TPAgg ) with GDP growth, and growth in real total payout per share (adjusted for the share decrease from buybacks; g TPexB ) with growth in GDP per capita. 14 We used GDP per capita as a proxy for productivity growth because it has a history since Figure 6 shows that over , as predicted, aggregate total payouts and GDP grew at roughly the same annualized rate: 3.27% and 3.36%, respectively. 16 Similarly, Figure 7 shows that over , the annualized growth in total payout per share (adjusted for the share decrease from buybacks) of 1.67% was approximately in line with the growth in GDP per capita of 1.83%. In our study, we ran statistical tests to determine whether the annual growth rates of total payouts are significantly different from those of their macroeconomic equivalent. The results of both statistical tests are insignificant, with a t-statistic of 0.75 for GDP growth versus aggregate total payout growth and a t-statistic of 0.61 for growth in GDP per capita versus growth in total payout per share (adjusted for the share decrease from buybacks), suggesting that the hypothesis that the two growth rates are the same cannot be rejected. Arnott and Bernstein (2002) assumed that the growth of dividends is lower than the growth of GDP per capita. We found no significant evidence in our study that growth in total payouts (adjusted for the share decrease from buybacks) is lower than growth in GDP per capita. Overall, Figures 6 and 7 provide empirical evidence that the long-run growth in total payouts can be approximated by the growth in the real economy. 17 Although the long-run growth rate trends depicted in Figures 6 and 7 are closely related, there are significant differences in the short run. On the one hand, these differences can be attributed to compositional differences between the stock market and real economic measures, with the stock market tracking listed companies only and real economic measures Figure 6. Growth in Aggregate Total Payout vs. GDP Growth, = Real GDP 8 4 g TPAgg Note: g TPAgg is the growth in real aggregate total payout. Volume 73 Number 3 cfapubs.org 43

13 Financial Analysts Journal A Publication of CFA Institute Figure 7. Growth in Total Payout per Share vs. Growth in GDP per Capita, = Real GDP per Capita 4 2 g TPexB Note: g TPexB is the growth in real total payout per share adjusted for the share decrease from buybacks. tracking the entire economy. On the other hand, there are obvious differences: Total payouts measure only the residual part of the economic process that accrues to owners of capital after all other claims have been paid, whereas GDP measures the economic value added by the overall economy. We refer to this driver as changes in factor share. Greenwald, Lettau, and Ludvigson (2014) identified changes in the factor share of workers versus shareholders as one of the three key stylized factors explaining stock price movements over time. Total factor productivity and changes in risk aversion are the other two. In this section, we provided both theoretical and empirical evidence of the relationship between corporate total payout growth and growth in the real economy. We identified macroeconomic growth measures that correspond to the two total payout growth measures. We further showed that long-run total payout growth rates are statistically indistinguishable from the macroeconomic growth rate. Forecasting Equity Returns Our total payout model can be used to generate forecasts of equity returns. In this section, we evaluate its ability to do so. Given that expected returns are time varying, we can distinguish between long-term (i.e., unconditional) and short-term (i.e., conditional) expected returns. We first discuss the long-term expected returns before turning to the model s ability to predict changes in expected returns (i.e., predicting short-term expected returns). The long-run expected return of any of the supply models discussed in this article can be expressed in terms of two basic components: payout yield and growth. Prior studies (e.g., Ibbotson and Chen 2003) have used dividends as the only measure of payout. Our study, along with Grinold et al. (2011), expands the definition of payouts to include stock buybacks. Consistent with Arnott and Bernstein (2002) and Ibbotson and Chen (2003), we exclude the changein-valuation term from long-term expected returns. Long-term expected real returns are given by Expected real return = Payout yield + Real growth + Interaction. To isolate the impact of using total yield as opposed to dividends as the basis of our long-term forecast, we apply the formula to both dividends and total payouts. In the case of dividends, payout is the 44 cfapubs.org Third Quarter 2017

