Greek evidence on post-earnings announcement drift ...

Greek evidence on post-earnings announcement drift: Old empirical tests in a new theoretical bottle. William Forbes, Loughborough Business School, Ashby Road, Loughborough, Leicester, Len Skerratt and George Yiannopoulos Department of Economics & Finance Brunel University, Uxbridge, England June 5, 2006

Abstract A large literature already attests to the failure of both analysts and investors to interpret earnings adequately. This underlies well known market anomalies, especially post–earnings–announcement–drift. While a number of theoretical models have examined information flows underlying anomalous behaviour this theoretical discussion has played little role in extant empirical work. We distinguish between the types of information received by investors in a study of post–earnings–announcement–drift, contradictions and negations of expectations regarding future earnings. In addition, this investigation is conducted in a market setting different from the mature markets of the UK or US. This allows us to assess the variation in PEAD across different types of national markets. Our evidence comes from a market known for the opacity of its financial reporting and a period of some somewhat frenzied activity, when rumors and resulting contagion seemed rampant.

1

Introduction

An extensive literature exists on the presence and causes of post-earnings announcement drift (PEAD) (for a review, Brown (1997), Kothari (2001)). More recently rival models have emerged to aid understanding of how PEAD might occur in competitive financial markets (see Daniel, Hirshleifer, and Subrahmanyam (1998),Barberis, Shleifer, and Vishny (1998), Hong and Stein (1999), Jackson and Johnson (2006)). Strangely, much empirical work seems to proceed without reference to this emerging theory. This may be because empirical work has for a long time run ahead of theory in this area. However, we believe that it is now appropriate for empirical work on PEAD to be more theoretically informed. In order to contribute to this process, we revisit an earlier empirical study which seems prescient of the recent theoretical models of why PEAD occurs. Specifically, we revisit the US paper by Mendenhall (1991), henceforth M91, in the light of work by Daniel, Hirshleifer, and Subrahmanyam (1998), henceforth DHS98, which models potential causes of PEAD in competitive markets. A feature of the DHS98 model is that overconfidence may cause investors to buy/sell stocks for which their expectations about earnings are confirmed, while failing to unwind their stock holdings for which their earnings expectations are contradicted. In order to test this overconfidence explanation of PEAD, we conduct our tests on a sample of Greek companies in the late mid 1990’s. Greek companies are used because of the opacity of their accounting methods which allow companies to manage their earnings (Bhattacharya, Daouk, and Welker (2003)). Table 3 of Bhattachayra et a (2003) reports that Greek accounting 1

is the most opaque amongst 34 countries they examined. Furthermore, the period covered is one in which the economy was in the grips of a speculative bubble. These features together provide a setting in which overconfidence may be likely to arise, with investors imperfectly interpreting earnings announcements and the actions of other investors. The rest of the paper is structured as follows. The next section re-visits the M91 empirical study and interprets it in the light of the DHS98’s model of the origins of PEAD. In the second section we test the resulting amended empirical model. A final section concludes the paper.

2 2.1

Mendenhall revisited Earnings announcements which confirm or negate prior expectations

DHS98 directly examine PEAD in a set-up where trading occurs between informed and uninformed agents over four periods. For details of the model readers are referred to the original paper. Here we focus on the effect of the self-attribution bias on investors’ response to confirmations, or negations, of prior expectations of earnings changes and the informational context in which this alleged bias arises. Self attribution bias has been characterised as having an attitude ”Heads I win, Tails its chance”. In particular, DHS98 suggest investors allow confirmations of prior expectations regarding earnings changes to increase their confidence in their ability to predict future earnings. However, negations of prior expectations do not cause investors to

2

lose confidence as might be expected. This asymmetric response of investors to earnings data leads to inefficient processing of new information and consequently investor overconfidence in their ability to predict earnings. Such overconfidence may cause investors to buy/sell stocks for which their expectations are confirmed, while failing to unwind their stock holdings for which their earnings expectations are contradicted. So we might expect underreaction/PEAD to be clustered in portfolios subject to contradictory earnings signals. Any attempt to unwind an unfavourable position might be particularly fraught in a very illiquid market, such as that in Athens. This might explain the persistence of the PEAD we observe. DHS98 investigate two variants of their model. In the first version the signal about earnings, issued at the second date, is public. It is the second variant of the model that we investigate here, in which the second date signal is a private signal received by managers of the firm alone. Both versions of the model imply the presence of short-term overreaction, followed by secular under-reaction. However, it is this second version of the model alone which predicts the occurrence of PEAD. The above framework implies (at least) two testable hypotheses. H 1 investors respond asymmetrically to confirmations and negations of their expectations of future earnings changes. Confirmations of prior expectations lead to increases in investors’ confidence in their ability to predict earnings, while negations of expectations fail to diminish investors’ confidence in their trading ability. Such reduced confidence results in under-reaction to earnings information at the earnings announcement date. H 2 any resulting investor overconfidence is particularly likely to induce

3

PEAD in stocks where private/”inside” information is rife. We test these two propositions below. We do so by revising and augmenting the empirical model of M91. Details of the M91 test method are given in the next sub-section. A subsequent sub-section extends the model to allow for the receipt of confirmatory and negatory earnings signals within public and private information contexts. 2.1.1

The M91 test

M91 investigate the possibility that investors underestimate the persistence of analysts’ forecast errors. Further, investors might try to judge the persistence of forecast errors from observing analysts’ revisions of their earnings forecasts. This failure to fully recognise the persistence of earnings is hypothesised to underly PEAD. Thus M91 argues that PEAD might be partly explained by the configuration of analysts’ forecasts errors and their forecast revisions. Specifically, M91 investigate the cycle of forecast errors represented by the time line of Figure 1 below.

1st forecast

1st actual

2nd forecast

2nd actual

t = F1

t = A1

t = F2

t = A2

Eˆ1,1 , Eˆ2,1

- E 1

-

Eˆ2,2

-

E2

Figure 1: Cycle of Earnings Announcement & Forecast Revision The cycle begins with an initial forecast of earnings, Eˆ for years 1 and 2,

4

made in the first year, Eˆ11 , Eˆ21 . Here the first subscript denotes the earnings announcement being predicted, while the second denotes the forecast date. Following the announcement of actual earnings for the first year of the cycle, E1 , a second forecast of earnings at date 2 is issued Eˆ2,2 . The cycle closes and re-commences with the announcement of year 2 earnings, E2 . Two variables are used to examine the forecast revision cycle and its relation to stock price movements. These are the forecast revision and the forecast error, both of which are scaled by price. So we have ERRi,n = REVn =

Ei,n − Eˆi,i,n Pi,n Eˆ2,2,n − Eˆ2,1,n

(1)

Pi,n

where ERRin = the price deflated forecast error for the ith earnings announcement (i=1,2) for the prediction made at the ith forecast date for the nth observation and REVn is the forecast revision between the first forecast date and the second for the nth, observation, Ein = the EPS reported on the ith announcement date for the nth observation, Eˆi,j,n = is the prediction of Ei made of the jth forecast date (i=1,2 and j=1,2) for the nth observation, Pi,n = the closing stock price ten days before the ith earnings announcement of the nth observation. So to undertake the M91 tests we require data on two consecutive earnings announcements and corresponding forecasts for earnings announced on those 5

two dates. Within this structure M91 test three related hypotheses. 1. H1M . A regression based test that successive forecast errors are related. This takes the form ERR2,n = α0 + α1 ERR1,n + errorn

(2)

where the null hypothesis of forecast rationality implies that as a minimum α1 = 0. 2. H2M . After controlling for the forecast revision there is no systematic relation between stock market returns around the forecast revision date and the most recent forecast error. M91 tests this by the following regression CARn (F2 −1, F2 +3) = γ0 +γ1 REVn +γ2s .Sn .ERR1,n +γ2d .(1−Sn ).ERR1,n +errorn (3)

where    1, If REVn & ERR1,n are the same sign and, Sn =  

0, If REVn & ERR1,n are of different sign

and CARn (F 2 − 1, F 2 + 3) is the 3 day event-window return to the stock portfolio. Under the null hypothesis of forecast rationality we would expect γˆ2s = γˆ2d = 0, in equation (3) i.e. past forecast errors to have no effect upon 6

investors’ response to forecast revisions. This is because past forecast errors have been fully impounded into both prices and analysts expectations of earnings. Conversely, if prior forecast revisions are regarded by the market as indicative that current forecast errors are likely to recur in the next period then we might expect γˆ2s > 0 and γˆ2d < 0. So that investors respond positively to revisions to counteract previous forecast errors and negatively to revisions which seem to exacerbate past errors. To illustrate what is happening here consider a stock whose price remains fixed at a value of 1 in Table 1. Suppose the value of earnings are forecast to be 1 this year and next year, (Eˆ11 = Eˆ12 = 1), but actually turns out to be 2, in period 1, E1 = 1 i.e. investors receive “good news” about earnings. So ERR takes the value (2-1)/1=1. In response to this analysts upgrade their forecast of next year’s earnings to remain at 2, (so Eˆ22 = 2). So the forecast revision in response to earnings is (2-1)/1=1. So ERR11 = REVn = 1 and a confirmatory information signal has been received from the upward revision of analysts’ earnings forecasts. The M91 framework predicts a positive price response to the forecast revision in this case. Conversely, when actual earnings are equal to a value of 2 (despite an expected value of 1) the resulting forecast revision falls from 2 to 0. Here a negative price response is predicted by M91. In this second case ERR remains at (2-1)/1=1, but REVn now takes a value (0-1)/1=-1. The analyst(s) following the firm have been pessimistic in the past and continue to exhibit pessimism. Hence the market discounts the credibility of the 7

forecast revision issued. When doubts persist in the market about the persistence of earnings, because of a contradiction expectations, greater PEAD is likely to result. Of course price movements will change the scale of both ERR and REV, but not their sign, so the underlying logic of our numerical example remains unchanged once prices are allowed to change. 3. H3M . Finally, M91 investigates whether there is any systematic relationship between prior earnings forecast revisions and PEAD. This is tested by the regression CAR(A2 − 1, A2 ) = β0 + β1 ERR1,n + β2 REVn + errorn

(4)

where under the null of investor rationality we expect β2 = 0, i.e. earning expectations fully impound analysts’ forecast revisions and PEAD is solely a function of any unpredicted ”earnings surprise”. That is the PEAD is purely a response to earnings “news”.

2.2

The M91 tests revisited following DHS98

Re-reading M91’s tests in light of the DHS98 interpretation of PEAD, one common theme which stands out is the importance of the distinction between confirmatory and negatory signals about earnings. Note that in DHS98’s framework confirmation of prior earnings expectations makes investors more confident. Nevertheless contradictions of prior earnings expectations do not diminish investor confidence in their ability to interpret earnings information. The presence of such “self-attribution bias” in investor’s earnings expectations implies γˆ2s > 0 and γˆ2d = 0 in equation (3). 8

However DHS98’s predictions relate to PEAD, as opposed to price responses to analysts forecast revisions. So in presenting the results we focus upon PEAD, as opposed to post forecast revision drift. Hence the importance of the distinction between confirmatory and negatory earnings announcements is in relation to the regression in equation (4), rather than equation (3). So in light of DHS98 we suggest a re-specification of equation (4) CARn (A2 − 1, A2 ) = γ0 + γ1s Sn REVn + γ1d (1 − Sn )REVn

(5)

+γ2s Sn ERR1,n + γ2d (1 − Sn )ERR1,n + errorn Sn =

   1, If REVn & ERR1,n are the same sign and,   0,

If REVn & ERR1,n are of different sign

Note that under DHS98 theoretical framework suggests γ2s > γ2d , at date A2 so announcements which confirm investors’ expectations lead to stronger market reactions than those which contradict them. Therefore if prior revisions have been positive, REV > 1 and those expectations are confounded by an optimistic forecast error, ERR < 0. the price response to“bad news” about earnings is relatively sluggish compared to the outcome if those expectations were confirmed. This induces the greater PEAD associated with such negation of earnings expectations. As in M91 we might also expect γ1d = γ2d = 0. 2.2.1

Extending the M91 tests to allow for the role of private information in PEAD

So far we have simply re-interpreted the M91 framework in light of DHS98. But the DHS98 model would not immediately suggest an empirical test of 9

the form M91 originally proposed. This is because of the important role of private information received only by managers in driving PEAD. This suggests a role for another element underlying PEAD not accounted for by the revised M91 tests. This is portrayed in the typology of price responses to earnings information as in Table 1. The DHS98 framework implies a concentration of PEAD in Cell B of Table 1 , where confirmatory news about earnings is received privately. Conversely, we would expect to observe least PEAD in Cell C, from firms receiving to public negatory signals about the future path of earnings. DHS98 are far from the first authors to stress the importance of the distinction between public and private information signals for the utility of accounting information and engaging in arbitrage using it. Indeed, the social value of publicly available accounting information has a central role in discussions of the normative value of alternative accounting/disclosure systems (Strong and Walker (1989)). Another element of the informational environment within which a firm’s earnings are announced which may affect PEAD has already been examined empirically by Bhushan (1994). Bhushan (1994) relates the incidence and magnitude of PEAD to variables capturing the cost for investors’ of processing earnings information. PEAD is hypothesised by Bhushan (1994) to be clustered in low-priced, low volume stocks. In low-priced stocks bid– ask spreads will constitute a greater proportion of value and so constitute a greater dis-advantage to arbitrage strategies designed to exploit PEAD. Similarly, low volume stocks are subject to greater liquidity problems, causing prices to move against arbitrage traders as they implement counter-PEAD

10

trading strategies. The relationship between company earnings announcements and the bid– ask spread is now well documented for the NYSE (Krinksy and Lee (1996)) and the NASDAQ (Affleck-Graves, Callahan, and Chipalklatti (2002)) in the U.S. Both studies confirm that as the earnings announcement draws near market-makers extract greater compensation for the risk of dealing with an corporate insider, who may have hidden knowledge about the veracity and reliability of announced earnings information. Such a risk-premium, often termed the adverse selection component of the bid-ask spread, is likely to be largest in low-volume, illiquid stocks. An adverse selection risk-premium will be charged by a market-maker faced with the prospect of dealing with someone more able to interpret the earnings announcement than himself (Glosten and Milgrom (1985)). Bhushan (1994) suggests increases in the adverse selection component of the bid-ask spread around earnings announcements causes PEAD to be clustered in the right–hand-side (B-D cells) of Table 1. But Bhushan (1994) has no prediction regarding the magnitude of PEAD in cell B relative to Panel D. Hence the relative magnitude of PEAD in Panels B and D is central to the validity of insights offered by the DHS98 model, as against that suggested by Bhushan (1994). The DHS98 framework suggests ranking of PEAD of the form B > D > A or C, while Bhushan (1994) suggests A or C < B or D. 2.2.2

PEAD and the stock market boom

We examine the Greek market in a stock market boom which anticipated the benefits of Greek entry into the European Union. This reached a crescendo 11

in the late 1990’s with a rise in the Athens Stock Exchange index of 86% in 1998 and 102% in 1999. Since The new millennium has seen a catastrophic decline in the market of about 70%. This somewhat frothy market provides an interestiing test bed for behavioural explanations of PEAD. Strong stock market booms, such as that seen in Greece in the late 90’s, see investors accumulate capital gains that become the ”reference point” for evaluating future trades and their effect on wealth. This fact has given rise to a ”Prospect Theory” justification of stock market momentum, including PEAD (i.e. stock price momentum in response to earnings changes, or expected changes, Barberis, Huang, and Santos (2001)). The accumulation of past gains, due to prior market rises make investors more risk-averse with respect future price gambles. Conversely, the accumulation of losses, in periodic market dips, make investors risk-loving in their desire to retrieve recent loses relative to the ”reference point” of recent price highs. A central point of perspectives on momentum is the asymmetric response of investors to gains and losses (Grinblatt and Han (2005)). The greater sensitivity with respect to trading losses may make them especially reluctant to sell stocks experiencing bad news at their earnings announcement. In an theoretical antecedent of DHS98 framework, Hirshleifer, Subrahmanyam, and Titman (1994) outline a model in which a market-makers aggregate trades from “early” and “late” informed investors and liquidity traders. In their model the “early” informed seek to unwind their position in the asset once their informational advantage has been eroded by the “late” informed. In the Greek market such unwinding of the position taken by the “early” informed may be rendered more difficult in the face of very thick

12

bid–ask spreads.

2.3

Data

In the M91 study, Value Line forecasts for the U.S. were employed. But here consensus forecasts of earnings in the Greek economy are used, to allow us to focus on a context where we suspect failures to impound earnings information into prices may be rife. Our sample is constructed from a set of Greek firms in the years 199299 for which we have a complete set of IBES consensus earnings forecasts, matching actuals, stock prices, volume and bid-ask spread data. This does serve to introduce an element of survivorship bias into the results reported below (Brown, Goetzmann, and Ross (1995)). Using these criteria we constructed a unbalanced panel of 47 firms, drawn from 33 industries, which yield 223 company/years. The industrial composition of the sample is summarised in Table 2. Food and household, Construction and textiles are the major source of the sample. No single industry contributes more than 20% of the sample and so reflects a broad range of industries. Table 3 shows the decomposition of the sample by earnings news and liquidity. Summary statistics for both forecast errors and revisions and errors (scaled by price) are given in Table 4. Recall we define the forecast error to be the actual value minus the forecast. So the negative mean values reflect the presence of a few large optimistic errors, skewing the distribution leftwards. At the median the forecast error is pessimistic. This pattern of forecast errors seems compatible with the idea that analysts in the Greek market forecast unmanaged earnings, as originally suggested by Abarbanell and Lehavy (2003). 13

Under the “unmanaged” earnings expectation hypothesis” analyst’s correctly forecast “true” earnings, based on cash flows and company activities, but not manager’s manipulation of those earnings as reflected in reported earnings. A particular problem for analysts is the initiation of “earnings baths” by managers in response to their inability to meet perceived earnings targets. As previously noted, international comparison studies suggest aggressive earnings management is far more extreme in Greece than in the US or UK. In order to study PEAD we collected earnings-announcement dates from the Greek financial press. Over half the reported earnings announcement dates were found in the weekend’s newspapers, with the largest number of announcements being made on a Sunday. Since financial markets are closed each weekend in Athens some allocation of these announcements to trading days had to be made. The procedure followed here is to allocated Saturday’s and Sunday’s announcements to the following Monday and all public holidays’ trading to the following day. DellaVigna and Pollet (2005) in a study of PEAD in the U.S. market report bad news about earnings is often released on a Friday, or at the weekend. DellaVigna and Pollet (2005) report results that suggest that on average Friday earnings announcements result in trading volume 10% below the mean of that on a weekday post-earnings announcement day and an abnormal return which is 50 basis points below earnings announcements on normal trading weekdays. Earnings announcements on a Friday constitute just under 6% of the DellaVigna and Pollet (2005) sample drawn from the years 1984-2003 in the U.S. market. Hence we might expect the delay in stock market impounding of earnings information to be far greater in Greece where earnings are normally announced in the

14

Sunday newspapers. During our period trade only occurred on the Athens exchange between 10.30am and 1.30pm each day. So any information in a Sunday earnings announcement would have to wait until 10.30am on Monday to be impounded into market prices. The liquidity of a stock is judged by reference to the volume of trade in the stock and the bid–ask spread charged on trades in that stock. For a stock to be classified as liquid in this study we require it to have both above median levels of volume and below the median bid-ask spread. Since each firm has both its volume and spread calculated each month any stock can transfer from the liquid to the illiquid portfolio in any given month. This is particularly important given the turmoil existing in the Greek market during our sample period. As can be seen from Table 5 both variables display considerable evidence of skewness, with lots of stocks not trading in the majority of sample months. No doubt this is partially because of the very high bid-ask spreads required of investors. For the most illiquid stocks investors seem guaranteed to make a loss by trading the stock because the bid-ask spread exceeds price over the period. Such a profile is quite consistent with our belief that family controlled firms, with high benefits of private control, are a prominent part of the Greek market. Poor market liquidity, trading execution risk, and specifically the fear of informed traders, has recently been shown to be priced within the US market. This was true even on a sample specifically trimmed to exclude excessive bid-ask spreads (Sadka (2006)). This has been advanced as a possible partial explanation of stock market momentum and PEAD in the U.S. market. If illiquidity can cause such distortions in the U.S. it is likely to have far greater impact in the illiquid, less regulated, markets of

15

Greece. 2.3.1

Choosing Greece as a case study

Reviewing various aspects of earnings “opacity” for 34 countries in the years 1984-1998 Bhattacharya, Daouk, and Welker (2003) report that Greece ranks in the least transparent group on all three measures used by the authors, earnings aggressiveness, loss-avoidance and earnings smoothing. Greek companies appear to engage in some of the most extreme earnings manipulation practices in the world. Leuz, Nanda, and Wysocki (2003) argue that this world-beating level of earnings management results from high private benefits control acquired under a legal regime where investor protection is weak. Our discussions with market professionals in the Greek market confirm that minority shareholder rights are weak, especially in the face of an entrenched family interest. Often the lack of effective regulation to control insider trading appears to induce a slightly conspiratorial ambiance to the “Sophokleous Street” culture in Athens1 . While cultural aspects of economic life are often seen as irrelevant to asset pricing recent empirical work appears to cast doubt on this conclusion ( DeBondt (2005)). Prior research already suggests the Greek stock market may be inefficient in the weak-form sense Dokery and Kavassanos (1995). Coutts, Kaplandis, and Roberts (2000) confirm that some of the most gross stock market distortions of pricing in the Greek stock market occur around holidays. It appears there is an inexplicable rise in returns just prior to a holiday in the Greek market. Semi-strong form inefficiency in the Athens exchange is also 1

The term is Greek equivalent of Wall Street in the U.S.

16

suggested by Kyriazis and Diacogiannis (2004) who find low price/earnings ratios, high dividend yields, small size, low market to book ratios, high market leverage and low betas subsequently generate significantly higher returns in the Athenian market. Trading became so frenzied in the Greek market during the 90’s that even at the inception of the system of electronic trading on the Athens Stock Exchange in 1992 a 8% price limit for movements in the Athens Stock Exchange index was imposed. If prices on the exchange moved by more than 8% trading was suspended. Suspension of trade closed the market until the next day. From 1996 onwards a tighter 4% price limit was imposed upon more illiquid stocks. As we shall see later, there are many highly illiquid stocks on the Athens Exchange (Diacogiannis, Patsalis, Tsangarakis, and Tsiritakis (2005) and Phylaktis, Kavussanos, and Manalis (1999)). As the 8% price limit became more frequently violated, the limit was raised to 12% in early 2000. But this innovation comes just after the end of our sample period. As well as being a period of general economic tumult within the Greek economy the Greek accounting profession itself undergoing a period of transition in the early 90’s ( Caramanis (1997), Caramanis (1999) ). The newly elected ”New Democracy” government followed a liberal Thatcher-Reagan style economic policy, dismantling much of the state control which characterised the postwar Greek economy. One part of this was the removal of a local monopoly on state mandated audits held by the Greek professional body (often denoted by the acronym SOL). In September 1991 Article 75 of the Greek legal code allowed entry into the Greek auditing market by the then “Big 8” international audit firms. The often aggressive accounting choices

17

made by our sample firms needs to be interpreted against the backdrop of this newly competitive environment for those supervising the reporting of earnings to the stock market.

3

Results

We begin our formal tests by reporting the results of M91 tests as implemented upon the sample data. In a separate subsection we present results for our own set of suggested tests using the theoretical framework of DHS98.

3.1

Results for the M91 tests

Hypothesis H M1 states in null form that successive forecast errors made in predicting earnings are unrelated. This is a fairly basic requirement for the informational efficiency of analysts’ forecasts; requiring that they learn from past mistakes. The results of the regression based tests suggested in M91 is presented in Table 6. Strong positive serial correlation in forecast errors is clearly present, suggesting analysts repeat their mistakes. Because of this Greek investors are likely to greet with scepticism forecast revisions that fail to reflect past errors. So if, for example, the analyst has just made an optimistic error in predicting announced earnings and then revises his/her forecast upwards they risk their forecast will be discounted by the market. This reasoning underlies the next two of M91’s hypotheses that we also test below. The second of the M91 hypotheses checks if post forecast-revision price movements are affected by the conditioning of current forecast revisions on 18

past forecast errors. We present our results in Table 7. Unlike M91 we find no relation between the post-revision price response and either past forecast errors or their relation to current revisions. Indeed the coefficient on confirmatory and negatory revisions is equal to three decimal places. The third of the M91 hypotheses concerns whether forecasts revisions, following the earnings announcement, affect PEAD. Informational efficiency requires that the price response at the earnings announcement is purely a response to “news”, rather than previously released forecast revisions. We present the results in Table 8. The results strongly confirm the hypothesis that past forecast revisions in themselves cannot explain the stock market response to earnings announcements in the Athenian market. But, as we show below, the relation between prior forecast revisions and current forecast errors is important for explaining PEAD.

3.2

Tests suggested by the DHS98 framework

One of the most obvious implications of the “self-attribution” bias is that confidence should be increased by a confirmation of the decision maker’s ability, while a refutation produces no such reduction in confidence. Fortunately the I/B/E/S consensus forecast data affords us an opportunity to test this hypothesis directly. We can do this by tracking the dispersion of the consensus forecast in the post-earnings announcement period in Table 9 and Figure 2. Consistent with self-attribution bias firms for which prior forecast revisions and current forecast errors have the same sign display increased confidence (reduced dispersion in forecasts). However such initial overconfidence is soon eroded and both groups of firms have a very similar forecast 19

dispersion after four months. Consistent with the revised version of M91’s second hypothesis (in equation (5)) presented in equation (6) we find negatory earnings announcements are greeted by a substantially greater price-response in the medium term than their confirmatory counterparts. If analysts have already correctly predicted the direction of earnings change there is less “news” value in an earnings announcement. Figure 3 shows this clearly for the case of liquid stocks, where a successful arbitrage trading strategy is likely to be most feasible. Liquid stocks subject to negatory earnings announcements, especially those receiving good news, show both greater anticipation of earnings and greater PEAD. The concentration of PEAD in good news stocks makes sense, given the legal restrictions on short-selling stocks in the Greek market (although we believe these restrictions are widely disregarded). While in Figure 3 the clustering of PEAD in liquid stocks subject to good news about earnings refuting analysts pessimism seems obvious the results using cumulative abnormal returns over a 3-day window (see Table 10) reject that conclusion. Suggesting either a delayed market response to the nature of the earnings announcement, or poor benchmarking of expected returns. At this early stage of our research ruling out the latter explanation would be premature. Table 11 reports the results of estimating equation (5), for a 3-day window, on our data. The coefficient on negatory response coefficients to earnings is larger in statistical, but perhaps not economic, terms. Negatory revisions do carry a substantially stronger response coefficient, but this is not reflected in the strong statistical significance of the coefficient. Similarly,

20

while the coefficient on errors which confirm recent revisions is twice that on confirmatory errors both are small in absolute terms. So while we find qualitative evidence to suggest the dominance of negatory signals about earnings in driving PEAD, that evidence does not have the strong quantitative support.

4

Conclusion

This paper revisits the earlier empirical study of M91 and recasts it in the light of the theoretical motivation for PEAD provided by DHS98. Our results provide some support for the idea that refutations of earnings expectations are associated with greater PEAD than confirmations of prior expectations. This supports the predictions of the DHS98 framework. While confirmations of expectations make analysts/investors more confident in forecasting earnings, as DHS98 suggest, this does not seem to induce a bigger price response to earnings. The refutation of investors expectations makes them less assertive in correcting PEAD induced mispricing. Our results are clouded by the clear role of trading costs and liquidity in inducing PEAD in the Greek market. As might be expected few stocks on the Athens Exchange enjoy the liquidity familiar to traders in the US or UK markets. While PEAD exists in the 20 % of “liquid” stocks (defined as having above median volume and below median spread) it is small compared that found in the majority of stocks which are fairly costly to trade. This makes differentiating the stories told in this paper from earlier transaction costs based explanations difficult. Future research should focus upon this

21

issue.

References Abarbanell, Jeffery S., and Rueven Lehavy, 2003, Biased forecasts or biased earnings: The role of reported earnings in explaining apparent bias and over/underreaction in analysts’s earnings forecasts, Journal of Accounting and Economics 36, 105–146. Affleck-Graves, J., C. