UNIVERSITY OF 
 
 ILLINOIS LIBRARY 
 
 AT URBANA-CHAMPAIGN 
 
 BOOKSTACKS 
 
UNIVERSITY LIBRARY 
 
 UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN 
 
 The person charging this material is responsible for its 
 
 renewal or return to the library on or before the due date. 
 
 The minimum fee for a lost item is $1 25.00, $300.00 for 
 
 bound journals. 
 
 Theft, mutilation, and underlining of books are reasons 
 
 for disciplinary action and may result in dismissal from 
 
 the University. Please note: self-stick notes may result in 
 
 torn pages and lift some inks. 
 
 Renew via the Telephone Center at 217-333-8400, 
 
 846-262-1510 (toll-free) orcirclib@uiuc.edu. 
 
 Renew online by choosing the My Account option at: 
 
 http://www.library.uiuc.edu/catalog/ 
 
Digitized by the Internet Archive 
 
 in 2011 with funding from 
 
 University of Illinois Urbana-Champaign 
 
 http://www.archive.org/details/fundamentalsstoc1664chan 
 
BEBR 
 
 FACULTY WORKING 
 PAPER NO. 90-1664 
 
 Fundamentals and Stock Returns 
 in Japan 
 
 Louis K.C. Chan 
 Yasushi Hamao 
 Josef Lakonishok 
 
 4uc '°W0 
 
 College of Commerce and Business Administration 
 Bureau of Economic and Business Research 
 University of Illinois Urbana-CHampaign 
 
NOTICE: Return or renew ell Library Materials! The Minimum Fee tor 
 
 each Lost Book is $50.00. 
 
 The person charging this material is responsible for 
 
 its return to the library from which it was withdrawn 
 
 on or before the Latest Date stamped below. 
 
 Theft, mutilation, and underlining of books are reasons for discipli- 
 
 na^acTn and ma, result in dismissal from the Un.versty. 
 
 To renew call Telephone Center, 333-8400 
 
 UNIVERSITY OF ILLINOIS LIBRARY AT URBANA-CHAMPAIGN 
 
BEBR 
 
 FACULTY WORKING PAPER NO. 90-1664 
 
 College of Commerce and Business Administration 
 
 University of Illinois at Urbana-Champaign 
 
 July 1990 
 
 Fundamentals and Stock Returns in Japan 
 
 Louis K.C. Chan 
 
 Yasushi Hamao 
 
 Josef Lakonishok 
 
 Department of Finance 
 
 Louis Chan and Josef Lakonishok are from the College of Commerce and Business Administration, 
 University of Illinois at Urbana-Champaign. Yasushi Hamao is from the Graduate School of 
 International Relations and Pacific Studies, University of California, San Diego. The article has been 
 presented at Purdue University, the University of California at Irvine, the Stockholm School of 
 economics, the London Business School, INSEAD, the CRSP Seminar on the analysis of Security 
 Prices at Chicago, and the Berkeley Program in Finance in Tokyo. The authors appreciate helpful 
 comments by Bill Bryan, Nai-Fu Chen, Don Keim, Claudio Loderer, Jay Ritter, Dennis Sheehan, 
 Neal Staughton, and Yuk Tse, and are grateful to Takeo Hoshi and Bing Yeh for their help with data 
 collection and to Brian Bielinski for research assistance. 
 
Abstract 
 
 This paper relates cross-sectional differences in returns on Japanese stocks to the underlying behavior of four 
 fundamental variables: earnings yield, size, book to market ratio, and cash flow yield. Alternative statistical 
 specifications and various estimation methods are applied to a comprehensive, high-quality data set that 
 extends from 1971 to 1988. The sample includes both manufacturing and non-manufacturing firms, 
 companies from both sections of the Tokyo Stock Exchange, and also delisted securities. Our findings reveal 
 a significant relationship between fundamental variables and expected returns in the Japanese market. Of 
 the four fundamental variables considered, the book to market ratio and cash flow yield have the most 
 significant positive impact on expected returns. 
 
The Japanese and the U.S. equity markets are, by far, the two largest in the world. As of March 
 1990, the two markets together accounted for 67 percent of the world's stock capitalization. In recent years, 
 money managers in the U.S., as well as in other countries, have substantially increased their exposure to the 
 Japanese market. In spite of the Japanese market's rapid growth and increasing significance in the 
 international financial arena, however, it is only recently that researchers have turned their attention to the 
 Tokyo market. 
 
 In contrast to the voluminous research in the U.S. relating stock returns to such fundamental 
 variables as firm size, earnings yield, cash flow yield and book to market value, there has been only very 
 limited research relating to the Japanese market. Moreover, the available Japanese evidence suffers from 
 methodological problems and a limited data base. For instance, previous studies focused on a subset of 
 stocks traded in the first section (the largest companies) for a relatively short time period and used only a 
 limited set of fundamental variables. To the best of our knowledge, information on delisted companies was 
 not available in previous studies; hence, the studies also suffer from a survivorship bias. 
 
 The purpose of our study is to fill this void by exploring the relationship between equity returns and 
 fundamental variables in Japan. The results provide evidence on which fundamental variables (if any) are at 
 work in the Japanese market. In contrast to earlier papers, we use a comprehensive, high quality data set. 
 Our data extends over a relatively long period of time (1971 to 1988) and contains most of the companies in 
 both the first and second sections of the Tokyo Stock Exchange, including delisted companies. The set of 
 fundamental variables employed is also more exhaustive than in previous papers; it includes earnings yield, 
 cash flow yield, size (market capitalization of equity) and book to market ratio (of equity). 
 
 Studying a market as important as the Japanese stock market is of interest in its own right. In 
 addition, our results, when placed alongside the evidence accumulated from studies of American data, may 
 also be useful in evaluating empirical models of the determinants of stock returns. Wherever possible, 
 therefore, we attempt to draw parallels between our findings and those from similar studies using American 
 data. In particular, the unprecedented appreciation of the Japanese stock market in recent years has raised 
 doubts as to whether fundamental valuation techniques developed in the U.S. apply to Japan. In a recent 
 Wall Street Journal article, M. Brauchli describes the Japanese market in the following terms: "Imagine a 
 stock market where investors ignore basic fundamentals such as earnings and corporate growth prospects. A 
 
market driven by floods of money. A market trading at mind-boggling price levels." (April 25, 1990) A 
 number of recent papers, surveyed in Frankel (1989), evaluate alternative explanations of the level of 
 Japanese equity prices. French and Poterba (1990), for example, conclude that differences in accounting 
 systems, discount rates and expected growth do not seem to explain the high level of Japanese stock prices. 
 On the one hand, finding that the same factors are at work in the two countries would strengthen our 
 confidence in the evidence accumulated for the U.S. An often-encountered criticism of studies using U.S. 
 data is that they are subject to data-snooping biases (e.g. Lakonishok and Smidt (1988) and Lo and 
 MacKinlay (1989)). Since stock prices have been studied so extensively, data-snooping biases are particularly 
 likely to be an issue. Fresh data, especially from one of the two largest stock markets in the world, is the 
 best remedy. On the other hand, finding that different factors are at work in the two countries would suggest 
 further research exploring institutional or behavioral differences between the two countries. For example, 
 Japanese regulations do not allow companies to use different reporting methods for tax purposes and for 
 public financial statements. As a result, companies predominantly use accelerated depreciation methods for 
 financial reporting. Firms with large capital investments thus tend to have substantially understated earnings. 
 Hence, we examine cash flow (earnings plus depreciation) yield in addition to earnings yield as an additional 
 fundamental variable. 
 
 In the U.S., the empirical relation between stock returns and fundamental variables has been 
 extensively studied. Furthermore, valuation models using fundamental variables have been widely used by 
 practitioners for at least 50 years. The results of previous studies are in general considered to be anomalous; 
 a positive relationship was found between equity returns and earnings yield, cash flow yield and book to 
 market ratio, and a negative relationship was found between equity returns and size. Especially voluminous 
 are the studies that document the size and the earnings yield effects and studies that try to disentangle the 
 two effects. See, for example, Basu (1977, 1983), Banz (1981), Reinganum (1981), Cook and Rozeff (1984), 
 Lakonishok and Shapiro (1986), Banz and Breen (1986), Jaffe, Keim and Westerfield (1989), and Ritter and 
 Chopra (1989). The shortcomings of accounting earnings motivated a number of recent papers to explore 
 the relationship between cash flow yields and stock returns (Bernard and Stober (1989) and Wilson (1986)). 
 Rosenberg, Reid, and Lanstein (1984) study the relationship between stock returns and the book to market 
 ratio. 
 
Despite extensive research, a consensus has not yet emerged on three issues. Are some or all of 
 these fundamental variables proxying for omitted risk factors? Does one fundamental variable (e.g. earnings 
 yield) subsume another (e.g. size)? Are the results robust to time period and sample composition? Evidence 
 from the Japanese market will shed some light on these issues. 
 
 Academic interest in the Japanese market is a relatively recent phenomenon. Hamao (1989) and 
 Hamao and Ibbotson (1989) characterize the Japanese capital markets by presenting summary statistics. 
 Tests of the Capital Asset Pricing Model for the Tokyo Stock Exchange by both Maru and Yonezawa (1984) 
 and Hawawini (1988) report that the model fits rather poorly. The multi- factor model or the Arbitrage 
 Pricing Theory is examined by Elton and Gruber (1988), Hamao (1988), and Brown and Otsuki (1990); they 
 all obtain significant supporting evidence for the model. Gultekin, Gultekin, and Penati (1989) apply the 
 APT framework to document the increasing degree of international capital market integration between the 
 U.S. and Japan. Campbell and Hamao (1989) examine the integration in terms of the long-run comovements 
 of expected returns, while Hamao, Masulis and Ng (1990) document short-run relationships between returns 
 and volatility across international markets. 
 
 In addition, the information content of the release of accounting reports is examined by Sakakibara, 
 Yamaji, Sakurai, Shiroshita, and Fukuda (1988) and Darrough and Harris (1989). Elton and Gruber (1989) 
 analyze the relationship between expectational data and actual stock performance. The difference between 
 U.S. and Japanese P/E multiples is examined by Bildersee, Chen and Lee (1990), and French and Poterba 
 (1990). 
 
 Finally, there are several studies that deal with anomalous price behavior on the Tokyo Stock 
 Exchange. Jaffe and Westerfield (1985) present day-of-the-week and turn-of-the-year patterns in daily index 
 returns. Daily patterns are also investigated using intraday index returns by Kato, Ziemba and Schwartz 
 (1990). Nakamura and Terada (1984) and Kato and Schallheim (1985) document size and seasonal 
 anomalies, while Nakamura and Terada (1984) and Aggarwal, Hiraki and Rao (1986) document P/E and size 
 effects. 
 
 Our findings reveal a significant relationship between fundamental variables and expected returns in 
 the Japanese market. Of the four variables considered, the book to market ratio and cash flow yield have a 
 reliably positive impact on expected returns. The performance of the size variable, although in general 
 
consistent with previous findings, turns out to be highly dependent on the specific model and time period. 
 While previous studies document a strong positive earnings yield effect, we find no such evidence after 
 controlling for the other variables. These conclusions are based on a variety of alternative statistical 
 specifications. In explaining expected stock returns, extra care should in general be taken to verify the 
 robustness of results before drawing any firm conclusions. 
 
 The remainder of the paper is structured as follows. In Section I, we describe our data. In Section 
 II, we present the methodology for testing the relationship between stock returns and fundamental variables. 
 Results from univariate analyses are in Section III. In Section IV, we present the results from regression 
 models. Section V examines the robustness of the results. Potential profits from trading strategies based on 
 the fundamental variables are documented in Section VI. The paper concludes in Section VII with a 
 summary of the findings. 
 
 I. Data Description 
 
 We use monthly data on stocks listed on the Tokyo Stock Exchange (TSE) from January 1971 to 
 December 1988. The Tokyo Stock Exchange, which accounts for more than 85 percent of total market 
 capitalization of Japanese equity, consists of two sections. The distinction between the two sections is similar 
 to that between the NYSE and AMEX. Typically, smaller firms are listed on the second section, and may 
 move to the first section when they satisfy the standards set by the exchange. The number of companies in 
 the sample changes over time, reflecting new listings and delistings. 
 
 Monthly returns (including dividends) and market capitalization are from the database compiled by 
 Hamao in collaboration with Daiwa Securities Co., Ltd., Tokyo. This database is comparable to the CRSP 
 (Center for Research in Security Prices at the University of Chicago) monthly database and is described in 
 Hamao and Ibbotson (1989). Two market indexes, an equally weighted and a value weighted portfolio of all 
 the stocks of the first and second sections of the TSE, are also derived from these data. Since there is no 
 Japanese equivalent to U.S. Treasury bills, we use a combined series of the call money rate (January 1971 - 
 November 1977) and the 30-day Gensaki (repo) rate (December 1977 - December 1988) as a risk-free 
 interest rate. This interest rate series is taken from Hamao and Ibbotson (1989). The data on the remaining 
 
three fundamental variables -- earnings yield, cash flow yield (earnings plus depreciation) and book to market 
 ratio -- are based on Nihon Keizai Shimbun Sha (1971 through 1988) as recorded by Daiwa Securities. 3 
 
 Aside from their limited database, previous studies of the Japanese market are subject to the 
 following two problems. First, it is typical for commercial databases to disregard delisted companies. This 
 problem is not unique to Japan. Therefore, studies using such databases suffer from survivorship bias which, 
 as Banz and Breen (1986) find, might lead to distorted results. Secondly, it has been customary in Japan to 
 limit the sample to manufacturing companies since accounting data on non-manufacturing (especially 
 financial) companies are difficult to obtain. However, the non-manufacturing sector is very large in Japan; 
 for example, in December 1988, the financial sector alone constituted 11 percent of the number of listed 
 firms and 30 percent of total market capitalization. 
 
 In contrast, our data set includes delisted stocks and non-manufacturing companies. There are 123 
 companies that were delisted from either section of the Tokyo Stock Exchange from 1971 to 1988. While this 
 includes firms that merged for strategic reasons, the majority experienced financial difficulties. In order to 
 warn investors, stocks that fail to satisfy the listing requirements are moved to the "warning post" by the 
 Tokyo Stock Exchange for three months prior to delisting. Our data on these companies are up to the last 
 month of being on the warning post. The accounting data on financial institutions are also obtained from 
 Nihon Keizai Shimbun Sha (1971 through 1988). As of December 1988 the entire sample includes 1570 
 companies, 1130 of which are in the first section. 
 
 The majority of Japanese firms have March as the end of their fiscal year. As of the end of 1988, 68 
 percent of the firms were on a March cycle. Some firms publish semi-annual statements; we do not use 
 these because of their tentative nature. The data available to us are from the parent-only financial 
 statements. Consolidated financial statements, which became mandatory only after 1978, are usually released 
 after a substantial delay - one to two months after the release of the parent-only statements. Also, 
 Darrough and Harris (1989) found, in a forecasting context, that the parent-only accounting information is 
 most relevant and Ishizuka (1987) found that the parent-only statements have more influence on stock prices 
 than consolidated statements. 
 
While a firm's stock price changes every month, its accounting numbers are updated only once a 
 year, in the month of announcement of the annual statements. All the fundamental variables change every 
 month because of the fluctuation in stock prices. 
 
 To summarize, our database is the most complete relative to prior studies: (1) it contains all 
 available firms from both the first and second sections, (2) the sample period is from 1971 to 1988, (3) it 
 includes delisted firms, and (4) it comprises the most comprehensive list of fundamental variables. 
 
 II. Methodology 
 
 The analysis of the relationship between stock returns and fundamental variables is conducted at the - 
 portfolio level. As stated, most companies in Japan have a fiscal year that ends on March 31, and as in the 
 U.S., practically all companies publish their financial statements within three months after the end of the 
 fiscal year. Accordingly, the portfolios we use are formed on the basis of the fundamental variables known to 
 investors as of the end of June for firms with March 31 and non-March 31 fiscal year ends. This ensures that 
 our tests are predictive in nature. 
 
 To form portfolios, a three-way grouping procedure is employed. Forming portfolios on the basis of 
 all the four fundamental variables would yield too few stocks per portfolio. In the first step, firms are ranked 
 by earnings yield as of the end of June in each year and placed into one of five groups. Group contains 
 stocks with negative earnings and groups one to four each contain an equal number of firms with positive 
 earnings. Group 1 comprises firms with the lowest positive earnings yields while group 4 comprises those 
 firms with the highest positive earnings yields. Each earnings yield group is then divided into four subgroups 
 on the basis of size. Subgroup 1 contains the stocks with the lowest capitalization while the largest stocks are 
 placed in subgroup 4. Each of these groups is in turn divided into five groups based on book to market ratio. 
 Stocks with negative book value are placed in subgroup 0; the remaining firms are sorted into four groups in 
 increasing order of book to market ratio. Each stock receives the same weight within its portfolio. The 
 result is a three-way grouping with a total number of 100 portfolios (5 X 4 X 5). In line with previous 
 studies, we focus in the regression analysis on firms with positive earnings and book values at the grouping 
 stage, thus a total of 64 portfolios (4 X 4 X 4). This portfolio formation procedure is updated in June of 
 each subsequent year. 
 
We employ the Seemingly Unrelated Regression (SUR) model to adjust simultaneously for portfolio 
 risk and test for the significance of the fundamental variables. Our basic model is 
 
 (1) Rp, - R,, = c^ + ^(RW, - R^) + J5p 2 (RE, - R^) + a, (E/P^ + ^(LS^ 
 
 + OjCB/MV, + a 4 (C/P)p, + ep, 
 P = 1, ... 64 t = 1, ... 210. 
 
 The dependent variable is the return on portfolio P in month t, Rp,, less the risk-free rate in month t, Rj,. 
 RW. is the return on a value weighted index in month t, while RE, is the return on an equally weighted index 
 in month t. The indexes contain all stocks from both the first and the second sections. The fundamental 
 variables analyzed include (E/P)p,, the average earnings yield for portfolio P in month t; (LS^, the average 
 of the natural logarithm of market capitalization for firms in portfolio P in month t; (B/M^, the average 
 book to market value for portfolio P in month t and (C/P)p,, the average cash flow yield for portfolio P in 
 month t. 4 All the fundamental variables are known at the beginning of month t. 
 
 Preliminary analysis showed that returns for typical companies track an equally weighted index 
 substantially better than a value weighted index. For example, in regressions of portfolio returns on index 
 returns, the R 2 averaged across the 64 portfolios is 74 percent for an equally weighted index and 27 percent 
 for a value weighted index. For larger companies, however, the R^ is substantially higher if a value weighted 
 index is used. Therefore, in the spirit of Huberman and Kandel (1985), we use two factors, the equally and 
 value weighted indexes, to account for market-wide risk. 
 
 Equation (1) provides a compact summary of the relation between stock returns and fundamental 
 variables. We also examine models based on subsets of the fundamental variables, and examine the 
 sensitivity of the results to the procedures used for portfolio formation (order of grouping) and sample 
 selection (by including or excluding second section stocks). 
 
 The SUR methodology used to estimate equation (1) is popular in the finance literature, with Jaffe, 
 Keim and Westerfield (1989) being a recent example. This methodology has two main advantages: (1) betas 
 are estimated simultaneously with the impact of the fundamental variables and (2) the procedure adjusts for 
 the cross-sectional correlation in the residual returns across portfolios. However, there is a strong 
 
possibility for bias in the way that the SUR methodology has been applied in prior research. The SUR 
 approach can be viewed as conducting a time-series regression for each of the 64 portfolios, requiring that 
 the coefficients of the fundamental variables be the same across all portfolios, and finally adjusting the 
 estimates for the cross-correlation in the residual returns. On the other hand, the underlying economic 
 hypothesis is that, at any point in time, cross-sectional differences in returns across stocks are determined by 
 cross-sectional differences in their fundamental variables. The cross-sectional distribution of fundamental 
 variables may vary over time, however. For example, suppose that size is an important fundamental variable. 
 Then a company whose absolute size is the same in 1971 and 1988 would not have the same market-adjusted 
 return, if its size relative to its peer firms has changed over time. Indeed, the level of fundamental variables 
 in Japan has changed substantially over time, due to the unprecedented appreciation in the value of Japanese 
 stocks. For example, large companies in 1971 (the earliest year in our sample) would not be considered 
 large in 1988; a P/E ratio of 15 would be considered very high in some years and low in others. An 
 adjustment for the level of the fundamental variables in any particular year is therefore essential. We 
 perform such an adjustment by dividing each fundamental variable every month by its corresponding cross- 
 sectional average from the previous June, thus preserving cross-sectional differences when pooling data across 
 time. 
 
 As an alternative to the SUR methodology, we also apply the Fama-MacBeth (1973) methodology. 
 The two methodologies are different and the sensitivity of the results to the various procedures is of interest. 
 The SUR methodology assumes that the betas for the portfolios are constant over time, whereas the Fama- 
 MacBeth procedure updates betas periodically. The Fama-Macbeth methodology, unlike the SUR model, 
 also allows the coefficients of the explanatory variables to vary across months. These coefficients are then 
 averaged across time. Such aggregation, however, assumes that the coefficients are drawn from an 
 underlying stationary distribution. Since the level of an independent variable affects the magnitude of its 
 coefficient, dramatic growth over time in the levels of the explanatory variables may invalidate the assumption 
 of a stationary distribution. In order to alleviate this potential non-stationarity, we again follow the 
 procedure of dividing the fundamental variables in each month by its corresponding cross-sectional average 
 from the previous June. In addition, estimating the monthly cross-sectional regressions by ordinary least 
 squares is likely to be inefficient, since the residuals across portfolio groups are in general correlated. In 
 
order to obtain a more powerful test, we also carry out the Fama-MacBeth methodology using generalized 
 least squares for each monthly cross-section. 
 
 In a related context, the findings of Chen and Kan (1989) further emphasize the importance of 
 verifying the robustness of results to econometric procedures. Chen and Kan found that a variable (the bid- 
 ask spread) that is an important predictor for portfolio returns using one approach becomes insignificant 
 using another approach. 
 
 We also check whether the results are sensitive to the methods used to form portfolios and to 
 sample selection. Our tests are replicated using portfolios formed by different orderings of the four 
 fundamental variables. Previous research on the Japanese market focused on the largest stocks -- those 
 stocks that are traded in the first section. We estimate our models for all the firms in the sample and 
 separately for the first and second section firms. Our time period is relatively long, enabling us also to 
 produce results for two subperiods and to observe possible changes over time. 
 
 All our tests hitherto use data on portfolios, where securities are ranked and grouped into portfolios 
 on the basis of various fundamental characteristics. In an important recent paper, Lo and MacKinlay (1989) 
 point out some difficulties with statistical tests based on data grouped by some empirically-motivated 
 fundamental attributes (such as size). In particular, the portfolio grouping approach may bias test statistics 
 and spuriously exaggerate the relationship between portfolio excess returns and the fundamental attributes. 
 The analysis of Lo and MacKinlay (1989), however, also suggests that the bias is less severe if data on 
 individual securities is used, without grouping into portfolios. In order to guard against the possibility that 
 our conclusions are distorted by biased test statistics, we apply the Fama-MacBeth procedure to data on 
 individual securities, instead of portfolios. 
 
 Prior research, especially a recent paper by Jaffe, Keim, and Westerfield (1989), indicates that the 
 relationship between equity returns and fundamental variables is different in the U.S. for January versus non- 
 January months. This issue will be addressed by introducing a dummy variable for January and non-January 
 months into equation (1). 
 
III. Univariate Analysis of Returns and Fundamental Variables 
 
 In Table I, we present average monthly returns for various groups of stocks formed by the four 
 fundamental variables (see also Figure 1). In each panel the stocks are grouped by a different fundamental 
 variable; the grouping procedure is repeated every year based on end of June values. For each of the 
 fundamental variables, four portfolios of equal size are formed for positive values and a special group for 
 negative values. In addition, for each grouping, we present the average values of the other fundamental 
 variables. The last two rows in a panel present the OLS beta calculated using an equally weighted market 
 portfolio based on all the first and second section stocks, and the average number of securities in each 
 portfolio. 
 
 From Table I, high E/P stocks outperform low E/P stocks with a difference of 0.40 percent per 
 month between the top and the bottom quartiles. Small stocks achieve substantially higher returns than large 
 stocks, registering a difference of 0.97 percent per month between the two extreme groups. Firms with large, 
 positive book to market ratios earn a premium of 1.10 percent over firms with low, positive book to market 
 ratios. Lastly, the difference in returns between the two extreme groups for the cash flow yield variable is 
 0.79 percent, which is larger than the spread in returns on the basis of earnings yield. These results are 
 consistent with U.S. findings (e.g., Jaffe, Keim and Westerfield (1989)) and with practitioners' views about 
 the relationship between stock returns and each of the four fundamental variables. 
 
 A relatively high return is achieved by stocks with negative fundamental values — stocks that most 
 likely performed poorly in the period preceding the portfolio formation period. Stocks with negative E/P, 
 B/M and C/P experience an average monthly return that exceeds 2.5 percent, relative to approximately 1.8 
 percent for the other groups. The difference in abnormal returns between "losers" and "non-losers" is smaller 
 because of the higher betas of the "losers", although the somewhat higher betas for losers cannot explain the 
 difference in returns. A similar result was obtained by Jaffe, Keim and Westerfield (1989), who concluded 
 that, "Surprisingly, stock with negative earnings outperform many of the positive earnings portfolios in the 
 other five columns." 
 
 The results in Table I also reveal that the various fundamental variables are correlated. For 
 example, stocks with high book to market ratios tend to have relatively high cash flow yield ratios. Hence, a 
 
 10 
 
multivariate analysis is required to disentangle the impact of the various fundamental variables on stock 
 returns. 
 
 IV. Regression Results 
 
 A. Seemingly Unrelated Regressions (SUR) with Unadjusted Fundamental Variables 
 
 The results in Table II are based on a SUR model. All the regressions use the same 64 portfolios 
 where first we group by earnings yield, then by size and last by book to market ratio. The first four models 
 include one of the four independent variables at a time. Models 5 to 7 include various combinations of the 
 fundamental variables and model 8 includes all our fundamental variables. The sample comprises stocks 
 from both sections and the fundamental variables, as in previous studies utilizing this approach, are not 
 adjusted for the change in levels over time. 
 
 The results are striking. They differ from the findings obtained from U.S. data and from previous 
 results on the Japanese market. Moreover, the results conflict with the univariate tests presented in Table I. 
 The earnings yield effect is not significant across the different models; it is not even significant if E/P is the 
 only independent variable. If all the fundamental variables are included, the coefficient on E/P has a 
 negative sign, implying that stocks with low earnings yields outperform stocks with high earnings yields. 
 
 Size, in general, is significant but, with an unexpected sign in all the models — large companies in 
 Japan tend to outperform small companies. This finding is very surprising and is also inconsistent with the 
 univariate results in Table I. Recall from Table I that a portfolio of low capitalization stocks outperforms a 
 portfolio of high capitalization stocks by as much as one percent per year. This discrepancy in results may be 
 due to the fact that the distribution of the fundamental variables changes over time, and the pooling 
 procedure fails to account for such changes. 
 
 While the C/P and B/M variables are widely used by the financial community as important 
 indicators of value, they have received relatively less attention in the academic literature. We find that the 
 cash flow yield variable and the book to market variable have the expected positive signs. The coefficient on 
 the C/P variable is marginally significant, whereas that on the B/M variable is at least two standard 
 deviations away from zero in all the models. 
 
 11 
 
All previous studies on the Japanese market examine only stocks from the first section. Hence, to 
 explore whether the inclusion of second section stocks is driving the unexpected results in Table II, we 
 present Table III, which excludes second section stocks. The results are, in general, similar. According to 
 model 8, which includes all the four variables, E/P and size are statistically significant, but with an 
 unexpected sign. On the other hand, the results for B/M and C/P parallel previous findings. The results for 
 some of the other models are somewhat sensitive to the sample composition. For example, in model 5, size 
 is only marginally significant if only first section stocks are included, whereas size is highly significant if data 
 from the first and second sections are pooled. 
 
 B. SUR with Adjusted Fundamental Variables. 
 
 The dramatic surge in Japanese stock prices has caused substantial changes in the levels of 
 fundamental variables. Therefore, the traditional approach of applying the SUR model without adjusting for 
 time series variation in the level of fundamental variables is questionable and may account for the unexpected 
 results in Tables II and III. Table IV presents the mean, median, 25th percentile and 75th percentile of the 
 four fundamental variables, for every third year. The results reveal a sharp decrease over time in E/P, C/P 
 and B/M and a sharp increase in size. For example, in 1973 the mean size is 28.39 billion yen and by 1988, 
 300.95 billion yen; the top quartile's boundary is 24.00 billion yen in 1973, which is below the median by 1985 
 and in the bottom quartile by 1988. With respect to E/P, the 25th percentile in 1973 is 3.96 percent, which is 
 in the top quartile by 1988. Similar changes over time occur with C/P and B/M. 
 
 To correct for the model misspecification in Tables II and III, each fundamental variable is deflated 
 each month by its cross-sectional mean value as of the preceding June. The adjusted value preserves the 
 cross-sectional ranking of each fundamental variable at successive points in time. The results with the 
 adjusted fundamental variables for all stocks in the first and second sections are presented in Table V. 
 These results are in sharp contrast to the results in Tables II and III; the findings are now compatible with 
 the univariate tests presented in Table I, and with previous evidence as well. Specifically, E/P, B/M, and 
 C/P each individually has a reliably positive influence on subsequent realized returns, while a reliably 
 negative relationship exists between size and future returns. Model 5 attempts to unravel the separate 
 influences of size and earnings yield on returns. While the t-statistics suggest that the coefficients are not 
 
 12 
 
estimated with a high degree of precision, neither variable dominates the other. Adding the book to market 
 ratio as the third independent variable in Model 6, however, robs both size and earnings yield of any 
 predictive power. It would thus appear that both the size and earnings yield effects are subsumed in our 
 sample by the influence of the book to market ratio. Such a model has not been previously tested in the 
 literature, and these results indicate the sensitivity of inferences to model specification. 
 
 In Model 7, earnings yield is replaced by the cash flow measure. For Japanese firms, the cash flow 
 variable may be more informative than earnings yields, since reported earnings are likely to be distorted by 
 the substantial divergence between economic and reported depreciation. In both Models 7 and 8, where cash 
 flow and earnings yield are simultaneously included, the cash flow variable has a positive coefficient and a t- 
 statistic slightly above two. These results suggest that C/P may be more important in predicting returns than 
 E/P. 
 
 Another variable that has been overlooked by academics, but which has a wide following among 
 practitioners, is the book to market ratio. This variable consistently has the largest coefficient 7 and the 
 highest t-statistic in every model in Table V. Since the book to market ratio is likely to be less noisy than 
 earnings and cash flow yields, which are very sensitive to the events of a single year, its influence can be 
 detected more reliably. In the full model (Model 8), only the coefficients on B/M and C/P are statistically 
 different from zero. If anything, the results for these two variables may be understated; in forming portfolios, 
 we group by B/M last and do not group by C/P, thereby dampening the dispersion across portfolios in the 
 values of these variables. 
 
 The magnitude of the coefficients suggests that the variables are not only statistically significant but 
 economically significant as well. For example, in 1988, for the book to market variable, the difference 
 between the boundaries of the upper and lower 10 percentiles is 1.28 (after dividing by the mean). 
 According to model 8 (Table V), the coefficient of the book to market ratio is 0.0034, which implies a 
 difference in returns of .44 percent per month. 
 
 To examine whether the results are sensitive to the composition of the sample, we present Table VI 
 which includes only stocks from the first section. The results are, in general, similar to those in Table V. In 
 the context of a comprehensive model (Model 8) the coefficients for E/P and size are statistically 
 
 13 
 
indistinguishable from zero; the effect of B/M is highly significant, while the coefficient for cash yield is 
 about 1.8 times its standard error. 
 
 The unprecedented appreciation of the Japanese market in recent years has motivated arguments on 
 the applicability of traditional fundamental variables in explaining cross-sectional differences in stock returns. 
 This issue is addressed in Table VII, where we provide results based on adjusted fundamental variables for 
 the second half of the period, 1980 to 1988. These results should be interpreted with caution because of the 
 relatively short time period. The results from Table V in general carry over, with the only exception being 
 that the effect of the cash flow yield variable becomes statistically weaker in Models 7 and 8. As before, 
 book to market is highly significant across the different models. 
 
 A number of previous papers, in studying the impact of fundamental variables on stock returns, 
 investigate January and non-January months separately. Cook and Rozeff (1984) find a significant E/P and 
 size effect in both January and non-January months; Jaffe, Keim and Westerfield (1989) find that size is 
 significant only in January while E/P is significant in both January and the rest of the year. 
 
 To address this issue for the Japanese market, we expand the SUR model described in Equation 1, 
 by introducing dummy variables for January and non-January months. Our basic model is: 
 
 ( 2 ) Rpt - Rft = %i Ck a + V 1 -^) + <W RW t - Rft^jt + W RW t - RftX 1 - D it ) 
 
 + ^(RE, - R^D, + ^(REt " W " D jt> + V E / P >Pt D |< 
 + o 1r (E/P) Pt (l - D. lX ) + ^(LS^Dj, + a, r (LSk(l - D jt ) + ^ (B/M^ D Jt 
 + a^B/M^l - Ci t ) + a 4j (C/PV,D Jt + ^(C/P^l - D Jt ) + ^ 
 P = 1, ... 64 t = 1, ... 210 
 
 where D rt is a dummy variable that takes the value of one in the month of January and zero otherwise. The 
 
 other variables are defined as in Equation 1. 
 
 In Table VIII results are presented for the comprehensive model (Equation 2) as well as models 
 
 based on partial sets of fundamental variables. We report results for adjusted fundamental variables. 
 
 Consistently across all the models, E/P has a negative sign in January and is in general significant. This is in 
 
 contrast to findings in the U.S. where Jaffe, Keim and Westerfield (1989) obtain a much higher positive 
 
 coefficient for E/P in January than in other months of the year (0.0633 vs 0.0209, their Table II). The results 
 
 for E/P for non-January months are similar to the results across all the months presented in Table V. 
 
 14 
 
The results for the size variable also differ from U.S. findings. Jaffe, Keim, and Westerfield (1989) 
 found that the size coefficient is significant only in January. In Japan, the size coefficient is in general not 
 significant and there is no noticeable difference between January and non-January months. With respect to 
 the remaining two variables, B/M and C/P, January does not seem to be a unique month. 
 
 The analysis that compares January to non-January months should be interpreted with care, since 
 there are only 17 Januaries in the sample period. Studies in the U.S., such as Jaffe, Keim and Westerfield 
 utilized much longer time periods. Based on our results, and in contrast to the U.S. studies, January does 
 not stand out as a special time period. 
 
 V. Sensitivity Analysis 
 
 In this section the sensitivity of the results presented above will be examined with respect to: (1) 
 various model specifications, (2) grouping procedures, and (3) sample composition. The results presented 
 are based on the SUR and Fama-MacBeth methodologies. 
 
 A. Sensitivity Analysis for the SUR Model 
 
 In order to reduce the burden on the reader, Table IX presents results for only three of the models 
 estimated: Models 5, 6, and 8. The first panel of the table investigates the sensitivity of the results to the 
 ranking procedure used to form portfolios. In particular, stocks are aggregated into portfolios first on the 
 basis of the book to market ratio, then by cash flow yield and finally by size. Panel A of Table IX yields 
 results that are qualitatively similar to the findings of the previous section. B/M and C/P continue to exert a 
 reliably positive effect on returns in the subsequent period. While size and returns are still negatively related, 
 the t-statistics for the size variable are sensitive to the specification of the model. The earnings yield variable 
 has a negative coefficient that, in contrast to the earlier results, is now highly significant. Portfolios were also 
 formed using alternative orderings of the variables by which securities are ranked; the basic findings of the 
 previous section are essentially unaltered. 
 
 Panel B investigates whether the results are sensitive to the generalized least squares correction 
 employed in the SUR procedure to correct for the contemporaneous correlation of the residuals. In the first 
 pass of the SUR procedure, the time series and cross-sectional observations are pooled, and the model is 
 
 15 
 
estimated under ordinary least squares, without adjusting for the residual correlations; it is these estimates 
 from the first pass that are reported in Panel B of Table IX. While the evidence there reveals that the 
 general results are robust, there are substantial changes in the magnitudes of individual coefficients and their 
 associated t-values. For example, under Model 8, the coefficient of B/M and its t-statistic are 0.0057 and 
 6.02, respectively, in Panel B of Table IX; the corresponding values from Table V are 0.0034 and 3.03. These 
 differences are consistent with non-negligible correlations across portfolio residuals, illustrating the potential 
 advantage of employing the SUR methodology. 
 
 The results in Panel C of Table IX exclude nonsurviving firms from the sample (a total of 123 
 companies between 1971 and 1988). Most studies of foreign stock markets suffer from such survivorship bias 
 - information on delisted firms is generally not available from the commercial services that typically furnish 
 the data for such studies. To the best of our knowledge, all prior studies of the Japanese market lack 
 information on delisted companies. In accordance with the results of Banz and Breen (1986) using U.S. data, 
 we also find that the survivorship bias appears to affect the outcome of the statistical tests. Model 5, which 
 includes both size and earnings yield, is a specification that has appeared in many studies. Applying this 
 model to surviving firms only leads to the conclusion that both E/P and size have the expected signs and are 
 highly significant. When delisted firms are included (as in Table V), however, E/P and size retain their 
 expected signs, but their t-statistics fall below two. In the sample containing only surviving firms, the 
 influence of size tends to be statistically stronger -- its t-statistic in Model 8 is -1.94, relative to -1.15 when 
 delisted firms are included. 
 
 B. Fama and MacBeth Procedure 
 
 As a final check on the robustness of the results, this section presents alternative estimates of 
 Models 5, 6, and 8 using the methodology developed in Fama and MacBeth (1973). 
 
 In Table X, the first four panels are based on portfolios formed from securities ranked first by 
 earnings yield, then by size and finally by book to market ratio. The results in the last panel are for 
 individual securities. In June of each year, the beta coefficients for each security (with respect to both the 
 equally and value weighted market indices) are estimated from the most recent thirty-six months of data. 
 The risk measures of the individual securities in a portfolio are averaged to obtain the portfolio risk 
 
 16 
 
estimates. Cross-sectional regressions relating returns to betas and fundamental variables from the prior 
 period are then conducted in each of the next twelve months. 
 
 As suggested by Litzenberger and Ramaswamy (1979), two possible ways of averaging the time series 
 of coefficients from the monthly cross-sectional regressions are presented: a simple average (Panel A of 
 Table X) and a weighted average, where the weights are inversely proportional to the variance of each 
 monthly estimate (Panel B). In line with the results from the SUR model, portfolio returns are reliably and 
 positively related to B/M and C/P. While the market capitalization variable has a negative coefficient, its t- 
 statistic is somewhat sensitive to the specification of the model. The sign of the coefficient for earnings yield, 
 however, is contrary to intuition. Evidently, at least two of the fundamental variables ~ the book to market 
 ratio and cash flow yield — have predictive power for portfolio returns, even after allowing the relationship 
 between returns and fundamentals to vary each month. 
 
 The results obtained earlier from the SUR model suggest that residual correlations across portfolios 
 may be substantively important (Table IX, Panel B). In Panel C of Table X, the results for each monthly 
 cross-sectional regression are obtained using generalized least squares. The estimated residual covariance 
 matrix used to form the cross-sectional estimates is the same matrix obtained from the SUR model. The 
 results in Panel C are similar to those in the previous panel; the main effect of the generalized least squares 
 adjustment is to lower the t-statistics for the size variable. 
 
 The Fama and MacBeth approach assumes that the coefficients estimated every period are drawn 
 from a stationary distribution. Changes over time in the levels of the explanatory variables will invalidate this 
 assumption. To alleviate this problem, we follow the same approach as for the SUR procedure and divide 
 the fundamental variables each month by the corresponding cross-sectional average from the previous June. 
 Panel D of Table X reports the simple time-series average of the monthly coefficients estimated using 
 ordinary least squares. No substantive changes arise as a result of adjusting the fundamental variables for 
 changes in levels over time; the results in Panel D essentially mimic those of Panel A. 
 
 In panel E we present results at the level of individual securities. In a recent paper Lo and 
 MacKinlay (1989) point out biases in statistical inferences when portfolios are formed by some empirically- 
 motivated fundamental attributes. They suggest that the bias is less severe when data on individual securities 
 is used. Another advantage of conducting the analysis at the level of individual securities is that arbitrary 
 
 17 
 
decisions on the order of grouping securities into portfolios do not have to be made. As a result, the 
 explanatory variables are all assigned equal importance beforehand. In comparing the relative importance of 
 size and earnings yield (Model 5), each variable has the expected sign. On the basis of their t-statistics, each 
 variable contributes some predictive power for returns that is not accounted for by the other. When 
 additional explanatory variables are introduced, however, as in Models 6 and 8, the coefficient for the E/P 
 switches sign and becomes negative. On the other hand, the coefficient for size is only slightly less than two 
 standard errors below zero. In both Models 6 and 8, the coefficients for B/M and C/P are positive, and 
 highly significant. Overall, the results with individual securities in Panel E suggest that our findings cannot 
 be entirely attributed to biased test procedures, nor to the methods used to group securities into portfolios. 
 
 VI. A Trading Strategy Based on Book to Market Ratio and Cash Flow Yield 
 
 After considering all the various specifications examined, the two fundamental variables with the 
 most significant association with stock returns are the book to market ratio and cash flow yield. There is of 
 course the possibility that some other variables, omitted from our list of fundamental variables, are driving 
 stock returns and B/M and C/P are correlated with these omitted variables. However, even in such a case, 
 profitable trading strategies based on B/M and C/P can be designed if the correlation between these 
 variables and the omitted variables can be assumed constant. 
 
 In Table XI average monthly returns, standard deviations of returns, and betas are presented for 
 portfolios formed each June on the basis of B/M and C/P. First, stocks with positive B/M are sorted into 
 four groups; then, within each B/M group, stocks with positive C/P are also sorted into four groups. This 
 procedure results in a time series of monthly observations on a total of 16 portfolios. The results are 
 graphed in Figure 2. 
 
 Within each of the four cash flow yield groups there is a positive monotonic relationship between 
 returns and book to market ratio. The relationship between returns and cash flow yield is nearly monotonic 
 for each of the four book to market groups as well. Comparing a portfolio of stocks with the highest B/M 
 and C/P ratios to a portfolio of stocks with the lowest B/M and C/P ratios gives rise to a substantial 
 difference in returns: 2.73 percent versus 1.15 percent respectively. The difference of 1.58 percent per month 
 is almost the size of the average monthly return on an equally weighted portfolio, which during this period of 
 
 18 
 
time was 1.85 percent. This difference cannot easily be explained by the differences in risk measures such as 
 betas and standard deviations. For example, the highest-performing portfolio has a beta of 0.97 versus a beta 
 of 0.89 for the lowest-performing portfolio. In general, there are substantial differences in returns across the 
 16 portfolios presented in Table XI, but much smaller differences in betas and standard deviations. These 
 results suggest that fundamental variables may be useful in designing potentially profitable trading strategies. 
 
 VII. Summary and Conclusions 
 
 The relationship between fundamental variables and stock returns has attracted a considerable 
 amount of research attention in the U.S. In contrast, there is very limited evidence for the Japanese market. 
 What available evidence there is suffers from methodological problems and a limited data base. In order to 
 fill this void, this paper relates cross-sectional differences in returns on Japanese stocks to the underlying 
 behavior of four fundamental variables: earnings yield, size, book to market ratio and cash yield. Because it 
 is difficult to pre-specify theoretically the correct relationship between observable fundamental variables and 
 stock returns, it is important to verify that any relationships that are uncovered are not merely statistical 
 artifacts. Accordingly, this paper takes great care to apply alternative statistical specifications, and various 
 estimation methods, to a comprehensive, high-quality data set that extends from 1971 to 1988. In contrast to 
 previous studies, our sample includes both manufacturing and non-manufacturing firms, companies from both 
 sections of the Tokyo Stock Exchange and also delisted securities, hence overcoming the survivorship bias. 
 The results are based on returns on individual securities, as well as returns on portfolios, where portfolios are 
 constructed under several different grouping schemes. Our analysis varies the sample period and allows the 
 relationship to differ in January from the rest of the year. 
 
 Our findings reveal a significant relationship between fundamental variables and expected returns in 
 the Japanese market. The performance of the book to market ratio is especially noteworthy; this variable is 
 statistically and economically the most important of the four variables investigated. For example, if all the 
 four fundamentals are considered together (Table V, Model 8), the expected difference in returns between 
 the 10th and the 90th percentiles of firms ranked by the book to market ratio is 0.44 percent per month. The 
 importance of this variable persists if the tests are applied to stocks from the first section only, and if only the 
 latter half of the sample period is considered. In addition, the predictive role of the book to market ratio is 
 
 19 
 
not a phenomenon specific to the month of January. Another variable, cash flow yield, also has a positive 
 and in general highly significant impact on expected returns. 
 
 The remaining two fundamental variables ~ earnings yield and size - have been the sole focus of 
 the few available cross-sectional studies on the Japanese market, and have been extensively analyzed in 
 studies on the U.S. market. Our findings confirm the existence of a "size effect"; small firms in our sample 
 tend to outperform larger firms, after adjusting for market risk and the other fundamental variables. 
 However, the statistical significance of the market capitalization variable is sensitive to the specification of the 
 model; indeed, in some cases it is not significant. 
 
 Of the four variables considered, it is hardest to disentangle the effect of the earnings yield variable. 
 In Table I, for example, a strategy of holding stocks with high E/P ratios would be profitable relative to a 
 strategy of holding stocks with low E/P ratios. In Table V, if earnings yield is considered in isolation, or if it 
 is included with size, it indeed has a positive and significant impact on returns. If the book to market ratio is 
 added to the model, however, the coefficient on earnings yield becomes insignificantly different from zero. In 
 the context of the full model, earnings yield even has a negative impact on stock returns, and is in some cases 
 statistically significant. 
 
 Overall, of the four variables we considered, the book to market ratio and cash flow yield have the 
 most significant impact on expected returns. This is intriguing, since these two variables, although popular in 
 the financial community, have received very little attention in the academic literature, in comparison to size 
 and earnings yield. 
 
 Our findings raise the possibility that earlier studies linking stock returns in Japan or the U.S. to 
 fundamental variables should be interpreted with caution; potentially important explanatory variables may 
 have been omitted. In our context, if we were to restrict attention to the model including only size and 
 earnings yield, in Table V, we would be led to conclude, as earlier studies have, that E/P has a positive and 
 nearly significant impact on stock returns. Yet this inference would be reversed when a larger set of 
 fundamental variables is considered. Our results suggest that, in general, such statistical inferences should be 
 made with caution since the correct specification is unknown. 
 
 20 
 
We do not claim, of course, that the list of fundamental variables employed in this paper is 
 exhaustive. While our sample period is long relative to studies of foreign stock markets, it is nonetheless 
 short and covers fewer different market cycles, in comparison with studies of American data. 
 
 Nevertheless, potentially profitable trading strategies based on our fundamental variables can be 
 designed if the correlation between these variables and any omitted variables is constant. In our sample, a 
 portfolio of stocks with the highest values for B/M and C/P earns 1.58 percent per month more than a 
 portfolio of stocks with the lowest B/M and C/P ratios. 
 
 Many U.S. studies have found January to be a unique month. Following their lead, we allow the 
 relationship between returns and fundamental variables to differ between January and non-January months. 
 In contrast to U.S. studies, however, January does not stand out as a special month. Our sample period, 
 however, extends over only seventeen Januaries. 
 
 In addition to using a larger set of fundamental explanatory variables, our results are based on a 
 SUR model that pools time-series observations on portfolios ranked by several fundamental characteristics. 
 Previous applications of this methodology, however, have failed to adjust for changes over time in market- 
 wide levels of the fundamental variables. Since the economic hypothesis predicts that relative returns in a 
 cross-section of portfolios are determined by the relative values of their fundamental variables, the failure to 
 perform this adjustment results in a mis-specified model. In light of the unprecedented rise in Japanese 
 stock prices, this adjustment is substantively important for our study. For example, the results using 
 unadjusted fundamental variables indicate that large companies in Japan significantly outperform small 
 companies. This is at odds with previous evidence and results based on adjusted fundamental variables. 
 Therefore, previous studies that failed to adjust for the level of fundamental variables may have to be re- 
 evaluated. 
 
 Another commonly used methodology, developed by Fama and MacBeth (1973), is less subject to the 
 above shortcoming because separate cross-sectional regressions are run every month. Indeed, our results 
 with the Fama-MacBeth procedure are compatible with the results from the SUR model using adjusted 
 variables. As an additional safeguard against the biases noted by Lo and MacKinlay (1989), we also apply 
 the Fama-MacBeth procedure to data on individual securities, without grouping into portfolios. There are no 
 essential differences between the results using individual securities and those using portfolio groups. 
 
 21 
 
Although we make no attempt to rationalize the general level of Japanese equity prices, it is perhaps 
 reassuring nonetheless to find that cross-sectional differences in Japanese stock returns are indeed related to 
 a small set of relatively standard fundamental variables. It is perhaps also understandable that the cash flow 
 yield variable has higher predictive power than does earnings yield, in light of the distortions in the earnings 
 of Japanese firms induced by accelerated depreciation allowances. The noise in reported earnings, related to 
 Japanese accounting standards, may also help to explain why the book to market ratio has such a strong 
 influence. 
 
 22 
 
Footnotes 
 
 1. The requirements include 20 million shares issued, average monthly trading volume of 200,000 shares and 
 continuous dividend payment of more than 5 yen per year over the last three years. See Tokyo Stock 
 Exchange (1988) for details. 
 
 2. While TOPIX (Tokyo Stock Price Index) is available from the Tokyo Stock Exchange, this value weighted 
 index is based on section one stocks and excludes dividends. 
 
 3. Nihon Keizai Shimbun Sha, the equivalent of Moody's Handbooks in the U.S., provides financial 
 information on first and second section stocks listed on the Tokyo Stock Exchange. 
 
 4. Cash flow is defined to be the sum of earnings and depreciation; the data were provided by Daiwa 
 Securities. 
 
 5. See Brown, Kleidon and Marsh (1983) for a discussion of the SUR methodology. 
 
 6. Note that previous studies of the Japanese market typically use a cross-sectional regression approach, 
 rather than the SUR methodology. 
 
 7. The variables are deflated by their mean value and hence the size of the coefficients is indicative of their 
 impact on returns. 
 
 8. Previous studies of the Japanese market find that small firms outperform large firms in the month of 
 January. These studies, however, rely on a subset of the first section stocks and exclude delisted firms. 
 When our results from Table VIII are replicated for only surviving firms, the coefficient of the size variable is 
 substantially higher in January than in other months. For example, in Model 8 of Table VIII, the coefficient 
 of the size variable is -0.0003 in January and -0.0002 in other months. When delisted firms are excluded, the 
 corresponding estimates are -0.0021 in January and -0.0003 in other months. The discrepancy between our 
 results in Table VIII and those of previous studies can therefore be attributed in part to survivorship bias. 
 
 9. Note that since portfolios are not formed by ranking securities on the basis of estimated betas, an extra 
 period for estimating portfolio betas is not necessary. 
 
 23 
 
References 
 
 Aggarwal, Raj, Takato Hiraki, and Ramesh Rao, 1986, The nature of risk adjusted returns on the 
 Tokyo Stock Exchange: P/E, size and seasonal influences, unpublished paper. 
 
 Banz, Rolf, 1981, The relationship between returns and market value of common stocks, Journal of 
 Financial Economics . 9, 3-18. 
 
 Banz, Rolf and William Breen, 1986, Sample-dependent results using accounting and market data: 
 some evidence, Journal of Finance . 41, 779-93. 
 
 Basu, Sanjoy, 1977, Investment performance of common stocks in relation to their price-earnings 
 ratios: a test of the efficient market hypothesis, Journal of Finance . 32, 663-82. 
 
 Basu, Sanjoy, 1983, The relationship between earnings yield, market value and return for NYSE 
 common stocks: further evidence, Journal of Financial Economics . 12, 129-56. 
 
 Bernard, Victor and Thomas Stober, 1989, The nature and amount of information in cash flows and 
 accruals, Accounting Review . 64, 624-652. 
 
 Bildersee, John, John Chen, and Changwoo Lee, 1990, The international price-earning ratio 
 phenomenon: A partial explanation, Working Paper, New York University. 
 
 Brown, Philip, Alan Kleidon, and Terry Marsh, 1983, New evidence on the nature of size-related 
 anomalies in stock prices, Journal of Financial Economics . 12, 33-56. 
 
 Brown, Stephen and Toshiyuki Otsuki, 1990, Macroeconomic factors and the Japanese equity 
 markets: The CAPMD project, in Edwin Elton and Martin Gruber(eds.), Japanese Capital Markets . New 
 York: Harper and Row. 
 
 Campbell, John and Yasushi Hamao, 1989, Predictable stock returns in the U.S. and Japan: A study 
 of long-term capital market integration, NBER Working Paper No. 3191. 
 
 Chen, Nai-fu and Raymond Kan, 1989, Expected returns and the bid-ask spread, Working paper, 
 University of Chicago. 
 
 Cook, Thomas and Michael Rozeff, 1984, Size and earnings/price ratio anomalies: One effect or 
 two?, Journal of Financial and Quantitative Analysis . 19, 449-466. 
 
 Darrough, Masako and Trevor Harris, 1989, Do management forecasts of earnings affect stock prices 
 in Japan?, Working Paper, Columbia University. 
 
 Elton, Edwin and Martin Gruber, 1988, A multi-index risk model of the Japanese stock market, 
 Japan and the World Economy . 1, 21-44. 
 
 Elton, Edwin and Martin Gruber, 1989, Analysts' expectation and Japanese stock prices, Japan and 
 the World Economy . 2. 
 
 Fama, Eugene and James MacBeth, 1973, Risk, return and equilibrium: empirical tests, Journal of 
 Political Economy . 81, 607-636. 
 
 Frankel, Jeffrey, 1989, Japanese finance: A survey, NBER Working Paper no. 3156. 
 
 French, Kenneth and James Poterba, 1990, Are Japanese stock prices too high?, Working Paper, 
 University of Chicago. 
 
 24 
 
Gultekin, Mustafa, Bulent Gultekin and Alessandro Penati, 1989, Capital controls and international 
 capital market segmentation: The evidence from the Japanese and American stock markets, Journal of 
 Finance . 44, 849-869. 
 
 Hamao, Yasushi, 1988, An empirical investigation of the Arbitrage Pricing Theory, Japan and the 
 World Economy . 1, 45-61. 
 
 Hamao, Yasushi, 1989, Japanese stocks, bonds, bills and inflation, Journal of Portfolio Management . 
 15, 20-26. 
 
 Hamao, Yasushi, in collaboration with Roger Ibbotson, 1989, Stocks. Bonds, and Inflation - Japan. 
 1988 Yearbook . Chicago: Ibbotson Associates. 
 
 Hamao, Yasushi, Ronald Masulis and Victor Ng, 1990, Correlations in price changes and volatility 
 across international stock markets, Review of Financial Studies , forthcoming. 
 
 Hawawini, Gabriel, 1988, Stock market anomalies and the pricing of equity on the Tokyo Stock 
 Exchange, Working Paper, The Wharton School, University of Pennsylvania. 
 
 Huberman, Gur and Shmuel Kandel, 1985, A size based stock returns model, Working paper, 
 University of Chicago. 
 
 Ishisuka, Hiroshi, 1987, Jissho Kaikei Joho to Kabuta (Accounting information and stock prices), 
 Tokyo: Dobunkan (in Japanese). 
 
 Jaffe, Jeffrey, Donald Keim and Randolph Westerfield, 1989, Earnings yields, market values and 
 stock returns, Journal of Finance . 44, 135-148. 
 
 Jaffe, Jeffrey and Randolph Westerfield, 1985, Patterns in Japanese common stock returns: Day of 
 the week and turn of the year effects, Journal of Financial and Quantitative Analysis . 20, 261-272. 
 
 Kato, Kiyoshi, William Ziemba and Sandra Schwartz, 1990, Day of the week effects in Japanese 
 stocks, in Edwin Elton and Martin Gruber (eds.), Japanese Capital Markets . New York: Harper and Row. 
 
 Kato, Kiyoshi and James Schallheim, 1985, Seasonal and size anomalies in the Japanese stock 
 market, Journal of Financial and Quantitative Analysis . 20, 243-260. 
 
 Lakonishok, Josef and Alan Shapiro, 1986, Systematic risk, total risk and size as determinants of 
 stock market returns, Journal of Banking and Finance . 10, 155-132. 
 
 Lakonishok, Josef and Seymour Smidt, 1988, Are seasonal anomalies real? A ninety-year perspective, 
 Review of Financial Studies . 1, 403-425. 
 
 Litzenberger, Robert and Krishna Ramaswamy, 1979, The effect of personal taxes and dividends on 
 capital asset prices: Theory and empirical evidence, Journal of Financial Economics . 7, 163-1%. 
 
 Lo, Andrew and A. Craig MacKinlay, 1989, Data-snooping biases in tests of financial asset pricing 
 models, NBER Working Paper no. 3001. 
 
 Maru, Junko and Yasuhiro Yonczawa, 1984, Nihon no Kabushiki Shijo (The Japanese Stock 
 Market) . Tokyo: Toyo Keizai Shimpo Sha (in Japanese). 
 
 Nakamura, Takeo and Noboru Terada, 1984, The size effect and seasonality in the Japanese stock 
 market, working paper. 
 
 Nihon Keizai Shimbun Sha, 1971 through 1988, Kaisha Joho (Company Handbook), in Japanese. 
 
 25 
 
Reinganum, Marc, 1981, Mis-specification of capital asset pricing: empirical anomalies based on 
 earnings yields and market values, Journal of Financial Economics . 9, 19-46. 
 
 Ritter Jay and Navin Chopra, 1989, Portfolio rebalancing and the turn-of-the-year effect, Journal of 
 Finance . 44, 149-166. 
 
 Rosenberg, Barr, Kenneth Reid, and Ronald Lanstein, 1984, Persuasive evidence of market 
 inefficiency, Journal of Portfolio Management . 11, 9-17. 
 
 Sakakibara, Shigeki, Hidetoshi Yamaji, Hisakatsu Sakurai, Kengo Shiroshita and Shimon Fukuda, 
 1988, The Japanese Stock Market . New York: Praeger. 
 
 Tokyo Stock Exchange, 1988, Fact Book . 
 
 Wilson, Pete, 1986, The relative information content of accruals and cash flows: Combined evidence 
 at the earnings announcement and annual report release date, Supplement to Journal of Accounting 
 Research. 165-200. 
 
 26 
 
Table I 
 
 Average monthly returns (standard deviations in parentheses), earnings to price (E/P) ratios, size (millions 
 of yen), book to market (B/M) ratios and cash flow to price (C/P) ratios, for portfolios sorted each June by 
 the four fundamental variables over the period 1971-1988 for both first and second section stocks. Also 
 reported is the estimated beta coefficient from an ordinary least squares regression of portfolio returns on an 
 equally weighted market portfolio of all first and second section stocks. N denotes the average number of 
 securities in each portfolio. 
 
 (a) Sorted by Earnings to Price ratio 
 < 1 (low) 
 
 4 (high) 
 
 Return .0277 
 
 .0154 
 
 .0174 
 
 .0176 
 
 .0194 
 
 (0.0665) 
 
 (0.0427) 
 
 (0.0414) 
 
 (0.0413) 
 
 (0.0425) 
 
 E/P -.2461 
 
 .0214 
 
 .0376 
 
 .0530 
 
 .0955 
 
 Size 27,668.5 
 
 84,470.4 
 
 107,703.0 
 
 86,685.0 
 
 64,496.4 
 
 B/M .2032 
 
 .4124 
 
 .4665 
 
 .5463 
 
 .6341 
 
 C/P -.1145 
 
 .0938 
 
 .1062 
 
 .1199 
 
 .1799 
 
 Beta 1.1905 
 
 .9678 
 
 .9529 
 
 .9389 
 
 .9423 
 
 N 105.2 
 
 275.1 
 
 279.4 
 
 279.0 
 
 276.2 
 
 (b) Sorted by Size 
 
 Return 
 
 E/P 
 
 Size 
 
 B/M 
 
 C/P 
 
 Beta 
 
 N 
 
 1 (low) 
 
 4 (high) 
 
 .0244 
 
 .0189 
 
 .0161 
 
 .0147 
 
 (0.0542) 
 
 (0.0457) 
 
 (0.0421) 
 
 (0.0408) 
 
 -.0083 
 
 .0305 
 
 .0361 
 
 .0402 
 
 4,587.8 
 
 13,699.0 
 
 36,831.4 
 
 262,112.0 
 
 .5013 
 
 .5086 
 
 .4870 
 
 .4518 
 
 .0977 
 
 .1080 
 
 .0997 
 
 .1076 
 
 1.0967 
 
 1.0297 
 
 .9638 
 
 .8146 
 
 308.3 
 
 310.8 
 
 312.2 
 
 315.0 
 
 (c) Sorted by Book to Market ratio 
 < 1 (low) 
 
 4 (high) 
 
 Return .0255 
 
 .0133 
 
 .0166 
 
 .0194 
 
 .0243 
 
 (0.0887) 
 
 (0.0431) 
 
 (0.0426) 
 
 (0.0427) 
 
 (0.0464) 
 
 E/P -.5151 
 
 .0211 
 
 .0370 
 
 .0404 
 
 .0441 
 
 Size 10,912.5 
 
 126,548.0 
 
 79,794.5 
 
 65,921.5 
 
 49,642.1 
 
 B/M -1.0706 
 
 .2659 
 
 .4292 
 
 .5612 
 
 .8031 
 
 C/P -.3439 
 
 .0746 
 
 .1035 
 
 .1240 
 
 .1486 
 
 Beta 1.1322 
 
 .9399 
 
 .9822 
 
 .9812 
 
 .9840 
 
 N 16.3 
 
 305.6 
 
 307.1 
 
 307.0 
 
 306.9 
 
 (d) Sorted by Cash Yield 
 
 
 
 
 
 < 
 
 1 (low) 
 
 2 
 
 3 
 
 4 (high) 
 
 Return .0262 
 
 .0143 
 
 .0168 
 
 .0190 
 
 .0222 
 
 (0.0705) 
 
 (0.0412) 
 
 (0.0408) 
 
 (0.0430) 
 
 (0.0464) 
 
 E/P -.4265 
 
 .0246 
 
 .0423 
 
 .0510 
 
 .0684 
 
 Size 17,149.6 
 
 121,298.0 
 
 75,508.4 
 
 62,510.9 
 
 72,388.9 
 
 B/M -.0208 
 
 .3783 
 
 .4920 
 
 .5650 
 
 .6170 
 
 C/P -.3303 
 
 .0523 
 
 .0860 
 
 .1242 
 
 .2340 
 
 Beta 1.1942 
 
 .9247 
 
 .9313 
 
 .9820 
 
 1.0135 
 
 N 50.1 
 
 293.5 
 
 295.8 
 
 294.4 
 
 294.0 
 
 27 
 
Table II 
 
 Estimated relationship between returns and fundamental variables for both first and second section stocks 
 using a SUR model. The basic model is, 
 
 Rp, - Rf, = c^ + Bbj(RW t - R„) + B p2 (RE t - R„) + a, (E/P^ + 0,(1^ + ^(B/M^ 
 + a/d/P^ + e^ 
 
 P = 1,... 64 t = 1,... 210 
 
 where, R^ - R^ is the excess return on portfolio P, RW ( - R^ and 
 
 RE - Rj, are the excess returns on the value weighted and equally weighted indexes, respectively, E/P and 
 C/F are the earnings and cash flow yields, respectively, LS is the natural logarithm of market capitalization 
 and B/M is the book to market ratio. Results for various subsets of independent variables are presented as 
 well. The numbers in parentheses are t-values. The order of the grouping is earnings yield, size and book to 
 market ratio. The sample period is July 1971 to December 1988. 
 
 Model Intercept E/P LS B/M C/P 
 
 (1) 
 
 -.0028 
 (-8.64) 
 
 -.0001 
 (-.01) 
 
 
 
 
 (2) 
 
 -.0074 
 (-3.69) 
 
 
 .0005 
 (2.28) 
 
 
 
 (3) 
 
 -.0043 
 (-5.79) 
 
 
 
 .0028 
 (2.00) 
 
 
 (4) 
 
 -.0035 
 (-7.60) 
 
 
 
 
 .0058 
 (1.59) 
 
 (5) 
 
 -.0079 
 (-3.65) 
 
 .0040 
 (0.62) 
 
 .0005 
 (2.37) 
 
 
 
 (6) 
 
 -.0152 
 (-5.05) 
 
 -.0054 
 (-0.81) 
 
 .0009 
 (3.83) 
 
 .0064 
 (3.80) 
 
 
 (7) 
 
 -.0156 
 (-5.14) 
 
 
 .0010 
 (3.91) 
 
 .0055 
 (2.90) 
 
 .0025 
 (0.51) 
 
 (8) 
 
 -.0161 
 
 (-5.31) 
 
 -.0202 
 (-1.85) 
 
 .0010 
 (4.13) 
 
 .0046 
 (2.25) 
 
 .0139 
 
 (1.74) 
 
 28 
 
Table III 
 
 Estimated relationship between returns and fundamental variables for first section stocks using a SUR model. 
 The basic model is, 
 
 Rp, - ^ = Qj, + JWRW, - R„) + /^(RE, - R„) + ^(E/P^ + 0,(1^, + a,(B/M)^ 
 + o 4 (6/P) h + ep, 
 
 P = 1,... 64 t = 1,... 210 
 
 where, Rp. - R^ is the excess return on portfolio P, RW ( - R^ and 
 
 RE - Rj- are the excess returns on the value weighted and equally weighted indexes, respectively, E/P and 
 C/P are the earnings and cash flow yields, respectively, LS is the natural logarithm of market capitalization 
 and B/M is the book to market ratio. Results for various subsets of independent variables are presented as 
 well. The numbers in parentheses are t-values. The order of the grouping is earnings yield, size and book to 
 market ratio. The sample period is July 1971 to December 1988. 
 
 Model Intercept E/P LS B/M C/P 
 
 (1) 
 
 -.0024 
 (-6.89) 
 
 -.0015 
 (-0.27) 
 
 
 
 
 (2) 
 
 -.0067 
 (-2.74) 
 
 
 .0004 
 (1.76) 
 
 
 
 (3) 
 
 -.0068 
 (-7.76) 
 
 
 
 .0078 
 (5.39) 
 
 
 (4) 
 
 -.0049 
 (-8.56) 
 
 
 
 
 .0170 
 (4.49) 
 
 (5) 
 
 -.0239 
 (-6.40) 
 
 .0001 
 (0.02) 
 
 .0004 
 (1.74) 
 
 • 
 
 
 (6) 
 
 -.0208 
 (-6.16) 
 
 -.0153 
 
 (-2.57) 
 
 .0013 
 (4.56) 
 
 .0119 
 (7.17) 
 
 
 (V) 
 
 -.0213 
 (-6.29) 
 
 
 .0013 
 (4.62) 
 
 .0116 
 (6.30) 
 
 -.0019 
 (-0.42) 
 
 (8) 
 
 -.0212 
 (-6.29) 
 
 -.0353 
 (-3.61) 
 
 .0013 
 (4.71) 
 
 .0089 
 (4.48) 
 
 .0199 
 (2.68) 
 
 29 
 
Table IV 
 
 Distribution of earnings yield, size, book to market ratio and cash flow yield for first and second section 
 stocks for every third year. 
 
 (a) Earnings Yield 
 
 
 First 
 
 Year 
 
 Quartile 
 
 1973 
 
 .0396 
 
 1976 
 
 .0253 
 
 1979 
 
 .0255 
 
 1982 
 
 .0316 
 
 1985 
 
 .0191 
 
 1988 
 
 .0093 
 
 Median 
 
 Mean 
 
 Third 
 Quartile 
 
 .0570 
 .0413 
 .0412 
 .0481 
 .0299 
 .0158 
 
 .0632 
 .0481 
 .0472 
 .0540 
 .0361 
 .0174 
 
 .0780 
 .0602 
 .0592 
 .0686 
 .0452 
 .0223 
 
 (b) Size (Billions of Yen) 
 
 
 First 
 
 Year 
 
 Quartile 
 
 1973 
 
 2.75 
 
 1976 
 
 3.90 
 
 1979 
 
 6.89 
 
 1982 
 
 7.15 
 
 1985 
 
 14.05 
 
 1988 
 
 35.55 
 
 Median 
 
 Mean 
 
 Third 
 Quartile 
 
 8.54 
 
 11.03 
 
 16.95 
 
 18.82 
 
 36.80 
 
 78.83 
 
 28.39 
 
 34.94 
 
 45.58 
 
 65.93 
 
 131.29 
 
 300.95 
 
 24.00 
 
 32.38 
 
 43.62 
 
 54.42 
 
 100.20 
 
 210.79 
 
 (c) Book to Market Ratio 
 
 Year 
 
 First 
 Quartile 
 
 Median 
 
 Mean 
 
 Third 
 Quartile 
 
 1973 
 1976 
 1979 
 1982 
 1985 
 1988 
 
 .3920 
 .4154 
 .3401 
 .4506 
 .2801 
 .1541 
 
 .5128 
 .5391 
 .4517 
 .6042 
 .3986 
 .2306 
 
 .5396 
 .5706 
 .4629 
 .6233 
 .4379 
 .2453 
 
 .6640 
 .6991 
 .5682 
 .7767 
 .5650 
 .3170 
 
 (d) Cash How Yield 
 
 First 
 Year Quartile 
 
 Median 
 
 Mean 
 
 Third 
 Quartile 
 
 1973 
 1976 
 1979 
 1982 
 1985 
 1988 
 
 .0941 
 .0758 
 .0631 
 .0748 
 .0427 
 .0216 
 
 .1336 
 .1114 
 .0901 
 .1046 
 .0648 
 .0327 
 
 .1549 
 .1360 
 .1069 
 .1269 
 .0816 
 .0409 
 
 .1852 
 .1706 
 .1290 
 .1552 
 .0971 
 .0477 
 
 30 
 
Table V 
 
 Estimated relationship between returns and fundamental variables for both first and second section stocks 
 using a SUR model. The basic model is, 
 
 Rpt - Rft = °b + Gpi ( RW t - Rft) + 'W* 1 ! - Rft) + <*i( E / p )pt + ^(LSVt + ^( B / M )f>t 
 
 + VC/P^ + ep, 
 
 P = 1,... 64 t = 1,... 210 
 
 where, R-, - R^ is the excess return on portfolio P, RW. - R^. and RE - R^ are the excess returns on the 
 value weighted and equally weighted indexes, respectively, E/P and C/P are the earnings and cash flow 
 yields, respectively, LS is the natural logarithm of market capitalization and B/M is the book to market ratio. 
 Results for various subsets of independent variables are presented as well. The numbers in parentheses are 
 t-values. The order of the grouping is earnings yield, size and book to market ratio. Each of the 
 fundamental variables is deflated by its cross sectional average from the previous June. The sample period is 
 July 1971 to December 1988. 
 
 Model Intercept E/P LS B/M C/P 
 
 (1) 
 (2) 
 (3) 
 (4) 
 (5) 
 (6) 
 (7) 
 (8) 
 
 -.0034 
 
 .0007 
 
 
 
 
 (-9.75) 
 
 (1.95) 
 
 
 
 
 -.0011 
 
 
 -.0005 
 
 
 
 (-1.23) 
 
 
 (-1.91) 
 
 
 
 -.0070 
 
 
 
 .0043 
 
 
 (-6.77) 
 
 
 
 (4.05) 
 
 
 -.0054 
 
 
 
 
 .0028 
 
 (-8.79) 
 
 
 
 
 (4.23) 
 
 -.0018 
 
 .0007 
 
 -.0005 
 
 
 
 (-1.80) 
 
 (1.88) 
 
 (-1.85) 
 
 
 
 -.0060 
 
 .0003 
 
 -.0003 
 
 .0042 
 
 
 (-4.22) 
 
 (0.91) 
 
 (-1.30) 
 
 (4.02) 
 
 
 -.0065 
 
 
 -.0003 
 
 .0036 
 
 .0014 
 
 (-4.49) 
 
 
 (-1.22) 
 
 (3.31) 
 
 (2.06) 
 
 -.0066 
 
 -.0005 
 
 -.0003 
 
 .0034 
 
 .0022 
 
 (-4.55) 
 
 (-0.92) 
 
 (-1.15) 
 
 (3.03) 
 
 (2.07) 
 
 31 
 
Table VI 
 
 Estimated relationship between returns and fundamental variables for first section stocks using a SUR model. 
 The basic model is, 
 
 Rp, - R„ = c^ + Bp^RW - R ft ) + B^RE, - R ft ) + a, (E/P^ + 0,(1^ + o^B/M^ 
 + ot 4 (C/P^, + ep, 
 
 P = 1,... 64 t = 1,... 210 
 
 where, Rp, - R^ is the excess return on portfolio P, RW. - Rj, and RE - R,, are the excess returns on the 
 value weighted and equally weighted indexes, respectively, E/P and C/P are the earnings and cash flow 
 yields, respectively, LS is the natural logarithm of market capitalization and B/M is the book to market ratio. 
 Results for various subsets of independent variables are presented as well. The numbers in parentheses are 
 t-values. The order of the grouping is earnings yield, size and book to market ratio. Each of the 
 fundamental variables is deflated by its cross-sectional average from the previous June. The sample period is 
 July 1971 to December 1988. 
 
 Model Intercept E/P LS B/M C/P 
 
 (1) 
 (2) 
 (3) 
 (4) 
 (5) 
 (6) 
 (7) 
 (8) 
 
 -.0034 
 (-9.61) 
 
 .0006 
 (1.79) 
 
 
 
 
 -.0008 
 (-0.49) 
 
 
 -.0007 
 (-1.38) 
 
 
 
 -.0066 
 (-6.45) 
 
 
 
 .0037 
 (3.72) 
 
 
 -.0066 
 (-8.42) 
 
 
 
 
 .0024 
 (3.87) 
 
 -.0013 
 (-0.83) 
 
 .0006 
 (1.76) 
 
 -.0007 
 (-1.38) 
 
 
 
 -.0057 
 (-3.01)' 
 
 .0003 
 (0.78) 
 
 -.0004 
 (-0.73) 
 
 .0037 
 (3.70) 
 
 
 -.0061 
 (-3.22) 
 
 
 -.0004 
 (-0.66) 
 
 .0032 
 (3.06) 
 
 .0011 
 (1.75) 
 
 -.0063 
 (-3.31) 
 
 -.0004 
 (-0.78) 
 
 -.0003 
 (-0.55) 
 
 . • .0030 
 (2.82) 
 
 .0017 
 (1.76) 
 
 32 
 
Table VII 
 
 Estimated relationship between returns and fundamental variables for both first and second section stocks 
 using a SUR model for the period 1980-1988. The basic model is, 
 
 Rp, - R,, = % + ^(RW- \) + Bp 2 (RE, - Rft) + ^ (E/P^ + o,(LS)^ + ^(B/M^ 
 + a 4 (W P )pt + «« 
 
 P = 1,... 64 t = 1,... 210 
 
 where, Rp. - R« is the excess return on portfolio P, RW. - R^and RE - R^ are the excess returns on the 
 value weighted and equally weighted indexes, respectively, E/P and C/P are the earnings and cash flow 
 yields, respectively, LS is the natural logarithm of market capitalization and B/M is the book to market ratio. 
 Results for various subsets of independent variables are presented as well. The numbers in parentheses are 
 t-values. The order of the grouping is earnings yield, size and book to market ratio. Each of the 
 fundamental variables is deflated by its cross-sectional average from the previous June. 
 
 Model Intercept E/P LS B/M C/P 
 
 (1) 
 (2) 
 (3) 
 (4) 
 (5) 
 (6) 
 (7) 
 (8) 
 
 -.0038 
 
 .0014 
 
 
 
 
 (-11.11) 
 
 (3.83) 
 
 
 
 
 .0002 
 
 
 -.0009 
 
 
 
 (0.12) 
 
 
 (-1.75) 
 
 
 
 -.0072 
 
 
 
 .0047 
 
 
 (-7.20) 
 
 
 
 (4.38) 
 
 
 -.0055 
 
 
 
 
 .0031 
 
 (-10.17) 
 
 
 
 
 (5.25) 
 
 -.0016 
 
 .0014 
 
 -.0007 
 
 
 
 (-1.00) 
 
 (3.75) 
 
 (-1.38) 
 
 
 
 -.0066 
 
 .0005 
 
 -.0003 
 
 .0045 
 
 
 (-3.50) 
 
 (1.44) 
 
 (-0.54) 
 
 (4.18) 
 
 
 -.0070 
 
 
 -.0002 
 
 .0042 
 
 .0011 
 
 (-3.67) 
 
 
 (-0.43) 
 
 (3.86) 
 
 (1.79) 
 
 -.0071 
 
 .0001 
 
 -.0002 
 
 .0042 
 
 .0010 
 
 (-3.70) 
 
 (0.13) 
 
 (-0.40) 
 
 (3.84) 
 
 (1.02) 
 
 33 
 
Table VIII 
 
 Estimated relationship between returns and fundamental variables for all stocks using a SUR model with 
 dummy variables for January and non-January months. The basic model is described in equation 2, in the 
 text. Results for subsets of independent variables are presented as well. The models are numbered in 
 accordance with the previous tables. The numbers in parentheses are t-values. The order of the grouping is 
 earnings yield, size and book to market ratio. Each of the fundamental variables is deflated by its cross 
 sectional average from the previous June. The sample period is July 1971 to December 1988. 
 
 Model 
 
 Intercept 
 
 E/P 
 
 LS 
 
 B/M 
 
 C/P 
 
 (1) 
 
 January 
 
 February- 
 December 
 
 .0040 
 
 (1-96) 
 -.0036 
 (-10.03) 
 
 -.0075 
 (-4.05) 
 .0009 
 (2.32) 
 
 
 
 
 (2) 
 
 January 
 
 February- 
 December 
 
 -.0017 
 (-0.43) 
 -.0015 
 (-1.49) 
 
 
 -.0003 
 (-0.25) 
 -.0004 
 (-1.52) 
 
 
 
 (3) 
 
 January 
 
 February- 
 December 
 
 .0048 
 (0.99) 
 -.0073 
 (-6.83) 
 
 
 
 -.0070 
 (-1.50) 
 .0045 
 (4.14) 
 
 
 (4) 
 
 January 
 
 February- 
 December 
 
 .0075 
 (2.64) 
 -.0059 
 (-9.32) 
 
 
 
 
 -.0105 
 (-3.89) 
 .0033 
 (4.83) 
 
 (5) 
 
 January 
 
 February- 
 December 
 
 .0049 
 (1.20) 
 -.0023 
 (-2.23) 
 
 -.0074 
 (-4.03) 
 .0009 
 (2.25) 
 
 -.0003 
 (-0.26) 
 -.0004 
 (-1.39) 
 
 
 
 (6) 
 
 January 
 
 February- 
 December 
 
 .0076 
 (1.33) 
 -.0069 
 (-4.60) 
 
 -.0074 
 (-3.87) 
 .0005 
 (1.32) 
 
 -.0003 
 (-0.26) 
 -.0002 
 (-0.80) 
 
 -.0024 
 (-0.50) 
 .0045 
 (4.15) 
 
 
 (7) 
 
 January 
 
 February- 
 December 
 
 .0088 
 T1.53) 
 -.0075 
 (-4.%) 
 
 
 -.0004 
 (-0.34) 
 -.0002 
 (-0.70) 
 
 .0007 
 (0.14) 
 .0036 
 (3.28) 
 
 -.0112 
 (-3.79) 
 .0019 
 (2.76) 
 
 (8) 
 
 January 
 
 February- 
 December 
 
 .0087 
 (1.51) 
 
 -.0076 
 (-5.01) 
 
 -.0049 
 (-1.90) 
 -.0006 
 (-1.07) 
 
 -.0003 
 (-0.30) 
 -.0002 
 (-0.63) 
 
 -.0001 
 (-0.02) 
 .0034 
 (2.96) 
 
 -.0057 
 (-1.43) 
 .0028 
 (2.58) 
 
 34 
 
Table IX 
 
 Estimated relationship between returns and fundamental variables for both first and second section stocks 
 using a SUR model. The basic model is, 
 
 Rpt - Rf. = °b + *W RW t - Rft) + ^(^ - Rft) + «i( e / p )r + ^(^Vt + M B / M )pt 
 
 + a^C/PV, + ep, 
 
 P = i v .. 64 t = 1,... 210 
 
 where, Rp, - R ft is the excess return on portfolio P, RW. - R^. and RE - R^ are the excess returns on the 
 value weighted and equally weighted indexes, respectively, E/P and C/P are the earnings and cash flow 
 yields, respectively, LS is the natural logarithm of market capitalization and B/M is the book to market ratio. 
 Results for various subsets of independent variables are presented as well. The numbers in parentheses are 
 t-values. Each of the fundamental variables is deflated by its cross sectional average from the previous June. 
 The sample period is July 1971 to December 1988. In panel A the order of the grouping is book to market, 
 cash yield, and size, whereas in panels B and C it is earnings yield, size and book to market. In panel B the 
 results are for the first pass regression under SUR, without adjustment for the cross correlation of residuals. 
 In panel C delisted companies are excluded from the sample. 
 
 Panel Model Intercept E/P LS B/M C/P 
 
 A 
 
 (5) 
 
 .0006 
 (0.56) 
 
 -.0011 
 (-2.68) 
 
 -.0006 
 (-2.31) 
 
 
 
 
 (6) 
 
 -.0054 
 (-3.53) 
 
 -.0017 
 (-3.97) 
 
 -.0005 
 (-1.83) 
 
 .0062 
 (5.19) 
 
 
 
 (8) 
 
 -.0063 
 (-4.03) 
 
 -.0030 
 (-5.46) 
 
 -.0004 
 (-1.47) 
 
 .0054 
 (4.51) 
 
 .0026 
 (3.54) 
 
 B 
 
 (5) 
 
 -.0006 
 (-0.52) 
 
 .0006 
 (1.37) 
 
 -.0007 
 (-2.37) 
 
 
 
 
 (6) 
 
 -.0079 
 (-5.43) 
 
 -.0004 
 (-0.82) 
 
 -.0003 
 (-1.13) 
 
 .0069 
 (8.84) 
 
 
 
 (8) 
 
 -.0089 
 (-5.87) 
 
 -.0015 
 (-2.25) 
 
 -.0002 
 (-0.83) 
 
 .0057 
 (6.02) 
 
 .0031 
 (2.28) 
 
 C 
 
 (5) 
 
 -.0012 
 (-1.00) 
 
 .0014 
 
 (2.74) 
 
 -.0008 
 (-2.62) 
 
 
 
 
 (6) 
 
 -.0059 
 (-3.77) 
 
 .0005 
 (0.89) 
 
 -.0006 
 (-2.09) 
 
 .0053 
 (4.48) 
 
 
 
 (8) 
 
 -.0068 
 (-4.21) 
 
 -.0007 
 (-1.01) 
 
 -.0006 
 (-1.94) 
 
 .0044 
 (3.58) 
 
 .0028 
 (2.57) 
 
 35 
 
Table X 
 
 Estimated relationship between returns and fundamental variables for both first and second section stocks 
 using the Fama-MacBeth approach. The basic model for month t is, 
 
 Rpt - Rft = %t + b it^i + ^^2 + aitCE/PVt + %t(^)pt + %t( B / M ** 
 
 where, Rp, - R^ is the excess return on portfolio (security) P, 6 pi and JBpg are betas for a two factor model 
 with the factors being excess returns on value and equally weighted indexes, respectively, E/P and C/P are 
 the earnings yield and cash flow yields, respectively, LS is the natural logarithm of market capitalization and 
 B/M is the book to market ratio. The numbers in parentheses are t-values. The order of the grouping is 
 earnings yield, size and book to market ratio. Results for various subsets of independent variables are 
 presented as well. The sample period is July 1971 to December 1988. The results in panel A are for a 
 simple average of the cross sectional coefficients, in panel B the weight of each monthly coefficient is 
 inversely proportional to the variance of the estimate, in panel C the monthly coefficients are estimated using 
 GLS; in panel D the fundamental variables are divided by cross sectional averages, and in panel E the results 
 are at the level of individual securities. 
 
 Panel Model Intercept E/P LS B/M C/P 
 
 A (5) 
 
 .0353 
 
 .0088 
 
 -.0025 
 
 
 
 
 (3.35) 
 
 (0.50) 
 
 (-2.67) 
 
 
 
 (6) 
 
 .0205 
 
 -.0266 
 
 -.0014 
 
 .0198 
 
 
 
 (1.74) 
 
 (-1.46) 
 
 (-1.45) 
 
 (4.56) 
 
 
 (8) 
 
 .0221 
 
 -.0740 
 
 -.0016 
 
 .0132 
 
 .0522 
 
 
 (1.87) 
 
 (-2.67) 
 
 (-1.60) 
 
 (3.37) 
 
 (2.86) 
 
 B (5) 
 
 .0226 
 
 .0068 
 
 -.0016 
 
 
 
 
 (5.63) 
 
 (0.98) 
 
 (-4.78) 
 
 
 
 (6) 
 
 .0119 
 
 -.0056 
 
 -.0011 
 
 .0147 
 
 
 
 (2.55) 
 
 (-0.78) 
 
 (-3.20) 
 
 (9.83) 
 
 
 (8) 
 
 .0118 
 
 -.0404 
 
 -.0011 
 
 .0099 
 
 .0342 
 
 
 (2.49) 
 
 (-3.45) 
 
 (-3.14) 
 
 (5.66) 
 
 (3.45) 
 
 C (5) 
 
 .0136 
 
 .0226 
 
 -.0005 
 
 
 
 
 (0.64) 
 
 (1.44) 
 
 (-1.98) 
 
 
 
 (6) 
 
 .0182 
 
 -.0112 
 
 -.0002 
 
 .0212 
 
 
 
 (0.87) 
 
 (-0.69) 
 
 (-0.55) 
 
 (5.12) 
 
 
 (8) 
 
 .0158 
 
 -.0350 
 
 -.0001 
 
 .0180 
 
 .0239 
 
 
 (0.72) 
 
 (-1.56) 
 
 (-0.48) 
 
 (4.34) 
 
 (1.97) 
 
 D (5) 
 
 .0353 
 
 -.0000 
 
 -.0081 
 
 
 
 
 (3.35) 
 
 (-0.01) 
 
 (-2.80) 
 
 
 
 (6) 
 
 .0205 
 
 -.0015 
 
 -.0055 
 
 .0103 
 
 
 
 (1.74) 
 
 (-1.89) 
 
 (-1.80) 
 
 (4.44) 
 
 
 (8) 
 
 .0221 
 
 -.0040 
 
 -.0059 
 
 .0063 
 
 .0071 
 
 
 (1.87) 
 
 (-2.87) 
 
 (-1.92) 
 
 (3.18) 
 
 (2.81) 
 
 E (5) 
 
 .0370 
 
 .0201 
 
 -.0026 
 
 
 
 
 (4.19) 
 
 (2.04) 
 
 (-3.27) 
 
 
 
 (6) 
 
 .0172 
 
 -.0071 
 
 -.0015 
 
 .0224 
 
 
 
 (1.83) 
 
 (-0.80) 
 
 (-1.93) 
 
 (6.56) 
 
 
 (8) 
 
 .0173 
 
 -.0299 
 
 -.0016 
 
 .0195 
 
 .0240 
 
 
 (1.84) 
 
 (-2.31) 
 
 (-1.99) 
 
 (6.09) 
 
 (3.10) 
 
 36 
 
Table XI 
 
 Average monthly returns, standard deviations(in parentheses) and betas (in brackets) for portfolios sorted 
 each June first to four groups based on B/M and then to four groups based on C/P. The results are for the 
 first and second section stocks and for the period 1971-88. To compute betas an equally weighted index of 
 the first and second sections stocks is used. 
 
 
 
 Cash Flow Yield 
 
 
 
 Highest 
 
 3 
 
 2 
 
 Lowest 
 
 Highest 
 
 .0273 
 
 .0230 
 
 .0238 
 
 .0230 
 
 
 (.0501) 
 [0.97] 
 
 (.0491) 
 [1.00] 
 
 (.0471) 
 [0.96] 
 
 (.0476) 
 [0.99] 
 
 3 
 
 .0212 
 
 .0194 
 
 .0180 
 
 .0177 
 
 Book to 
 Market 
 
 (.0477) 
 [1.03] 
 
 (.0449) 
 [1.00] 
 
 (.0417) 
 [0.92] 
 
 (.0428) 
 [0.95] 
 
 2 
 
 .0176 
 
 .0163 
 
 .0152 
 
 .0165 
 
 
 (.0477) 
 [1.04] 
 
 (.0434) 
 [0.96] 
 
 (.0423) 
 [0.92] 
 
 (.0438) 
 [0.97] 
 
 Lowest 
 
 .0156 
 
 .0125 
 
 .0116 
 
 .0115 
 
 
 (.0471) 
 [0.99] 
 
 (.0454) 
 [0.94] 
 
 (.0446) 
 [0.90] 
 
 (.0451) 
 [0.89] 
 
 37 
 
O 
 
 i> 
 
 <5 
 
 
 o»c»«» 
 
 o % 
 
 P-, 
 
 Si 
 
 0-. 
 
 N 
 
 •rH 
 
 'd 
 
 o 
 xn 
 
V) 
 
 «**"***■** 
 
 \ 
 
 P-H 
 
 -3 
 
 \ 
 
 i 
 $ 
 
 a 
 
 PQ 
 
 CD 
 
 o 
 
 C2 
 
 CVJ 
 
 LU 
 
 (J 
 
 Hi 
 
ArOT/CE: Return or renew all Library Materials! The Minimum Fee for 
 each Lost Book Is $50.00. 
 
 The person charging this material is responsible for 
 its return to the library from which it was withdrawn 
 on or before the Latest Date stamped below. 
 
 Theft, mutilation, and underlining of books are reasons for discipli- 
 nary action and may result in dismissal from the University. 
 To renew call Telephone Center, 333-8400 
 
 UNIVERSITY OF ILLINOIS LIBRARY AT URBANA-CHAMPAIGN 
 
 JUil 
 
 L16I— O-I096 
 
■nJ-To-Pk-w N MANCHESTER.