Forthcoming, Journal of Finance

A Better Three-Factor Model That Explains More Anomalies

Long Chen John M. Olin Business School Washington University in St. Louis Lu Zhang Stephen M. Ross School of Business University of Michigan and NBER

June 2009

Abstract A new three-factor model consisting of the market factor and common factors formed on investment and return on assets goes a long way in summarizing the cross-sectional variation of expected stock returns. The model substantially outperforms traditional asset pricing models in describing average returns across testing portfolios formed on short-term prior returns, nancial distress, net stock issues, asset growth, and earnings surprises. The model also performs roughly as well as the Fama-French model in accounting for average returns across portfolios formed on valuation ratios, industry, and CAPM betas. The models performance, combined with its economic intuition, suggests that it can be used to obtain expected return estimates in practice.

John M. Olin Business School, Washington University in St. Louis, 212 Simon Hall, 1 Olympian Way, St. Louis MO 63130-4899. Tel: (314) 935-8374 and e-mail: lchen29@wustl.edu. Finance Department, Stephen M. Ross School of Business, University of Michigan, 701 Tappan Street, R 4336, Ann Arbor MI 48109-1234; and NBER. Tel: (734) 615-4854, fax: (734) 936-0282, and e-mail: zhanglu@bus.umich.edu. For helpful discussions, we thank Andrew Ang, Sreedhar Bharath, Ken French, Gerald Garvey (BGI discussant), Joao Gomes (AFA discussant), Hong Liu (FIRS discussant), Scott Richardson, Tyler Shumway, Richard Sloan, Alan Timmermann, Motohiro Yogo (UBC discussant), and other seminar participants at Barclays Global Investors, Hong Kong University of Science and Technology, Renmin University of China, Rutgers Business School, Tel Aviv University, University of California at San Diego, University of Michigan, Sanford C. Bernstein Conference on Controversies in Quantitative Finance and Asset Management, UBC PH&N Summer Finance Conference in 2007, the 2008 Financial Intermediation Research Society Conference on Banking, Corporate Finance, and Intermediation, and the 2009 American Finance Association Annual Meetings. Cynthia Jin provided valuable research assistance. Cam Harvey (the Editor), an anonymous Associated Editor, and an anonymous referee deserve special thanks. In particular, the title of the paper is due to the referee. An Internet Appendix containing supplementary results is available at http://www.afajof.org/supplements.asp. Previous drafts of the paper have been circulated under the titles Neoclassical Factors, An Equilibrium Three-Factor Model, and Production-based Factors.

Electronic copy available at: http://ssrn.com/abstract=1418117

Introduction

Although an elegant theoretical contribution, the empirical performance of the Sharpe (1964) and Lintner (1965) Capital Asset Pricing Model (CAPM) has been abysmal.1 Fama and French (1993), among others, have tried to augment the CAPM with certain factors to explain what the CAPM cannot.2 However, over the past two decades, it has become increasingly clear that even the famous Fama-French model cannot explain many cross-sectional patterns. Prominent examples include the positive relations of average returns with short-term prior returns and earnings surprises as well as the negative relations of average returns with nancial distress, net stock issues, and asset growth.3 We motivate a new three-factor model from q-theory, and show that it goes a long way in accounting for many patterns in cross-sectional returns that the Fama-French model cannot. In the new model the expected return on a portfolio in excess of the risk-free rate, denoted E[r j ]rf , is described by the sensitivity of its return to three factors: the market excess return (rM KT ), the dierence between the return on a portfolio of low-investment stocks and the return on a portfolio of highinvestment stocks (rIN V ), and the dierence between the return on a portfolio of stocks with high returns on assets and the return on a portfolio of stocks with low returns on assets (rROA ). Specically,
j j j E[r j ] rf = M KT E[rM KT ] + IN V E[rIN V ] + ROA E[rROA ]

(1)

j j in which E[rM KT ], E[rIN V ], and E[rROA ] are expected premiums, and the loadings, M KT , IN V , j and ROA , are time series slopes from regressing portfolio excess returns on rM KT , rIN V , and rROA .

In our 19722006 sample, rIN V and rROA earn average returns of 0.43% (t = 4.75) and 0.96% per month (t = 5.10), respectively. These average returns subsist after adjusting for their exposures to the Fama-French factors and the Carhart (1997) factors. Most important, the new factor model does a good job in describing the average returns of 25 size and momentum portfolios. None of the
DeBondt and Thaler (1985), Rosenberg, Reid, and Lanstein (1985), Fama and French (1992), and Lakonishok, Shleifer, and Vishny (1994) show that average returns covary with book-to-market, earnings-to-price, cash ow-toprice, dividend-to-price, long-term past sales growth, and long-term prior returns, even after one controls for market betas. Jegadeesh and Titman (1993) show that stocks with higher short-term prior returns earn higher average returns. 2 Specically, Fama and French (1993, 1996) show that their three-factor model, which includes the market excess return, a factor mimicking portfolio based on market equity, SM B, and a factor mimicking portfolio based on book-to-market, HM L, can explain many CAPM anomalies such as average returns across portfolios formed on size and book-to-market, earnings-to-price, cash ow-to-price, dividend-to-price, and long-term prior returns. 3 See, for example, Ritter (1991), Ikenberry, Lakonishok, and Vermaelen (1995), Loughran and Ritter (1995), Spiess and Aeck-Graves (1995), Chan, Jegadeesh, and Lakonishok (1996), Fama and French (1996, 2008), Dichev (1998), Grin and Lemmon (2002), Daniel and Titman (2006), Campbell, Hilscher, and Szilagyi (2008), and Cooper, Gulen, and Schill (2008). Many of these papers argue that the evidence is driven by mispricing due to investors over- or underreaction to news. For example, Campbell et al. (2008) suggest that their evidence is a challenge to standard models of rational asset pricing in which the structure of the economy is stable and well understood by investors (p. 2934).
1

Electronic copy available at: http://ssrn.com/abstract=1418117

winner-minus-loser portfolios across ve size quintiles have signicant alphas. The alphas, ranging from 0.08% to 0.54% per month, are all within 1.7 standard errors from zero. For comparison, the alphas vary from 0.92% (t = 3.10) to 1.33% per month (t = 5.78) in the CAPM and from 0.92% (t = 2.68) to 1.44% (t = 5.54) in the Fama-French model. The new model fully captures the negative relation between average returns and nancial distress as measured by Campbell, Hilscher, and Szilagyis (2008) failure probability. The high-minuslow distress decile earns an alpha of 0.32% per month (t = 1.09) in our model, which cannot be rejected across the distress deciles by the Gibbons, Ross, and Shanken (1989, GRS) test at the 5% signicance level. In contrast, the alpha is 1.87% (t = 5.08) in the CAPM and 2.14% (t = 6.43) in the Fama-French model, and both models are strongly rejected by the GRS test. Using Ohlsons (1980) O-score to measure distress yields largely similar results. Intuitively, more distressed rms have lower returns on assets (ROA) and therefore have lower rROA -loadings and earn lower expected returns than less distressed rms. Prior studies overlook the positive ROA-expected return relation, and, not surprisingly, nd the negative distress-expected return relation anomalous. Several other anomaly variables including net stock issues, asset growth, and earnings surprises also have received much attention since Fama and French (1996). We show that the new model outperforms traditional asset pricing models in accounting for these eects, often by a big margin. For example, the high-minus-low net stock issues portfolio earns an insignicant alpha of 0.28% per month (t = 1.39) in our model. In contrast, the CAPM alpha is 1.06% (t = 5.07) and the Fama-French alpha is 0.82% (t = 4.33). Finally, the new model performs roughly as well as the Fama-French model in explaining portfolios formed on valuation ratios such as book-tomarket equity. Stocks with low valuation ratios (suggesting low growth opportunities) invest less and therefore load more on the low-minus-high rIN V factor and earn higher average returns than stocks with high valuation ratios (suggesting high growth opportunities).4 We motivate the investment and ROA factors from q-theory. Intuitively, investment predicts returns because given expected cash ows, high costs of capital mean low net present values of new capital, which in turn mean low investment, and low costs of capital mean high net present values of new capital, which in turn mean high investment. ROA predicts returns because high expected
More generally, our models performance is comparable with that of the Fama-French model in capturing the average returns of testing portfolios which Fama and French (1996) show that their three-factor model is capable of explaining. The list includes earnings-to-price, dividend-to-price, prior 1360 month returns, ve-year sales rank, and market leverage (total assets-to-market equity). We only report the results of the 25 size and book-to-market portfolios to save space because Fama and French (1996) show that book-to-market largely subsumes the aforementioned variables in predicting future returns. The online appendix reports detailed factor regressions for all the other testing portfolios.
4

ROA relative to low investment means high discount rates. The high discount rates are necessary to counteract the high expected ROA to produce low net present values of new capital and thereby low investment. If the discount rates are not high enough to oset the high expected ROA, rms would observe high net present values of new capital and invest more instead. Similarly, low expected ROA relative to high investment (such as small-growth rms in the late 1990s) means low discount rates. If the discount rates are not low enough to oset the low expected ROA, these rms would observe low net present values of new capital and invest less instead. Our central contribution is to provide a new workhorse factor pricing model for estimating expected returns. In particular, we oer an update of Fama and French (1996), who show that their three-factor model summarizes what we know about the cross-section of returns by the mid-1990s. Similarly, we show that the new factor model summarizes what we know about the cross-section of returns by the late 2000s. In doing so we also elaborate a simple conceptual framework in which many anomalies can be interpreted simultaneously in a unied and economically meaningful way. The models performance, combined with its economic intuition, suggests that it can be used in many practical applications such as evaluating mutual fund performance, measuring abnormal returns in event studies, estimating expected returns for portfolio choice, and obtaining the cost of equity capital estimates for capital budgeting and stock valuation. Most asset pricing studies motivate common factors from the consumption side of the economy (e.g., Breeden, Gibbons, and Litzenberger (1989), Ferson and Harvey (1992, 1993), and Lettau and Ludvigson (2001)). We instead exploit a direct link between rm-level returns and characteristics from the production side. Cochrane (1991) rst uses this production-based approach to study stock market returns, but we study anomalies in cross-sectional returns. Liu, Whited, and Zhang (2008) explore the return-characteristics link via structural estimation, but we use the Fama-French portfolio approach to produce a workhorse factor model. A factor pricing model is more practical because of its powerful simplicity and the availability of high-quality monthly returns data. Section 2 motivates the new factors from q-theory, Section 3 constructs the new factors, Section 4 tests the new factor model, and Section 5 summarizes and interprets the results.

Hypothesis Development

We develop testable hypotheses from q-theory (e.g., Tobin (1969) and Cochrane (1991)). We outline a two-period structure to x the intuition, but the basic insights hold in more general settings. 4

There are two periods, 0 and 1, and heterogeneous rms, indexed by j. Firm js operating prots are given by j0 Aj0 in date 0 and j1 Aj1 in date 1, in which Aj0 and Aj1 are the rms scale of productive assets and j0 and j1 are the rms return on assets in dates 0 and 1, respectively. Firm j starts with assets, Aj0 , invests in date 0, produces in both dates, and exits at the end of date 1 with a liquidation value of (1 )Aj1 , in which is the rate of depreciation. Assets evolve as Aj1 = Ij0 + (1 )Aj0 , in which Ij0 is investment. Investment entails quadratic adjustment costs of (a/2)(Ij0 /Aj0 )2 Aj0 , in which a > 0 is a constant parameter. Firm j has a gross discount rate of r j . The discount rate varies across rms due to, for example, rm-specic loadings on macroeconomic risk factors. The rm chooses Aj1 to maximize the market value at the beginning of date 0: max j0 Aj0 [Aj1 (1 )Aj0 ] a Aj1 (1 ) 2 Aj0
2

{Aj1 }

Aj0 +

1 [j1 Aj1 + (1 )Aj1 ] rj

(2)

The market value is date 0s free cash ow, j0 Aj0 Ij0 (a/2)(Ij0 /Aj0 )2 Aj0 , plus the discounted value of date 1s cash ow, [j01 Aj1 + (1 )Aj1 ] /r j . With only two dates the rm does not invest in date 1, so date 1s free cash ow is the sum of operating prots and the liquidation value. The tradeo of rm j is simple: forgoing date 0s free cash ow in exchange for higher date 1s free cash ow. Setting the rst-order derivative of equation (2) with respect to Aj1 to zero yields: rj = j1 + 1 1 + a(Ij0 /Aj0 ) (3)

This optimality condition is intuitive. The numerator in the right-hand side is the marginal benet of investment including the marginal product of capital (return on assets), j1 , and the marginal liquidation value of capital, 1 . The denominator is the marginal cost of investment including the marginal purchasing cost of investment (one) and the marginal adjustment cost, a(Ij0 /Aj0 ). Because the marginal benet of investment is in date 1s dollar terms and the marginal cost of investment is in date 0s dollar terms, the rst-order condition says that the marginal benet of investment discounted to date 0s dollar terms should be equal to the marginal cost of investment. Equivalently, the investment return, dened as the ratio of the marginal benet of investment divided by the marginal cost of investment, should equal the discount rate, as in Cochrane (1991).

2.1

The Investment Hypothesis

We use the rst-order condition (3) to develop testable hypotheses for cross-sectional returns. H1 : Given the expected ROA, the expected return decreases with investment-to-assets.

This mechanism drives the negative relations of average returns with net stock issues, asset growth, valuation ratios, and long-term past sales growth and prior returns. Figure 1 illustrates the investment hypothesis.

Figure 1: The Investment Hypothesis in the Cross-Section of Returns

Y -axis: the discount rate

Low investment-to-assets rms Matching nonissuers Low net stock issues rms Low asset growth rms Value rms with high book-to-market High market leverage rms Firms with low long-term prior returns Firms with low long-term past sales growth

Firms with high long-term past sales growth Firms with high long-term prior returns Low market leverage rms Growth rms with low book-to-market High asset growth rms High net stock issues rms SEO rms, IPO rms, convertible bond issuers High investment-to-assets rms

X-axis: investment-to-assets

Intuition The negative relation between the expected return and investment is intuitive. Firms invest more when their marginal q (the net present value of future cash ows generated from one additional unit of capital) is high. Given expected ROA or cash ows, low discount rates give rise to high marginal q and high investment, and high discount rates give rise to low marginal q and low investment. This intuition is probably most transparent in the capital budgeting language of Brealey, Myers, and Allen (2006). In our simple setting capital is homogeneous, meaning that there is no dierence between project-level costs of capital and rm-level costs of capital. Given expected cash ows, high costs of capital imply low net present values of new projects and therefore low investment, and low costs of capital imply high net present values of new projects and therefore high investment. Without uncertainty, it is well-known that the interest rate and investment are negatively correlated, meaning that the investment demand curve is downward sloping (e.g., Fisher (1930) and 6

Fama and Miller (1972, Figure 2.4)). With uncertainty, more investment leads to lower marginal product of capital under decreasing returns to scale, giving rise to lower expected returns (e.g., Li, Livdan, and Zhang (2008)). The real options models of Berk, Green, and Naik (1999) and Carlson, Fisher, and Giammarino (2004) also imply the negative investment-expected return relation. In their models expansion options are riskier than assets in place. Investment converts riskier expansions options into less risky assets in place, meaning that high-investment rms are less risky and earn lower expected returns than low-investment rms. Portfolio Implications The negative investment-expected return relation is conditional on expected ROA. ROA is not disconnected with investment: more protable rms tend to invest more than less protable rms. This conditional relation provides a natural portfolio interpretation of the investment hypothesis. Sorting on net stock issues, asset growth, book-to-market, and other valuation ratios is closer to sorting on investment than sorting on expected ROA. Equivalently, these sorts produce higher magnitudes of spread in investment than in expected ROA. As such, we can interpret the average return spreads generated from these diverse sorts using their common implied sort on investment. The negative relations of average returns with net stock issues and asset growth is consistent with the negative investment-expected return relation. The balance-sheet constraint of rms says that the uses of funds must equal the sources of funds, meaning that issuers must invest more and earn lower average returns than nonissuers.5 Cooper, Gulen, and Schill (2008) document that asset growth negatively predicts future returns and interpret the evidence as investor underreacting to overinvestment. However, asset growth is the most comprehensive measure of investment-to-assets, in which investment is simply dened as the change in total assets. This observation means that the asset growth eect is potentially consistent with optimal investment. The value premium also can be interpreted using the negative investment-expected return relation: investment-to-assets is an increasing function of marginal q, which is closely related to average q (market-to-book). With constant returns to scale the marginal q equals the average q. But the average q of the rm and market-to-book equity are highly correlated, and are identical without debt nancing. As such, value rms with high book-to-market invest less and earn higher average returns than growth rms with low book-to-market. In general, rms with higher valuation ratios have more
Lyandres, Sun, and Zhang (2008) show that adding the investment factor into the CAPM and the Fama-French model greatly reduces the magnitude of the underperformance following initial public oerings, seasoned equity oerings, and convertible debt oerings. Lyandres et al. also report the part of Figure 1 related to the new issues puzzle.
5

growth opportunities, invest more, and should earn lower expected returns, and rms with lower valuation ratios have less growth opportunities, invest less, and should earn higher expected returns. We also include market leverage into this category. Fama and French (1992) measure market leverage as the ratio of total assets divided by market equity. Empirically, the new factor model captures the market leverage-expected return relation roughly as well as the Fama-French model (see the online appendix). Intuitively, because market equity is in the denominator, high leverage signals low growth opportunities, low investment, and high expected returns, and low leverage signals high growth opportunities, high investment, and low expected returns. This investment mechanism diers from the leverage eect in standard corporate nance texts. According to the leverage eect, high leverage means high proportion of asset risk shared by equity holders, inducing high expected equity returns. This mechanism assumes that the investment policy is xed and that asset risk does not vary with investment. In contrast, the investment mechanism allows investment and leverage to be jointly determined, giving rise to a negative relation between market leverage and investment and thereby a positive relation between market leverage and expected returns. High valuation ratios can result from a stream of positive shocks on fundamentals and low valuation ratios from a stream of negative shocks on fundamentals. As such, high valuation ratios of growth rms can be manifested as high past sales growth and high long-term prior returns. Firms with high long-term prior returns and high past sales growth should have higher valuation ratios, invest more, and earn lower average returns than rms with low long-term prior returns and low past sales growth. As such, the investment mechanism also helps capture DeBondt and Thalers (1985) reversal eect and Lakonishok, Shleifer, and Vishnys (1994) sales growth eect.

2.2

The ROA Hypothesis

The rst-order condition (3) also implies the following ROA hypothesis: H2 : Given investment-to-assets, rms with high expected ROA should earn higher expected returns than rms with low expected ROA. This positive ROA-expected return relation drives the positive relations of average returns with short-term prior returns and earnings surprises and the negative relation of average returns with nancial distress. Intuition Why do high expected ROA rms earn higher expected returns than low expected ROA rms? We explain the intuition in two ways, the discounting intuition and the capital budgeting intuition. 8

First, the marginal cost of investment in the denominator of the right-hand side of the rst-order condition (3) equals marginal q, which in turn equals average q or market-to-book. As such, equation (3) in eect says that the expected return is the expected ROA divided by market-to-book equity, or equivalently, the expected cash ow divided by the market equity. This relation is analogous to the Gordon (1962) Growth Model. In a two-period world price equals the expected cash ow divided by the discount rate: high expected cash ows relative to low market equity (or high expected ROA relative to low market-to-book) mean high discount rates, and low expected cash ows relative to high market equity (or low expected ROA relative to high market-to-book) mean low discount rates. This discounting intuition from valuation theory also is noted by Fama and French (2006). Using the residual income model, Fama and French argue that expected stock returns are related to three variables: the book-to-market equity, expected protability, and expected investment, and that controlling for book-to-market and expected investment, more protable rms have higher expected returns. In particular, they argue that (p. 492): Controlling for Bt /Mt and expected growth in book equity due to reinvestment of earnings, more protable rmsspecically, rms with higher expected earnings relative to current book equityhave higher expected returns. However, Fama and French do not motivate the ROA eect from q-theory or construct the ROA factor and use it to capture the momentum and distress eects, as we do in Section 4. In addition to the discounting intuition, q-theory also provides capital budgeting intuition for the positive ROA-expected return relation. Equation (3) says that the expected return equals the expected ROA divided by an increasing function of investment-to-assets. High expected ROA relative to low investment must mean high discount rates. The high discount rates are necessary to oset the high expected ROA to produce low net present values of new capital and thereby low investment. If the discount rates are not high enough to counteract the high expected ROA, rms would instead observe high net present values of new capital and thereby invest more. Similarly, low expected ROA relative to high investments (such as small-growth rms in the 1990s) must mean low discount rates. If the discount rates are not low enough to oset the low expected ROA, these rms would instead observe low net present values of new capital and thereby invest less. Portfolio Implications The positive ROA-expected return relation has important portfolio implications: for any sorts that generate higher magnitudes of spread in expected ROA than in investment, their average return patterns can be interpreted using the common implied sort on expected ROA. We explore three such 9

sorts in Section 4, sorts on short-term prior returns, on nancial distress, and on earnings surprises. First, sorting on short-term prior returns should generate a spread in expected ROA. Intuitively, shocks to earnings are positively correlated with contemporaneous shocks to stock returns. Firms with positive earnings surprises are likely to experience immediate stock price increases, whereas rms with negative earnings surprises are likely to experience immediate stock price decreases. As such, winners with high short-term prior returns should have higher expected ROA and earn higher average returns than losers with low short-term prior returns. Second, less distressed rms are more protable (have higher expected ROA) and, all else equal, should earn higher average returns. More distressed rms are less protable (have lower expected ROA) and, all else equal, should earn lower average returns. As such, the distress eect can be interpreted using the positive ROA-expected return relation. Finally, sorting on earnings surprises can generate an expected ROA spread between extreme portfolios. Intuitively, rms that have experienced large positive earnings surprises should be more protable than rms that have experienced large negative earnings surprises.

The Explanatory Factors

We test the investment and ROA hypotheses via the Fama-French portfolio approach. We construct new common factors based on investment-to-asset and ROA in a similar way that Fama and French (1993, 1996) construct their size and value factors. Because the new factors are motivated from the production side of the economy, we also include the market factor from the consumption side, and use the resulting three-factor model as a parsimonious description of cross-sectional returns. In the same way that Fama and French test their three-factor model, we use factor regressions to evaluate the new models performance. The simplicity of the portfolio approach allows us to implement the new model on a wide range of testing portfolios. Monthly returns, dividends, and prices are from the Center for Research in Security Prices (CRSP) and accounting information from Compustat Annual and Quarterly Industrial Files. The sample is from January 1972 to December 2006. The starting date is restricted by the availability of quarterly earnings and assets data. We exclude nancial rms and rms with negative book equity.

3.1

The Investment Factor

We dene investment-to-assets (I/A) as the annual change in gross property, plant, and equipment (Compustat annual item 7) plus the annual change in inventories (item 3) divided by the

10

lagged book value of assets (item 6). Changes in property, plant, and equipment capture capital investment in long-lived assets used in operations over many years such as buildings, machinery, furniture, and other equipment. Changes in inventories capture capital investment in short-lived assets used in a normal operating cycle such as merchandise, raw materials, supplies, and work in progress. This denition is consistent with the practice of National Income Accounting: Bureau of Economic Analysis measures gross private domestic investment as the sum of xed investment and the net change in business inventories. Also, investment and growth opportunities are closely related: growth rms with high market-to-book equity invest more than value rms with low market-to-book equity. However, growth opportunities can manifest in other forms such as high employment growth and large R&D expense that are not captured by I/A. We construct the investment factor, rIN V , from a two by three sort on size and I/A. Fama and French (2008) show that the magnitude of the asset growth eect varies across dierent size groups: it is strong in microcaps and small stocks, but is largely absent in big stocks. To the extent that asset growth is eectively the most comprehensive measure of investment (divided by assets), it seems necessary to control for size when constructing rIN V . The two by three sort also is used by Fama and French (1993) in constructing SM B and HM L to control for the correlation between size and book-to-market. In June of each year t we break NYSE, Amex, and NASDAQ stocks into three I/A groups based on the breakpoints for the low 30%, middle 40%, and high 30% of the ranked values. We also use the median NYSE market equity (stock price times shares outstanding) to split NYSE, Amex, and NASDAQ stocks into two groups. We form six portfolios from the intersections of the two size and the three I/A groups. Monthly value-weighted returns on the six portfolios are calculated from July of year t to June of t+1, and the portfolios are rebalanced in June of t+1. Designed to mimic the common variation in returns related to I/A, the investment factor is the dierence (low-minus-high), each month, between the simple average of the returns on the two low-I/A portfolios and the simple average of the returns on the two high-I/A portfolios. From Table 1, the average rIN V return in the 19722006 sample is 0.43% per month (t = 4.75). Regressing rIN V on the market factor generates an alpha of 0.51% per month (t = 6.12) and an R2 of 16%. The average return also subsists after controlling for the Fama-French and Carhart factors (data from Kenneth Frenchs Web site.) rIN V also has a high correlation of 0.51 with HM L. The evidence is consistent with Titman, Wei, and Xie (2004), Anderson and Garcia-Feijo (2006), and o Xing (2008).6 Sorting on I/A produces a large I/A spread. For example, the small and low-I/A
6

Titman, Wei, and Xie (2004) sort stocks on CEt1 /[(CEt2 +CEt3 +CEt4 )/3], in which CEt1 is capital expen-

11

portfolio has an average I/A of 4.27% per annum, whereas the small and high-I/A portfolio has an average of 30.15% (see the Internet Appendix). The impact of industries on the investment factor is relatively small (see the Internet Appendix). We conduct an annual two by three sort on industry size and I/A using Fama and Frenchs (1997) 48 industries. Following Fama and French (1995), we dene industry size as the sum of market equity across all rms in a given industry and industry I/A as the sum of investment for all rms in a given industry divided by the sum of assets for the same set of rms. We construct the industry-level investment factor as the average low-I/A industry returns minus the average high-I/A industry returns. If the industry eect is important for the rm-level investment factor, the industry-level investment factor should earn signicant average returns. Moskowitz and Grinblatt (1999) use a similar test design to construct industry-level momentum and show that it accounts for much of the rm-level momentum. However, the average return for the industry-level investment factor is only 0.17% per month (t = 1.50). The CAPM alpha, Fama-French alpha, and Carhart alpha are 0.19%, 0.14%, and 0.22% per month, respectively, none of which are signicant at the 5% level. Finally, each of the six rm-level size-I/A portfolios draws observations from a wide range of industries, and the industry distribution of rm-month observations does not vary much across the portfolios.

3.2

The ROA Factor

We construct rROA based on return on assets. We sort on current ROA (as opposed to expected ROA) because ROA is highly persistent. Fama and French (2006) show that current protability is the strongest predictor of future protability, and that adding more regressors in the specication of the expected protability decreases its explanatory power for future stock returns. Also, because rROA is most relevant for explaining earnings surprises, prior returns, and distress eects that are constructed monthly, we use a similar approach to construct rROA .7
diture (Compustat annual item 128) scaled by sales as the scal year t1. The prior three-year moving average of CE is designed to capture the benchmark investment level. We sort stocks directly on I/A because it is more closely connected to q-theory. Xing (2008) shows that an investment growth factor contains information similar to HM L and can explain the value premium roughly as well as HM L. The average return of the investment growth factor is only 0.20% per month, albeit signicant. Our investment factor is more powerful for several reasons. In principle, q-theory (see equation (3)) says that investment-to-assets is a more direct predictor of returns than past investment growth. Empirically, rm-level investment can often be zero or negative, making investment growth ill-dened. Xing measures investment as capital expenditure, in eect ignoring rms with zero or negative capital investment. By using annual change in property, plant, and equipment, we include these rms in our factor construction. Finally, we also use a more comprehensive measure of investment that includes both long-term investment and short-term working capital investment. 7 The Internet Appendix shows that the original earnings surprises, momentum, and the distress eects do not exist in portfolios that are annually rebalanced. Specically, in June of each year t we sort all NYSE, Amex, and NASDAQ stocks into ten deciles based on, separately, the Standardized Unexpected Earnings measured at the scal yearend of t1, the 12-month prior return from June of year t1 to May of year t, and Campbell, Hilscher, and Szilagyis (2008)

12

We measure ROA as income before extraordinary items (Compustat quarterly item 8) divided by last quarters total assets (item 44). Each month from January 1972 to December 2006, we categorize NYSE, Amex, and NASDAQ stocks into three groups based on the breakpoints for the low 30%, middle 40%, and high 30% of the ranked values of quarterly ROA from at least four months ago. We impose the four-month lag to ensure that the required accounting information is known before forming the portfolios. The choice of the four-month lag is conservative: using shorter lags only strengthens our results. We also use the NYSE median each month to split NYSE, Amex, and NASDAQ stocks into two groups. We form six portfolios from the intersections of the two size and three ROA groups. Monthly value-weighted returns on the six portfolios are calculated for the current month, and the portfolios are rebalanced monthly. Meant to mimic the common variation in returns related to rm-level ROA, the ROA factor is the dierence (high-minus-low), each month, between the simple average of the returns on the two high-ROA portfolios and the simple average of the returns on the two low-ROA portfolios. From Panel A of Table 1, rROA earns an average return of 0.96% per month (t = 5.10) from January 1972 to December 2006. Controlling for the market factor, the Fama-French factors, and the Carhart factors does not aect the average rROA return. This evidence means that, like rIN V , rROA also captures average return variation not subsumed by existing common factors. From Panel B, rROA and the momentum factor have a correlation of 0.26, suggesting that shocks to earnings are positively correlated with contemporaneous shocks to returns. The correlation between rIN V and rROA is only 0.10 (p = 0.05), meaning that there is no need to neutralize the two factors against each other. From the Internet Appendix, sorting on ROA generates a large spread in ROA: the small and low-ROA portfolio has an average ROA of 13.32% per annum, whereas the small and high-ROA portfolio has an average ROA of 13.48%. The large ROA spread only corresponds to a modest spread in I/A: 11.49% versus 12.56% per annum, helping explain the low correlation between rIN V and rROA . The ROA spread in small rms corresponds to a large spread in prior 212 month returns: 9.55% versus 34.44%, helping explain the high correlation between rROA and the momentum factor. The industry eect on the ROA factor is small (see the Internet Appendix). We conduct a monthly two by three sort on industry size and ROA using Fama and Frenchs (1997) 48 indusfailure probability and Ohlsons (1980) O-score measured at the scal yearend of t 1. We calculate monthly valueweighted returns from July of year t to June of t + 1 and rebalance the portfolios in June. None of these strategies produce mean excess returns or CAPM alphas that are signicantly dierent from zero. Because the targeted eects only exist in monthly frequency, it seems natural to construct the explanatory ROA factor in the same frequency.

13

tries. We dene industry ROA as the sum of earnings across all rms in a given industry divided by the sum of assets across the same set of rms. The industry ROA factor is constructed as the average high-ROA industry returns minus the average low-ROA industry returns. If the industry eect is important for the rm-level ROA factor, the industry ROA factor should show signicant average returns. The evidence says otherwise. The average return of the industry ROA factor is only 0.19% per month (t = 1.63), and the CAPM alpha, Fama-French alpha, and Carhart alpha are 0.21% (t = 1.77), 0.31% (t = 2.57), and 0.14% (t = 1.20), respectively. Relative to the rm-level ROA factor with an average return of 0.96% (t = 5.10), the industry eect seems small in magnitude. Finally, each of the six rm-level size-ROA portfolios draws observations from a wide range of industries, and the industry distribution of observations does not vary much across the portfolios.

Factor Regressions

We use simple time series regressions to confront the new factor model with testing portfolios formed on a wide range of anomaly variables:
j j j r j rf = j + M KT rM KT + IN V rIN V + ROA rROA + j . q

(4)

If the models performance is adequate, j should be statistically indistinguishable from zero. q

4.1

Short-Term Prior Returns

Following Jegadeesh and Titman (1993), we construct the 25 size and momentum portfolios using the 6/1/6 convention. For each month t, we sort stocks on their prior returns from month t2 to t7, skip month t1, and calculate the subsequent portfolio returns from month t to t+5. We also use NYSE market equity quintiles to sort all stocks independently each month into ve size portfolios. The 25 portfolios are formed monthly as the intersection of the ve size quintiles and the ve quintiles based on prior 27 month returns.8 Table 2 reports large momentum prots. From Panel A, the winner-minus-loser (W-L) average return varies from 0.85% (t = 3.01) to 1.25% per month (t = 5.49). The CAPM alphas for the W-L portfolios are signicantly positive across all ve size quintiles. The small-stock W-L strategy, in particular, earns a CAPM alpha of 1.33% per month (t = 5.78). Consistent with Fama and French
8 Using the 25 portfolios with the 11/1/1 convention from Kenneth Frenchs Web site yields largely similar results (see the Internet Appendix). The 11/1/1 convention means that, for each month t, we sort stocks on their prior returns from month t 2 to t 12, skip month t 1, and calculate portfolio returns for the current month t.

14

(1996), their three-factor model exacerbates momentum. In particular, the small-stock W-L portfolio earns a Fama-French alpha of 1.44% per month (t = 5.54). Losers have higher HM L-loadings than winners, so their model counterfactually predicts that losers should earn higher average returns. Panel B reports the new models factor regressions. None of the W-L strategies across ve size quintiles earn signicant alphas. The small-stock W-L strategy has an alpha of 0.54% per month (t = 1.70), which represents a reduction of 59% in magnitude from its CAPM alpha and 63% from its Fama-French alpha. The average magnitude of the W-L alphas in our model is 0.37% per month, whereas it is 1.08% in the CAPM and 1.17% in the Fama-French model. The new factor models relative success derives from two sources. First, from Table 2, winners have higher rROA -loadings than losers across all size groups, going in the right direction in explaining the average returns. The loading spreads range from 0.22 to 0.45, which, given an average rROA return of 0.96% per month, explain 0.21% to 0.43% per month of momentum prots. Second, somewhat surprisingly, the rIN V -loadings also go in the right direction because winners have higher rIN V -loadings than losers. The loading spreads, ranging from 0.57 to 0.83, are all signicant across the size groups. Combined with an average rIN V return of 0.43% per month, the loadings explain 0.25% to 0.36% per month of momentum prots. This loading pattern is counterintuitive. One would expect that winners with high valuation ratios should invest more and have lower loadings on the low-minus-high rIN V factor than losers with low valuation ratios. To understand what drives these loading patterns, we use the event-study approach of Fama and French (1995) to examine how I/A varies across momentum portfolios. We nd that winners indeed have higher contemporaneous I/A than losers at the portfolio formation month. More important, winners also have lower I/A than losers starting from two to four quarters prior to the portfolio formation. Because rIN V is rebalanced annually, the higher rIN V -loadings for winners accurately reect their lower I/A several quarters prior to the portfolio formation. Specically, for each portfolio formation month t from January 1972 to December 2006, we calculate annual I/As for t+m, m = 60, . . . , 60. The I/As for t+m are then averaged across portfolio formation months t. For a given portfolio, we plot the median I/As among the rms in the portfolio. From Panel A of Figure 2, although winners have higher I/As at the portfolio formation month t, winners have lower I/As than losers from month t60 to month t8. Panel B also shows that winners have higher contemporaneous I/As than losers in the calendar time in the smallest-size quintile. The contemporaneous I/A is dened as the I/A at the current scal yearend. For example, if the current

15

month is March or September 2003, the contemporaneous I/A is the I/A at the scal yearend of 2003. More important, Panel C shows that winners also have lower lagged (sorting-eective) I/As than losers in the smallest-size quintile. The sorting-eective I/A is dened as the I/A on which an annual sort on I/A in each June is based. For example, if the current month is March 2003, the sorting-eective I/A is the I/A at the scal yearend of 2001 because the annual sort on I/A is in June 2002. If the current month is September 2003, the sorting-eective I/A is the I/A at the scal yearend of 2002 because the corresponding sort on I/A is in June 2003. Because rIN V is rebalanced annually, the lower sorting-eective I/As of winners explain their higher rIN V -loadings than losers. Finally, as expected, Figure 2 also shows that winners have higher ROAs than losers for about ve quarters before and 20 quarters after the portfolio formation month (Panel D). In the calendar time, winners have consistently higher ROAs than losers, especially in smallest-size quintile (Panels E and F). This evidence explains the higher rROA -loadings for winners documented in Table 2.

4.2

Distress

The new factor model fully accounts for the negative relation between nancial distress and average returns. We form ten deciles based on Ohlsons (1980) O-score and Campbell, Hilscher, and Szilagyis (2008) failure probability. Appendix A details the variable denitions.9 Each month from June 1975 to December 2006, we sort all stocks into ten deciles on failure probability from at least four months ago. The starting point of the sample is restricted by the availability of data items required to construct failure probability: for comparison, Campbell, Hilscher, and Szilagyi (2008) start their sample in 1981. Monthly value-weighted portfolio returns are calculated for the current month. Panel A of Table 3 reports that more distressed rms earn lower average returns than less distressed rms. The high-minus-low (H-L) distress portfolio has an average return of 1.38% per month (t = 3.53). Controlling for traditional risk measures only makes things worse: more distressed rms are riskier per traditional factor models. The H-L portfolio has a market beta of 0.73 (t = 5.93) in the CAPM, giving rise to an alpha of 1.87% per month (t = 5.08). The portfolio also has a loading of 1.10 (t = 7.46) on the size factor and a market beta of 0.57 (t = 4.57) in the Fama-French model, producing an alpha of 2.14% per month (t = 6.43). The new factor model reduces the H-L alpha to an insignicant level of 0.32% per month (t = 1.09). Although two out of ten deciles have signicant alphas, the model is not rejected by the
We also have experimented with portfolios formed on Altmans (1968) Z-score, but the CAPM adequately captures the average returns of these portfolios in our sample.
9

16

GRS test. In contrast, both CAPM and the Fama-French model are rejected at the 5% signicance level. The rROA -loading goes in the right direction in explaining the distress eect. More distressed rms have lower rROA -loadings than less distressed rms, and the loading spread is 1.40, which is more than 14 standard errors from zero. This evidence makes sense because failure probability has a strong negative relation with protability (see equation (A.1)), meaning that more distressed rms are less protable than less distressed rms. The average portfolio ROA decreases monotonically from 11.20% per annum for the low distress decile to 12.32% for the high distress decile, and the ROA-spread of 23.52% is more than ten standard errors from zero (see the Internet Appendix). Panel B of Table 3 reports similar results for deciles formed on O-score. The high O-score decile underperforms the low O-score decile by an average of 0.92% per month (t = 2.84), even though the high decile has a higher market beta than the low decile, 1.38 versus 1.02. The CAPM alpha for the H-L portfolio is 1.10% per month (t = 3.56). The high decile also has signicantly higher size and value factor loadings than the low decile, producing a H-L Fama-French alpha of 1.44% per month (t = 6.49). More important, the new model eliminates the abnormal return: the alpha is reduced to a tiny 0.09% per month (t = 0.32). Again, the driving force is the large and negative rROA -loading of 1.07 (t = 11.03) for the H-L portfolio. The average portfolio ROA decreases monotonically from 9.68% per annum for the low O-score decile to 20.60% for the high decile, and the ROA-spread of 30.16% is more than ten standard errors from zero (see the Internet Appendix). In all, the evidence suggests that the distress eect is largely driven by the positive ROA-return relation. Once we control for ROA in factor regressions, the distress eect largely disappears.

4.3

Net Stock Issues

In June of each year t, we sort all NYSE, Amex, and NASDAQ stocks into ten deciles based on net stock issues at the last scal yearend. Following Fama and French (2008), we measure net stock issues as the natural log of the ratio of the split-adjusted shares outstanding at the scal yearend in t1 divided by the split-adjusted shares outstanding at the scal yearend in t2. The split-adjusted shares outstanding is shares outstanding (item 25) times the adjustment factor (item 27). Monthly value-weighted portfolio returns are calculated from July of year t to June of year t+1. From Panel A of Table 4, rms with high net issues earn lower average returns than rms with low net issues, 0.16% vs. 1% per month. The H-L portfolio earns an average return of 0.84% per month (t = 4.64), a CAPM alpha of 1.06% (t = 5.07), and a Fama-French alpha of 0.82% per month (t = 4.33).

17

The new model outperforms traditional factor models in explaining the net issues eect. Although the model is rejected by the GRS test, the H-L net issues decile earns an alpha of 0.28% per month (t = 1.39). The H-L portfolio has an rIN V -loading of 0.55 (t = 4.25), going in the right direction in explaining the average returns. This loading pattern is consistent with the underlying investment pattern. The average portfolio I/A increases virtually monotonically from 6.26% per annum for the low decile to 30.83% for the high decile, and the I/A-spread of 24.58% is more than ten standard errors from zero (see the Internet Appendix). Intriguingly, the rROA -loading also goes in the right direction: the H-L portfolio has an rROA -loading of 0.39 (t = 6.53). At the portfolio formation the high decile has a signicantly lower average ROA than the low decile. This evidence diers from Loughran and Ritters (1995) that equity issuers are more protable than nonissuers. While Loughran and Ritter only examine new issues, net stock issues also include share repurchases. Our evidence makes sense in light of Lie (2005), who shows that rms announcing repurchases exhibit superior operating performance relative to industry peers.

4.4

Asset Growth

In June of each year t we sort all NYSE, Amex, and NASDAQ stocks into ten deciles based on asset growth at the scal yearend of t1. Following Cooper, Gulen, and Schill (2008), we measure asset growth as total assets (Compustat annual item 6) at scal yearend of t 1 minus total assets at scal yearend of t 2 divided by total assets at scal yearend of t 2. Panel B of Table 4 reports that the high asset growth decile earns a lower average return than the low decile with a spread of 1.04% per month (t = 5.19). The H-L portfolio earns a CAPM alpha of 1.16% (t = 5.92) and a Fama-French alpha of 0.65% per month (t = 3.57). The new model reduces the magnitude of the H-L alpha to 0.55% per month (t = 3.06). While the Fama-French model gets its explanatory power from the value factor, our model works through the investment factor. The H-L portfolio has an rIN V -loading of 1.38 (t = 15.04). The average portfolio I/A increases monotonically from 8.83% per annum for the low asset growth decile to 6.39% for the fth decile and to 42.56% per annum for the high asset growth decile. The spread of 51.40% per annum is highly signicant (see the Internet Appendix). Both asset growth and I/A capture rm-level investments, and rIN V fails to fully capture the asset growth eect probably because asset growth is a more comprehensive measure of investment than I/A.

18

4.5

Earnings Surprises

The new factor model outperforms traditional models in capturing the earnings surprise eect. Following Chan, Jegadeesh, and Lakonishok (1996), we dene Standardized Unexpected Earnings (SU E) as the change in quarterly earnings (Compustat quarterly item 8) per share from its value four quarters ago divided by the standard deviation of the change in quarterly earnings over the prior eight quarters. We rank all NYSE, Amex, and NASDAQ stocks each month based on their most recent past SU E. Monthly value-weighted portfolio returns are calculated for the current month, and the portfolios are rebalanced monthly. From Panel A of Table 5, the H-L SU E portfolio earns an average return of 1.18% per month (t = 8.34), a CAPM alpha of is 1.22% (t = 8.76), and a FamaFrench alpha of 1.22% (t = 8.19). The new factor model reduces the alpha to 0.90% (t = 6.52), which only represents a modest reduction of 27% from the Fama-French alpha. The H-L portfolio has a positive rIN V -loading of 0.16 (t = 1.87) and a positive rROA -loading of 0.23 (t = 4.61). While we follow Chan, Jegadeesh, and Lakonishok (1996) in constructing the SU E portfolios on the most recent past earnings, we impose a four-month lag between the sorting-eective earnings and the return holding period in constructing the ROA factor. Our conservative timing (to guard against look-ahead bias) partially explains why rROA is only modestly successful in accounting for the SU E eect. In Panel B of Table 5 we reconstruct the SU E portfolios while imposing the four-month lag. The H-L SU E portfolio earns only 0.52% per month (t = 3.61), but it still cannot be explained by the CAPM or the Fama-French model with alphas of 0.57% and 0.62% (t = 3.98 and 4.03), respectively. Both models are rejected by the GRS test at the 1% level. The new model reduces the H-L alpha to 0.33% (t = 2.24), and the model is not rejected by the GRS test.

4.6

Book-to-Market Equity

Table 6 reports factor regressions of Fama and Frenchs (1993) 25 size and book-to-market portfolios. The portfolio data are from Kenneth Frenchs Web site. Value stocks earn higher average returns than growth stocks. The average H-L return is 1.09% per month (t = 5.08) in the small-size quintile and is 0.25% (t = 1.20) in the big-size quintile. The small-stock H-L portfolio has a CAPM alpha of 1.32% per month (t = 7.10). The Fama-French model reduces the small-stock H-L alpha to 0.68% per month, albeit still signicant (t = 5.50). The new model performs roughly as well as the Fama-French model: the small-stock H-L earns an alpha of 0.57% per month (t = 2.72). The new model does exceptionally well in capturing the low average returns of the small-growth portfolio. This portfolio earns a CAPM alpha of 0.63% per month (t = 2.61), a Fama-French 19

alpha of 0.52% (t = 4.48), but only a tiny alpha of 0.08% (t = 0.27) in our model.10 The rIN V - and rROA -loadings shed light on the explanatory power. From Panel B of Table 6, value stocks have higher rIN V -loadings than growth stocks. The loading spreads, ranging from 0.68 to 0.93, are all at least ve standard errors from zero. This evidence shows that growth rms invest more than value rms, consistent with Fama and French (1995). The rROA -loading pattern is more complicated. In the small-size quintile, the H-L portfolio has a positive loading of 0.39 (t = 4.53) because the small-growth portfolio has a large negative loading of 0.62 (t = 5.65). However, in the big-size quintile, the H-L portfolio has an insignicantly negative rROA -loading of 0.11. It is somewhat surprising that the small-growth portfolio has a lower rROA -loading than the small-value portfolio. Using an updated sample through 2006, the Internet Appendix documents that growth rms indeed have persistently higher ROAs than value rms in the big-size quintile both in the event time and in the calendar time. In the small-size quintile, however, growth rms have higher ROAs than value rms before, but lower ROAs after, the portfolio formation. In the calendar time, a dramatic downward spike of ROA appears for the small-growth portfolio over the past decade. This downward spike explains their abnormally low rROA -loadings.

4.7

Industries, CAPM Betas, and Market Equity

Lewellen, Nagel, and Shanken (2008) argue that asset pricing tests are often misleading because apparently strong explanatory power (such as high cross-sectional R2 s) provides quite weak support for a model. Our tests are largely immune to this critique because we focus on the intercepts from factor regressions as a yardstick for evaluating competing models. Following Lewellen et al.s prescription, we also confront our model with a wide array of testing portfolios formed on characteristics other than size and book-to-market. We test the model further with industry and CAPM beta portfolios. Because these portfolios do not display much cross-sectional variation in average returns, the models performance is roughly comparable with that of the CAPM and the Fama-French Model. From Panel A of Table 7, the CAPM captures the returns of ten industry portfolios with an insignicant GRS statistic of 1.35. Both the Fama-French model and our model are rejected by the GRS test, probably because the regression R2 s are higher than those from the CAPM, so even an economically small deviation from the null is statistically signicant. The average magnitude of the alphas is comparable across three models: 0.16% in the CAPM, 0.20% in the Fama-French model,
The small-growth eect is notoriously dicult to explain. Campbell and Vuolteenaho (2004), for example, show that the small-growth portfolio is particularly risky in their two-beta model: it has higher cash-ow and discount-rate betas than the small-value portfolio. As a result, their two-beta model fails to explain the small-growth eect.
10

20

and 0.23% in the new model. Panel B shows that none of the models are rejected by the GRS test using the ten portfolios formed on pre-ranking CAPM betas. The average magnitude of the alphas is again comparable: 0.20% in the CAPM, 0.14% in the Fama-French model, and 0.16% in our model. Panel C reports a weakness of the new model. Small rms earn slightly higher average returns than big rms. The average return, CAPM alpha, and the Fama-French alpha for the small-minusbig portfolio are smaller than 0.30% in magnitude and are within 1.2 standard errors of zero. The new model delivers an alpha of 0.53%, albeit insignicant, and the model is not rejected by the GRS test. The new model inates the size premium because small rms have lower rROA -loadings than big rms, going in the wrong direction in explaining returns. However, this weakness also is the strength that allows the new model to fully capture the low average returns of small-growth rms.

Summary and Interpretation

We oer a new factor model consisting of the market factor, a low-minus-high investment factor, and a high-minus-low ROA factor. The models empirical performance is remarkable. With only three factors, the model captures many patterns anomalous to the Fama-French model, and performs roughly as well as their model in capturing the portfolio returns which Fama and French (1996) show that their model is capable of explaining. Our pragmatic approach means that the new factor model can be used in many applications that require expected return estimates. The list includes evaluating mutual fund performance, measuring abnormal returns in event studies, estimating expected returns for asset allocation, and calculating costs of equity for capital budgeting and stock valuation. These applications primarily depend on the models performance, and the q-theory intuition also raises the likelihood that the performance can persist in the future. We interpret our new factor model as providing a parsimonious description of the cross-section of expected stock returns. In particular, diering from Fama and French (1993, 1996), who interpret their similarly constructed SM B and HM L as risk factors in the context of ICAPM or APT, we do not interpret our new factors as risk factors. On the one hand, q-theory allows us to tie expected returns with rm characteristics in an economically meaningful way, without assuming mispricing. Unlike size and book-to-market that directly involve market equity, which behaviorists often use as a proxy for mispricing (e.g., Daniel, Hirshleifer, and Subrahmanyam (2001)), the new factors are constructed on economic fundamentals that are less likely to be aected by mispricing, at least directly. On the other hand, our tests are only (intuitively) motivated from q-theory, and

21

are not formal structural evaluation of the theory. More important, q-theory is silent on investors behavior, which can be rational or irrational. As such, our tests are not aimed to, and cannot, distinguish whether the anomalies are driven by rational or behavioral forces. We also have conducted horse races between covariances and characteristics following the research design of Daniel and Titman (1997, Table III). We nd that after controlling for investmentto-assets, investment factor loadings are not related to average returns, but controlling for investment factor loadings does not aect the explanatory power of investment-to-assets. Similarly, after controlling for ROA, ROA factor loadings are not related to average returns, but controlling for ROA factor loadings does not aect the explanatory power of ROA (see the Internet Appendix). Consistent with Daniel and Titman, the evidence suggests that low-investment stocks and high-ROA stocks have high average returns whether or not they have similar return patterns (covariances) of other low-investment and high-ROA stocks. And factor loadings do not explain the high returns associated with low-investment and high-ROA stocks beyond the extent to which the loadings act as proxies for the characteristics. We reiterate that, deviating from Fama and French (1993, 1996) but echoing Daniel and Titman (1997), we do not interpret investment and ROA factor loadings as risk measures in the context of the ICAPM or APT. As noted, we view the new factor model agnostically as a parsimonious description of cross-sectional returns. The factor loadings explain returns because the factors are constructed on characteristics. In our view time series and cross-sectional regressions as largely equivalent ways of summarizing empirical correlations. If a characteristic is signicant in crosssectional regressions, its factor is likely to be signicant in time series regressions. And if a factor is signicant in time series regressions, its characteristic is likely to be signicant in cross-sectional regressions. Factor loadings are no more primitive than characteristics, and characteristics are no more primitive than factor loadings. The evidence in Daniel and Titman (1997) is sometimes interpreted as saying that risk does not determine expected returns. In our view this interpretation is too strong. Theoretically, q-theory predicts an array of relations between characteristics and expected returns, as observed in the data (see equation (3) and Section 2). The simple derivation of that equation is not based on mispricing, and is potentially consistent with the risk hypothesis. (We rush to point out that the theoretical analysis does not refute mispricing either.) Empirically, it is not inconceivable that characteristics provide more precise estimates of the true betas than the estimated betas (e.g., Miller and Scholes

22

(1972)). In particular, the betas are estimated with rolling-window regressions run between 42 months and 6 months prior to the formation date (June of year t) (Daniel and Titman, p. 18), and are in eect average betas at 24 months prior to portfolio formation. It seems reasonable to imagine that it would be hard, for example, for the 24-month-lagged ROA factor loading to compete with four-month-lagged ROA in explaining monthly returns.11 Future work can sort out the dierent interpretations. However, because true conditional betas are unobservable in reality, reaching a denitive verdict is virtually impossible.

References Altman, Edward I., 1968, Financial ratios, discriminant analysis and the prediction of corporate bankruptcy, Journal of Finance 23, 589609. Anderson, Christopher W., and Luis Garcia-Feijo, 2006, Empirical evidence on capital investo ment, growth options, and security Journal of Finance 61, 171194. Ang, Andrew and Joseph Chen, 2007, CAPM over the long run: 19262001, Journal of Empirical Finance 14, 140. Berk, Jonathan B, Richard C. Green, and Vasant Naik, 1999, Optimal investment, growth options, and security returns, Journal of Finance 54, 11531607. Brealey, Richard A., Stewart C. Myers, and Franklin Allen, 2006, Principles of corporate nance, 8th edition, Irwin McGraw-Hill. Breeden, Douglas T., Michael R. Gibbons, and Robert H. Litzenberger, 1989, Empirical tests of the consumption-oriented CAPM, Journal of Finance 44, 231262. Campbell, John Y. and Tuomo Vuolteenaho, 2004, Bad beta, good beta, American Economic Review 94, 12491275. Campbell, John Y., Jens Hilscher, and Jan Szilagyi, 2008, In search of distress risk, Journal of Finance 63, 28992939. Carhart, Mark M. 1997, On persistence in mutual fund performance, Journal of Finance 52, 5782. Carlson, Murray, Adlai Fisher, and Ron Giammarino, 2004, Corporate investment and asset price dynamics: Implications for the cross section of returns, Journal of Finance 59, 25772603. Chan, Louis K. C., Narasimhan Jegadeesh, and Josef Lakonishok, 1996, Momentum strategies, Journal of Finance 51, 16811713.
The conditioning approach uses up-to-date information to estimate betas (e.g., Harvey (1989, 1991), Shanken (1990), and Ferson and Harvey (1993, 1999)). However, linear specications likely contain specication errors due to nonlinearity (e.g., Harvey (2001)), and the conditional CAPM often performs no better than the unconditional CAPM (e.g., Ghysels (1998) and Lewellen and Nagel (2006)). Ang and Chen (2007) and Kumar, Srescu, Boehme, and Danielsen (2008) document better news for the conditional CAPM, however.
11

23

Cochrane, John H., 1991, Production-based asset pricing and the link between stock returns and economic uctuations, Journal of Finance 46, 209237. Cooper, Michael J., Huseyin Gulen, and Michael J. Schill, 2008, Asset growth and the cross-section of stock returns, Journal of Finance 63, 16091652 Daniel, Kent D., David Hirshleifer, and Avanidhar Subrahmanyam, 2001, Overcondence, arbitrage, and equilibrium asset pricing, Journal of Finance 56, 921965. Daniel, Kent D. and Sheridan Titman, 1997, Evidence on the characteristics of cross sectional variation in stock returns, Journal of Finance 52, 133. Daniel, Kent D. and Sheridan Titman, 2006, Market reactions to tangible and intangible information, Journal of Finance 61, 16051643. DeBondt, Werner F. M. and Richard Thaler, 1985, Does the stock market overreact? Journal of Finance 40, 793805. Dichev, Ilia, 1998, Is the risk of bankruptcy a systematic risk? Journal of Finance 53, 11411148. Fama, Eugene F. and Kenneth R. French, 1992, The cross-section of expected stock returns, Journal of Finance 47, 427465. Fama, Eugene F. and Kenneth R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, 356. Fama, Eugene F. and Kenneth R. French, 1995, Size and book-to-market factors in earnings and returns, Journal of Finance 50, 131155. Fama, Eugene F. and Kenneth R. French, 1996, Multifactor explanation of asset pricing anomalies, Journal of Finance 51, 5584. Fama, Eugene F. and Kenneth R. French, 1997, Industry cost of equity, Journal of Financial Economics 43, 153193. Fama, Eugene F. and Kenneth R. French, 2006, Protability, investment, and average returns, Journal of Financial Economics 82, 491518. Fama, Eugene F. and Kenneth R. French, 2008, Dissecting anomalies, Journal of Finance 63, 16531678. Fama, Eugene F. and Merton H. Miller, 1972, The Theory of Finance, Holt, Rinehart, and Winston. Ferson, Wayne E. and Campbell R. Harvey, 1992, Seasonality and consumption-based asset pricing, Journal of Finance 47, 511552. Ferson, Wayne E. and Campbell R. Harvey, 1993, The risk and predictability of international equity returns, Review of Financial Studies 6, 527566. Ferson, Wayne E. and Campbell R. Harvey, 1999, Conditional variables and the cross section of stock returns, Journal of Finance 54, 13251360. Fisher, Irving, 1930, The Theory of Interest: As Determined by Impatience to Spend Income and Opportunity to Invest It, 1st ed., New York: The Macmillan Co. 24

Ghysels, Eric, 1998, On stable factor structures in the pricing of risk: Do time-varying betas help or hurt? Journal of Finance 53, 549573. Gibbons, Michael R., Stephen A. Ross, and Jay Shanken, 1989, A test of the eciency of a given portfolio, Econometrica 57, 11211152. Gordon, Myron J., 1962, The Investment, Financing, and Valuation of the Corporation, The Irwin Series in Economics, Homewood, Illinois. Grin, John M. and Michael L. Lemmon, 2002, Book-to-market equity, distress risk, and stock returns, Journal of Finance 57, 23172336. Harvey, Campbell R., 1989, Time-varying conditional covariances in tests of asset pricing models, Journal of Financial Economics 24, 289317. Harvey, Campbell R., 1991, The world price of covariance risk, Journal of Finance 46, 111157. Harvey, Campbell R., 2001, The specication of conditional expectations, Journal of Empirical Finance 8, 573637. Ikenberry, David, Josef Lakonishok, and Theo Vermaelen, 1995, Market underreaction to open market share repurchases, Journal of Financial Economics 39, 181208. Jegadeesh, Narasimhan and Sheridan Titman, 1993, Returns to buying winners and selling losers: Implications for stock market eciency, Journal of Finance 48, 6591. Kumar, Praveen, Sorin M. Sorescu, Rodney D. Boehme, and Bartley R. Danielsen, 2008, Estimation risk, information, and the conditional CAPM: Theory and evidence, Review of Financial Studies 21, 10371075. Lakonishok, Josef, Andrei Shleifer, Robert W. Vishny, 1994, Contrarian investment, extrapolation, and risk, Journal of Finance 49, 15411578. Lettau, Martin and Sydney Ludvigson, 2001, Resurrecting the (C)CAPM: A cross-sectional test when risk premia are time-varying, Journal of Political Economy 109, 12381287. Lewellen, Jonathan, Stefan Nagel, and Jay Shanken, 2008, A skeptical appraisal of asset-pricing tests, forthcoming, Journal of Financial Economics. Li, Erica X. N., Dmitry Livdan, and Lu Zhang, 2008, Anomalies, forthcoming, Review of Financial Studies. Lie, Erik, 2005, Operating performance following open market share repurchase announcements, Journal of Accounting and Economics 39, 411436. Lintner, John, 1965, The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets, Review of Economics and Statistics 47, 1337. Liu, Laura Xiaolei, Toni M. Whited, and Lu Zhang, 2008, Investment-based expected stock returns, working paper, University of Michigan. Loughran, Tim, and Jay R. Ritter, 1995, The new issues puzzle, Journal of Finance 50, 2351. Lyandres, Evgeny, Le Sun, and Lu Zhang, 2008, The new issues puzzle: Testing the investmentbased explanation, Review of Financial Studies 21, 28252855. 25

Miller, Merton H. and Myron S. Scholes, 1972, Rate of return in relation to risk: A reexamination of some recent ndings, in Studies in the Theory of Capital Markets, Michael C. Jensen ed. New York: Praeger. Moskowitz, Tobias J. and Mark Grinblatt, 1999, Do industries explain momentum? Journal of Finance 54, 12491290. Ohlson, James A., 1980, Financial ratios and the probabilistic prediction of bankruptcy, Journal of Accounting Research 18, 109131. Ritter, Jay R., 1991, The long-run performance of initial public oerings, Journal of Finance 46, 327. Rosenberg, Barr, Kenneth Reid, and Ronald Lanstein, 1985, Persuasive evidence of market inefciency, Journal of Portfolio Management 11, 917. Shanken, Jay, 1990, Intertemporal asset pricing: An empirical investigation, Journal of Econometrics 45, 99120. Sharpe, William F., 1964, Capital asset prices: A theory of market equilibrium under conditions of risk, Journal of Finance 19, 425442. Spiess, Katherine D. and John Aeck-Graves, 1995, Underperformance in long-run stock returns following seasoned equity oerings, Journal of Financial Economics 38, 243267. Titman, Sheridan, K. C. John Wei, and Feixue Xie, 2004, Capital investments and stock returns, Journal of Financial and Quantitative Analysis 39, 677700. Tobin, James, 1969, A general equilibrium approach to monetary theory, Journal of Money, Credit, and Banking 1, 1529. Xing, Yuhang, 2008, Interpreting the value eect through the Q-theory: An empirical investigation, Review of Financial Studies 21, 17671795.

The Distress Measures

We construct the distress measure following Campbell, Hilscher, and Szilagyi (2008, the third column in Table 4): Distress(t) 9.164 20.264 N IM T AAV Gt + 1.416 T LM T At 7.129 EXRET AV Gt + 1.411 SIGM At 0.045 RSIZEt 2.132 CASHM T At + 0.075 M Bt 0.058 P RICEt (A.1) 1 2 N IM T AAV Gt1,t12 N IM T At1,t3 + + 9 N IM T At10,t12 (A.2) 1 12 1 EXRET AV Gt1,t12 EXRETt1 + + 11 EXRETt12 (A.3) 1 12 The coecient = 21/3 , meaning that the weight is halved each quarter. N IM T A is net income (Compustat quarterly item 69) divided by the sum of market equity and total liabilities 26

(item 54). The moving average N IM T AAV G is designed to capture the idea that a long history of losses is a better predictor of bankruptcy than one large quarterly loss in a single month. EXRET log(1 + Rit ) log(1 + RS&P500,t ) is the monthly log excess return on each rms equity relative to the S&P 500 index. The moving average EXRET AV G is designed to capture the idea that a sustained decline in stock market value is a better predictor of bankruptcy than a sudden stock price decline in a single month. T LM T A is the ratio of total liabilities divided by the sum of market equity and total liabilities. SIGM A is the volatility of each rms daily stock return over the past three months. RSIZE is the relative size of each rm measured as the log ratio of its market equity to that of the S&P 500 index. CASHM T A, used to capture the liquidity position of the rm, is the ratio of cash and short-term investments divided by the sum of market equity and total liabilities. M B is the market-to-book equity. P RICE is the log price per share of the rm. We follow Ohlson (1980, Model One in Table 4) to construct O-score: 1.32 0.407 log(M KT ASSET /CP I) + 6.03 T LT A 1.43 W CT A + 0.076 CLCA 1.72 OEN EG 2.37 N IT A 1.83 F U T L + 0.285 IN T W O 0.521 CHIN (A.4) M KT ASSET is market assets dened as book asset with book equity replaced by market equity. We calculate M KT ASSET as total liabilities + Market Equity + 0.1(Market Equity Book Equity), where total liabilities are given by Compustat quarterly item 54. The adjustment of M KT ASSET using ten percent of the dierence between market equity and book equity follows Campbell, Hilscher, and Szilagyi (2008) to ensure that assets are not close to zero. The construction of book equity follows Fama and French (1993). CP I is the consumer price index. T LT A is the leverage ratio dened as the book value of debt divided by M KT ASSET . W CT A is working capital divided by market assets, (item 40 item 49)/M KT ASSET . CLCA is current liability (item 40) divided by current assets (item 49). OEN EG is one if total liabilities exceeds total assets and is zero otherwise. N IT A is net income (item 69) divided by assets, M KT ASSET . F U T L is the fund provided by operations (item 23) divided by liability (item 54). IN T W O is equal to one if net income (item 69) is negative for the last two years and zero otherwise. CHIN is (N It N It1 )/(|N It | + |N It1 |), where N It is net income (item 69) for the most recent quarter.

27

Table 1 : Properties of the Investment Factor, rIN V , and the ROA Factor, rROA , 1/197212/2006, 420 Months
Investment-to-assets (I/A) is annual change in gross property, plant, and equipment (Compustat annual item 7) plus annual change in inventories (item 3) divided by lagged book assets (item 6). In each June we break NYSE, Amex, and NASDAQ stocks into three I/A groups using the breakpoints for the low 30%, middle 40%, and high 30% of the ranked I/A. We also use median NYSE size to split NYSE, Amex, and NASDAQ stocks into two groups, small and big. Taking intersections, we form six size-I/A portfolios. Monthly value-weighted returns on the six portfolios are calculated from July of year t to June of year t+1, and the portfolios are rebalanced in June of year t+1. rINV is the dierence (low-minus-high), each month, between the average returns on the two low-I/A portfolios and the average returns on the two high-I/A portfolios. Return on assets (ROA) is quarterly earnings (Compustat quarterly item 8) divided by one-quarter-lagged assets (item 44). Each month from January 1972 to December 2006, we sort NYSE, Amex, and NASDAQ stocks into three groups based on the breakpoints for the low 30%, middle 40%, and the high 30% of the ranked quarterly ROA from at least four months ago. We also use the NYSE median each month to split NYSE, Amex, and NASDAQ stocks into two size groups. We form six portfolios from the intersections of the two size and the three ROA groups. Monthly value-weighted returns on the six portfolios are calculated for the current month, and the portfolios are rebalanced monthly. rROA is the dierence (high-minus-low), each month, between the simple average of the returns on the two high-ROA portfolios and the simple average of the returns on the two low-ROA portfolios. In Panel A we regress rINV and rROA on traditional factors including the market factor, SM B, HM L, and W M L (from Kenneth Frenchs Web site). The t-statistics (in parentheses) are adjusted for heteroscedasticity and autocorrelations. Panel B reports the correlation matrix of the new factors and the traditional factors. The p-values (in parentheses) test the null hypothesis that a given correlation is zero.

28
Panel A: Means and factor regressions of rINV and rROA Mean rINV 0.43 (4.75) 0.51 (6.12) 0.33 (4.23) 0.22 (2.87) 1.05 (5.61) 1.01 (5.60) 0.74 (4.16) M KT 0.16 (8.83) 0.09 (4.79) 0.08 (4.11) 0.16 (4.00) 0.05 (1.23) 0.02 (0.38) SM B HM L W M L R
2

Panel B: Correlation matrix (p-value in parenthesis) rROA rINV rROA rM KT SM B HM L 0.10 (0.05) rM KT 0.40 (0.00) 0.19 (0.00) SM B 0.09 (0.07) 0.38 (0.00) 0.26 (0.00) HM L 0.51 (0.00) 0.22 (0.00) 0.45 (0.00) 0.29 (0.00) W ML 0.20 (0.00) 0.26 (0.00) 0.07 (0.14) 0.02 (0.62) 0.11 (0.02) 0.16

0.06 (2.27) 0.05 (2.29)

0.27 (9.47) 0.29 (10.65)

0.31 0.10 (5.89) 0.36 0.04

rROA

0.96 (5.10)

0.40 (7.14) 0.41 (7.56)

0.11 (1.74) 0.18 (2.81)

0.31 0.26 (6.43) 0.24

Table 2 : Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on 25 Size and Momentum Portfolios, 1/197212/2006, 420 Months
The data for the one-month Treasury bill rate (rf ) and the Fama-French factors are obtained from Kenneth Frenchs Web site. The monthly constructed size and momentum portfolios are the intersections of ve portfolios formed on market equity and ve portfolios formed on prior 27 month returns. The monthly size breakpoints are the NYSE market equity quintiles. For each portfolio formation month t, we sort stocks on their prior returns from month t2 to t7 (skipping month t1), and calculate the subsequent portfolio returns from month t to t+5. All portfolio returns are value-weighted. Panel A reports mean percent excess returns and their t-statistics, CAPM alphas () and their t-statistics, and the intercepts (j F ) and their t-statistics from Fama-French three-factor regressions. F
j j j Panel B reports the new three-factor regressions: r j rf = j + M KT rM KT + INV rINV + ROA rROA + j . See Table 1 for the description of rINV and rROA . q

All the t-statistics are adjusted for heteroscedasticity and autocorrelations. FGRS is the Gibbons, Ross, and Shanken (1989) F -statistic testing that the intercepts of all 25 portfolios are jointly zero, and pGRS is its associated p-value. We only report the results of quintiles 1, 3, and 5 for size and momentum to save space. W-L is the winner-minus-lower quintile.

29
Small 3 Big Small 3 Big Small 3 Big

Loser

Winner

W-L

Loser

Winner

W-L

Loser
j

3 q

Winner j q +

W-L
j M KT rM KT

Loser +
j INV

3 rINV +

Winner
j ROA rROA

W-L + j j 1.70 0.77 1.13 4.66 3.75 3.35 3.48 2.62 1.72

Panel A: Means, CAPM alphas, and Fama-French alphas Mean 0.04 0.03 0.22 0.59 0.51 0.69 0.93 0.62 0.60 0.80 0.58 0.29 0.40 0.18 0.07 F F 0.05 0.17 0.05 0.73 0.48 0.25 0.51 0.47 0.46 1.33 1.00 0.94 1.44 1.09 1.06 1.21 0.98 0.68 1.25 0.95 0.90 tMean 0.09 2.78 0.08 2.29 0.65 1.37 t (FGRS = 3.28, 3.54 5.49 3.03 3.63 2.46 3.17 pGRS = 0) 5.78 3.67 3.15 = 0) 5.54 3.57 3.19

Panel B: r rf = 0.38 0.35 0.10 0.31 0.58 0.67 0.80 0.53 0.21 0.49 0.12 0.10 INV

tq (FGRS = 2.20, pGRS = 0) 0.92 0.63 0.31 0.54 0.28 0.41 0.60 0.58 0.57 0.45 0.39 0.22 1.04 1.23 0.38 1.70 3.90 5.16 6.62 5.36 2.41 2.40 0.94 1.15 tIN V 3.75 3.12 1.99

2.01 2.22 3.39 2.21 1.56 2.88 3.08 0.91 1.86 tF F (FGRS = 3.40, pGRS 3.67 2.58 2.41 0.54 1.91 0.74 4.89 4.26 3.31

0.25 0.29 0.05 0.00 0.11 0.10 ROA 0.24 0.03 0.09 0.35 0.14 0.00

2.21 2.25 0.59 0.04 2.56 1.38 tROA 3.56 0.67 2.89 4.01 1.75 0.07

Table 3 : Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on Deciles Formed on Campbell, Hilscher, and Szilagyis (2008) Failure Probability Measure and Deciles Formed on Ohlsons (1980) O-Score
The data on the one-month Treasury bill rate (rf ), the Fama-French three factors are from Kenneth Frenchs Web site. See Table 1 for the description of rINV and rROA . We sort all NYSE, Amex, and NASDAQ stocks at the beginning of each month into deciles based on failure probability and on O-score from at least four months ago. Monthly valueweighted returns on the portfolios are calculated for the current month, and the portfolios are rebalanced monthly. We report the average return in monthly percent and its t-statistics, the CAPM regression (r j rf = j + j rM KT + j ), the Fama-French three-factor regression (r j rf = j F +bj rM KT +sj SM B +hj HM L+j ), and the new three-factor F
j j j regression (r j rf = j + M KT rM KT + INV rINV + ROA rROA + j ). For each asset pricing model, we also report q

the Gibbons, Ross, and Shanken (1989) F -statistic (FGRS ) testing that the intercepts are jointly zero and its p-value (in parenthesis). All the t-statistics are adjusted for heteroscedasticity and autocorrelations. We only report the results of deciles 1 (Low), 5, 10 (High), and high-minus-low (H-L) to save space. Low 5 High H-L FGRS (p) Low 5 High H-L FGRS (p)

Panel A: The failure probability deciles (6/197512/2006, 379 months) Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 1.03 4.07 0.39 0.95 2.60 0.39 0.91 0.17 0.04 2.46 0.19 0.99 0.00 0.18 1.09 25.21 0.04 2.46 0.72 2.93 0.01 1.06 0.13 0.01 1.06 0.01 0.03 0.07 0.13 1.03 0.01 0.10 1.14 36.96 0.15 2.46 0.35 0.72 1.48 1.69 4.57 1.75 1.48 1.27 0.09 6.39 0.13 1.42 0.02 1.22 0.49 18.17 0.16 13.42 1.38 3.53 1.87 0.73 5.08 2.14 0.57 1.10 0.13 6.43 0.32 0.43 0.03 1.40 1.09 5.78 0.18 14.64 3.01 (0) 4.75 (0) 0.48 2.04 0.04 1.02 0.51 0.12 0.99 0.15 0.21 1.68 0.02 1.00 0.21 0.05 0.20 50.55 4.29 2.74

Panel B: The O-score deciles (1/197212/2006, 420 months) 0.50 2.03 0.00 1.00 0.01 0.24 1.03 0.33 0.32 2.36 0.02 1.00 0.07 0.06 0.19 31.23 0.93 1.55 0.44 1.04 1.14 1.38 3.96 1.32 1.16 1.35 0.10 6.39 0.07 1.21 0.01 1.02 0.29 17.43 0.07 10.48 0.92 2.84 1.10 0.36 3.56 1.44 0.17 1.50 0.32 6.49 0.09 0.22 0.20 1.07 0.32 2.93 1.18 11.03 2.49 (0.01) 6.33 (0)

1.78 (0.06)

1.10 (0.36)

30

Table 4 : Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on the Net Stock Issues Deciles and the Asset Growth Deciles, 1/197212/2006, 420 Months
The data on the one-month Treasury bill rate (rf ) and the Fama-French three factors are from Kenneth Frenchs Web site. See Table 1 for the description of rINV and rROA . We measure net stock issues as the the natural log of the ratio of the split-adjusted shares outstanding at the scal yearend in t1 (Compustat annual item 25 times the Compustat adjustment factor, item 27) divided by the split-adjusted shares outstanding at the scal yearend in t2. In June of each year t, we sort all NYSE, Amex, and NASDAQ stocks into ten deciles based on the breakpoints of net stock issues measured at the end of last scal yearend. Monthly value-weighted returns are calculated from July of year t to June of year t + 1. In June of each year t, we sort all NYSE, Amex, and NASDAQ stocks into ten deciles based on asset growth measured at the end of last scal year t 1. Asset growth for scal year t 1 is the change of total assets (item 6) from the scal yearend of t 2 to yearend of t 1 divided by total assets at the scal yearend of t 2. Monthly value-weighted returns are calculated from July of year t to June of year t + 1. We report the average return in monthly percent and its t-statistics, the CAPM regression (r j rf = j + j rM KT + j ), the Fama-French three-factor regression (r j rf = j F + bj rM KT + sj SM B + hj HM L + j ), and the new three-factor F
j j j regression (r j rf = j + M KT rM KT + INV rINV + ROA rROA + j ). For each asset pricing model, we also report q

the Gibbons, Ross, and Shanken (1989) F -statistic (FGRS ) testing that the intercepts are jointly zero and its p-value (in parenthesis). All the t-statistics are adjusted for heteroscedasticity and autocorrelations. We only report the results of deciles 1 (Low), 5, 10 (High), and high-minus-low (H-L) to save space. Low 5 High H-L FGRS (p) Low 5 High H-L FGRS (p)

Panel A: The net stock issues deciles Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 1.00 4.73 0.42 0.88 3.68 0.22 0.99 0.01 0.32 2.39 0.09 0.96 0.11 0.21 0.90 45.73 1.66 5.06 0.82 3.61 0.17 0.99 1.98 0.13 1.01 0.00 0.08 1.36 0.24 0.96 0.17 0.02 2.49 42.35 3.47 0.53 0.16 0.55 0.64 1.21 4.34 0.59 1.14 0.26 0.07 3.89 0.19 1.08 0.43 0.18 1.10 29.85 4.74 4.09 0.84 4.64 1.06 0.33 5.07 0.82 0.15 0.25 0.39 4.33 0.28 0.12 0.55 0.39 1.39 2.67 4.25 6.53 3.97 (0) 3.10 (0) 1.10 3.48 0.49 1.21 2.92 0.17 1.20 0.65 0.40 1.15 0.45 1.26 0.59 0.25 2.49 27.15 5.99 4.06

Panel B: The asset growth deciles 0.63 3.03 0.18 0.89 2.59 0.01 0.98 0.00 0.26 0.10 0.03 0.94 0.24 0.03 0.41 45.42 5.19 0.85 0.05 0.15 0.67 1.43 4.77 0.48 1.27 0.31 0.33 3.84 0.10 1.28 0.79 0.16 0.72 43.03 9.47 4.25 1.04 5.19 1.16 0.23 5.92 0.65 0.07 0.34 0.72 3.57 0.55 0.02 1.38 0.09 3.06 0.44 15.04 1.30 5.82 (0) 3.71 (0)

2.67 (0)

3.05 (0)

31

Table 5 : Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on Deciles Formed on Most Recent (and Four-Month-Lagged) Standardized Unexpected Earnings (SU E), 1/197212/2006, 420 Months
The data on the one-month Treasury bill rate (rf ) and the Fama-French three factors are from Kenneth Frenchs Web site. See Table 1 for the description of rINV and rROA . We dene SU E as the change in quarterly earnings per share from its value announced four quarters ago divided by the standard deviation of earnings change over the prior eight quarters. In Panel A we rank all NYSE, Amex, and NASDAQ stocks into ten deciles at the beginning of each month by their most recent past SU E. Monthly value-weighted returns on the SU E portfolios are calculated for the current month, and the portfolios are rebalanced monthly. In Panel B we use the same procedure but instead of the most recent SU E we sort on the SU E from at least four months ago. We report the average return in monthly percent and its t-statistics, the CAPM regression (r j rf = j + j rM KT + j ), the Fama-French three-factor regression (r j rf = j F + bj rM KT + sj SM B + hj HM L + j ), and the new three-factor regression F
j j j (r j rf = j + M KT rM KT + INV rINV + ROA rROA + j ). For each asset pricing model, we also report the q

Gibbons, Ross, and Shanken (1989) F -statistic (FGRS ) testing that the intercepts are jointly zero and its p-value (in parenthesis). All the t-statistics are adjusted for heteroscedasticity and autocorrelations. We only report the results of deciles 1 (Low), 5, 10 (High), and high-minus-low (H-L) to save space. Low 5 High H-L FGRS (p) Low 5 High H-L FGRS (p)

Panel A: Deciles on most recent SU E Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 0.10 0.41 0.62 1.02 6.65 0.58 1.02 0.03 0.04 6.16 0.43 0.98 0.17 0.10 4.47 40.67 2.86 3.08 0.26 1.09 0.25 1.01 2.86 0.32 1.02 0.08 0.09 3.65 0.17 0.99 0.02 0.09 1.88 41.41 0.32 2.20 1.08 4.84 0.61 0.94 7.22 0.64 0.95 0.10 0.03 7.14 0.47 0.96 0.01 0.14 5.53 39.62 0.22 4.47 1.18 8.34 1.22 0.08 8.76 1.22 0.07 0.07 0.01 8.19 0.90 0.02 0.16 0.23 6.52 0.54 1.87 4.61 10.65 (0) 11.01 (0)

Panel B: Deciles on four-month-lagged SU E 0.34 1.36 0.18 1.04 1.83 0.16 1.05 0.06 0.02 1.57 0.02 1.00 0.14 0.12 0.22 31.24 1.97 3.45 0.32 1.37 0.18 1.00 2.10 0.20 1.00 0.03 0.04 2.28 0.11 0.98 0.01 0.08 1.26 41.53 0.18 2.01 0.86 3.86 0.39 0.94 4.52 0.45 0.94 0.13 0.08 5.21 0.30 0.96 0.04 0.13 3.51 39.09 0.73 4.40 0.52 3.61 0.57 0.10 3.98 0.62 0.11 0.08 0.06 4.03 0.33 0.04 0.10 0.25 2.24 0.88 1.14 5.03 3.65 (0) 4.60 (0)

5.56 (0)

1.79 (0.06)

32

Table 6 : Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on 25 Size and Book-to-Market Portfolios, 1/197212/2006, 420 Months
The data for the one-month Treasury bill rate (rf ), the Fama-French factors, and the 25 size and book-to-market portfolios are obtained from Kenneth Frenchs Web site. For all testing portfolios, Panel A reports mean percent excess returns and their t-statistics, CAPM alphas () and their t-statistics, and the intercepts
j j (F F ) and their t-statistics from Fama-French three-factor regressions. Panel B reports the new three-factor regressions: r j rf = j +M KT rM KT +INV rINV + q j ROA rROA + j . See Table 1 for the description of rINV and rROA . All the t-statistics are adjusted for heteroscedasticity and autocorrelations. FGRS is the

Gibbons, Ross, and Shanken (1989) F -statistic testing that the intercepts of all 25 portfolios are jointly zero, and pGRS is its associated p-value. We only report the results of quintiles 1, 3, and 5 for size and book-to-market to save space. H-L is the high-minus-low quintile. Low 3 High H-L Low 3 tMean 1.19 1.07 0.65 Small 3 Big Small 3 Big 0.63 0.27 0.11 0.52 0.03 0.17 0.37 0.70 0.27 0.59 0.16 0.25 F F 0.09 0.12 0.02 0.16 0.02 0.26 1.32 0.86 0.36 0.68 0.01 0.43 1.09 0.66 0.25 0.25 3.10 4.21 5.08 1.22 3.14 4.12 2.86 1.67 2.75 2.80 1.20 t (FGRS = 4.25, pGRS = 0) 2.61 2.15 3.82 1.74 2.32 3.71 1.29 1.54 1.61 tF F (FGRS = 3.08, pGRS 4.48 0.37 2.75 1.35 1.50 0.28 2.16 0.22 2.34 7.10 3.96 1.81 = 0) 5.50 0.08 3.34 0.08 0.19 0.11 0.11 0.43 0.26 0.62 0.23 0.12 High H-L Low 3
j

High j E q

H-L
j

Low
j I

3 rINV +
j R

High rROA +
j

H-L

33
Small 3 Big 0.10 0.41 0.40

Panel A: Means, CAPM alphas, and Fama-French alphas Mean 0.88 0.74 0.59

Panel B: r rf = 0.46 0.64 0.07 0.31 0.04 0.03 INV 0.35 0.58 0.24 0.50 0.14 0.42 ROA 0.26 0.07 0.12 0.23 0.02 0.01

+ M KT + 0.57 0.13 0.14 0.69 0.93 0.68

tq (FGRS = 2.72, pGRS = 0) 0.27 1.05 1.17 0.76 4.43 5.17 5.65 2.95 4.75 2.23 3.31 0.60 1.86 0.39 0.16 tIN V 3.22 4.68 3.39 4.20 2.19 3.76 tROA 3.00 1.55 2.61 3.50 0.25 0.11 2.72 0.57 0.61 5.63 7.03 5.05 4.53 2.10 1.23

0.39 0.24 0.11

Table 7 : Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on Ten Industry Portfolios and Deciles Formed on Pre-ranking CAPM Betas and the Market Equity, 1/197212/2006, 420 Months
The one-month Treasury bill rate (rf ), the Fama-French three factors, ten industry portfolio returns, and ten market equity portfolio returns are from Kenneth Frenchs Web site. See Table 1 for the description of rINV and rROA . We estimate pre-ranking CAPM betas on 60 (at least 24) monthly returns prior to July of year t. In June of year t we sort all stocks into ten deciles based on the pre-ranking betas. The value-weighted monthly returns on the resulting ten portfolios are calculated from July of year t to June of year t + 1. For each portfolio we report the average return in monthly percent and its t-statistics, the CAPM regression (r j rf = j + j rM KT + j ), the Fama-French three-factor regression (r j rf = j F + bj rM KT + sj SM B + hj HM L + j ), and the new three-factor regression F
j j j (r j rf = j + M KT rM KT + INV rINV + ROA rROA + j ). For each asset pricing model, we also report the q

Gibbons, Ross, and Shanken (1989) F -statistic (FGRS ) testing that the intercepts are jointly zero and its p-value (in parenthesis). All the t-statistics are adjusted for heteroscedasticity and autocorrelations. In Panels B and C we only report the results for deciles 1, 5, 10, and zero-investment portfolios to save space. Panel A: Ten industry portfolios NoDur Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 0.67 3.04 0.27 0.81 1.99 0.10 0.91 0.08 0.27 0.76 0.24 0.92 0.32 0.33 1.89 29.52 4.39 7.47 Low Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 0.48 2.26 0.16 0.62 0.95 0.16 0.75 0.08 0.48 0.94 0.07 0.67 0.15 0.15 0.39 12.22 1.61 2.26 Durbl 0.41 1.43 0.11 1.03 0.64 0.47 1.17 0.11 0.53 2.75 0.25 1.07 0.24 0.02 1.25 23.18 2.20 0.19 5 0.57 2.55 0.10 0.93 1.10 0.09 1.03 0.01 0.30 1.10 0.14 0.98 0.10 0.18 1.44 39.76 1.87 4.30 Manuf 0.56 2.31 0.05 1.01 0.49 0.08 1.08 0.03 0.20 0.90 0.20 1.06 0.07 0.20 2.08 45.36 1.10 5.91 High 0.37 0.80 0.53 1.79 2.23 0.31 1.50 0.82 0.45 1.54 0.47 1.57 0.70 0.62 1.93 26.35 4.84 8.16 Enrgy 0.76 2.82 0.37 0.77 1.79 0.17 0.91 0.20 0.33 0.82 0.30 0.76 0.22 0.18 1.27 14.75 1.56 2.75 H-L 0.10 0.24 0.69 1.17 2.10 0.15 0.75 0.74 0.92 0.53 0.54 0.91 0.85 0.77 1.58 9.74 4.72 7.32 1.60 (0.10) 1.23 (0.27) HiTec 0.49 1.44 0.17 1.32 0.98 0.22 1.09 0.21 0.64 1.45 0.47 1.18 0.51 0.37 2.70 29.79 5.32 7.23 FGRS (p) Telcm 0.55 2.35 0.17 0.75 1.02 0.16 0.81 0.22 0.04 0.90 0.26 0.76 0.25 0.21 1.32 18.57 2.15 3.27 Shops 0.55 2.03 0.02 1.04 0.16 0.09 1.06 0.12 0.16 0.63 0.21 1.08 0.02 0.22 1.46 25.34 0.19 4.64 Small 0.73 2.42 0.21 1.03 1.08 0.04 0.88 1.18 0.22 0.40 0.46 1.02 0.34 0.40 2.00 17.56 2.84 4.44 Hlth 0.58 2.31 0.15 0.85 0.91 0.41 0.81 0.35 0.35 2.50 0.10 0.88 0.08 0.28 0.51 17.39 0.80 3.89 5 0.71 2.60 0.15 1.12 1.30 0.02 1.05 0.68 0.16 0.36 0.29 1.10 0.02 0.15 2.29 30.06 0.35 2.72 Utils 0.51 2.52 0.25 0.50 1.47 0.13 0.72 0.15 0.61 0.89 0.01 0.56 0.22 0.15 0.07 12.06 2.26 2.14 Big 0.46 2.15 0.02 0.94 0.31 0.06 0.97 0.31 0.08 2.47 0.07 0.95 0.05 0.08 1.20 59.37 1.55 3.51 Other 0.59 2.36 0.06 1.04 0.64 0.17 1.16 0.04 0.35 1.81 0.24 1.11 0.25 0.17 2.38 45.91 4.89 4.28 S-B 0.28 1.16 0.23 0.09 0.96 0.10 0.10 1.49 0.31 1.11 0.53 0.08 0.39 0.48 1.91 1.07 2.68 4.39 1.79 (0.06) 1.82 (0.06) 1.35 (0.20) 2.88 (0.00) FGRS (p)

2.17 (0.02)

Panel B: Ten pre-ranking CAPM beta deciles

Panel C: Ten market equity deciles FGRS (p)

1.77 (0.06)

1.57 (0.11)

34

Figure 2 : Investment-to-Assets (I/A, Contemporaneous and Lagged) and ROA for the 25 Size and Momentum Portfolios, 1972:Q1 to 2006:Q4, 140 Quarters
We measure I/A as the annual change in gross property, plant, and equipment (Compustat annual item 7) plus the annual change in inventories (item 3) divided by the lagged book assets (item 6). ROA is quarterly earnings (Compustat quarterly item 8) divided by one-quarter-lagged assets (item 44). The 25 size and momentum portfolios are constructed monthly as the intersections of ve quintiles formed on market equity and ve quintiles formed on prior 27 month returns (skipping one month). For each portfolio formation month t = January 1972 to December 2006, we calculate annual I/As and quarterly ROAs for t + m, m = 60, . . . , 60. The I/A and ROA for month t + m are averaged across portfolio formation months t. ROA is the most recent ROA relative to formation month t. Panel A plots the median I/As across rms for the four extreme portfolios. In Panel B I/A is the current yearend I/A relative to month t. In Panel C the lagged I/A is the I/A on which an annual sorting on I/A in each June is based. Panel A: Event-time, annual I/A
10 9 8 7 6 BigWinner

Panel B: Calendar-time, annual I/A

20 18 16 14 12 10 8 SmallWinner

5 4 SmallWinner 3 SmallLoser 2 20 15 10 5 0 Quarter 5 10 15 20 BigLoser

6 4 2 0 2 1975 SmallLoser 1980 1985 1990 Year 1995 2000 2005

Panel C: Calendar-time, annual I/A (lagged)

20 18 16 14 12
1.2 2 1.8

Panel D: Event-time, quarterly ROA

BigWinner

SmallLoser

1.6 1.4 BigLoser

10 8 6 4 2 0 2 1975 SmallWinner 1980 1985 1990 Year 1995 2000 2005

1 0.8 0.6 0.4 0.2 SmallLoser 0 20 15 10 5 0 Quarter 5 10 15 20 SmallWinner

Panel E: Calendar-time, quarterly ROA

6 SmallWinner 4 4 6

Panel F: Calendar-time, quarterly ROA

BigWinner

35
4 1975 SmallLoser 1980 1985 1990 Year 1995 2000 2005 4 1975 1980 1985 1990 Year

BigLoser

1995

2000

2005

Internet Appendix for A Better Three-Factor Model That Explains More Anomalies

This Internet Appendix documents: the unabridged Tables 1 to 7 (the abridged tables are in the main text); the performance of the q-theory factor model using additional testing portfolios; the new three-factor regressions of the 25 size and momentum portfolios but with the investment factor constructed on quarterly investment data; additional evidence on the cross-sectional variation of investment-to-assets and ROA across additional testing portfolios; the evidence that the industry eect on the investment and ROA factors is relatively small; the evidence that the annually rebalanced SU E, short-term prior returns, and distress portfolios fail to produce signicant average returns for zero-investment strategies; and the results of the covariances vs. characteristics tests following the design of Daniel and Titman (1997).

Citation format: Chen, Long and Lu Zhang, 2009, A Better Three-Factor Model That Explains More Anomalies, Journal of Finance [-volume-], [-pages-], Internet Appendix http://www.afajof.org/supplements.asp. Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

I. Unabridged Tables 1 to 7

Tables I to VII report the unabridged Tables 1 to 7 in the main text. The unabridged Table 1 reports additional details of the six size-I/A portfolios and the six size-ROA portfolios. And the unabridged Tables 2 to 7 report the results for every decile or quintile of the testing portfolios. In contrast, the abridged tables only report the results for deciles 1, 5, and 10 and quintiles 1, 3, and 5 to save space in the main text.

II. Additional Testing Portfolios

A. The 25 I/A and ROA Portfolios We sort all NYSE, Amex, and NASDAQ stocks into ve ROA quintiles each month based on quarterly ROA from at least four months ago, and we sort all stocks independently in June of each year into ve quintiles based on I/A at the last scal yearend. Taking intersections yields 25 I/A and ROA portfolios. Their value-weighted returns are calculated for the current month, and the portfolios are rebalanced monthly. Table VIII reports that high ROA stocks earn higher average returns than low ROA stocks, especially among high investment rms, and that high investment stocks earn lower average returns than low investment stocks, especially among low ROA rms. The average high-minus-low ROA portfolio return varies from 0.80% per month (t = 2.20) in the lowest-I/A quintile to 1.65% (t = 4.97) in the highest-I/A quintile. The average low-minus-high I/A portfolio return varies from an insignicant 0.26% per month (t = 1.23) in the highest-ROA quintile to 1.11% (t = 3.41) in the lowest-ROA quintile. Traditional factor models cannot explain the 25 I/A and ROA portfolio returns. Ten out of 25 portfolios have signicant CAPM alphas. The alpha for the high-minus-low ROA portfolio ranges from 0.58% to 1.82% per month and are mostly signicant across the ve I/A quintiles. The CAPM alpha for the low-minus-high I/A portfolio ranges from 0.34% to 1.25% per month and are mostly 1

signicant across the ve ROA quintiles. Despite their higher average returns, high ROA rms have mostly lower SM B and HM L loadings than low ROA rms. As a result, 11 out of 25 portfolios have signicant Fama-French alphas. In particular, the alpha for the high-minus-low ROA portfolio ranges from 0.63% to 2.04% per month and are all signicant across the ve I/A quintiles. The new factor model does a much better job in explaining the average returns of these portfolios. From Panel B of Table VIII, although the model is rejected overall with a GRS statistic of 2.42 (p = 0), only three out of 25 alphas are signicant. Two out of ve high-minus-low ROA portfolios have signicant alphas, whereas four out of ve alphas are signicant in the CAPM, and all ve of them are signicant in the Fama-French model. More important, the average magnitude of the high-minus-low ROA alphas is lower in our model: 0.39% per month versus 1.15% in the CAPM and 1.27% in the Fama-French model. Further, only one out of the ve low-minus-high I/A alphas is signicant, whereas four out of ve are signicant in both the CAPM and the Fama-French model. The average magnitude of the low-minus-high I/A alphas is also lower in our model: 0.26% per month versus 0.79% in the CAPM and 0.59% in the Fama-French model. As expected, high ROA rms have signicantly higher rROA -loadings than low ROA rms, and low-investment rms have signicantly higher rIN V -loadings than high-investment rms. The systematic variation in the loadings across the portfolios in the same direction as their average returns explains the better performance of our model than the CAPM and the Fama-French (1993) model. B. The 25 Size and 11/1/1-Momentum Portfolios Table IX shows that using the 11/1/1 convention of momentum yields largely similar results as those with the 6/1/6 convention reported in the paper. C. Testing Portfolios Motivated from Fama and French (1996) Fama and French (1996) show that, except for momentum, their three-factor model captures well average return variations across portfolios sorted on earnings-to-price (E/P ), cash ow-to2

assets (C/P ), dividend-to-price (D/P ), past sales growth, and long-term prior returns. Table X shows that our models performance in explaining these average return variations is largely comparable to that of the Fama-French model. For ease of comparison with Fama and French (1996), we use portfolio data from Kenneth Frenchs Web site whenever possible. French provides portfolio data for the one-way deciles sorted on E/P, C/P, D/P , and prior 1360 month returns. We form the deciles on past ve-year sales growth (5-Yr SR) and market leverage (A/M E).

C.1.

The E/P, C/P , and D/P Deciles

From Panel A of Table X, the high-minus-low E/P portfolio is protable from January 1972 to December 2006. This portfolio generates an average return of 0.68% per month (t = 2.81) and a CAPM alpha of 0.81% (t = 3.42). The alpha disappears in the Fama-French (1993) three-factor regression, which produces an insignicant intercept of 0.11% per month. The reason is that high E/P stocks have higher loadings on HM L than low E/P stocks. Although its magnitude is higher than that from the Fama-French model, our model also delivers an insignicant alpha of 0.31% per month (t = 1.31). The main driving force is that high E/P stocks have higher loadings on rIN V than low E/P stocks with the spread of 0.56 signicant at the 1% level. The C/P and D/P results are largely similar to those on the E/P portfolios. The high-minuslow C/P and D/P average returns are lower, 0.50% and 0.10% per month, respectively, and the latter is insignicant. But both strategies generate signicant positive CAPM alphas, 0.64% and 0.45% per month. The Fama-French (1993) model reduces these alphas to insignicant levels because high C/P and D/P portfolios have signicantly higher loadings on HM L than low C/P and D/P loadings. Our model generates an insignicant alpha of 0.18% per month for the high-minuslow D/P portfolio (t = 0.76), but a marginally signicant alpha of 0.48% for the high-minus-low C/P portfolio (t = 2.09). The driving force is that high C/P and D/P stocks have signicantly

higher loadings on rIN V than low C/P and D/P stocks. C.2. The Long-Term Prior Returns and 5-Yr SR Deciles

Consistent with DeBondt and Thaler (1985), stocks with high prior 1360 month returns (longterm winner) earn lower average returns than stocks with low prior 1360 month returns (long-term losers). From Panel D of Table X, the average-return spread is 0.41% per month, and that the long-term winner-minus-loser portfolio has a marginally signicant alpha of 0.45%. The Fama and French (1993) model reduces the magnitude of the intercept to 0.24%. The long-term winners load negatively and long-term losers load positively on
IN V .

As a result, the long-term winner-

minus-loser has a negative rIN V -loading of 0.74, which goes in the right way in explaining its low average return. But the zero-investment portfolio has a signicantly positive rROA -loading of 0.44, which goes in the wrong way in explaining its low average return. Overall, the new factor model produces an alpha of 0.41% per month (t = 1.58) for the zero-investment portfolio. Consistent with Lakonishok, Shleifer, and Vishny (1994), stocks with high past ve-year sales growth (5-Yr SR) earn lower average returns than stocks with low past ve-year sales growth. The CAPM alpha of the high-minus-low 5-Yr SR portfolio is 0.55% per month (t = 2.53), although its average return of 0.35% is insignicant at the 5% level. The Fama-French (1993) model performs extremely well: the alpha of the zero-investment portfolio is only six basis points per month because of its large negative loading on HM L. The zero-investment portfolio has an insignicant positive alpha of 0.28% per month in the our three-factor model. The driving force is a large negative rIN V -loading of 1.17 and a small negative rROA -loading of 0.27. C.3. The Market Leverage Deciles

Bhandari (1988) and Fama and French (1992) report that stocks with high market leverage earn higher average returns than stocks with low market leverage. Following Fama and French, we 4

measure market leverage, A/M E, as the ratio of the year-end book assets (Compustat annual item 6) to the year-end market equity. (We also have constructed portfolios based on book leverage, the ratio of year-end book assets to year-end book equity. But the high-minus-low book-leverage portfolio has an insignicant average return and an insignicant CAPM alpha.) From Panel F of Table X, high A/M E stocks earn higher average returns than low A/M E stocks, and the average-return spread of 0.57% per month is signicant at the 5% level. The zerocost high-minus-low A/M strategy generates a CAPM alpha of 0.65% (t = 2.62). The Fama-French (1993) model produces a negative alpha of 0.32% (t = 1.95). In our model, high A/M E stocks have higher rIN V -loadings but slightly lower rROA -loadings than low A/M E stocks. As a result, the zero-cost portfolio has an insignicantly positive alpha of 0.35% per month.

C.4.

The Book-to-Market Deciles

From Panel G of Table X, the results are largely similar to those for the 25 size and book-tomarket portfolios reported in the main text.

III. Quarterly Investment Factor

To verify that the annual rebalancing of rIN V is indeed the driving force of the rIN V -loading patterns across momentum portfolios, we experiment with an alternative investment factor, deQ noted rIN V , which is constructed on quarterly investment data. We measure quarterly I/A as the

change in gross property, plant, and equipment (Compustat quarterly item 42) plus the change in inventory (item 38) divided by lagged total assets (item 44). This denition is the exact quarterly counterpart of our denition based on annual data. Each month from January 1975 to December 2006, we categorize NYSE, Amex, and NASDAQ stocks into three groups based on the breakpoints for the low 30%, middle 40%, and high 30% of 5

the ranked values of quarterly I/A from four months ago. (The starting point of the sample is restricted by the availability of quarterly investment data.) We also use the NYSE median market equity each month to split all stocks into two size groups. We form six portfolios from the intersections of the two size and three I/A portfolios and calculate monthly value-weighted returns
Q on the six portfolios for the current month. rIN V is the dierence (low-minus-high), each month,

between the simple average of the returns on the two low-I/A portfolios and the simple average of the returns on the two high-I/A portfolios.
Q We nd that rIN V earns an average return of 0.49% per month (t = 3.77). More important, TaQ ble XI shows that once we replace rIN V with rIN V in our three-factor regression, the W-L portfolios Q have insignicantly negative rIN V -loadings. This evidence is in contrast to the signicantly positive

rIN V -loadings from Table 2 in the main text. (The rROA -loadings are not materially aected.) As a result, the magnitude of E is in general higher. IV. The Cross-Sectional Variations of Economic Fundamentals (I/A and ROA)

Figure 2 in the main text illustrates the cross-sectional variation of I/A and ROA across the 25 size and momentum. Using an updated sample through 2006, Figure 1 in this Internet Appendix documents that growth rms indeed have persistently higher ROAs than value rms in the big-size quintile both in the event time (Panel A) and in the calendar time (Panel C). In the small-size quintile, however, growth rms have higher ROAs than value rms before, but lower ROAs after the portfolio formation (Panel A). In the calendar time, a dramatic downward spike of ROA appears for the small-growth portfolio over the past decade (Panel B). This downward spike explains their abnormally low rROA -loadings. Table XII furnishes the evidence for other testing portfolios. Specically, the table shows that ROA varies across the deciles sorted on distress measures, and that I/A varies across various deciles 6

formed on valuation ratios. Portfolio ROA and I/A are value-weighted across all stocks in the portfolio, where the weights are given by their market equity to be consistent with the calculations of portfolio returns. From Panel A, ROA decreases monotonically from the less distressed rms to more distress rms. The ROA of the high F -Prob decile is lower than the ROA of the low F -Prob decile by 5.88% per quarter (t = 14.72). Similarly, the ROA of the high O-score decile is lower than the ROA of the low O-score decile by 7.54% per quarter (t = 13.89). This evidence on ROA explains the higher rROA -loadings and thus higher average returns of less distressed rms than those of more distressed rms. Panel A also shows that the ROA-spread between the two extreme SU E deciles is only 1.22% per quarter, albeit highly signicant (t = 11.86). This low ROA-spread explains why our model is only partially successful in explaining the post-earnings announcement drift. The rest of Table XII shows that I/A decreases with various valuation ratios including E/P , C/P , D/P , and A/M E, and increases with prior 13-60 month returns and past ve-year sales growth, meaning again that value rms invest less and load more on rIN V than growth rms. V. The Industry Eect on the Investment and ROA Factors

The main text reports that the six size-I/A and the six size-ROA portfolios draw their rmmonth observations from a wide range of portfolios. In what follows we report the detailed evidence. Specically, we use Fama and Frenchs (1997) classication of ten industries. For each of the six size-I/A portfolios Figure 2 reports the percentage of rm-month observations that belong to a given industry. The gure shows that the industry distributions of the rm-month observations are not drastically dierent across the six size-I/A portfolios. Manufacturing rms account for about 1525% of the observations for most portfolios. Overall, each portfolio seems to draw observations from a wide range of industries. Using Fama and Frenchs (1997) classication of ten industries, we plot in Figure 3 the per-

centage of rm-month observations belonging to a given industry for each of the six size-ROA portfolios. The evidence is largely similar to that of the size-I/A portfolios. Each portfolio again draws observations from a wide range of industries. Manufacturing and high tech industries each account for around 15%20% of the rm-month observations for most portfolios.

VI. Annually Rebalanced Earnings Surprise, Short-Term Prior Returns, and Distress Portfolios

This section shows that the original earnings surprise, short-term prior return, and distress eects do not exist once we change the rebalancing frequency of the underlying testing portfolios from monthly to annual frequency. Because these eects only exist in monthly frequency, it seems reasonable to construct the explanatory ROA factor in the same frequency. Specically, in June of each year t we sort all NYSE, Amex, and NASDAQ stocks into ten deciles based on, separately, the Standardized Unexpected Earnings (SU E, dened in the same way as in Chan, Jegadeesh, and Lakonishok (1996)) measured at the scal yearend of t 1 (using the average SU E across the four quarters in the scal year of t 1 yields quantitatively similar results (not reported)) the 12-month prior return from June of year t 1 to May of year t (note that we skip one month in June before calculating returns from July of year t) the Campbell, Hilscher, and Szilagyis (2008) failure probability measured at the scal yearend of t 1 the Ohlsons (1980) O-score measured at the scal yearend of t 1 We calculate monthly value-weighted returns from July of year t to June of t + 1 and rebalance the portfolios in June.

Table XIII shows that none of the zero-investment annually rebalanced portfolios earn signicant average returns or CAPM alphas. The zero-cost portfolios average returns are all within one standard error of zero and the CAPM alphas are all within 1.6 standard errors of zero. The CAPM is rejected by the GRS test using ten annually rebalanced SU E portfolios, but is not rejected by the three other sets of annualized rebalanced portfolios.

VII. Covariances vs. Characteristics

Table XIV reports the horse races between covariances and characteristics following the research design of Daniel and Titman (1997, Table III). In Panel A we rst rank all NYSE rms by their market capitalization at the end of June of year t and independently rank all NYSE, Amex, and NASDAQ stocks by I/A at the end of scal year t 1. We sort all stocks into three size groups based on the 20%, 30%, and 50% NYSE breakpoints (per Fama and Frenchs (2008) categorization of microcaps, small stocks, and big stocks). We also sort all stocks into three I/A groups based on the 30%, 40%, and 30% breakpoints. Taking interactions forms nine size and I/A portfolios. The rms remain in these portfolios from July of year t to June of year t+1. The individual rms in each of these nine portfolios are further divided into one of ve equal-numbered subgroups based on their rIN V -loadings, j V , from the new three-factor regression. The regression is run with 36 IN months of returns (at least 24 months) prior to the formation date in June of year t. We report the mean excess monthly value-weighted returns in percentage for all the 45 portfolios, the nine high-minus-low IN V portfolios, and the averages across the nine size and I/A portfolios. Panel A shows that after we control for the I/A characteristic, IN V is not related to average returns. None of the nine high-minus-low IN V portfolios earn signicant average returns. In fact, their average returns are often negative. Averaging across the nine size and I/A portfolios, we see that the high-minus-low IN V portfolio earns an average return of 0.15% per month (t = 1.05).

In contrast, controlling for IN V does not aect the predictive power of the I/A characteristic. The test design in Panel B is similar to that of Panel A, except that we rst do independent sorts on size and IN V and then subdivide each of the nine resulting portfolios into ve equal-numbered I/A subgroups. Panel B shows that the high-minus-low I/A portfolios earn large negative returns on average, which are mostly signicant. Averaging across the nine size and IN V portfolios, we observe that the high-minus-low I/A portfolio earns an average return of 0.62% per month, which is more than 6.5 standard errors from zero. Panels C and D document similar results that the ROA characteristic dominates the ROA factor loading in predicting future returns.

10

references Bhandari, Laxmi Chand, 1988, Debt/equity ratio and expected common stock returns: Empirical evidence, Journal of Finance 43, 507528. Campbell, John Y., Jens Hilscher, and Jan Szilagyi, 2008, In search of distress risk, Journal of Finance 63, 28992939. Chan, Louis K. C., Narasimhan Jegadeesh, and Josef Lakonishok, 1996, Momentum strategies, Journal of Finance 51 (5), 16811713. Daniel, Kent D., and Sheridan Titman, 1997, Evidence on the characteristics of cross sectional variation in stock returns, Journal of Finance 52, 133. DeBondt, Werner F. M., and Richard Thaler, 1985, Does the stock market overreact? Journal of Finance 40 (3), 793805. Fama, Eugene F., and Kenneth R. French, 1992, The cross-section of expected stock returns, Journal of Finance 47, 427465. Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, 356. Fama, Eugene F., and Kenneth R. French, 1996, Multifactor explanation of asset pricing anomalies, Journal of Finance 51, 5584. Fama, Eugene F., and Kenneth R. French, 1997, Industry cost of equity, Journal of Financial Economics 43, 153193. Fama, Eugene F., and Kenneth R. French, 2008, Dissecting anomalies, Journal of Finance 63, 16531678. Lakonishok, Josef, Andrei Shleifer, Robert W. Vishny, 1994, Contrarian investment, extrapolation, and risk, Journal of Finance 49, 15411578. Ohlson, James A., 1980, Financial ratios and the probabilistic prediction of bankruptcy, Journal of Accounting Research 18, 109131.

11

Table I

Properties of the Investment Factor, rIN V , and the ROA Factor, rROA , 1/197212/2006, 420 Months
Investment-to-assets (I/A) is annual change in gross property, plant, and equipment (Compustat annual item 7) plus annual change in inventories (item 3) divided by lagged book assets (item 6). In each June we break NYSE, Amex, and NASDAQ stocks into three I/A groups using the breakpoints for the low 30%, middle 40%, and high 30% of the ranked I/A. We also use median NYSE size to split NYSE, Amex, and NASDAQ stocks into two groups, small and big. Taking intersections, we form six size-I/A portfolios, denoted SLI , SM I , SH I , BLI , BM I , and BH I . Monthly value-weighted returns on the six portfolios are calculated from July of year t to June of year t+1, and the portfolios are rebalanced in June of year t+1. rINV is the dierence (low-minus-high), each month, between the average returns on the two low-I/A portfolios (SLI and BLI ) and the average returns on the two high-I/A portfolios (SH I and BH I ). Return on assets (ROA) is quarterly earnings (Compustat quarterly item 8) divided by one-quarter-lagged assets (item 44). Each month from January 1972 to December 2006, we sort NYSE, Amex, and NASDAQ stocks into three groups based on the breakpoints for the low 30%, middle 40%, and the high 30% of the ranked quarterly ROA from at least four months ago. We also use the NYSE median each month to split NYSE, Amex, and NASDAQ stocks into two size groups. We form six portfolios from the intersections of the two size and the three ROA groups, denoted SLP , SM P , SH P , BLP , BM P , and BH P . Monthly value-weighted returns on the six portfolios are calculated for the current month, and the portfolios are rebalanced monthly. rROA is the dierence (high-minus-low), each month, between the simple average of the returns on the two high-ROA portfolios (SH P and BH P ) and the simple average of the returns on the two low-ROA portfolios (SLP and BLP ). In Panel A we regress rINV and rROA on traditional factors including the market factor, SM B, HM L, and W M L (from Kenneth Frenchs Web site). The t-statistics (in parentheses) are adjusted for heteroscedasticity and autocorrelations. Panel B reports the correlation matrix of the new factors and the traditional factors. The p-values (in parentheses) test the null hypothesis that a given correlation is zero. Panel A: Means and factor regressions of rINV and rROA Mean rINV 0.43 (4.75) 0.51 (6.12) 0.33 (4.23) 0.22 (2.87) 1.05 (5.61) 1.01 (5.60) 0.74 (4.16) t(Mean) 3.13 3.13 1.37 3.31 2.61 1.44 M KT 0.16 (8.83) 0.09 (4.79) 0.08 (4.11) 0.16 (4.00) 0.05 (1.23) 0.02 (0.38) # Firms 902 1,075 856 137 353 217 SM B HM L W M L R2 0.16 0.06 (2.27) 0.05 (2.29) 0.27 (9.47) 0.29 (10.65) 0.31 0.10 (5.89) 0.36 0.04 0.40 (7.14) 0.41 (7.56) Size 257 287 288 8,277 9,108 7,646 0.11 (1.74) 0.18 (2.81) B/M 1.43 1.09 1.00 0.83 0.67 0.57 0.31 0.26 (6.43) I/A 4.27 7.05 30.15 2.48 7.18 25.71 0.24 Panel D: Details of the six size-ROA portfolios ROA 2.16 3.60 4.16 6.24 7.76 8.04 SL SM P SH P BLP BM P BH P
P

Panel B: Correlation matrix (p-value in parenthesis) rROA rINV rROA rM KT SM B HM L 0.10 (0.05) rM KT 0.40 (0.00) 0.19 (0.00) SM B 0.09 (0.07) 0.38 (0.00) 0.26 (0.00) HM L 0.51 (0.00) 0.22 (0.00) 0.45 (0.00) 0.29 (0.00) W ML 0.20 (0.00) 0.26 (0.00) 0.07 (0.14) 0.02 (0.62) 0.11 (0.02)

12
rROA SL SM I SH I BLI BM I BH I
I

0.96 (5.10)

Panel C: Details of the six size-I/A portfolios Mean 0.94 0.88 0.45 0.76 0.56 0.39

Mean 0.22 0.73 1.29 0.09 0.43 0.50

t(Mean) 0.59 2.67 4.25 0.29 1.99 2.17

# Firms 842 975 677 74 334 281

Size 253 297 302 6,629 9,391 11,546

B/M 1.03 1.17 0.70 0.89 0.83 0.40

I/A 11.49 11.35 12.56 9.96 9.40 11.07

ROA 13.32 4.28 13.48 6.92 4.64 13.36

Table II

Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on 25 Size and Momentum Portfolios, 1/197212/2006, 420 Months
The data for the one-month Treasury bill rate (rf ) and the Fama-French factors are obtained from Kenneth Frenchs Web site. The monthly constructed size and momentum portfolios are the intersections of ve portfolios formed on market equity and ve portfolios formed on prior 27 month returns. The monthly size breakpoints are the NYSE market equity quintiles. For each portfolio formation month t, we sort stocks on their prior returns from month t 2 to t 7 (skipping month t1), and calculate the subsequent portfolio returns from month t to t+5. All portfolio returns are value-weighted. For all testing portfolios, Panel A reports mean percent excess returns and their t-statistics, CAPM alphas () and their t-statistics, and the intercepts (j F ) and their t-statistics from F
j j Fama-French three-factor regressions. Panel B reports the new three-factor regressions: r j rf = j + j KT rM KT + j q M INV rINV + ROA rROA + . See Table

I for the description of rINV and rROA . All the t-statistics are adjusted for heteroscedasticity and autocorrelations. FGRS is the Gibbons, Ross, and Shanken (1989) F -statistic testing that the intercepts of all 25 portfolios are jointly zero, and pGRS is its associated p-value. L 2 3 Mean S 0.04 2 0.11 3 0.03 4 0.05 B 0.22 S 2 3 4 B S 2 3 4 B 0.60 0.47 0.39 0.36 0.21 0.80 0.71 0.58 0.47 0.29 0.95 0.81 0.71 0.63 0.41 0.53 0.38 0.30 0.24 0.05 0.15 0.06 0.03 0.02 0.07 1.21 1.06 0.98 0.90 0.68 0.73 0.55 0.48 0.43 0.25 0.51 0.47 0.47 0.46 0.46 4 W W -L L 2 3 tMean 4 W W -L L 2
j

3 q

4 j q +

W W -L j KT rM KT M +

L j INV

2 rINV +

W W -L + j j 3.75 3.27 3.12 2.57 1.99 1.70 1.62 0.77 0.19 1.13

Panel A: Means, CAPM alphas, and Fama-French alphas

Panel B: r rf =

j ROA rROA

13
0.93 0.87 0.62 0.47 0.60

tq (FGRS = 2.20, pGRS = 0) 0.92 0.75 0.63 0.48 0.31 0.54 1.04 1.96 2.40 2.99 0.53 0.76 1.43 1.90 2.17 0.28 1.23 1.11 0.94 1.65 0.08 1.29 0.65 0.03 0.76 0.41 0.38 0.07 1.15 1.50 tIN V 0.60 0.62 0.58 0.83 0.57 1.70 3.85 3.90 4.84 5.16

1.25 0.09 1.89 2.78 3.29 3.54 5.49 0.38 0.45 0.49 0.60 1.17 0.27 1.54 2.60 2.94 3.07 4.75 0.22 0.25 0.30 0.35 0.95 0.08 1.39 2.29 2.80 3.03 3.63 0.35 0.18 0.12 0.20 0.85 0.13 1.31 1.93 2.62 2.97 3.01 0.40 0.10 0.00 0.07 0.90 0.65 0.86 1.37 1.95 2.46 3.17 0.10 0.01 0.10 0.11 t (FGRS = 3.28, pGRS = 0) INV 1.33 1.23 1.00 0.92 0.94 1.44 1.34 1.09 0.92 1.06 2.01 0.77 2.22 3.00 3.39 5.78 2.82 0.07 2.06 2.68 2.81 4.89 2.21 0.35 1.56 2.71 2.88 3.67 2.02 0.52 0.72 2.65 2.88 3.10 3.08 1.40 0.91 0.78 1.86 3.15 tF F (FGRS = 3.40, pGRS = 0) 3.67 3.82 2.58 1.64 2.41 2.37 3.05 2.62 2.12 0.97 0.54 1.05 1.91 2.47 0.74 1.88 0.84 0.45 0.30 1.01 4.89 4.15 4.26 3.65 3.31 5.54 4.72 3.57 2.68 3.19 0.31 0.56 0.58 0.80 0.67 0.80 0.59 0.53 0.44 0.21

0.59 0.15 0.40 0.69 0.01 0.29 0.51 0.05 0.18 0.49 0.07 0.07 0.69 0.17 0.07 F F 0.30 0.34 0.35 0.31 0.13 0.05 0.08 0.17 0.22 0.05

0.13 0.25 0.31 0.29 0.05 0.11 0.17 0.07 0.16 0.05 0.12 0.00 0.19 0.03 0.11 0.03 0.22 0.11 0.07 0.10 ROA

1.02 2.21 2.71 2.25 4.66 0.44 1.18 1.90 0.57 4.60 1.76 0.59 1.61 0.04 3.75 2.29 0.39 1.95 0.30 4.57 3.02 2.56 1.84 1.38 3.35 tROA 4.95 3.56 3.75 4.01 3.48 3.58 1.30 1.25 2.72 2.50 2.39 0.67 0.91 1.75 2.62 1.06 1.29 3.07 0.86 2.29 0.68 2.89 4.82 0.07 1.72

0.37 0.24 0.24 0.35 0.45 6.62 0.22 0.07 0.07 0.24 0.35 5.98 0.14 0.03 0.04 0.14 0.39 5.36 0.07 0.06 0.10 0.06 0.38 3.97 0.04 0.09 0.13 0.00 0.22 2.41

Table III

Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on Deciles Formed on Campbell, Hilscher, and Szilagyis (2008) Failure Probability Measure and Deciles Formed on Ohlsons (1980) O-Score
The data on the one-month Treasury bill rate (rf ), the Fama-French three factors are from Kenneth Frenchs Web site. See Table I for the description of rINV and rROA . We sort all NYSE, Amex, and NASDAQ stocks at the beginning of each month into deciles based on failure probability and on O-score from at least four months ago. Monthly valueweighted returns on the portfolios are calculated for the current month, and the portfolios are rebalanced monthly. For each portfolio we report the average return in monthly percent and its t-statistics, the CAPM regression (r j rf = j + j rM KT + j ), the Fama-French three-factor regression (r j rf = j F + bj rM KT + sj SM B + hj HM L + j ), and F
j j the new three-factor regression (r j rf = j + j KT rM KT + j q M INV rINV + ROA rROA + ). For each asset pricing

model, we also report the Gibbons, Ross, and Shanken (1989) F -statistic (FGRS ) testing that the intercepts are jointly zero and its p-value (in parenthesis). All the t-statistics are adjusted for heteroscedasticity and autocorrelations. Low 2 3 4 5 6 7 8 9 High H-L FGRS (p)

Panel A: The failure probability deciles, 6/197512/2006, 379 months Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 1.03 4.07 0.39 0.95 2.60 0.39 0.91 0.17 0.04 2.46 0.19 0.99 0.00 0.18 1.09 25.21 0.04 2.46 0.48 2.04 0.04 1.02 0.51 0.12 0.99 0.15 0.21 1.68 0.02 1.00 0.21 0.05 0.20 50.55 4.29 2.74 0.82 3.69 0.21 0.90 1.99 0.36 0.87 0.14 0.19 3.30 0.01 0.95 0.07 0.17 0.12 26.98 1.06 3.52 0.62 2.54 0.09 1.05 1.15 0.09 1.02 0.11 0.01 0.98 0.14 1.03 0.10 0.00 1.50 31.89 1.84 0.10 0.72 3.33 0.10 0.93 1.13 0.19 0.93 0.18 0.10 2.18 0.11 0.97 0.03 0.17 1.29 37.60 0.65 5.07 0.48 1.91 0.05 1.05 0.50 0.23 1.09 0.21 0.25 2.41 0.07 1.06 0.02 0.02 0.61 27.74 0.29 0.38 0.63 2.79 0.01 0.95 0.14 0.10 0.94 0.16 0.13 0.95 0.01 0.95 0.08 0.04 0.08 24.40 1.48 0.81 0.53 2.24 0.04 0.98 0.42 0.21 1.03 0.30 0.35 2.49 0.01 1.00 0.14 0.02 0.05 33.31 2.04 0.52 0.72 2.93 0.01 1.06 0.13 0.01 1.06 0.01 0.03 0.07 0.13 1.03 0.01 0.10 1.14 36.96 0.15 2.46 0.50 2.03 0.00 1.00 0.01 0.24 1.03 0.33 0.32 2.36 0.02 1.00 0.07 0.06 0.19 31.23 0.93 1.55 0.45 1.57 0.33 1.17 2.50 0.49 1.21 0.12 0.20 3.15 0.07 1.10 0.19 0.14 0.44 20.84 2.38 2.32 0.48 1.97 0.00 0.95 0.03 0.22 0.95 0.45 0.28 2.11 0.19 0.93 0.04 0.20 1.40 25.87 0.53 4.31 0.58 1.80 0.30 1.30 1.68 0.27 1.24 0.20 0.09 1.65 0.33 1.16 0.27 0.43 2.23 30.46 2.40 6.77 0.43 1.64 0.07 0.99 0.51 0.43 1.03 0.55 0.47 3.61 0.14 0.97 0.15 0.28 0.97 22.19 1.63 5.86 0.28 0.80 0.61 1.32 2.75 0.71 1.24 0.48 0.02 3.24 0.34 1.11 0.34 0.69 2.00 26.98 3.22 9.92 0.26 0.92 0.27 1.06 1.70 0.52 1.02 0.66 0.29 3.86 0.10 1.00 0.00 0.36 0.63 20.58 0.04 7.58 0.16 0.39 0.86 1.51 3.31 1.06 1.38 0.85 0.09 4.66 0.24 1.27 0.34 0.82 1.03 21.85 2.21 11.54 0.19 0.59 0.40 1.18 2.06 0.74 1.12 0.92 0.40 5.04 0.17 1.09 0.00 0.55 0.91 20.29 0.02 9.04 0.35 0.72 1.48 1.69 4.57 1.75 1.48 1.27 0.09 6.39 0.13 1.42 0.02 1.22 0.49 18.17 0.16 13.42 0.44 1.04 1.14 1.38 3.96 1.32 1.16 1.35 0.10 6.39 0.07 1.21 0.01 1.02 0.29 17.43 0.07 10.48 1.38 3.53 1.87 0.73 5.08 2.14 0.57 1.10 0.13 6.43 0.32 0.43 0.03 1.40 1.09 5.78 0.18 14.64 0.92 2.84 1.10 0.36 3.56 1.44 0.17 1.50 0.32 6.49 0.09 0.22 0.20 1.07 0.32 2.93 1.18 11.03

3.01 (0) 4.75 (0)

1.78 (0.06)

Panel B: The O-score deciles, 1/197212/2006, 420 months

2.49 (0.01) 6.33 (0)

1.10 (0.36)

14

Table IV

Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on the Net Stock Issues Deciles and the Asset Growth Deciles, 1/197212/2006, 420 Months
The data on the one-month Treasury bill rate (rf ) and the Fama-French three factors are from Kenneth Frenchs Web site. See Table I for the description of rINV and rROA . We measure net stock issues as the the natural log of the ratio of the split-adjusted shares outstanding at the scal yearend in t1 (Compustat annual item 25 times the Compustat adjustment factor, item 27) divided by the split-adjusted shares outstanding at the scal yearend in t2. In June of each year t, we sort all NYSE, Amex, and NASDAQ stocks into ten deciles based on the breakpoints of net stock issues measured at the end of last scal yearend. Monthly value-weighted returns are calculated from July of year t to June of year t + 1. In June of each year t, we sort all NYSE, Amex, and NASDAQ stocks into ten deciles based on asset growth measured at the end of last scal year t 1. Asset growth for scal year t 1 is the change of total assets (item 6) from the scal yearend of t 2 to yearend of t 1 divided by total assets at the scal yearend of t 2. Monthly value-weighted returns are calculated from July of year t to June of year t + 1. For each portfolio we report the average return in monthly percent and its t-statistics, the CAPM regression (r j rf = j + j rM KT + j ), the Fama-French three-factor regression (r j rf = j F +bj rM KT +sj SM B +hj HM L+j ), and the new three-factor F
j j regression (r j rf = j + j KT rM KT + j q M INV rINV + ROA rROA + ). We also report the Gibbons, Ross, and

Shanken (1989) F -statistic (FGRS ) testing that the intercepts are jointly zero and its p-value (in parenthesis). The t-statistics are adjusted for heteroscedasticity and autocorrelations. Low 2 3 4 5 6 7 8 9 High H-L FGRS (p)

Panel A: The net stock issues deciles Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 1.00 4.73 0.42 0.88 3.68 0.22 0.99 0.01 0.32 2.39 0.09 0.96 0.11 0.21 0.90 45.73 1.66 5.06 1.10 3.48 0.49 1.21 2.92 0.17 1.20 0.65 0.40 1.15 0.45 1.26 0.59 0.25 2.49 27.15 5.99 4.06 0.77 3.65 0.17 0.90 1.84 0.08 0.96 0.01 0.15 0.88 0.00 0.95 0.07 0.10 0.05 43.08 1.47 3.01 0.79 3.06 0.27 1.04 2.22 0.02 1.11 0.21 0.34 0.19 0.07 1.12 0.56 0.09 0.57 41.93 8.44 1.77 0.39 1.53 0.30 1.05 2.13 0.28 1.03 0.03 0.03 2.12 0.01 0.97 0.22 0.15 0.06 29.22 3.17 2.47 0.84 3.66 0.36 0.95 3.84 0.11 1.05 0.08 0.37 1.31 0.15 1.02 0.40 0.02 1.34 43.85 5.89 0.36 0.85 3.88 0.25 0.92 2.25 0.15 0.99 0.06 0.17 1.37 0.12 0.94 0.10 0.14 1.07 35.43 1.69 3.29 0.75 3.63 0.31 0.87 3.81 0.13 0.97 0.07 0.28 1.82 0.06 0.93 0.28 0.10 0.66 49.88 6.14 2.24 0.82 3.61 0.17 0.99 1.98 0.13 1.01 0.00 0.08 1.36 0.24 0.96 0.17 0.02 2.49 42.35 3.47 0.53 0.63 3.03 0.18 0.89 2.59 0.01 0.98 0.00 0.26 0.10 0.03 0.94 0.24 0.03 0.41 45.42 5.19 0.85 0.88 3.59 0.18 1.06 1.73 0.16 1.04 0.18 0.04 1.55 0.35 1.01 0.12 0.08 2.90 34.29 1.77 1.61 0.64 3.00 0.17 0.92 2.38 0.07 0.98 0.06 0.16 1.11 0.05 0.93 0.03 0.13 0.74 15 54.12 0.72 3.82 0.72 2.68 0.04 1.16 0.36 0.00 1.12 0.08 0.06 0.01 0.24 1.07 0.40 0.06 1.96 37.39 5.36 1.58 0.60 2.58 0.09 1.02 1.37 0.11 1.02 0.04 0.03 1.66 0.16 0.99 0.23 0.04 2.11 56.76 5.96 1.23 0.68 2.35 0.13 1.23 0.94 0.01 1.14 0.12 0.19 0.07 0.43 1.06 0.51 0.23 3.14 35.95 5.80 6.78 0.49 1.85 0.08 1.14 0.98 0.04 1.06 0.09 0.21 0.54 0.25 1.04 0.56 0.05 3.32 60.54 12.82 2.12 0.27 0.89 0.55 1.24 3.21 0.41 1.13 0.16 0.23 2.45 0.16 1.05 0.50 0.36 1.10 33.72 6.00 7.23 0.31 1.06 0.32 1.24 2.94 0.11 1.11 0.15 0.33 1.15 0.17 1.10 0.74 0.11 1.66 52.69 11.96 3.28 0.16 0.55 0.64 1.21 4.34 0.59 1.14 0.26 0.07 3.89 0.19 1.08 0.43 0.18 1.10 29.85 4.74 4.09 0.05 0.15 0.67 1.43 4.77 0.48 1.27 0.31 0.33 3.84 0.10 1.28 0.79 0.16 0.72 43.03 9.47 4.25 0.84 4.64 1.06 0.33 5.07 0.82 0.15 0.25 0.39 4.33 0.28 0.12 0.55 0.39 1.39 2.67 4.25 6.53 1.04 5.19 1.16 0.23 5.92 0.65 0.07 0.34 0.72 3.57 0.55 0.02 1.38 0.09 3.06 0.44 15.04 1.30

3.97 (0) 3.10 (0)

2.67 (0)

Panel B: The asset growth deciles

5.82 (0) 3.71 (0)

3.05 (0)

Table V

Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on Deciles Formed on Most Recent (and Four-Month-Lagged) Standardized Unexpected Earnings (SU E), 1/197212/2006, 420 Months
The data on the one-month Treasury bill rate (rf ) and the Fama-French three factors are from Kenneth Frenchs Web site. See Table I for the description of rINV and rROA . We dene SU E as the change in quarterly earnings per share from its value announced four quarters ago divided by the standard deviation of earnings change over the prior eight quarters. In Panel A we rank all NYSE, Amex, and NASDAQ stocks into ten deciles at the beginning of each month by their most recent past SU E. Monthly value-weighted returns on the SU E portfolios are calculated for the current month, and the portfolios are rebalanced monthly. In Panel B we use the same procedure but instead of the most recent SU E we sort on the SU E from at least four months ago. For each portfolio we report the average return in monthly percent and its t-statistics, the CAPM regression (r j rf = j + j rM KT + j ), the Fama-French three-factor regression (r j rf = j F + bj rM KT + sj SM B + hj HM L + j ), and the new three-factor regression F
j j (r j rf = j + j KT rM KT + j q M INV rINV + ROA rROA + ). For each asset pricing model, we also report the

Gibbons, Ross, and Shanken (1989) F -statistic (FGRS ) testing that the intercepts are jointly zero and its p-value (in parenthesis). All the t-statistics are adjusted for heteroscedasticity and autocorrelations. Low 2 3 4 5 6 7 8 9 High H-L FGRS (p)

Panel A: Most recent SU E Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 0.10 0.41 0.62 1.02 6.65 0.58 1.02 0.03 0.04 6.16 0.43 0.98 0.17 0.10 4.47 40.67 2.86 3.08 0.34 1.36 0.18 1.04 1.83 0.16 1.05 0.06 0.02 1.57 0.02 1.00 0.14 0.12 0.22 31.24 1.97 3.45 0.21 0.88 0.29 0.99 3.07 0.25 0.98 0.04 0.05 2.45 0.21 0.96 0.22 0.03 1.98 32.30 3.52 0.78 0.34 1.44 0.16 1.00 1.87 0.13 0.98 0.00 0.05 1.41 0.06 0.96 0.19 0.01 0.64 35.05 3.00 0.23 0.11 0.43 0.42 1.05 4.17 0.41 1.02 0.10 0.03 3.77 0.12 0.98 0.25 0.17 1.11 31.55 4.10 4.04 0.30 1.19 0.23 1.06 2.42 0.22 1.03 0.07 0.03 2.31 0.00 0.99 0.19 0.17 0.01 37.40 3.54 4.30 0.13 0.53 0.39 1.02 3.92 0.34 0.97 0.14 0.09 3.48 0.17 0.98 0.08 0.17 1.45 39.48 1.64 3.43 0.26 1.04 0.26 1.04 2.60 0.26 1.00 0.18 0.04 2.40 0.06 0.99 0.10 0.19 0.56 39.37 2.01 3.11 0.26 1.09 0.25 1.01 2.86 0.32 1.02 0.08 0.09 3.65 0.17 0.99 0.02 0.09 1.88 41.41 0.32 2.20 0.32 1.37 0.18 1.00 2.10 0.20 1.00 0.03 0.04 2.28 0.11 0.98 0.01 0.08 1.26 41.53 0.18 2.01 0.61 2.68 0.12 0.98 1.57 0.14 0.98 0.04 0.02 1.69 0.11 0.98 0.01 0.00 1.23 50.54 0.30 0.15 0.58 2.52 0.09 0.98 1.11 0.10 0.99 0.06 0.01 1.12 0.04 0.99 0.03 0.05 0.41 42.33 0.53 1.91 0.61 2.68 0.11 0.98 1.54 0.15 0.98 0.09 0.04 2.05 0.01 1.00 0.01 0.11 0.10 54.29 0.29 4.25 0.66 2.90 0.17 0.96 2.08 0.14 1.00 0.12 0.06 1.86 0.04 1.00 0.07 0.13 0.48 48.75 1.56 4.35 0.90 3.87 0.41 0.98 4.38 0.43 0.98 0.05 0.03 4.86 0.24 1.01 0.11 0.10 2.66 47.16 2.10 2.27 0.67 3.06 0.20 0.93 2.48 0.23 0.93 0.07 0.04 2.90 0.08 0.96 0.06 0.12 0.98 48.04 1.23 3.90 0.96 4.14 0.47 0.98 5.45 0.46 1.00 0.06 0.03 5.03 0.34 1.00 0.03 0.11 3.42 44.70 0.56 3.30 0.72 3.04 0.22 0.98 2.25 0.22 0.99 0.02 0.01 2.22 0.16 1.00 0.00 0.08 1.60 43.54 0.08 1.99 1.08 4.84 0.61 0.94 7.22 0.64 0.95 0.10 0.03 7.14 0.47 0.96 0.01 0.14 5.53 39.62 0.22 4.47 0.86 3.86 0.39 0.94 4.52 0.45 0.94 0.13 0.08 5.21 0.30 0.96 0.04 0.13 3.51 39.09 0.73 4.40 1.18 8.34 1.22 0.08 8.76 1.22 0.07 0.07 0.01 8.19 0.90 0.02 0.16 0.23 6.52 0.54 1.87 4.61 0.52 3.61 0.57 0.10 3.98 0.62 0.11 0.08 0.06 4.03 0.33 0.04 0.10 0.25 2.24 0.88 1.14 5.03

10.65 (0) 11.01 (0)

5.56 (0)

Panel B: Four-month-lagged SU E

3.65 (0) 4.60 (0)

1.79 (0.06)

16

Table VI

Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on 25 Size and Book-to-Market Portfolios, 1/197212/2006, 420 Months
The data for the one-month Treasury bill rate (rf ), the Fama-French factors, and the 25 size and book-to-market portfolios are obtained from Kenneth Frenchs Web site. For all testing portfolios, Panel A reports mean percent excess returns and their t-statistics, CAPM alphas () and their t-statistics, and the intercepts (F F ) and their t-statistics from Fama-French three-factor regressions. j q + j KT rM KT M + j INV rINV + j ROA rROA
j

Panel B reports the new three-factor regressions: r j rf =

+ . See Table I for the description of rINV and rROA . All the t-statistics are adjusted for heteroscedasticity

and autocorrelations. FGRS is the Gibbons, Ross, and Shanken (1989) F -statistic testing that the intercepts of all 25 portfolios are jointly zero, and pGRS is its associated p-value. L 2 3 Mean S 2 3 4 B S 2 3 4 B 0.10 0.34 0.41 0.51 0.40 0.63 0.38 0.27 0.13 0.11 0.81 0.66 0.72 0.58 0.61 0.21 0.09 0.17 0.04 0.13 0.88 0.90 0.74 0.79 0.59 1.07 1.00 0.84 0.84 0.65 1.19 1.04 1.07 0.92 0.65 0.70 0.53 0.59 0.45 0.25 1.09 0.69 0.66 0.42 0.25 1.32 0.91 0.86 0.58 0.36 0.25 0.93 1.22 1.68 1.67 t 4 H H-L L 2 3 tMean 2.40 3.10 4.05 4.21 2.27 3.51 4.06 3.77 2.70 3.14 3.67 4.12 2.28 3.30 3.72 3.65 2.68 2.75 3.13 2.80 (FGRS = 4.25, pGRS = 0) 4 H H-L L 2 3
j

H j E

H-L
j

L j I

2 rINV +

3 j R

4 rROA +
j

H-L

Panel A: Means, CAPM alphas, and Fama-French alphas

Panel B: r rf = q 5.08 0.08 0.64 0.46 0.59 3.27 0.14 0.19 0.32 0.40 2.86 0.19 0.14 0.07 0.15 1.93 0.19 0.12 0.05 0.15 1.20 0.11 0.13 0.04 0.03 INV 0.11 0.36 0.43 0.36 0.26 0.15 0.05 0.03 0.08 0.12 0.35 0.45 0.25 0.32 0.24 0.33 0.26 0.39 0.14 0.26 ROA

+ M KT +

tq (FGRS = 2.72, pGRS = 0) 0.64 0.57 0.27 2.49 2.23 3.10 0.38 0.24 0.63 1.09 2.18 2.75 0.31 0.13 1.05 1.05 0.60 1.17 0.08 0.11 1.23 1.18 0.40 1.32 0.03 0.14 1.17 1.51 0.39 0.26 tIN V 0.58 0.51 0.50 0.52 0.42 0.69 0.87 0.93 0.87 0.68 0.76 3.26 4.43 5.75 5.17 1.20 0.54 0.46 1.24 2.51 3.22 4.18 2.98 3.67 3.39 4.06 3.88 5.14 2.19 3.54 tROA 3.31 2.72 2.19 1.21 1.86 0.57 0.52 0.45 0.16 0.61 4.68 4.43 4.20 5.41 3.76 5.63 6.97 7.03 7.30 5.05

17

0.37 0.60 0.40 0.53 0.27 0.40 0.30 0.39 0.16 0.26 F F

2.61 1.03 2.15 3.64 3.82 7.10 2.07 0.57 2.96 3.78 3.18 4.83 1.74 1.45 2.32 3.16 3.71 3.96 1.14 0.37 2.68 3.33 3.06 2.82 1.29 1.48 1.54 2.18 1.61 1.81 tF F (FGRS = 3.08, pGRS = 0)

S 0.52 0.08 0.09 0.23 0.16 0.68 4.48 0.88 1.35 3.31 2.16 5.50 2 0.21 0.12 0.05 0.09 0.07 0.15 2.63 1.55 0.67 1.23 0.93 1.42 3 0.03 0.05 0.12 0.09 0.02 0.01 0.37 0.58 1.50 1.13 0.22 0.08 4 0.11 0.17 0.07 0.05 0.11 0.22 1.33 1.87 0.83 0.56 1.06 1.84 B 0.17 0.04 0.02 0.13 0.26 0.43 2.75 0.55 0.28 1.75 2.34 3.34

0.62 0.47 0.26 0.21 0.23 0.39 0.31 0.12 0.04 0.03 0.11 0.21 0.23 0.02 0.07 0.07 0.02 0.24 0.13 0.12 0.12 0.04 0.10 0.23 0.12 0.19 0.12 0.16 0.01 0.11

5.65 4.21 3.00 2.91 3.50 4.53 3.71 1.82 0.70 0.58 1.63 2.64 2.95 0.36 1.55 1.27 0.25 2.10 2.03 2.74 2.10 0.68 1.49 2.10 4.75 5.63 2.61 2.58 0.11 1.23

Table VII

Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on Ten Industry Portfolios and Deciles Formed on Pre-ranking CAPM Betas and the Market Equity, 1/197212/2006, 420 Months
The one-month Treasury bill rate (rf ), the Fama-French three factors, ten industry portfolio returns, and ten market equity portfolio returns are from Kenneth Frenchs Web site. See Table I for the description of rINV and rROA . We estimate pre-ranking CAPM betas on 60 (at least 24) monthly returns prior to July of year t. In June of year t we sort all stocks into ten deciles based on the pre-ranking betas. The value-weighted monthly returns on the resulting ten portfolios are calculated from July of year t to June of year t + 1. For each portfolio we report the average return in monthly percent and its t-statistics, the CAPM regression (r j rf = j + j rM KT + j ), the Fama-French three-factor regression (r j rf = j F + bj rM KT + sj SM B + hj HM L + j ), and the new three-factor regression F
j j (r j rf = j + j KT rM KT + j q M INV rINV + ROA rROA + ). For each asset pricing model, we also report the

Gibbons, Ross, and Shanken (1989) F -statistic (FGRS ) testing that the intercepts are jointly zero and its p-value (in parenthesis). All the t-statistics are adjusted for heteroscedasticity and autocorrelations. Panel A: Ten industry portfolios NoDur Mean tMean t F F b s h tF F q M KT INV ROA tq t M KT t IN V t ROA 0.67 3.04 0.27 0.81 1.99 0.10 0.91 0.08 0.27 0.76 0.24 0.92 0.32 0.33 1.89 29.52 4.39 7.47 Low Mean tMean t F F b s h tF F q M KT INV ROA tq t M KT t IN V t ROA 0.48 2.26 0.16 0.62 0.95 0.16 0.75 0.08 0.48 0.94 0.07 0.67 0.15 0.15 0.39 12.22 1.61 2.26 Durbl 0.41 1.43 0.11 1.03 0.64 0.47 1.17 0.11 0.53 2.75 0.25 1.07 0.24 0.02 1.25 23.18 2.20 0.19 2 0.69 3.73 0.37 0.63 3.04 0.14 0.77 0.11 0.37 1.21 0.05 0.70 0.16 0.23 0.43 20.05 2.27 4.27 Manuf 0.56 2.31 0.05 1.01 0.49 0.08 1.08 0.03 0.20 0.90 0.20 1.06 0.07 0.20 2.08 45.36 1.10 5.91 3 0.62 3.32 0.27 0.70 2.53 0.09 0.83 0.20 0.29 0.93 0.09 0.78 0.25 0.22 0.85 25.79 4.65 5.73 Enrgy 0.76 2.82 0.37 0.77 1.79 0.17 0.91 0.20 0.33 0.82 0.30 0.76 0.22 0.18 1.27 14.75 1.56 2.75 4 0.66 3.31 0.26 0.79 2.63 0.09 0.91 0.13 0.28 0.96 0.10 0.86 0.15 0.27 1.03 40.48 3.16 7.70 HiTec 0.49 1.44 0.17 1.32 0.98 0.22 1.09 0.21 0.64 1.45 0.47 1.18 0.51 0.37 2.70 29.79 5.32 7.23 5 0.57 2.55 0.10 0.93 1.10 0.09 1.03 0.01 0.30 1.10 0.14 0.98 0.10 0.18 1.44 39.76 1.87 4.30 Telcm 0.55 2.35 0.17 0.75 1.02 0.16 0.81 0.22 0.04 0.90 0.26 0.76 0.25 0.21 1.32 18.57 2.15 3.27 6 0.54 2.20 0.01 1.05 0.09 0.12 1.10 0.02 0.19 1.40 0.13 1.07 0.00 0.13 1.35 48.99 0.07 3.60 Shops 0.55 2.03 0.02 1.04 0.16 0.09 1.06 0.12 0.16 0.63 0.21 1.08 0.02 0.22 1.46 25.34 0.19 4.64 7 0.57 2.14 0.00 1.14 0.04 0.13 1.17 0.12 0.18 1.37 0.03 1.13 0.11 0.02 0.27 37.96 1.69 0.51 Hlth 0.58 2.31 0.15 0.85 0.91 0.41 0.81 0.35 0.35 2.50 0.10 0.88 0.08 0.28 0.51 17.39 0.80 3.89 8 0.54 1.81 0.10 1.26 0.95 0.15 1.21 0.34 0.04 1.47 0.19 1.20 0.23 0.17 1.52 35.24 3.03 3.68 Utils 0.51 2.52 0.25 0.50 1.47 0.13 0.72 0.15 0.61 0.89 0.01 0.56 0.22 0.15 0.07 12.06 2.26 2.14 9 0.51 1.40 0.23 1.49 1.44 0.12 1.31 0.51 0.25 0.81 0.34 1.37 0.29 0.41 2.01 31.21 3.02 7.43 Other 0.59 2.36 0.06 1.04 0.64 0.17 1.16 0.04 0.35 1.81 0.24 1.11 0.25 0.17 2.38 45.91 4.89 4.28 High 0.37 0.80 0.53 1.79 2.23 0.31 1.50 0.82 0.45 1.54 0.47 1.57 0.70 0.62 1.93 26.35 4.84 8.16 H-L 0.10 0.24 0.69 1.17 2.10 0.15 0.75 0.74 0.92 0.53 0.54 0.91 0.85 0.77 1.58 9.74 4.72 7.32 FGRS (p)

1.35 (0.20) 2.88 (0.00)

2.17 (0.02)

Panel B: Ten pre-ranking CAPM beta deciles FGRS (p)

1.60 (0.10) 1.23 (0.27)

1.77 (0.06)

18

Panel C: Ten market equity deciles Small Mean tMean t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 0.73 2.42 0.21 1.03 1.08 0.04 0.88 1.18 0.22 0.40 0.46 1.02 0.34 0.40 2.00 17.56 2.84 4.44 2 0.73 2.37 0.14 1.16 0.86 0.06 1.01 1.07 0.17 1.04 0.38 1.14 0.22 0.33 1.94 23.20 2.04 4.05 3 0.74 2.55 0.16 1.15 1.10 0.04 1.04 0.90 0.18 0.84 0.33 1.13 0.09 0.22 2.14 26.27 1.04 3.50 4 0.68 2.42 0.11 1.13 0.84 0.08 1.03 0.80 0.18 1.47 0.25 1.11 0.04 0.15 1.72 27.08 0.48 2.60 5 0.71 2.60 0.15 1.12 1.30 0.02 1.05 0.68 0.16 0.36 0.29 1.10 0.02 0.15 2.29 30.06 0.35 2.72 6 0.63 2.47 0.08 1.08 0.90 0.08 1.04 0.48 0.18 1.25 0.13 1.07 0.03 0.06 1.28 37.05 0.42 1.67 7 0.70 2.77 0.15 1.09 1.90 0.02 1.07 0.38 0.15 0.34 0.13 1.10 0.10 0.03 1.29 48.96 2.14 0.72 8 0.63 2.54 0.08 1.08 1.17 0.04 1.07 0.28 0.15 0.68 0.14 1.08 0.06 0.08 1.66 65.39 1.43 2.27 9 0.60 2.65 0.09 1.00 1.70 0.00 1.03 0.06 0.13 0.07 0.04 1.01 0.05 0.02 0.73 75.97 1.53 0.96 Big 0.46 2.15 0.02 0.94 0.31 0.06 0.97 0.31 0.08 2.47 0.07 0.95 0.05 0.08 1.20 59.37 1.55 3.51 S-B 0.28 1.16 0.23 0.09 0.96 0.10 0.10 1.49 0.31 1.11 0.53 0.08 0.39 0.48 1.91 1.07 2.68 4.39 1.79 (0.06) 1.82 (0.06) FGRS (p)

1.57 (0.11)

19

Table VIII

Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on 25 I/A and ROA Portfolios, 1/197212/2006, 420 Months
The data for the one-month Treasury bill rate (rf ) and the Fama-French (1993) factors are from Kenneth Frenchs Web site. We sort all stocks into ve quintiles each month based on quarterly ROA from four months ago, and all stocks independently in June of each year into ve quintiles based on investment-to-assets (I/A) at the last scal yearend. Taking intersections yields 25 investment and protability portfolios. We calculate value-weighted returns for the current month. Panel A reports mean percent excess returns and their t-statistics, the CAPM alphas () and their t-statistics, and the Fama-French alphas (F F ) and their j j t-statistics. Panel B reports the equilibrium three-factor regressions: r j rf = j + j KT rM KT + j q M INV rINV + ROA rROA + . All the t-statistics are adjusted for heteroscedasticity and autocorrelations. FGRS is the Gibbons, Ross, and Shanken (1989) F -statistic testing that the intercepts of all 25 portfolios are jointly zero, and pGRS is the p-value. LP 2 3 Mean L 0.13 0.67 2 0.44 0.62 3 0.26 0.29 4 0.58 0.31 H I 0.98 0.42 L-H I 1.11 1.09 LI 2 3 4 HI L-H I LI 2 3 4 HI L-H I
I

4 H P H-LP

LP

3 t(Mean)

H P H-LP

LP

2
j

3 q

4 j q

H P H-LP + j KT M

LP j INV

2 rINV +

3 j ROA

H P H-LP

Panel A: Means, CAPM alphas, and Fama-French regressions 0.71 0.83 0.45 0.48 0.22 0.49 1.03 0.93 0.57 0.84 0.71 0.68 0.43 0.58 0.41 0.67 0.62 0.26 0.80 0.30 2.05 2.52 3.84 3.02 0.40 1.06 2.03 3.46 2.43 3.30 0.95 0.64 0.97 1.73 3.02 2.91 1.16 1.37 0.99 1.72 1.79 2.07 1.65 2.06 1.19 0.74 1.43 2.08 3.41 4.63 2.39 3.10 1.23 t (FGRS = 4.20, pGRS = 0) 0.91 0.58 1.18 1.26 1.82

Panel B: r rf =

rM KT +

rROA + j

tq (FGRS = 2.42, pGRS = 0)

2.20 0.25 0.15 0.12 0.00 0.04 0.21 0.79 0.67 0.69 0.00 0.20 0.63 1.24 0.67 0.52 0.23 0.16 0.05 0.62 2.54 2.46 1.54 1.39 0.37 2.23 3.13 0.19 0.20 0.15 0.09 0.08 0.12 0.87 1.00 0.86 0.67 0.66 0.52 3.62 0.10 0.47 0.04 0.08 0.20 0.30 0.38 2.48 0.25 0.76 1.63 1.06 4.97 0.46 0.18 0.15 0.05 0.23 0.68 1.58 0.86 0.79 0.30 1.62 2.28 0.71 0.33 0.03 0.05 0.19 1.97 1.48 0.16 0.23 0.87 INV tIN V

20

0.56 0.07 0.19 0.52 0.35 0.25 0.05 0.37 0.10 0.33 0.97 0.26 0.04 0.24 0.21 1.26 0.27 0.06 0.07 0.00 1.81 1.07 0.35 0.15 0.01 1.25 1.14 0.53 0.67 0.34 F F 0.67 0.24 0.16 0.27 0.25 0.32 0.12 0.11 0.08 0.31 0.93 0.37 0.33 0.11 0.33 1.24 0.35 0.27 0.14 0.28 1.82 1.09 0.58 0.21 0.22 1.15 0.85 0.43 0.49 0.04

1.76 0.36 1.12 3.47 1.98 2.51 0.56 0.45 0.47 0.54 0.28 0.29 3.28 4.37 4.28 5.72 2.32 1.48 0.90 0.30 2.92 0.91 2.57 1.85 0.27 0.09 0.37 0.40 0.13 0.15 2.12 0.79 4.20 5.57 1.92 1.12 3.71 1.44 0.25 2.09 1.89 4.11 0.46 0.08 0.06 0.07 0.05 0.40 3.71 0.72 0.74 0.86 0.82 3.20 4.28 1.44 0.43 0.67 0.02 4.12 0.39 0.34 0.18 0.21 0.52 0.13 2.43 3.53 1.82 3.70 7.75 0.75 5.98 4.94 1.99 1.01 0.08 5.70 0.83 0.60 0.33 0.66 0.75 0.08 5.41 5.58 2.93 7.24 8.91 0.51 4.00 4.72 2.56 3.36 1.61 1.39 1.06 0.80 1.20 1.03 7.67 8.69 6.63 9.48 7.69 ROA tROA tF F (FGRS = 4.36, pGRS = 0) 2.44 1.23 1.04 1.81 1.32 0.63 0.86 0.74 3.69 2.01 2.17 0.95 4.36 1.79 1.84 1.29 6.40 5.15 3.27 1.30 3.45 3.56 1.99 2.28 1.42 2.24 3.11 2.37 1.66 0.17 2.83 2.15 4.69 5.00 6.61 1.05 0.30 0.06 1.01 0.40 0.04 0.89 0.41 0.08 0.92 0.54 0.01 0.89 0.56 0.03 0.16 0.26 0.09 0.23 0.16 0.06 0.20 0.11 0.16 0.11 0.06 0.14 0.16 0.10 0.00 1.21 9.12 3.93 1.04 1.21 8.15 4.69 0.75 1.05 13.69 6.01 1.17 0.98 11.61 7.65 0.18 1.05 9.08 6.39 0.40 1.16 3.33 1.35 5.18 1.46 1.92 2.71 2.47 1.79 3.40 11.40 4.48 9.59 4.60 14.19 1.31 10.31 3.38 9.11 0.04

0.93 0.63 1.26 1.51 2.04

Table IX

Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on 25 Size and 11/1/1-Momentum Portfolios, 1/197212/2006, 420 Months
The data for the one-month Treasury bill rate (rf ), the Fama-French (1993) factors, and the 25 size and momentum (11-1-1) portfolio returns are obtained from Kenneth Frenchs Web site. All portfolio returns are value-weighted. For all testing portfolios, Panel A reports mean percent excess returns and their t-statistics, CAPM alphas () and their t-statistics, and the intercepts (j F ) and their t-statistics from Fama-French (1993) three-factor regressions. Panel B reports the F
j j new three-factor regressions: r j rf = j + j KT rM KT + j q M INV rINV + ROA rROA + . See Table 1 in the main text for the description of the investment

factor rINV and the ROA factor rROA . All the t-statistics are adjusted for heteroscedasticity and autocorrelations. FGRS is the Gibbons, Ross, and Shanken (1989) F -statistic testing that the intercepts of all 25 portfolios are jointly zero, and pGRS is its associated p-value. L 2 3 Mean S 0.19 2 0.07 3 0.12 4 0.14 B 0.17 S 2 3 4 B S 2 3 4 B 0.65 0.55 0.46 0.54 0.45 0.91 0.82 0.66 0.57 0.31 1.08 1.02 0.79 0.79 0.55 0.61 0.51 0.31 0.30 0.11 0.26 0.21 0.02 0.06 0.08 1.52 1.32 1.23 1.07 0.81 0.94 0.68 0.61 0.50 0.28 0.78 0.64 0.60 0.51 0.41 4 W W -L L 2 3 tMean 1.72 0.49 2.40 3.61 4.16 4.68 7.63 1.40 0.19 2.00 3.28 3.96 3.96 5.99 1.11 0.34 1.76 2.78 3.34 3.94 4.17 0.93 0.40 2.03 2.42 3.37 3.69 3.19 0.64 0.53 1.90 1.45 2.56 3.07 2.16 t (FGRS = 3.28, pGRS = 0) 1.78 1.45 1.15 1.00 0.69 1.96 1.61 1.31 1.17 0.85 3.43 0.99 2.95 3.82 4.61 8.23 3.76 0.16 2.47 3.75 3.77 6.30 2.71 0.49 1.59 2.64 3.83 4.34 2.38 0.06 0.86 3.10 3.51 3.43 2.06 0.00 1.49 1.17 2.17 2.33 tF F (FGRS = 3.40, pGRS = 0) 5.74 5.43 3.35 2.78 2.00 2.42 3.04 3.52 2.09 0.74 0.64 0.10 1.74 2.30 2.40 3.22 2.74 0.23 0.76 0.91 6.98 5.86 5.05 3.95 2.97 7.97 6.30 4.42 3.55 2.58 0.12 0.09 0.28 0.32 0.03 4 W W -L L 2
j

3 q

4 j q +

W -L rM KT +

L j INV

2 rINV +

3 j ROA

W
j

W -L

Panel A: Means, CAPM alphas, and Fama-French alphas

Panel B: r rf =

j KT M

rROA +

tq (FGRS = 2.20, pGRS = 0) 0.37 0.38 1.27 1.36 0.12 1.86 2.72 3.06 4.10 3.25 1.21 1.81 2.71 3.25 2.36 0.21 0.70 0.81 2.77 0.73 0.46 0.37 0.46 1.52 0.18 0.05 2.81 2.60 0.27 0.19 tIN V 2.98 3.50 2.41 2.57 2.21 2.88 2.62 4.27 1.50 4.89 tROA 2.15 0.99 0.76 2.12 1.98 5.09 4.88 4.12 4.72 3.19 5.38 5.40 5.38 4.91 3.54

21

0.38 0.50 0.57 1.01 0.89 0.21 0.27 0.42 0.70 0.61 0.03 0.09 0.11 0.51 0.23 0.07 0.04 0.05 0.26 0.06 0.01 0.29 0.24 0.04 0.07 INV 0.33 0.41 0.24 0.26 0.18 0.23 0.18 0.28 0.09 0.26 ROA 0.31 0.12 0.08 0.20 0.18 0.73 0.71 0.73 0.96 0.68 0.78 0.72 0.82 0.88 0.64

0.84 0.16 0.46 0.76 0.02 0.33 0.54 0.06 0.18 0.50 0.01 0.09 0.41 0.00 0.13 F F 1.18 0.97 0.71 0.66 0.44 0.26 0.33 0.38 0.27 0.10 0.06 0.01 0.15 0.21 0.21

0.42 0.24 0.59 0.02 0.65 0.04 0.76 0.06 0.50 0.04 1.04 0.80 0.74 0.67 0.31

2.18 1.87 4.18 0.20 4.62 0.39 5.20 0.70 3.47 0.47 7.28 8.28 7.47 5.90 2.78

0.41 0.24 0.18 0.26 0.26 0.05 0.03 0.08 0.14 0.02 0.14 0.08 0.11 0.08 0.18 0.21 0.01 0.16 0.31 0.33

4.07 2.76 2.36 2.52 2.87 0.64 0.46 0.88 1.85 0.25 2.61 1.03 1.29 1.12 3.48 2.44 0.14 3.18 7.04 3.84

Table X

Summary Statistics and Factor Regressions for Monthly Percent Excess Returns on Deciles Formed on Earnings-to-Price (E/P ), Dividend-to-Price (D/P ), Cash ow-to-Price (C/P ), Prior 1360 Month Returns (Reversal), Past Five-Year Sales Growth (5-Yr SR), Market Leverage (A/M E), and Book-to-Market Equity (B/M ) 1/197212/2006, 420 Months
For each portfolio we report the average return and volatility in monthly percent, the CAPM regression (r j rf = j + j rM KT +j ), the Fama-French three-factor regression (r j rf = j F +bj rM KT +sj SM B +hj HM L+j ), and F
j j the new three-factor regression (r j rf = j + j KT rM KT + j q M INV rINV + ROA rROA + ). For each asset pricing model, we also report the Gibbons, Ross, and Shanken (1989) F -statistic (FGRS ) testing that the intercepts are jointly

zero and its p-value (in parenthesis). All the t-statistics are adjusted for heteroscedasticity and autocorrelations. The data on one-month Treasury bill, rf , the Fama-French three factors, and portfolio returns on E/P, D/P, B/M , and reversal deciles are from Kenneth Frenchs Web site. Following Lakonishok, Shleifer, and Vishny (1994), we measure the ve-year sales rank for June of year t, 5-Yr SR(t), as the weighted average of the annual sales growth ranks for P5 the prior ve years, j=1 (6 j) Rank(t j). The sales growth for year t j is the percentage change in sales (COMPUSTAT annual item 12) from year t j 1 to t j, log[Sales(t j)/Sales(t j 1)]. Only rms with data for all ve prior years are used to determine the annual sales growth ranks for years t 5 to t 1. To create the sales growth ranks, we sort rms into ten deciles in an ascending order on sales growth in each June, and then assign rank i, where i = 1, . . . , 10, to a rm if its sales growth falls into the ith decile. Following Fama and French (1992), we measure market leverage as the ratio of the book assets (item 6) to the market equity (price times number of shares). To form 5-Yr SR and A/M E deciles, we sort all NYSE, Amex, and NASDAQ stocks at the beginning of June of year t using the breakpoints of 5-Yr SR and A/M E measured at the scal year-end of t1 for the stocks traded on all three exchanges. Monthly value-weighted returns on the portfolios are calculated from July of year t to June of t+1. Low Mean Std t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA Mean Std t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 0.30 6.07 0.30 1.23 2.59 0.06 1.05 0.02 0.57 0.71 0.06 1.16 0.37 0.06 0.47 34.88 5.84 1.48 0.51 5.85 0.09 1.20 0.86 0.11 1.13 0.09 0.30 1.14 0.04 1.17 0.37 0.18 0.36 46.88 7.23 5.38 2 0.41 4.92 0.09 1.02 1.09 0.03 1.00 0.15 0.16 0.35 0.20 1.03 0.06 0.18 2.48 52.22 1.36 7.01 0.48 5.19 0.04 1.06 0.47 0.07 1.03 0.11 0.16 0.80 0.09 1.06 0.19 0.18 1.08 48.88 3.97 7.12 3 0.58 4.75 0.11 0.95 1.12 0.09 0.99 0.16 0.05 0.93 0.08 0.98 0.03 0.26 0.83 42.15 0.60 6.10 0.61 5.12 0.11 1.02 1.03 0.16 1.05 0.25 0.05 1.56 0.07 1.05 0.07 0.27 0.67 39.81 1.09 6.01 4 0.56 4.49 0.11 0.91 1.35 0.07 0.96 0.13 0.09 0.91 0.04 0.94 0.01 0.18 0.51 45.65 0.32 6.11 0.58 4.86 0.11 0.96 0.98 0.00 1.05 0.14 0.19 0.03 0.22 1.03 0.15 0.30 2.43 48.24 3.22 7.08 5 0.53 4.60 0.07 0.92 0.79 0.08 1.02 0.13 0.25 0.85 0.07 0.94 0.01 0.17 0.73 35.81 0.19 3.79 0.50 4.72 0.06 0.90 0.51 0.09 1.02 0.23 0.26 0.80 6 0.64 4.46 0.22 0.86 2.06 0.02 0.99 0.04 0.39 0.27 0.00 0.91 0.15 0.18 0.03 33.66 2.26 3.53 0.58 4.61 0.15 0.88 1.31 0.07 1.03 0.20 0.37 0.71 7 0.83 4.38 0.41 0.84 4.00 0.13 1.00 0.10 0.45 1.59 0.11 0.91 0.23 0.22 1.12 33.08 4.40 4.63 0.65 4.41 0.24 0.84 2.20 0.01 1.00 0.20 0.38 0.14 0.11 0.92 0.30 0.24 1.00 36.78 5.12 4.76 8 0.80 4.47 0.38 0.85 3.54 0.05 1.00 0.04 0.51 0.59 0.18 0.90 0.19 0.13 1.62 23.42 3.37 2.80 0.75 4.34 0.36 0.80 3.13 0.06 0.97 0.13 0.48 0.63 0.13 0.86 0.18 0.16 1.05 28.52 2.76 2.79 9 0.82 4.74 0.39 0.87 3.12 0.03 1.06 0.06 0.65 0.29 0.25 0.91 0.17 0.06 1.88 19.66 2.10 0.94 0.70 4.04 0.36 0.69 2.85 0.00 0.88 0.09 0.59 0.03 0.13 0.75 0.33 0.07 1.01 21.76 4.52 1.28 High 0.98 5.38 0.50 0.97 3.29 0.05 1.19 0.23 0.84 0.48 0.32 1.02 0.19 0.10 2.08 21.89 2.01 1.21 0.61 4.00 0.36 0.50 2.24 0.13 0.74 0.04 0.77 0.99 0.15 0.55 0.30 0.07 0.81 10.94 3.28 0.89 H-L 0.68 4.88 0.81 0.25 3.42 0.11 0.13 0.25 1.41 0.79 0.38 0.14 0.56 0.16 1.57 1.96 4.06 1.44 0.10 5.53 0.45 0.71 2.08 0.24 0.39 0.05 1.07 1.37 0.18 0.62 0.67 0.11 0.76 9.37 5.71 1.13

Panel A: E/P Deciles

Panel B: D/P Deciles

0.31 22 0.2 0.98 0.96 0.18 0.24 0.34 0.27 2.75 1.81 32.20 33.27 3.11 4.13 6.74 5.52

Low Mean Std t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA Mean Std t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 0.34 5.87 0.26 1.20 2.48 0.08 1.06 0.07 0.52 1.04 0.11 1.16 0.29 0.01 1.06 39.96 4.70 0.42 0.93 6.77 0.35 1.18 1.78 0.06 1.18 0.87 0.51 0.40 0.47 1.18 0.33 0.37 2.30 21.31 3.05 4.52

2 0.43 4.90 0.06 1.01 0.78 0.06 0.98 0.16 0.18 0.81 0.16 1.02 0.10 0.18 1.90 45.26 2.54 6.41 0.86 5.42 0.35 1.03 2.64 0.04 1.09 0.40 0.43 0.33 0.29 1.06 0.24 0.08 1.99 24.79 3.06 1.32

3 0.57 4.70 0.09 0.96 1.10 0.06 1.00 0.11 0.06 0.75 0.07 0.99 0.00 0.20 0.85 49.76 0.07 6.70 0.83 4.92 0.36 0.94 2.90 0.10 1.01 0.22 0.38 0.92 0.19 0.99 0.26 0.05 1.40 25.49 3.55 0.82

4 0.58 4.77 0.11 0.97 1.21 0.00 1.04 0.09 0.17 0.06 0.05 1.00 0.02 0.18 0.57 43.51 0.36 4.00 0.68 4.50 0.25 0.88 2.55 0.00 0.98 0.10 0.38 0.01 0.09 0.93 0.21 0.06 0.86 31.54 3.77 1.35

5 0.68 4.56 0.23 0.91 2.42 0.05 1.01 0.02 0.28 0.57 0.01 0.97 0.17 0.16 0.07 38.90 3.20 3.91 0.72 4.45 0.29 0.88 2.97 0.07 0.99 0.05 0.35 0.84 0.06 0.93 0.21 0.16 0.56 35.41 3.75 3.59

6 0.59 4.49 0.16 0.87 1.52 0.00 0.99 0.15 0.28 0.00 0.15 0.94 0.21 0.25 1.43 37.31 4.12 5.58 0.64 4.34 0.22 0.86 2.26 0.03 0.97 0.09 0.32 0.32 0.01 0.91 0.18 0.17 0.10 32.25 3.42 3.65

7 0.69 4.44 0.27 0.85 2.43 0.00 0.97 0.03 0.42 0.02 0.06 0.90 0.10 0.20 0.50 30.48 1.55 3.91 0.61 4.58 0.16 0.91 1.71 0.03 1.01 0.15 0.23 0.31 0.07 0.96 0.12 0.21 0.74 42.56 1.73 5.52

8 0.66 4.50 0.26 0.83 2.13 0.09 1.01 0.09 0.55 0.90 0.00 0.89 0.19 0.19 0.03 23.38 2.72 3.31 0.58 4.53 0.14 0.90 1.43 0.07 0.98 0.20 0.14 0.79 0.09 0.95 0.09 0.22 0.93 36.69 1.87 5.57

9 0.87 4.39 0.47 0.80 3.93 0.05 0.99 0.05 0.65 0.60 0.32 0.84 0.15 0.09 2.54 20.09 1.96 1.55 0.54 4.87 0.05 0.99 0.54 0.06 1.03 0.20 0.01 0.73 0.12 1.02 0.08 0.26 1.38 47.75 1.89 8.18

High 0.84 5.05 0.38 0.92 2.80 0.09 1.10 0.21 0.71 0.80 0.36 0.93 0.05 0.01 2.44 19.10 0.62 0.16 0.52 6.13 0.10 1.25 0.92 0.18 1.13 0.06 0.43 1.80 0.06 1.20 0.41 0.07 0.53 37.39 7.14 1.82

H-L 0.50 4.48 0.64 0.28 3.03 0.17 0.04 0.29 1.23 1.17 0.48 0.23 0.35 0.03 2.09 3.2 2.54 0.26 0.41 5.21 0.45 0.07 1.81 0.24 0.05 0.93 0.94 1.18 0.41 0.02 0.74 0.44 1.58 0.24 5.60 4.18

Panel C: C/P Deciles

Panel D: Deciles Formed on Prior 1360 Month Returns

23

Low Mean Std t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA Mean Std t() F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 0.35 5.02 0.11 0.93 0.82 0.42 1.02 0.25 0.45 3.30 0.38 1.00 0.41 0.06 2.64 23.62 5.22 1.30

2 0.22 4.73 0.22 0.90 1.90 0.45 1.01 0.00 0.36 3.98 0.55 0.99 0.47 0.11 4.60 34.90 6.75 2.22

3 0.15 4.55 0.28 0.87 2.56 0.51 1.00 0.07 0.37 5.25 0.53 0.94 0.30 0.11 4.54 34.38 4.49 2.08

4 0.10 4.57 0.35 0.91 3.66 0.45 0.97 0.06 0.17 4.49 0.38 0.92 0.10 0.02 3.23 39.73 1.68 0.40 0.12 4.93 0.37 1.00 4.10 0.42 1.04 0.04 0.09 4.60 0.43 1.01 0.04 0.05 4.15 41.13 0.67 1.14

5 0.01 4.29 0.41 0.85 4.38 0.48 0.92 0.18 0.14 5.61 0.58 0.89 0.12 0.13 5.85 35.85 2.09 3.38 0.21 4.74 0.25 0.95 2.57 0.38 1.02 0.04 0.20 3.91 0.37 0.97 0.09 0.08 3.59 34.66 1.48 1.79

6 0.07 4.50 0.37 0.90 4.13 0.41 0.97 0.20 0.09 4.69 0.44 0.91 0.06 0.12 4.50 38.29 1.22 2.89 0.16 4.61 0.29 0.91 2.92 0.53 1.02 0.04 0.37 5.82 0.33 0.92 0.07 0.00 3.10 32.85 1.28 0.04

7 0.00 4.74 0.48 0.98 5.99 0.44 0.99 0.12 0.04 5.71 0.45 0.97 0.10 0.02 5.58 51.96 2.11 0.70 0.26 4.34 0.15 0.83 1.42 0.46 0.97 0.04 0.48 5.37 0.28 0.87 0.21 0.03 2.53 28.31 3.30 0.55

8 0.19 4.90 0.30 1.00 3.36 0.26 1.01 0.13 0.04 2.94 0.31 1.00 0.15 0.11 3.37 48.71 3.20 3.15 0.23 4.47 0.19 0.84 1.67 0.59 1.02 0.06 0.63 7.56 0.32 0.87 0.17 0.05 2.45 25.15 2.38 0.81

High 0.00 6.80 0.66 1.35 4.60 0.36 1.18 0.08 0.49 2.76 0.10 1.19 0.76 0.21 0.71 37.50 10.6 3.81 0.37 5.69 0.14 1.03 0.90 0.63 1.18 0.38 0.73 5.02 0.15 1.05 0.32 0.19 0.80 19.63 3.50 2.35

H-L

Panel E: 5-Year SR 0.06 5.40 0.60 1.11 6.22 0.45 1.04 0.04 0.24 4.78 0.38 1.04 0.43 0.01 4.18 43.67 9.30 0.27 0.35 4.90 0.09 0.89 0.70 0.60 1.08 0.25 0.75 6.35 0.15 0.91 0.19 0.06 0.99 21.69 2.23 0.80 0.35 4.74 0.55 0.42 2.53 0.06 0.16 0.17 0.94 0.31 0.28 0.19 1.17 0.27 1.42 3.51 10.07 3.59 0.57 5.15 0.65 0.17 2.62 0.32 0.18 0.50 1.45 1.95 0.35 0.07 0.76 0.12 1.28 0.89 5.06 1.10

Panel F: A/M E Deciles 0.20 6.30 0.79 1.20 4.98 0.32 1.00 0.13 0.73 2.56 0.50 1.12 0.44 0.07 2.97 26.64 4.67 1.24 0.13 5.24 0.66 1.07 7.14 0.50 1.02 0.10 0.23 5.50 0.52 1.03 0.31 0.03 5.29 47.90 5.43 0.92 0.01 4.85 0.50 1.00 5.90 0.50 0.99 0.03 0.00 5.59 0.51 0.99 0.11 0.09 5.65 46.69 2.31 2.65

24

Low Mean Std t F F b s h tF F q M KT INV ROA tq tM KT tIN V tROA 0.31 5.36 0.24 1.10 2.51 0.13 0.97 0.17 0.55 2.04 0.11 1.05 0.44 0.09 1.17 41.15 8.02 2.97

2 0.52 4.84 0.01 1.03 0.08 0.06 1.02 0.06 0.07 0.85 0.12 1.05 0.00 0.12 1.67 57.78 0.04 4.94

3 0.58 4.86 0.07 1.02 0.87 0.01 1.07 0.05 0.12 0.08 0.13 1.06 0.09 0.15 1.60 56.60 1.91 4.53

4 0.67 4.74 0.18 0.97 1.94 0.00 1.05 0.02 0.27 0.04 0.04 1.01 0.14 0.14 0.48 36.54 2.82 3.66

5 0.62 4.50 0.17 0.90 1.76 0.07 1.02 0.04 0.37 0.76 0.06 0.95 0.19 0.13 0.63 36.83 3.20 3.02

6 0.69 4.38 0.24 0.88 2.72 0.00 1.00 0.01 0.38 0.05 0.09 0.93 0.21 0.05 0.92 33.97 3.89 1.20

7 0.77 4.31 0.35 0.82 3.28 0.03 1.00 0.01 0.59 0.42 0.14 0.88 0.27 0.08 1.25 25.87 3.64 1.46

8 0.74 4.26 0.34 0.81 3.11 0.12 1.00 0.11 0.68 2.02 0.06 0.88 0.34 0.09 0.60 28.72 5.38 1.89

9 0.81 4.56 0.38 0.86 3.20 0.11 1.05 0.15 0.73 1.46 0.17 0.92 0.39 0.01 1.39 24.77 4.93 0.12

High 0.97 5.34 0.50 0.94 3.16 0.14 1.16 0.35 0.93 1.41 0.22 1.04 0.65 0.05 1.26 20.50 5.47 0.75

H-L 0.66 4.68 0.74 0.16 3.29 0.27 0.20 0.52 1.48 2.47 0.33 0.01 1.09 0.14 1.41 0.11 7.29 1.61

Panel G: B/M

25

Table XI

New Three-Factor Regressions with the Quarterly Investment Factor Q (r j rf = j + j KT rM KT + j V Q rIN V + j q M ROA rROA + j ) for Monthly Percent Excess Returns on 25 Size and IN 11/1/1-Momentum Portfolios and on 25 Size and 6/1/6-Momentum Portfolios, 1/197512/2006, 420 Months
The data for the one-month Treasury bill rate (rf ), the market factor, and the 25 size and 11/1/1-momentum portfolios are from Kenneth Frenchs Web site. See Table IX for the description of the 25 size and 11/1/1-momentum portfolios. See Section ?? in this appendix for the construction of the quarterly investment factor,
Q denoted rINV and Table 1 in the main text for the description of the ROA factor rROA . The t-statistics are adjusted for heteroscedasticity and autocorrelations.

FGRS is the Gibbons, Ross, and Shanken (1989) F -statistic testing that the intercepts of all 25 portfolios are jointly zero, and pGRS is its associated p-value.

3 q

W W -L

W W -L

3 M KT

W W -L

W W -L

Panel A: 25 Size and 6/1/6-Momentum Portfolios tq (FGRS = 4.61, pGRS = 0) 1.14 0.73 0.55 0.49 0.26 0.24 0.33 0.32 0.27 0.33 0.65 1.59 2.46 2.68 3.06 0.57 0.63 1.17 1.31 1.75 0.34 0.92 0.24 0.21 0.83 0.27 0.88 0.18 0.86 0.33 0.35 0.40 0.46 2.20 2.02 tIN V Q 0.23 0.23 0.24 0.19 0.34 0.05 0.55 0.52 0.59 0.08 1.24 1.52 1.91 2.61 1.29 1.46 1.59 2.88 3.91 2.90 0.99 1.15 2.09 4.02 1.46 4.33 3.07 2.76 2.85 2.17 1.53 2.09 2.27 2.73 5.32 2.54 2.01 1.10 0.79 1.21 1.11 1.25 1.46 1.01 2.13 1.25 1.30 1.23 1.18 1.00 0.95 0.72 0.62 0.56 0.31 1.05 1.11 1.07 1.04 0.89 S 0.50 0.51 0.54 0.65 2 0.15 0.18 0.20 0.28 3 0.21 0.03 0.03 0.10 4 0.22 0.02 0.09 0.03 B 0.09 0.06 0.18 0.15 INV Q S 2 3 4 B 0.01 0.10 0.07 0.08 0.01 0.24 0.21 0.24 0.24 0.11 0.22 0.21 0.29 0.32 0.12 q S 0.09 0.40 0.56 0.76 2 0.00 0.16 0.26 0.50 3 0.17 0.07 0.04 0.07 4 0.14 0.07 0.12 0.08 B 0.08 0.09 0.35 0.22 INV Q S 2 3 4 B 0.01 0.09 0.11 0.03 0.14 0.35 0.26 0.30 0.36 0.27 0.32 0.28 0.33 0.37 0.12 0.16 0.16 0.23 0.35 0.10 1.36 1.01 0.69 0.45 0.06 0.25 0.40 0.31 0.24 0.17 0.14 0.14 0.19 0.24 0.05 1.01 1.03 1.06 1.09 1.04 1.05 1.02 1.03 0.88 0.92 ROA 1.19 0.06 1.28 0.02 1.23 0.00 1.18 0.00 1.08 0.08 tROA 0.57 0.56 0.57 0.59 0.38 6.80 6.19 5.53 4.53 3.12 4.79 3.51 3.10 3.29 4.53 2.89 1.06 0.43 1.51 3.95 1.87 1.14 1.62 0.61 3.69 1.14 1.68 3.25 0.39 3.52 0.12 2.67 5.63 1.34 2.87

26

0.42 0.30 0.28 0.38 0.20 0.07 0.03 0.15 0.11 0.06 0.08 0.05 0.08 0.09 0.14 0.03 0.01 0.11 0.19 0.07 M KT

Panel B: 25 Size and 11/1/1-Momentum Portfolios tq (FGRS = 3.40, pGRS = 0) 1.28 0.27 2.11 3.05 3.64 1.01 0.01 0.94 1.71 2.94 0.51 0.71 0.48 0.29 0.49 0.31 0.51 0.46 1.06 0.81 0.14 0.29 0.57 3.25 2.10 tIN V Q 0.24 0.31 0.21 0.21 0.31 0.06 0.53 0.77 0.17 0.90 1.85 1.84 2.87 3.93 2.85 2.16 2.40 3.42 5.22 2.15 1.34 1.55 3.34 5.84 1.23 4.74 3.77 3.14 2.27 0.38 1.51 2.34 2.31 2.03 1.54 3.83 3.06 1.37 0.75 0.35 1.07 1.53 1.03 0.85 1.28 1.20 1.30 1.23 1.21 1.08 1.05 0.82 0.73 0.67 0.32 0.94 1.03 1.04 1.06 0.87 0.90 0.94 1.00 1.03 0.97 1.00 0.99 1.03 0.89 0.91 ROA 1.13 0.08 1.24 0.06 1.23 0.01 1.16 0.05 1.08 0.00 tROA 0.74 0.67 0.77 0.85 0.62 7.16 7.67 6.22 4.80 2.39 4.37 2.45 1.92 2.48 4.51 2.64 0.46 0.30 1.27 4.13 1.07 1.79 4.28 0.45 4.13 0.34 2.90 6.07 1.73 3.86 0.64 3.70 6.97 3.25 3.05

0.35 0.18 0.16 0.31 0.20 0.03 0.02 0.15 0.07 0.10 0.22 0.04 0.03 0.17 0.28 0.18 0.05 0.19 0.34 0.31

Table XII

Economic Fundamentals for Deciles Formed on Campbell-Hilscher-Szilagyis (2008) Failure Probability Measure (F -Prob), Ohlsons (1980) O-Score, Standardized Unexpected Earnings (SU E), Earnings-to-Price (E/P ), Cash Flow-to-Price (C/P ), Dividend-to-Price (D/P ), Prior 1360 Month Returns (Reversal), and Past Five-Year Sales Growth (5-Yr SR), and Market Leverage (A/M E)
ROA is quarterly percent earnings (Compustat quarterly item 8) divided by one-quarter-lagged assets (item 44). I/A (in annual percent) is the annual change in gross property, plant, and equipment (Compustat annual item 7) plus the annual change in inventories (item 3) divided by the lagged book value of assets (item 6). Portfolio ROA and I/A are value-weighted ROAs and I/As of all the stocks in the portfolio, respectively, where the weights are provided by their market equity. The breakpoints for the E/P, C/P, D/P , and reversal portfolio deciles are from Kenneth Frenchs Web site. See this appendix or the main text for the description of the formation of the F -Prob and O-score portfolios, the SU E portfolios, the 5-Yr SR, net stock issues, asset growth, and A/M E portfolios. Low 2 3 4 5 6 7 8 9 High H-L

Panel A: Means and standard deviations (in annual percent) of ROA The F -prob deciles (19752006) Mean(ROA) Std(ROA) Mean(ROA) Std(ROA) Mean(ROA) Std(ROA) 11.20 1.04 9.68 0.82 4.00 2.06 9.48 0.74 5.96 1.12 5.64 1.22 8.04 0.74 4.92 1.02 5.76 1.18 6.32 0.80 3.76 1.24 6.32 0.98 5.20 1.12 2.92 1.36 6.56 0.86 3.52 1.12 1.64 1.64 6.88 0.86 2.32 1.42 0.04 2.58 4.40 3.70 12.32 9.54 23.52 9.32

The O-score deciles (19722006) 0.32 2.18 7.52 0.82 2.84 3.50 7.68 0.84 5.96 3.72 7.68 0.82 20.60 12.88 8.88 1.10 30.16 12.66 4.88 2.40

The SU E deciles (19722006)

Panel B: Means and standard deviations (in annual percent) of I/A The net stock issues deciles (19722006) Mean(I/A) Std(I/A) Mean(I/A) Std(I/A) Mean(I/A) Std(I/A) Mean(I/A) Std(I/A) Mean(I/A) Std(I/A) Mean(I/A) Std(I/A) Mean(I/A) Std(I/A) Mean(I/A) Std(I/A) 6.26 2.92 7.64 3.13 7.42 3.48 1.72 2.65 9.81 3.58 9.95 4.50 8.85 3.52 5.86 4.48 5.40 3.18 10.17 3.36 5.54 1.94 4.26 2.55 9.05 4.46 9.05 3.94 8.24 4.16 6.46 4.02 6.87 3.29 9.48 4.38 8.31 2.75 6.39 2.59 8.03 6.32 8.95 4.84 7.98 4.10 7.38 3.60 7.45 3.27 8.76 3.76 9.17 2.82 8.31 2.85 8.24 4.34 9.26 5.08 6.76 3.82 8.00 3.50 8.34 5.19 7.85 2.81 9.30 3.25 10.80 3.03 8.33 3.59 8.55 4.76 7.70 4.34 8.93 3.53 8.87 3.42 7.40 2.70 10.76 3.41 14.56 3.49 8.83 5.23 9.56 6.35 6.66 4.92 9.23 3.70 10.07 3.42 5.95 4.09 14.60 3.90 21.59 5.29 8.23 3.42 8.88 4.48 6.93 3.18 11.36 4.09 11.53 4.13 5.07 2.83 30.83 12.70 42.56 11.91 8.59 4.31 7.70 3.89 5.80 4.29 16.34 5.43 17.32 6.37 5.30 4.67 24.58 14.01 51.40 13.72

The asset growth deciles (19722006) 8.83 4.79 11.58 4.66 13.54 4.91 15.11 4.57 3.93 5.49 4.86 5.60 15.70 6.78 1.69 3.26 10.65 4.03 10.25 3.86 11.02 3.58 4.09 5.85 5.87 3.95 11.79 4.23

The E/P deciles (19722006) 2.99 6.24

The C/P deciles (19722006) 5.84 6.50

The D/P deciles (19722006) 9.3 5.66 12.41 7.01 12.46 8.01

Deciles formed on prior 1360 month returns (19722006)

The 5-Yr SR deciles (19722006)

The A/M E deciles (19722006) 10.4 8.74

27

Table XIII

Summary Statistics and CAPM Regressions for Monthly Percent Excess Returns on Annually Rebalanced Deciles Formed on Standardized Unexpected Earnings, Prior 12-Month Returns, Campbell, Hilscher, and Szilagyis (2008) Failure Probability, and Ohlsons (1980) O-Score, 1/197212/2006, 420 Months
For each portfolio we report the average return and volatility in monthly percent and the CAPM regression (r j rf = j + j rM KT + j ). For each asset pricing model, we also report the Gibbons, Ross, and Shanken (1989) F -statistic (FGRS ) testing that the intercepts are jointly zero and its p-value (in parenthesis). All the tstatistics are adjusted for heteroscedasticity and autocorrelations. The data on one-month Treasury bill, rf , are from Kenneth Frenchs Web site. To construct the testing portfolios, in June of each year t we sort all NYSE, Amex, and NASDAQ stocks into ten deciles based on, separately, the Standardized Unexpected Earnings measured at the scal yearend of t 1, the 12-month prior return from June of year t 1 to May of year t, and the Campbell, Hilscher, and Szilagyis (2008) failure probability and Ohlsons (1980) O-score measured at the scal yearend of t 1. Monthly value-weighted returns are calculated from July of year t to June of t + 1 and the portfolios are rebalanced in June. Low 2 3 4 5 6 7 8 9 High H-L FGRS (p)

Panel A: Prior returns from June of year t 1 to May of year t Mean Std t Mean Std t Mean Std t Mean Std t 0.44 8.77 0.31 1.48 1.13 0.46 5.25 0.08 1.06 0.73 0.7 5.06 0.09 1.01 0.73 0.46 4.75 0.04 1 0.55 0.49 7.04 0.13 1.24 0.62 0.39 4.92 0.12 1.01 1.23 0.61 4.73 0.02 0.98 0.16 0.47 4.93 0.04 1.03 0.59 0.33 5.88 0.21 1.08 1.28 0.44 5.2 0.08 1.03 0.7 0.41 4.53 0.15 0.93 1.37 0.68 5.08 0.16 1.05 1.79 0.46 5.5 0.06 1.03 0.39 0.52 5.48 0.04 1.12 0.39 0.46 4.81 0.12 0.96 0.98 0.52 4.99 0.02 1 0.22 0.43 4.87 0.06 0.97 0.54 0.56 4.97 0.05 1.02 0.55 0.67 4.87 0.05 1.03 0.53 0.68 4.99 0.18 0.99 1.68 0.44 4.46 0.01 0.9 0.16 0.69 4.7 0.2 0.97 2.41 0.47 5.44 0.19 1.1 1.48 0.71 4.85 0.23 0.96 2.12 0.57 4.35 0.13 0.87 1.37 0.61 4.49 0.14 0.94 1.83 0.51 6.12 0.22 1.21 1.36 0.59 5.17 0.1 1 0.81 0.54 4.79 0.04 0.98 0.46 0.7 4.56 0.22 0.94 2.6 0.77 6.96 0.02 1.31 0.08 0.46 5.62 0.06 1.04 0.43 0.62 5.84 0.06 1.12 0.39 0.6 5.06 0.08 1.03 0.81 0.63 8.17 0.25 1.47 1.01 0.53 6.42 0.04 1.15 0.24 0.51 7.2 0.17 1.35 0.89 0.59 4.48 0.13 0.91 1.46 0.71 8.69 0.17 1.46 0.57 0.3 8.45 0.41 1.43 1.59 0.07 7.28 0.13 0.13 0.37 0.13 2.81 0.21 0.15 1.52 0.01 6.73 0.26 0.45 0.81 0.16 6.17 0.37 0.43 1.34

0.79 (0.64)

Panel B: SU E measured as the scal yearend of t 1

2.09 (0.02)

Panel C: Failure probability measured at the scal yearend of t 1, 6/197512/2006, 379 months

1.05 (0.40)

Panel D: O-score measured at the scal yearend of t 1

1.84 (0.05)

28

Table XIV

Mean Percentage Excess Monthly Returns of the 45 Portfolios Formed on Size, Investment-to-Assets, and Investment Factor Loadings and the 45 Portfolios Formed on Size, ROA, and ROA Factor Loadings, 1/197212/2006, 420 Months
In Panel A we rst rank all NYSE rms by their market capitalization at the end of June of year t and independently rank all NYSE, Amex, and NASDAQ stocks by their investment-to-assets (I/A) at the end of scal year t1. We form 20%, 30%, and 50% breakpoints for size and 30%, 40%, and 30% breakpoints for I/A. Starting in July of year t, we place all NYSE, Amex, and NASDAQ stocks into the three size groups and three I/A groups based on these breakpoints. Taking interactions forms nine size and I/A portfolios. The rms remain in these portfolios from July of year t to June of year t+1. The individual rms in each of these nine portfolios are further sorted into one of ve equal-numbered subgroups based on the investment factor loadings ( j INV ) from the regression:
j j r j rf = j + j KT rM KT + j q M INV rINV + ROA rROA + . The regression is run with 36 months of returns (at least 24 months) prior to the formation date in June of year t. We report the mean excess monthly value-weighted returns in percentage for all the 45 portfolios, the nine high-minus-low INV portfolios,

and the averages across the nine size and I/A portfolios. The t-statistics test whether the average high-minus-low INV portfolio returns equal zero. Panel B is similar to Panel A, except that we rst do independent sorts on size and INV and then subdivide each of the nine resulting portfolios further into ve equal-numbered I/A subgroups. In Panel C we do an independent three-by-three sort on size and ROA each month to form nine portfolios. We form 20%, 30%, and 50% NYSE breakpoints for size at the end of the portfolio formation month and 30%, 40%, and 30% NYSE, Amex, and NASDAQ breakpoints for quarterly ROA with earnings announced at least four months ago. Each of the nine portfolios is further divided into ve equal-numbered subgroups based on their ROA factor loadings ( j ROA ) estimated with 36 months of data (at least 24 months) prior to the portfolio formation month. Panel D is similar to Panel C, except that we rst do independent sorts on size and j ROA and then subdivide each of the nine resulting portfolios into ve equal-numbered ROA subgroups. Panel A: Sorts on size and I/A, then on INV INV Size I/A Low 1.18 1.18 0.42 1.11 1.03 0.74 0.92 0.72 0.70 0.89 2 1.29 1.13 0.68 1.05 1.03 0.57 0.99 0.58 0.41 0.86 3 1.15 1.18 0.57 1.19 1.00 0.68 0.87 0.65 0.34 0.85 4 1.24 1.11 0.71 0.97 1.06 0.75 0.80 0.66 0.50 0.87 ROA Size ROA Low 0.23 1.02 2.22 0.21 0.76 1.19 0.32 0.72 0.74 0.73 2 0.13 1.17 1.95 0.07 0.82 1.33 0.21 0.49 0.51 0.73 3 0.16 1.05 2.28 0.09 0.95 1.14 0.21 0.46 0.72 0.75 4 0.09 0.91 1.91 0.20 0.78 1.26 0.04 0.71 0.65 0.70 High 0.29 0.85 1.99 0.34 0.72 1.10 0.16 0.65 0.63 0.69 H-L 0.06 0.17 0.22 0.55 0.04 0.09 0.16 0.07 0.10 0.04 t 0.19 0.73 0.79 1.38 0.16 0.37 0.39 0.25 0.37 0.18 Size ROA Low 0.49 0.23 0.49 0.31 0.26 0.32 0.37 0.34 0.47 0.03 2 0.14 0.31 0.00 0.31 0.68 0.83 0.33 0.33 0.61 0.39 3 0.66 0.93 0.76 0.61 1.03 0.80 0.55 0.47 0.59 0.71 Micro Low Micro Medium Micro High Small Low Small Medium Small High Big Low Big Medium Big High Average Micro Low Micro Medium Micro High Small Low Small Medium Small High Big Low Big Medium Big High Average High 1.15 1.16 0.54 0.80 0.84 0.45 0.81 0.57 0.31 0.73 H-L 0.03 0.02 0.12 0.32 0.20 0.29 0.11 0.15 0.39 0.15 t 0.15 0.13 0.59 1.34 0.97 1.14 0.45 0.66 1.42 1.05 Size INV Low 1.32 1.21 1.24 1.09 1.10 0.88 0.77 0.86 0.79 1.03 2 1.12 1.25 1.19 0.87 0.99 0.96 0.71 0.71 0.75 0.95 3 1.22 1.20 1.02 0.99 1.02 0.97 0.81 0.62 0.58 0.94 Panel B: Sorts on size and INV , then on I/A I/A 4 1.05 1.07 1.14 0.93 0.92 0.76 0.74 0.46 0.54 0.84 ROA 4 1.33 1.27 1.41 0.77 1.04 0.92 0.84 0.50 0.53 0.96 High 2.13 2.12 2.06 1.18 1.35 1.30 1.04 0.76 0.65 1.40 H-L 2.61 2.35 2.54 1.49 1.10 0.98 0.67 0.42 0.17 1.37 t 9.42 11.39 8.95 4.89 6.60 4.49 2.16 2.07 0.91 9.65 High 0.31 0.51 0.36 0.42 0.63 0.39 0.45 0.26 0.35 0.41 H-L 1.00 0.70 0.88 0.67 0.47 0.50 0.32 0.60 0.43 0.62 t 5.37 5.26 5.42 3.31 3.71 2.35 1.52 4.62 1.89 6.58 Micro Low Micro Medium Micro High Small Low Small Medium Small High Big Low Big Medium Big High Average

29

Micro Low Micro Medium Micro High Small Low Small Medium Small High Big Low Big Medium Big High Average

Panel C: Sorts on size and ROA, then on ROA

Panel D: Sorts on size and ROA , then on ROA

Panel A: Event-time
3

Panel B: Calendar-time, small rms

4 4

Panel C: Calendar-time, big rms

2.5 BigGrowth

1.5 BigValue 1 SmallValue

SmallValue

BigValue

BigGrowth

4 SmallGrowth

30

0.5 SmallGrowth 0 20 15 10 5 0 Quarter 5 10 15 20

1975

1980

1985

1990 Year

1995

2000

2005

1975

1980

1985

1990 Year

1995

2000

2005

Figure 1. Quarterly ROA for the 25 Size and Book-to-Market Portfolios, 1972:Q1 to 2006:Q4, 140 Quarters. We measure
ROA as quarterly earnings (Compustat quarterly item 8) divided by one-quarter-lagged assets (item 44). For each portfolio formation year t = 1972 to 2005, we calculate the quarterly ROAs for t + q, q = 20, . . . , 20. The ROA for t + q are then averaged across portfolio formation years t. We follow Fama and French (1996) in constructing the 25 size and book-to-market portfolios. We plot the median ROA among the rms in a given portfolio.

Panel A: Portfolio SLI

30 30

Panel B: Portfolio BLI

25

25

20 Percentage Percentage 1 2 3 4 5 6 7 Industry Identifier 8 9 10

20

15

15

10

10

4 5 6 7 Industry Identifier

10

Panel C: Portfolio SM I
30 30

Panel D: Portfolio BM I

25

25

20 Percentage Percentage 1 2 3 4 5 6 7 Industry Identifier 8 9 10

20

15

15

10

10

4 5 6 7 Industry Identifier

10

Panel E: Portfolio SH I
30 30

Panel F: Portfolio BH I

25

25

20 Percentage Percentage 1 2 3 4 5 6 7 Industry Identifier 8 9 10

20

15

15

10

10

4 5 6 7 Industry Identifier

10

Figure 2. Industry Distributions of Firm-Month Observations for the Six Size-I/A Portfolios Used to Construct the Investment Factor. The classication of ten industries follows
Fama and French (1997): industry 1 is consumer nondurables, industry 2 is consumer durables, industry 3 is manufacturing, industry 4 is energy, industry 5 is high tech business equipment, industry 6 is telecommunications, industry 7 is wholesale, retail, and similar services, industry 8 is health care, industry 9 is utilities, and industry 10 is all others.

31

Panel A: Portfolio SLI

30 30

Panel B: Portfolio BLI

25

25

20 Percentage Percentage 1 2 3 4 5 6 7 Industry Identifier 8 9 10

20

15

15

10

10

4 5 6 7 Industry Identifier

10

Panel C: Portfolio SM I
30 30

Panel D: Portfolio BM I

25

25

20 Percentage Percentage 1 2 3 4 5 6 7 Industry Identifier 8 9 10

20

15

15

10

10

4 5 6 7 Industry Identifier

10

Panel E: Portfolio SH I
30 30

Panel F: Portfolio BH I

25

25

20 Percentage Percentage 1 2 3 4 5 6 7 Industry Identifier 8 9 10

20

15

15

10

10

4 5 6 7 Industry Identifier

10

Figure 3. Industry Distributions of Firm-Month Observations for the Six Size-ROA Portfolios Used to Construct the Investment Factor. The classication of ten industries follows
Fama and French (1997): industry 1 is consumer nondurables, industry 2 is consumer durables, industry 3 is manufacturing, industry 4 is energy, industry 5 is high tech business equipment, industry 6 is telecommunications, industry 7 is wholesale, retail, and similar services, industry 8 is health care, industry 9 is utilities, and industry 10 is all others.

32