What Determines Price-Earnings Ratios?
What Determines Price-Earnings Ratios?
What Determines Price-Earnings Ratios?
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
CFA Institute<strong>What</strong> <strong>Determines</strong> <strong>Price</strong>-<strong>Earnings</strong> <strong>Ratios</strong>?Author(s): William Beaver and Dale MorseSource: Financial Analysts Journal, Vol. 34, No. 4 (Jul. - Aug., 1978), pp. 65-76Published by: CFA InstituteStable URL: http://www.jstor.org/stable/4478160Accessed: 30/03/2010 11:37Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available athttp://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay use content in the JSTOR archive only for your personal, non-commercial use.Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained athttp://www.jstor.org/action/showPublisher?publisherCode=cfa.Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact support@jstor.org.CFA Institute is collaborating with JSTOR to digitize, preserve and extend access to Financial AnalystsJournal.http://www.jstor.org
y William Beaver and Dale Morse<strong>What</strong><strong>Price</strong><strong>Determines</strong>-<strong>Earnings</strong> <strong>Ratios</strong>?> Recent studies on the behavior of earningsgrowth over time raise doubt about the ability of pastgrowth to explain differences in price-earningsratios. Either future growth is difficult to predict, orinvestors are basing their predictions on informationother than past growth.Grouping common stocks into portfolios on thebasis of price-earnings ratios, the authors find thatthe initial P/E differences among the portfoliospersist up to 14 years. Growth appears to explainlittle of the persisting P/E differences, however.<strong>Price</strong>-earnings ratios correlate negatively withearnings growth in the year of the portfolio'sformation, but positively with earnings growth in thesubsequent year, suggesting that investors areforecasting only short-lived earnings distortions.Nor does risk supply the explanation for thesedifferences. Although price-earnings ratios can varyeither positively or negatively with market risk,depending on the market conditions in a given year,market risk is of little assistance in explaining theobserved persistence in price-earnings ratios overperiods longer than two or three years.The authors conclude that the most likelyexplanation of the evident persistence in priceearningsratios is not growth or risk, but differencesin accounting method. >proach, we examine the behavior of P/E ratios andexplore the ability of earnings growth (hereaftergrowth) and risk to explain P/E ratio differencesacross stocks. We find that, although differences inP/E ratios persist for up to 14 years, growth and riskappear to explain little of this persistency. In particular,growth appears to have virtually no effectbeyond two years.>Valuation TheoryUnder perfect markets and certainty, the price of asecurity is equal to the present value of the futurecash flows. Over an infinite horizon, the currentprice will reflect the stream of dividends. Under thefurther assumptions of (1) a constant dividend payoutratio (K), (2) constant growth in earnings pershare (g) and (3) a constant riskless rate (r), P/E isgiven by the Gordon-Shapiro valuation equation:P/E=rKg ( (1 )In a certainty world, earnings per share (E) can bedefined as that constant cash flow whose presentvalue is equivalento the present value of the cashflows generated from current equity investment.Where the investment involves assets with finitelives, this definition implicitly reflects the fact thatthe value of the assets will depreciate over theirlives.3 We adopt this definition, which is often referredto as permanent earnings. Absent further investment,or if the earnings rate on future investmentT HE PRICE-EARNINGS ratio (hereafter P/E 1. Footnotes appear at end of article.ratio) is of considerable interest, yet little isknown about how it behaves over time or William Beaver is Thomas D. Dee, ll Professor of Acaboutthe relative importance of the factors believed counting at the Graduate School of Business, StanfordUniversity. Dale Morse is a Ph.D. candidate at theto influence its behavior. Differences in expectedGraduate School of Business, Stanford. Financial supgrowthare commonly offered as a major explanation port for their research was provided by the Stanfordfor differences in P/E ratios. Yet recent research Program in Professional Accounting, the major spon -raises doubt about this interpretation; past growth sors of which are Arthur Andersen & Co., Arthurand analysts' forecasts appear to have little ability to Young & Co., Coopers & Lybrand, Ernst & Ernst,explain subsequent growth.' Using a portfolio ap- Peat, Marwick, Mitchell & Co. and <strong>Price</strong> Waterhouse.FINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978 O 65
is r, the P/E ratio is simply the reciprocal of the risklessrate (I/r). The P/E ratio will reflect a growth"premium"(or discount) only when the rate ofreturn on future investment exceeds (or falls below)the riskless rate, r.4When the world is no longer certain, it is nolonger clear what the "earnings term" in Equation 1is intended to represent. The earnings concept underlyingmarket prices is future-oriented, hence isdefined in terms of the expectations of market participants.As such, it is not directly observable, butpresumably representsome form of expected permanentearnings per share attributable to the currentequity investment.A second consequence of uncertainty is that,along with E, the actual values of the variables r, gand K are also unknown. Each symbol in Equation 1is often interpreted as the expected value of the correspondingvariable. When Equation 1 is used toanalyze the behavior of current prices, these variablesare commonly interpreted as a "consensus" expectationacross investors.5 While there are problemsin using Equation I in this manner, it may stillbe a reasonable approximation of a more complexvaluation process.A third consequence of uncertainty is that the ex-pected return is no longer the riskless rate, but rathera risky rate. Since stocks will differ with respect torisk, the expected risky rate for stock i will bedenoted ri. In the one-period capital asset pricingmodel (CAPM), differences in the expected riskyrate of return are due solely to differences in betathesensitivity of the stock to return on the generalmarket rm. In particular:ri - rf + bi(rni - rf), (2)where ri is the expected rate of return on security i, rfthe riskless rate, rm the expected rate of return onthe market portfolio and bi security i's sensitivity tomarket risk, or beta.Moreover, actual earnings per share (EPS) willvary from year to year because of transitory (i.e.,temporary) factors peculiar to a particular year.Therefore, actual earnings may differ from the expectedearnings upon which market prices are based.This leads to the distinction between the transitoryversus the permanent component of EPS. This distinctionwill become crucial in interpreting our results.6Research DesignA portfolio approach potentially diversifies outsome of the "noise" at the individual stock level.We selected stocks that satisfied the following criteria:(1) five consecutive years of data on the Compustatand CRSP tapes (the latter implies New YorkStock Exchange membership) and (2) a fiscal yearending on December 31.For each year from 1956 through 1974 we computedthe P/E for each stock with data available inTABLE 1: Median Values of VariablesMedianPercentageNo. of Median Growth MedianYear Stocks P/E in EPS Beta*1956 270 11.55 0.9811957 279 10.08 -3.63 0.9521958 284 17.61 -8.94 0.9591959 295 15.75 20.73 0.9541960 354 15.29 -4.83 0.9641961 373 19.68 1.56 0.9591962 398 14.98 8.39 1.0071963 409 15.46 9.00 0.9811964 435 14.45 19.42 0.9601965 464 15.18 18.19 0.9521966 493 11.60 13.01 0.9751967 514 16.91 -0.69 0.9671968 548 19.43 7.52 0.9911969 581 14.04 5.31 0.9981970 600 15.58 -10.16 0.9391971 600 17.06 9.921972 600 14.07 20.461973 600 7.45 23.101974 600 5.01 12.371975 600 8.04 -1.21*A stock's beta for 1956 is computed over the 60-month period following December1956 (January 1957 through December 1961), according to the method described inFootnote 8.66 O FINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978
that year. We defined P/E as price per share on December31 divided by earnings per share for the year,computed on a pre-extraordinary item basis. Usingdata from the Compustatape, we defined earningsgrowth as the percentage change in the year's earningsper share relative to the previous year andmeasured risk as the stock's beta, computed frommonthly stock price return data available on theCRSP tape.8median annual growth rates reported in Table 1 andthe growth in aggregate EPS for S&P Industrialsyielded a positive correlation of 0.89.12How Do P/E <strong>Ratios</strong> Behave Over Time?Table 2 reports the rank correlation between P/Eratios in the year of formation and P/E ratios in subsequentyears.'3 The first row of Table 2 displays thecorrelations between the P/E ratios of portfoliosWe then ranked each year's stocks according to formed in 1956 and the P/E ratios of the same port-P/E and formed 25 portfolios, with Portfolio One folios in subsequent years. The second row displayscomprising those four per cent with the highest P/E's results for portfolios formed in 1957 and the finaland Portfolio 25 comprising the stocks with the row the correlation between the P/E ratios of thelowest P/E's. We then compared the median P/E for portfolios formed in 1974 and the P/E ratios of thoseeach portfolio in its base year (year of formation) same portfolios in 1975. For example, for the portwiththe median P/E, median realized growth and folios formed at the end of 1956, the correlation bemedianrisk for the portfolio in subsequent years.9 tween portfolio P/E ratios in 1956 (the year of for-Note that, in all cases, once formed the portfolio's mation) and 1957 (one year later) is 0.96. The rankcomposition was fixed (i.e., a buy and hold strategy correlation between the P/E ratios in 1956 and 1966was used).'0(10 years later) is 0.74.Table 1 report some summary statistics." Once a The median correlation of each column is restockappears on the tape in a given year, its data are ported at the bottom of the table. The median correavailablefrom that year onward. The similarity of lations are not strictly comparable for several reastocksappearing later relative to those appearing sons. First, the group of calendar years over whichearlier is supported by the median beta, which shows the median is computed gradually changes as oneno trend over time and is close to one. When we cor- moves across the columns: One year after formationrelated the median P/E for each year with the aggre- includes calendar years 1957 through 1975, whilegate P/E ratio for Standard & Poor's Composite 14 years after formation includes only calendarstocks for the years 1956 through 1975, we obtained years 1970 through 1975. Then, too, the median P/Ea positive rank correlation of 0.85, which is reason- ratio varies considerably by calendar years, as Tableable, given the differences in the stocks and the 1 indicates, dropping sharply in 1973, 1974 andmethods used to compute the average P/E for a given 1975. Table 1 also shows that the average number ofyear. Furthermore, the rank correlation between the stocks per portfolio differs; in 1956 the number ofTABLE 2: Rank Correlations of Portfolios Formed By P/E <strong>Ratios</strong> With P/E <strong>Ratios</strong> in Subsequent YearsBaseYears Following Base YearYear 1 2 3 4 5 6 7 8 9 10 11 12 13 141956 0.96 0.87 0.88 0.65 0.78 0.70 0.82 0.85 0.69 0.74 0.62 0.36 0.41 0.591957 0.85 0.91 0.83 0.84 0.89 0.89 0.90 0.81 0.86 0.72 0.51 0.67 0.78 0.441958 0.95 0.73 0.64 0.52 0.43 0.55 0.49 0.30 0.60 0.22 0.24 0.49 0.41 0.181959 0.96 0.91 0.91 0.73 0.57 0.88 0.74 0.69 0.33 0.46 0.69 0.56 0.40 0.561960 0.94 0.94 0.93 0.89 0.88 0.79 0.70 0.63 0.80 0.73 0.61 0.61 0.50 0.241961 0.98 0.96 0.86 0.89 0.85 0.76 0.74 0.87 0.83 0.87 0.76 0.72 0.55 0.641962 0.92 0.89 0.93 0.94 0.87 0.69 0.86 0.73 0.78 0.76 0.87 0.78 0.771963 0.99 0.98 0.95 0.89 0.71 0.77 0.75 0.61 0.79 0.93 0.77 0.761964 0.95 0.96 0.94 0.72 0.88 0.89 0.72 0.88 0.90 0.81 0.821965 0.99 0.93 0.83 0.83 0.77 0.86 0.93 0.91 0.80 0.711966 0.96 0.89 0.95 0.96 0.84 0.95 0.89 0.80 0.791967 0.98 0.98 0.94 0.89 0.85 0.58 0.57 0.691968 0.98 0.95 0.88 0.84 0.63 0.53 0.351969 0.89 0.92 0.95 0.74 0.80 0.821970 0.95 0.79 0.63 0.72 0.731971 0.96 0.80 0.70 0.671972 0.96 0.96 0.961973 0.99 0.971974 0.97MedianCorrelation 0.96 0.92 0.91 0.83 0.80 0.78 0.74 0.76 0.79 0.73 0.69 0.64 0.50 0.44FINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978 L 67
Portfoliostocks is 10, while the base years 1970 through 1974average 24 per portfolio. On purely statisticalgrounds, the correlation coefficient should rise asthe number of stocks per portfolio increases. However,Table 2 does not display any obvious tendencyfor the correlations to increase systematically in thelater base years.'4With these caveats in mind, we interpret thecorrelations in Table 2 as supporting a long-termpersistency in the portfolios' P/E ratios. With onlytwo minor disruptions, the median correlation declinessteadily with the number of years since portfolioformation. Five years after formation the mediancorrelation is 0.80, while 10 years after formationthe median correlation is 0.73. Fourteen yearsafter formation, the median is 0.44. We tentativelyconclude that, although much of the effect of the factorsthat determine P/E ratios dissipates over the 14years, a portion still clearly remains after five, 10 orperhaps even 14 years.This conclusion is supported by Table 3, whichdisplays a composite picture of six of the 25 portfolios.'5We averaged the P/E ratios across the baseyears, weighting each year by the number of stocks inthat year.'6 The striking feature of Table 3 is theshrinkage over time in the P/E differences among theportfolios. Not surprisingly, this tendency is mostevident in the extreme portfolios (i.e., Portfolios Oneand 25). The P/E of Portfolio One is less than half itsvalue one year after formation, and less than onethirdits value two years after formation. Portfolio 25shows a similar reversion toward a central value, asdo other portfolios, for which the pattern is, however,less pronounced.As a convenient summary, Table 3 reports theratio of the P/E values for Portfolio One relative tothose for Portfolio 25. In the year of formation,Portfolio One's P/E is over eight times that of Portfolio25's, while in the next year it has shrunk toslightly over three times and by three years after formationit is less than twice Portfolio 25's P/E. Apartfrom this dramatic convergence of the P/E ratios,their most striking feature is the stability of the relativedifference from the third year through the elev-TABLE 3: <strong>Price</strong>-<strong>Earnings</strong> <strong>Ratios</strong>enth year after formation. After the eleventh year,further convergence occurs until, in the fourteenthyear, Portfolio 25 has a P/E greater than that of PortfolioOne's.We take this to mean that the effect of the factorsdetermining P/E ratios in the year of formation dissipatesdramatically by the third year after formation.On the other hand, the fact that the convergenceof P/E ratios is by no means complete impliesthat certain factors are still causing differences inP/E ratios through at least the eleventh year afterformation.The pattern of reversion toward a central value isa common phenomenon among economic variables.Research on the behavior of beta indicates a similarpattern. Two factors explain such behavior: (1) Thevariable being ranked normally contains a transitorycomponent; in our context, this means that earningsin a given year result in part from factors peculiar tothat year whose effects will either not persist beyondthat year or will dissipate in subsequent years. (2)The underlying, permanent value is reverting towardthe average. Such a reversion in the ratio of price toexpected earnings per share could be caused by achange in expected earnings growth or by a changein risk.Examining the time series behavior of the P/Eratio providesome insight into the nature of the factorsthat influence it. Apparently, some of these factorsdissipate substantially within the first three yearsafter formation. On the other hand, some continueeffective through at least the eleventh year after formation.We will consider three potential factorsgrowth,risk and accounting method.Does Growth Explain Differencesin P/E <strong>Ratios</strong>?Table 4 displays correlations between median P/Eratios in the year of formation and median earningsgrowth in the year of and in the years subsequentoformation.'I Column zero indicates the correlationbetween the P/E ratio and growth in the year of formation.The correlation of earnings growth in 1957with the P/E ratio computed at the end of 1957 isYears After Formation0 1 2 3 4 5 6 7 8 9 10 11 12 13 141 50.0 22.7 16.4 13.8 12.3 13.2 13.5 13.2 17.2 14.9 13.0 13.2 10.5 9.3 8.35 20.8 17.5 16.9 15.9 15.9 13.7 13.0 12.8 12.5 11.8 11.9 10.9 10.1 10.2 8.410 14.3 11.9 11.5 11.1 10.3 10.1 9.4 9.0 10.0 10.0 9.9 11.0 10.6 9.5 8.315 11.1 10.8 10.4 10.8 10.0 10.0 9.4 9.7 9.3 9.5 8.6 9.2 8.3 8.6 7.120 8.9 9.1 9.6 9.3 9.4 9.3 9.3 9.0 8.8 8.8 9.0 8.2 7.6 7.0 7.725 5.8 6.9 8.0 7.8 7.9 7.9 8.2 8.8 8.3 8.5 7.8 7.5 7.5 7.5 8.9Port. 1Port. 258.6 3.3 2.1 1.8 1.6 1.7 1.6 1.5 2.1 1.8 1.7 1.8 1.4 1.2 0.968 O FINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978
negative 0.28. The median correlation over the 19years from 1957 through 1975 is also negative 0.28,and 16 of the 19 correlations are negative.The negative correlation implies that stocks withrelatively low earnings growth during the year tendto have relatively high P/E ratios. This is consistentwith the contention that market participants perceivethat earnings contain transitory componentsand price stocks accordingly.'8 Since we formedportfolios on the basis of the ratio of price to realizedearnings, we expect that the ranking will systematicallygroup together stocks with transitory earningsof the same sign. In other words, the portfoliowith the highest P/E ratio will tend to include firmswith negative transitory components (i.e., realizedearnings below expected earnings) and converselyfor the portfolio with the lowest P/E ratio.Table 4 displays a strong correlation between P/Eand earnings growth in the year subsequento portfolioformation. The median correlation for baseyears 1957 through 1975 is 0.53, and all 19 correlationsare positive. Market participants' perceptionsof the transitory nature of earnings were confirmedby actual earnings behavior. While, in the year offormation, current earnings were abnormally lowrelative to expected permanent earnings, in the subsequentyear earnings tended to "catch up" to investors'expectations about permanent earnings.'9In the second year after formation, the mediancorrelation is 0.25 and, from there on, growth is essentiallyuncorrelated with P/E in the year of formation.The rapid dissipation in subsequent growthrates is similar to the pattern for P/E ratios observedin Table 2. In general, the pattern behaves as if marketparticipants, in determining prices, cannot forecastdifferential growth beyond two years.20One may ask what this tells us about the ability ofmarket participants to isolate and detect transitoryelements in current earnings versus their ability toforecast unusual earnings situations with respect toadditional investment. Although the distinction mayseem arbitrary, the implications can differ substantially.The first process concerns unusual factors dueto events of this year that will not persist (e.g., an abnormallyhigh rate of inflation or abnormally highinterest rates), while the second asks questions relatedto future unusual earnings opportunities.2'As far as we can tell, the data provide no basis forassessing how much of the observed growth differentialis due to each factor (although the introductionof other evidence may permit such a basis).2' As aresult, we cannot preclude the possibility that the resultsmay be entirely due to the detection of transitorycomponents that take more than one year to disappearfrom reported earnings. It is not our intentionto be unduly pessimistic, but rather to cautionagainst interpreting the figures as solely the result ofmarket participants' ability to forecast unusual earningsopportunities on future investments.Magnitude of Differential GrowthTable 4's correlation matrix does not provide informationon the magnitude of differental earningsgrowth. Table 5 displays this information for theTABLE 4: Rank Correlations of Portfolios Formed By P/E <strong>Ratios</strong> With <strong>Earnings</strong> Growth in Subsequent YearsBaseYears After FormationYear 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 151956 0.12 0.13 -0.30 0.48 0.00 -0.12 -0.62 -0.17 -0.19 -0.32 -0.08 -0.31 0.07 0.05 -0.481957 -0.28 0.53 -0.23 0.51 -0.07 0.02 -0.73 -0.26 0.26 -0.23 -0.01 -0.14 0.22 0.21 -0.33 -0.031958 -0.28 0.52 0.32 0.03 0.18 -0.15 0.19 0.29 0.11 -0.54 0.16 0.32 -0.20 -0.40 0.39 0.231959 -0.37 0.62 0.40 -0.12 -0.16 0.27 0.15 0.30 0.04 -0.22 0.32 0.04 -0.48 0.17 0.21 0.361960 -0.10 0.49 -0.15 -0.04 -0.04 0.04 -0.01 0.26 0.29 0.26 0.06 -0.13 0.09 -0.05 -0.45 0.101961 -0.45 0.53 0.13 0.07 -0.05 -0.03 0.27 -0.11 0.22 0.40 0.08 0.36 -0.16 -0.51 0.271962 -0.35 0.17 0.12 0.05 -0.26 -0.40 0.20 -0.39 0.49 -0.17 -0.19 -0.44 -0.46 0.311963 -0.42 0.26 0.24 0.13 0.35 0.08 0.13 0.56 -0.03 -0.27 -0.12 -0.47 0.281964 -0.43 0.11 -0.14 0.68 0.08 0.44 0.71 0.06 -0.21 -0.30 -0.31 0.041965 0.12 0.47 0.55 0.04 0.22 0.61 -0.16 -0.14 -0.15 -0.26 0.261966 -0.02 0.77 0.02 0.30 0.66 -0.35 -0.14 -0.38 -0.16 0.351967 0.18 0.74 0.41 0.12 0.05 0.39 0.01 -0.31 -0.071968 -0.02 0.77 0.74 0.54 0.42 -0.14 0.15 0.321969 0.15 0.75 0.48 0.05 -0.17 -0.42 0.581970 -0.19 0.90 0.58 -0.09 -0.63 0.341971 -0.22 0.71 0.27 -0.02 0.321972 -0.18 0.45 -0.09 0.391973 -0.31 0.52 0.511974 -0.66 0.891975 -0.66MedianCorrelation -0.28 0.53 0.25 0.05 0.06 0.02 0.14 -0.11 0.01 -0.22 0.02 -0.08 -0.18 0.07 0.24 0.10FINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978 E 69
Portfoliosame six composite portfolios reported in Table 3,using the same composite process. The results supportthe contentions made earlier. Portfolio One (thehighest P/E portfolio) experienced a median drop inearnings of 4.1 per cent, while Portfolio 25 experienceda median earnings increase of 26.4 per cent.In simplest terms, the prices of the stocks in PortfolioOne did not change proportionately with theirearnings; as a result, their P/E ratios were relativelyhigh. Similarly, the stocks in Portfolio 25 experienceda price change that, on average, was less than26 per cent, and their P/E ratios were relatively low.Again, this implies a price formation processwhereby participants view changes in earnings ascontaining a transitory element.In Table 5, Portfolio One shows a median earningsgrowth of 95.3 per cent in the first year after formation,while Portfolio 25 shows a drop in earningsof 3.3 per cent. This is consistent with the results reportedin Table 4: The perceptions of market participantsregarding the transitory element in earningswere confirmed by subsequent earnings behavior.The results in Table 5 differ from those in Table 4in one major respect-the highest P/E portfoliomaintains its distinctive earnings growth behaviorfor seven years after formation. This is neither contradictorynor surprising. Whereas correlations inTable 4 reflect the strength of the relationship for all25 portfolios, where growth and P/E are essentiallyunrelated after two years, the comparison describedabove involves only one of those portfolios. For thatone " extreme" portfolio we would expect nonnormalgrowth to be larger and to last longer.The results in Table 5 can be deceptive in at leasttwo respects. First, they reflect the average effectacross a number of base years and do not revealvariation from one base year to another. A moredetailed examination, not reported here, revealedconsiderable variation across base years. Second,while it is intuitively appealing to focus solely on thehigh P/E portfolio, it constitutes only four per centof the observations. It is importanto remember that,with the remaining portfolios included, there is littleor no apparent relation beyond the second year.TABLE 5: <strong>Earnings</strong> GrowthComparing the P/E analysis with the growthanalysis, we conclude that some of the initial dissipationof the P/E ratio in the first three years after formationcan be explained by differential growth inearnings. Beyond that, however, there clearly exists aP/E differential that cannot be explained by differentialearnings growth.Before leaving growth analysis, we'd like to commenton one aspect of the data. In contrast to P/Eratios, which exhibit a high degree of correlationover time, previous evidence indicates that earningsgrowth rates possess near-zero correlation over time.To ensure that the same behavior held for the stocksin our sample, we constructed a portfolio strategybased on earnings growth in the year of formationand then observed subsequent growth. Our resultsconfirmed previous findings of near-zero correlationof earnings growth rates. While a mechanical processrelying on past growth rates is largely unsuccessful inpredicting future differential growth, the P/E ratio issuccessful because price reflects a process wherebymarket participants rely on more information thanpast earnings in distinguishing the transitory andpermanent components of earnings.Risk AnalysisThe expected sign of the correlation between P/Eand beta may be either positive or negative. The argumentis developed in greater detail in the appendix,but it essentially proceeds as follows: Stocks'earnings move together because of economy-widefactors. In years of transitorily low earnings, themarket-wide P/E will tend to be high, but stocks withhigh betas will tend to have even higher P/E ratiosbecause their earnings are most sensitive to economy-wideevents. Conversely, in years of transitorilyhigh earnings, high beta stocks will have even lowerP/E ratios than most. Therefore we expect a positivecorrelation in "high" P/E years and a negative correlationin "low" years.Table 6 reports the rank correlations between P/Eand beta. We compared beta for a given base yearover the 60 months subsequento formation; thusthe beta for 1956 was computed over the years 1957Years After Formation0 1 2 3 4 5 6 7 8 9 10 11 12 13 141 -4.1 95.3 37.2 28.2 16.4 18.9 18.1 19.7 13.1 14.8 15.3 10.8 10.9 10.2 11.85 10.7 14.9 12.1 13.1 14.2 10.9 10.4 11.8 10.5 11.6 8.0 11.9 8.3 13.3 18.110 9.6 12.9 11.5 12.3 12.6 9.2 10.1 10.8 12.8 8.3 12.9 22.2 16.6 20.6 29.615 10.0 8.8 8.5 8.1 8.2 14.3 11.6 5.4 13.3 10.3 11.0 10.8 11.9 12.8 33.420 10.8 5.2 9.3 12.6 12.4 6.0 8.4 13.0 10.2 11.3 11.1 25.0 12.9 17.7 18.025 26.4 -3.3 7.5 10.8 8.3 12.9 17.1 13.6 18.0 12.8 16.7 14.2 10.9 12.4 10.1Port. 1Port. 25-0.155 * 5.0 2.61 1.98 1.47 1.06 1.45 0.73 1.16 0.92 0.76 1.00 0.82 1.17*Not meaningful because of negative growth in denominator.70 D FINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978
TABLE 6: Rank Correlations of Portfolio Median P/E<strong>Ratios</strong> and Median BetaRank of PredictedRank Median Sign ofYear Correlation P/Es Correlation1956 -0.34 14 -1957 -0.23 15 -1958 0.22 3 +1959 0.41 5 +1960 0.50 8 +1961 0.55 1 +1962 -0.48 10 -1963 -0.42 7 + b1964 -0.63 11 -1965 -0.26 9 -1966 -0.44 13 -1967 0.50 4 +1968 0.53 2 +1969 0.58 12 _b1970 0.28 6 +Median Adjustedfor Predicted 0.41Signa Computed from Table 1.b 1963 and 1969 are the two years incorrectly predicted.through 1961.To predict the sign of the correlation in a givenbase year, we ranked the market-wide P/E ratios (asreported in Table 1) from high to low.23 We hypothesizedthat the years with the eight highestvalues of market-wide P/E ratios would have a positivecorrelation between P/E and beta, while theyears with the seven lowest values of market-wideP/E would have a negative correlation. Over the 15base years 1956 through 1970, the actual correlationsare positive eight times and negative sevenand those in which the correlation was negative. Thethird column reports the beta differences for theyears of positive correlation. The differences aresmall for Portfolios Five through 25, where betaranges from 1.09 to 0.94. The largest difference occursin the highest P/E portfolio, with its beta of1.28. In the fourth column, negative correlation isevident for Portfolios Five through 25, with a pronouncedaberrant behavior for Portfolio One. Forthis set of stocks, a "U-shaped" relationship is present.Given the consistently high betas of the highestP/E portfolio, it is imperative that some form of riskadjustedperformance standard be introduced toavoid spuriously inferring superior stock price performance.While beta clearly holds some explanatory power,the crucial issue is, to what extent does it explain theP/E ratio behavior reported in Tables 2 and 3? Wethink it explains little: If beta were an important explanatoryvariable, then the predicted behavior ofP/E over time would be much different from whatTables 2 and 3 report. Stocks in Portfolio One duringyears of high market-wide P/E would tend tomove to Portfolio 25 (or its neighbors) in years oflow market-wide P/E. Looking across a row of Table2, we would expect to see a pattern of positive andnegative correlationsimilar to that reported in thelast column of Table 6. Instead, we observe a strongpositive serial correlation throughout.24 Furthermore,the relative differences in betas are not of thesame magnitude as the relative differences in P/Eratios. Before considering another source of P/E differences,however, we report the results of a regressionanalysis that combines both growth and riskanalysis.times. We correctly predicted the sign of the correlationfor 13 of the 15 years. This is impressive, given Regression Analysisthe crudeness of the test. (The test's limitations are Table 8 displays the results of a simple linear rediscussedin the appendix.)gression that included beta and earnings growth asTable 7 reports the magnitude of the betas for the independent variables. We used the EIP, rather thansix portfolios presented in Tables 3 and 5. Because P/E, ratio because the Litzenberger and Rao modelthe relation between P/E and beta can be either posi- posits linearity in E/P (not in P/E).25 The expectedtive or negative, we averaged results over two sets of sign of the E/P and beta relationship is thus the reyears-thosein which the correlation was positive verse of that shown in the final column of Table 6.The actual regression coefficients have the predictedTable 7: Relation of P/E and Beta sign in 13 of the 15 years; again, 1963 and 1969 areAverage Average exceptions.Beta Beta The predicted signs of the growth coefficients areAverage in Years of in Years of also negative, since we're using E/P as the dependentBeta in Positive Negativevariable. For growth in the year subsequento port-Portfolio All Years Correlation Correlationfolio formation (gl), all 15 coefficients have the pre-1 1.22 1.28 1.13dicted sign. Growth two years subsequento forma-5 1.01 1.03 0.9810 1.05 1.09 1.00tion (g2) has the predicted sign in 12 years. By the15 0.96 0.94 1.00 third year (g3), however, the signs of the coefficients20 1.03 0.96 1.11 are evenly divided. Table 4 suggests there is little25 1.04 0.95 1.14 merit to introducing additional growth variables.FINANCIAL ANALYSTS JOURNAL I JULY-AUGUST 1978 O 71
TABLE 8: E/P RegressionResultsRegression Coefficients Adjusted F(t-Statistic) R2 StatisticBaseYear Constant Beta 91 92 931956 0.0700.030 --0.046 -0.035 0.0531956 0.070 (0.71) (-1.38) (-0.73) (0.93)01523 0.185 2.361957 0.3480.086* -0. 142' -0.066 -0.123'.8 .(1957 0.348 (1.74) (-4.13) (-1.57) (-2.29) 0.581 930*1958 0.136 0000 -0.01 9) -0.070 0.013 0.270 3.96(0) (-2.57) (-1.69) (0.26)1959 0.076 -0.053' -0.157* 0.098 0.086 0.505 7.13*(-1.76) (-4.45) (1.29) (1.62)-0.075* -0. 161'* 0.046 0.0921960 0.1550.502 7.06*1960 0.155 ~~(-2.83) (-3.55) (0.71) (1.17) 0.27061961 0.077 -0.054* -0.063 0.064 0.023 0.289 3.44*(-1.89) (-1.25) (0.93) (0.57)1962 0.119 0.1 16* --0.055' -0.026 --0.064 0.524 7.61*(4.68) (-2.10) (-0.50) (-1.02)1963 02390. 097* -0. 106* -0.089' -0.0341963 0.239 (3.10) (-1.93) (-1.84) (-0.62) 037452 0370 4.52*1964 1964 0.260 0. 0076* -0.085 -0.045 -0 114' 0.572 9.03*~~(2.36) (-1.26) (-0.67) (-2.24) 0.79031965 0.071* -0. 167* -0.091' -0-0221965 0.300 ?(2.00) (-2.54) (-1.73) (-0.32) 047644 0.475 6.44*1966 0.501 0. 112' -0.304' -0.068 -0. 138* 0.783 22.63*(3.53) (-7.06) (-0.96) (-1.80)1967 -0.065* -0. 164' -0.106 --00341967 0.447 (_-2.15) (-2.87) (-1.67) (-0.94) 05791, 0.575 9.10'1968 0.285 -0.031 -0.033 -0. 108; -0.060 0.738 17.91*(-1.71) (-1.71) (-5.23) (-1.62)1969 0.3801970 0.185-0.054 -0. 159' -0. 134' 0.030(-1.40) (-4.64) (-2.09) (0.57)-0.029 -0.003 -0. 1 67* 0.089(-0.71) (-0.28) (-2.58) (1.42)*Significant at five per cent level (one-tail test on regression coeff icients).0.658 12.56*0.391 4.86*The R2 (proportion of variance explained) adjusted stated when measurement error is present. (2) Betterfor degrees of freedom ranges from 18.5 per cent to specification of the denominator of the P/E ratio78.3 per cent, with a median of 50.5 per cent. The F- might yield better results. ' (3) Accounting rulesstatistic, which tests the null hypothesis that all of the could be creating P/E differences; our final comcoefficientsate zero, is significant at the five per cent ments are devoted to this area.level in 13 of the 15 years."'iWhile risk and growth on the average explain approximately50 per cent of the variance of the EIPAccounting Methodratio, they obviously leave an equal proportion un- The finding that P/E ratio differences persist wellexplained. Thus the regression results presented in beyond three years after portfolio formation suggestsTable 8 provide only the crude beginnings of an at- the influence of some factor other than risk ortempt to explain cross-sectional P/E differences. We growth. Accounting effects are obvious candidates.suggest further research in a number of areas: (1) Accounting method effects are of two types- use ofEven if the realized values of beta and growth are different rules (e.g., depreciation methods) by differunbiasedestimates of expectations, they may still ent firms for essentially the same or similar circummeasurethose expectations with error; better speci- stances and errors introduced by applying a uniformfication could lead to higher R2, which is under- accounting rule (e.g., historical cost) to differing72 O FINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978
economic circumstances (e.g., current value ofassets).The P/E ratio will be influenced by the effect onearnings of differing accounting methods. Assumingprices are not dependent on the accounting methodused in annual reports, firms that use conservativeaccounting methods (e.g., accelerated depreciationor LIFO inventory valuation) would tend to havehigher P/E ratios than firms that use less conserva-tive methods, holding constanthe effects of risk andgrowth.'8 For example, Beaver and Dukes found theP/E ratios of a portfolio of firms using accelerateddepreciation were greater than the P/E ratios of a(a) we have:(E/Pt - E/P) = b(Mt - M) + Utwhere E/P = expected value of E/Pt andM = expected value of MtIgnoring ut and taking b equal to b, we have:(E/Pt - E/P) = b(Mt - M) .pressed as a deviation from the expected value).However, when the realized M1 is below its expectedportfolio of firms using straight-line depreciation.'9 value, stocks with higher betas will have lower E/P'sThe two portfolios were essentially the same with (expressed in terms of a deviation from its mean).respect to risk (beta) and growth. Moreover, whenthe earnings of the straight-line portfolio were con- Limitations of the Empirical Testverted according to the accelerated method, the P/E One obvious limitation is the failure to expressratios were essentially the same for both portfolios. P/E (or E/P) as a deviation from its expected value,In other words, the P/E differences in the two port- as indicated by Equation (c). This test implicitlyfolios disappeared when earnings were computed on assumes that inter-stock P/E differences are zero.a uniform depreciation method. We suggest an ex- This is obviously not the case, as Tables 2 and 3 intensionof this type of analysis to other accounting dicate. However, we did not take this latter stepmethods as an obvious candidate for future re- since our concern throughout has been with the risksearch.30 Udifferences of a simple, P/E-oriented portfoliostrategy, not a strategy that expresses P/E as a deviationfrom its expected value. A second limitation isthe assumption that 15 years taken as a whole con-APPENDIXtain approximately an equal number of realizationsIf the EPS used to compute P/E ratios contained no above and below the expected value, which in turn istransitory elements, we would expect a positive rela- assumed to be a constant over 15 years.tion between E/P and beta and a negative relation In the absence of any other evidence, this assumpbetweenP/E and beta. However, the evidence sug- tion seems as reasonable as any other. However, thegests that transitory elements in EPS are present. assumption of a constant expected value cannot beHow does this affect the analysis of risk? strictly true. Factors such as changing interest ratesPrevious empirical research indicates that a would lead us to expect that market-wide P/E ratiosstock's E/P ratio can be characterized by the follow- change over time. Third, the test ignores the influing(linear) process:ence of ut (the unsystematic component) as expressedin Equation (b).E/Pt=a+bMt+ut(a)where E/Pt = earnings-price ratio for a stock in Further Analysis of Table 7year t,Evidence in our article supports the contentionMt = a market-wide E/P ratio for year t, that transitory elements in earnings exist. We wouldexpect high P/E stocks to have greater earnings variut= a non-market residual for a stock in ability to the extent that transitory elements accountyear tandfor P/E differences. Research by Beaver and Manegold,among others, confirms that stocks with greater= the intercept and slope of the linear earnings variability have higher betas. However, the6 } relationship.,argument is not complete: If earnings volatility wereexclusively systematic, we would have observed noMoreover, this research has shown that, at the port- U-shaped behavior in the years with negative correfoliolevel, b (the earnings-price "beta") is highly lation. If unsystematic earnings volatility (the varicorrelatedwith beta.ance of ut from Equation (a)) is positively correlatedIn a given year where the actual, realized earnings with beta, and if high P/E stocks have greater unsysmaydiffer from the expected earnings, what relation tematic volatility, the result would be a consistentlycan we expect between E/P and beta? Rearranging higher beta for the highest P/E portfolio. However, a(b)(c)When the realized Mt is above its expected value,stocks with higher betas will have higher E/P's (ex-FINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978 FO 73
similar argument could be offered for the lowest P/Eportfolio. Yet consistently higher betas are not observedhere. We offer no explanation for this result.Footnotes1. See I. Little, "Higgledy Piggledy Growth" (Instituteof Statistics, Oxford, November 1962), the seminalwork. See also R. Ball and R. Watts, "Some TimeSeries Properties of Accounting <strong>Earnings</strong> Numbers,"Journal of Finance (June 1972), pp. 663-682; J.Cragg and B. Malkiel, "The Consensus and Accuracyof Some Predictions of the Growth of Corporate<strong>Earnings</strong>," Journal of Finance (March 1968), pp.67-84; I. Little and Raynor, Higgledy PiggledyGrowth Again (Oxford: Basil Blackwell, 1966); andJ. Murphy, "Relative Growth in <strong>Earnings</strong> PerShare-Past and Future," Financial Analysts Journal(November/December 1966), pp. 73-76. Excellentsummaries appear in R. Brealey, An Introductionto Risk and Return from Common Stocks(Cambridge: MIT Press, 1969) and in J. Lorie and M.Hamilton, The Stock Market: Theories and Evidence(Homewood, Illinois: Irwin, 1973).2. Previous studies have attempted to use the P/E ratioitself as a growth predictor-with little success. Twoexamples are Cragg and Malkiel, "Consensus andAccuracy" and J. Murphy and H. Stevenson,"<strong>Price</strong>/<strong>Earnings</strong> <strong>Ratios</strong> and Future Growth of <strong>Earnings</strong>and Dividends," Financial Analysts Journal(November/December 1967), pp. 1 11- 114. The portfoliostrategy adopted in our study provides an opportunityto uncover relations that may have beenundetected by previous research.3. Let CFt equal the cash flow generated t periods fromnow from current equity investment. Assuming r isconstant:E xcr CFt (1 + r)-t =rPresent Value of Current1 Equity Investment,E r l CFt (I + r)A-t =Permanent, t= I <strong>Earnings</strong>4 It appears to us that a basic inconsistency may existwhen perfect markets are invoked to motivate presentvalue formulas and yet abnormal returns in productiveopportunities are posited to explain growth premiumsor discounts in P/E ratios. However, this is apuzzle we are not prepared to resolve. Stocks may stillbe priced "as if' a discounted cash flow model wereapplied even in the presence of abnormal returns inthe productive sector (i.e., the product and factormarkets).5. Consensus expectations in general depend on thewealth, risk preferences and beliefs of market participants.("Market participants" is a generic term intendedto include individuals whose expectations directlyor indirectly influence market prices-i.e.,analysts as well as investors.) In a one-period setting,expressing price as a function of expected values is anarbitrary, although innoucuous, way to view valuation.In a multi-period setting, however, such a valuationscheme will not necessarily hold.6. Previous attempts by researchers to remove transitoryelements from earnings vary from subjective adjustmentsof the components of earnings to statisticaldata fitting via Box-Jenkins techniques. For an exampleof the latter, see P. Griffin, "The Time Series Behaviorof Quarterly <strong>Earnings</strong>," Journal of AccountingResearch (Spring 1977), 71-83.We would like to be able to say a portfolio approachwill permit us to diversify out the transitoryearnings components. Clearly we cannot make such astatement. To the contrary, since the portfolios willbe formed on the basis of the ratio of price to realizedearnings, we expect the ranking will systematicallygroup together stocks with transitory earnings ofthe same sign. In other words, the portfolio with thehighest P/E ratios will tend to include firms where thetransitory component is negative (i.e., realized earningsare below expected earnings) and conversely forthe portfolio with the lowest P/E ratios. Our evidencewill supporthese contentions, and it is a major pointto keep in mind in interpreting the results.7. For a more detailed discussion of the effects of aggregationinto portfolios see W. Beaver and J. Manegold,"The Association Between Market-Determinedand Accounting-Determined Risk Measures," Journalof Financial and Quantitative Analysis (June1975), pp. 231-284. In this context, "noise" refers tothe fact that our growth and risk measures may differfrom the growth and risk expected at the time of portfolioformation.8. Beta was estimated as the slope of a linear regressionof the form:Rit = ai + biRmt + eit , t = 1,60where Rit equals monthly percentage change in price(adjusted for dividends) for security i in month t andRmt equals monthly percentage change in a marketindex of price changes (adjusted for dividends) of allNYSE firms (provided as part of CRSP tape).9. The median was used because it was a nonparametricmeasure that would place less demands on the data.Various weighting schemes (e.g., weighting by marketvalue) were also applied with essentially the same resultsas reported here. Since it is unclear how to definegrowth off of negative earnings, the portfolioswere formed only over those stocks with positiveearnings in the base year. However, the sign of earningswas unrestricted in the years subsequento formation.Hence earnings could be negative in lateryears.10. The Compustatape contains those firms that havesurvived mergers and bankruptcy. We would expectthat this could induce a potential bias in the levels ofthe variables to be studied. However, the studyfocuses on differences in these variables across portfoliosof stocks. It is not obvious to what extent survivorshipimparts a bias for this purpose. This wouldrequire knowledge of how non-surviving firms sys-74 O FINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978
tematically differ from surviving ones. In any event,because of the survivorship criteria, the portfoliostrategies described here could not have been literallyfollowed by an investor.11. The number of stocks increases over time because ofavailability on the Compustat tape. In virtually all instances,the firms with incomplete histories on Compustatexisted throughou the 1956-75 period butwere picked up at some later date by Compustat. Inother words, the firms being added do not tend to benew firms.12. These aggregate statistics were obtained from S&P'sTrade and Securities Statistics, 1976.13. It is importanto realize that a correlation matrix likethat of Table 2 invariably leaves out certain information.For example, does a correlation coefficient ofone mean the P/E difference between portfolios remainsapproximately the same as it was in the year offormation? Not necessarily. We can draw no inferenceabout the magnitude of portfolio differences; therank correlation coefficient merely indicates asimilarity in the rankings of the portfolios. A coefficientof one indicates that the rankings of two portfoliosremain the same, but it does not tell us anythingabout the spread between the portfolios. Thespread between the portfolios may have shrunk orgrown. However, we do know that, even if the spreadhas changed, at least the relative positions remainedunaltered.14. Since the P/E ratios are highly correlated, we cannotview elements in one cell as unrelated to the othercells; the results in two adjacent rows should be highlyrelated. In other words, the results for the base year1957 are, not surprisingly, similar to the results forthe base year 1956. Because the results are not perfectlycorrelated, however, some additional informationis conveyed by repeating the portfolio strategyfor different base years.15. Table 3 is subjecto the same caveats as Table 2. Forexample, the years 1971 through 1975 play a relativelymore important role in the later years after formation,so the P/E ratios exhibit a downward drift.The real focal point of the table is the difference inP/E ratios across portfolios, rather than the commonmovement by all portfolios. Moreover, the results ofa more extensive examination of the data, which heldconstanthe calendar year composition of each year,did not differ from the findingshown in Table 3. Wechose Table 3's composite because it was the mostcomprehensive and the simplest method of presentation.16. For example, for Year 0 (the year of formation), theaverage P/E ratio for Portfolio One was computed bytaking a weighted average of each of the median P/Eratios for Portfolio One for the years 1956-74 inclusive.The weights were determined by the number ofsecurities in that portfolio in that year. 1956, with 10securities per portfolio, carried a weight of 0.029,while 1970-74 carried weights of 0.070 each with thesum of the weights over the 1956-74 period equallingone.17. Since earnings could be negative in any subsequentyear, a problem rose as to how to define growth whenthe denominator is negative. When earnings changedfrom negative to positive, the growth was defined tobe greater than the median (i.e.,"very large"). Whenearnings remained negative in both years, the observationwas deleted. Typically, this caused a deletionof less than two per cent of the observations, with theexception of Portfolio One, where deletions rangedfrom two to three per cent of the observations.18. We regard the transitory earnings argument as one,but by no means the only, interpretation to place onthe results.19. Suppose expected earnings per share are $1.00. Forconvenience, assume that realized earnings per sharewere $1.00 and $0.75 in 19xl and 19x2. The actualgrowth rate of 19x2 is -25 per cent, while the expectedgrowth in 19x3 is 33 per cent. Assuming priceremains essentially unchanged (because permanentearnings are unchanged) at $10.00, the P/E ratiowould be 10 and 13.3 for 19xl and 19x2, while theexpected P/E in 19x3 would be 10. Note that in 19x2there was an abnormal low earnings growth associatedwith a high P/E (negative correlation) and thata high P/E in l9x2 was followed by abnormal highgrowth in l9x3 (positive correlation). This is discussedin more detail in W. Beaver, "The InformationContent of the Magnitude of Unexpected <strong>Earnings</strong>"(Stanford Research Seminar, 1974).20. This conclusion is contingent upon the way we choseto measure the variables and to rank the portfolios.For example, a longer term measure of growth (i.e.,five or 10 years) might produce different results.21. By unusual earnings opportunities we essentiallymean opportunities to earn a return on future investmentsgreater than the cost of capital. Valuationtheory tells us that this form of future earnings growthwill induce a growth premium in the P/E. Note that itis crucial to distinguish between unusual earnings opportunitieson future, rather than current, investment.To the extent there are abnormal returnsearned on the current investment, this will not affectthe P/E ratio, although the ratio of market price tobook value per share would be affected.22. For example, if one believes that product and factormarkets are reasonably competitive, then there existlittle or no unusual earnings opportunities and theobserved growth differences are almost entirely dueto transitory factors.23. Operationally, the market-wide P/E was defined to bethe median P/E as reported in Table 1.24. Although not reported here, the analysis of the serialcorrelation in P/E behavior was augmented by ananalysis of a transition matrix. This analysis also confirmedthe lack of any material tendency to movefrom one extreme portfolio to extreme portfolios atthe other end of the P/E spectrum.25. Since our study is concerned with differences acrossP/E ratios at a given point in time, beta (bi) will be thesole determinant of differences in P/E resulting fromdifferences in ri. Assuming a finite growth horizon,Litzenberger and Rao ("Estimates of the MarginalRate of Time Preference and Average Risk AversionFINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978 a 75
of Investors in Electric Utility Shares: 1960-66," TheBell Journal of Economics and ManagementScience (Spring 1971), pp. 265-277) have shown theE/P ratio (the reciprocal of P/E) is a simple linearfunction of beta and a growth variable, with the followingform:l=Yo+ 'Ylbl =Y2f(g)PiThe sign of yI is expected to vary, f(g) is some functionof growth and Y2 is expected to be negative.26. An analysis of the residuals indicated that they arewell approximated by normality. This is not surprisingin the sense that each observation is an"average" for a portfolio of stocks. By the CentralLimit Theorem, we would expect the sampling distributionsof averages to approach normality. Note,however, that the results from any one year's regressionare not independent of those of other years.27. Two approaches immediately come to mind-the applicationof Box-Jenkins techniques to the past earningsseries or the use of analysts' forecasts of earnings.Either method might produce better assessments ofexpected earnings per share.28. This ordinal statement holds for both depreciationand inventory, even though inventory also implies a"real" difference due to taxes. Obviously, the specificadjustments to be made would have to distinguish betweenaccounting differences that imply tax differencesversus those that do not.29. W. Beaver and R. Dukes, "Delta-DepreciationMethods: Some Empirical Results," AccountingReview (April 1972), pp. 320-332. In this study, allsample firms were using an accelerated method fortax purposes. Therefore the difference was solely dueto the depreciation method used for annual reportpurposes.30. Another approach would be to introduce variablesrelated to the accounting effect. Recent works byWatts and Zimmerman ("Toward a Positive Theoryof the Determination of Accounting Standards," AccountingReview (April 1978) ) suggesthat conservativenessof accounting method varies positivelywith firm size. Van Breda ("The Prediction of Corporate<strong>Earnings</strong>" (Stanford University, 1976)) indicatesthat average age of assets is one of the variablesthat explains cross-sectional differences in return onequity (i.e., earnings available to common dividendby the book value of common equity).recommendations, issued on Novem- aspects of the new AICPA self-regu-Securitiesber 14, 1977, involved much less far- latory program. Although Moss isreaching federal intervention than the rumored to be on the verge of introandRegulation January 1977 staff report recom- ducing a new bill, he has indicatedmended. The subcommittee report on that he would not seek re-election.continued from page 38"Improving the Accountability ofPublicly-Owned Corporations and AICPA Actionstheir Auditors" recommended con- In September 1977, the Council of"... the Metcalf hearings conveyed one tinued emphasis on self-regulation for the AICPA adopted a proposal tovery definite and clear message- a sense accountants and the establishment of create a new division of CPA firmsof expectation and urgency for the profes- a self-regulatory organization, possi- that would comprise two sections withsion and, as necessary, the Commission, bly with mandatory membership, sub- voluntary firm membership-theto act to build the public's confidence in ject to SEC oversight. The report con- SEC Practice section, for CPA firmsthe independence of accountants, in their cluded that the accounting profession whose clients are SEC registrants, andresolve and ability to engage in meaning- and the SEC should "act in a timely the Private Companies Practice secfulself-discipline, and in the process by manner to implementhe policy goalswhich accounting standards are estabtion,for firms serving companies notlished.in this report."registered with the SEC. (A firmHowever, Representative Moss,"I have very little desire tocould be a member of bothpreside,sections.)during my five years as Chairman, over Chairman of the House Subcom- Each section would be administeredincreased regulation of the accounting mittee on Oversight and Investiga- by a 21 -member executive committee.profession. Similarly, I have no wish to tions, speaking in early March 1978, A five-member Public Oversightsee legislation passed that would place the warned the accounting profession that Board will have power to monitor acresponsibilityon the Commission-or he would seek legislation to impose tivities of the SEC Practice sectionsome other government body-to regu- federal regulation of accountants un- and report its findings publicly. Itlate accountants. Nevertheless, time is less the accounting profession im- could also impose sanctions such asrapidly running out on the opportunity proved its self-regulatory efforts. requiring a firm to take internal acforvoluntary initiatives."Moss seemed particularly concerned tion to improve its controls or opera-In fact, the Metcalf Subcommittee's about the possible anti-competitiveconcluded on page 8576 O FINANCIAL ANALYSTS JOURNAL / JULY-AUGUST 1978