14 The Long-Run Drivers of Stock Returns average dividend income return (Inc) over , and the growth term is the growth in real dividends per share ( g DIV ). Conversely, for total payouts, the payout is the historical total yield over the sample period, whereas the growth term is the growth in total payout per share (adjusted for the share decrease from buybacks; g TPexB ). Again following Ibbotson and Chen (2003), we include the Interaction term. Table 3 reports the real return forecasts based on the historical payout yield and historical growth over The implied historical return forecasts are 6.21% and 6.79% for the dividend model and the total payout model, respectively. The difference can be attributed to the fact that the dividend model does not account for buybacks in the payout yield term, and the growth in dividend per share was lower than the growth in total payout per share (adjusted for the share decrease from buybacks) because buybacks have been substituted for dividends in recent decades. Therefore, using historical payout yields and growth rates, the dividend model leads to an expected real return that is 0.58% lower than that of the total payout model. Next, we estimate an expected return on the basis of current yields and historical growth rates. Because current yields are sensitive to current market valuations, we technically need to include an additional term to account for the potential future change in valuations (e.g., g P/ TP in Equation 6). For this analysis, however, we assume no change in valuations in order to focus on the impact of the different definitions of payout on the forward-looking equity return. As of December 2014, the dividend yield was 1.92% and the cyclically adjusted total yield (CATY) was 3.21%. Total payouts tend to be more cyclical than dividends owing to the greater volatility of buybacks, so we estimate the total yield on the basis of the average of the past-10-year real total payouts. The current (i.e., not cyclically adjusted) total yield is 4.41% as of year-end We use the same growth and interaction terms as in the historical scenario. The combination of current yields and historical growth rates leads to significantly different estimates of forward equity returns by the two models. The expected real return based on the dividend model is 3.63%, whereas the expected real return based on the total payout model is 5.11%. Not surprisingly, the bulk of the 1.48 percentage point (pp) difference can be attributed to the fact that the dividend yield is 1.29 pps lower than the total yield. The difference in historical growth rates of 0.21 pps is small by comparison. Overall, the dividend model leads to a lower estimate of expected equity returns both for the historical scenario and when current yields are combined with historical growth rates. The dividend model not only excludes buybacks from the payout yield but also underestimates historical growth owing to the substitution of buybacks for dividends. In contrast, the total payout model provides a more accurate forecast of equity returns because each supply component is independent of changes in the payout method, even if current yields are combined with historical growth rates. Having discussed long-term expected returns, we now turn to the total payout model s ability to forecast shorter-term changes in expected returns. We compare the short-term predictability of total payout yields with other valuation measures, such as the dividend yield and the price-to-earnings ratio. The variance decomposition of 10-year log returns in Table 2 highlights that most of the variability in shorter-term Table 3. Long-Run Expected Returns Historical ( ) Current Yield and Historical Growth Dividend Total Payout Dividend Total Payout Payout yield 4.50% 4.89% 1.92% 3.21% Growth 1.46% 1.67% 1.46% 1.67% Interaction 0.26% 0.24% 0.26% 0.24% Expected real return 6.21% 6.79% 3.63% 5.11% Dividend model minus total payout model expected real return 0.58 pps 1.48 pps Volume 73 Number 3 cfapubs.org 45

15 Financial Analysts Journal A Publication of CFA Institute returns is due to changes in valuation. Total payout yields (i.e., TY and NTY) capture the variability in the payout term and the change-in-valuation term because the latter is just the inverse of the total payout yield (P/TP = 1/TY). To evaluate the extent to which our variables predict changes in future returns, we ran predictive regressions of current market measures, such as yield, on non-overlapping one-yearahead real equity total returns as measured by the S&P Composite Index. The results of the predictive regressions are reported in Table 4. The first set of columns in Table 4 shows the regression results for simple valuation ratios, such as the price-to-earnings ratio (P/E), dividend yield, total yield, and net total yield. The last two terms are identical to the total yield and net payout terms in Equations 6 and 8, respectively. The regression results show Table 4. Predictive Regressions of One-Year-Ahead Returns (t-statistics in parentheses) Simple Variables Cyclically Adjusted Variables P/E Div. Yield (DY) Total Yield (TY) Net Total Yield (NTY) CAPE CATY Since 1871 Intercept (2.98)** (0.32) ( 0.47) Coefficient ( 0.83) (1.80) (2.18)* R % 2.24% 3.27% Since 1881 Intercept (2.69)** (0.14) ( 0.69) (4.24)** ( 0.50) Coefficient ( 0.71) (1.82) (2.26)* ( 2.49)* (2.56)* R % 2.46% 3.77% 4.51% 4.78% Since 1901 Intercept (2.48)* (0.18) ( 0.63) (0.77) (3.92)** ( 0.40) Coefficient ( 0.64) (1.66) (2.09)* (2.44)* ( 2.29)* (2.35)* R % 2.43% 3.78% 5.08% 4.53% 4.72% Since 1970 Intercept (1.73) (0.18) ( 0.68) (1.25) (2.14)* ( 0.74) Coefficient ( 0.38) (1.05) (1.56) (1.55) ( 0.99) (1.70) R % 2.57% 5.45% 5.43% 2.28% 6.44% Notes: Our data sample starts in 1871; 1881 is the first year we can calculate the cyclically adjusted (10-year average) valuation measures CAPE and CATY; 1901 is the earliest we have data to estimate NTY. The period since 1970 captures the time frame when buybacks became prevalent. *Significant at the 5% level. **Significant at the 1% level. 46 cfapubs.org Third Quarter 2017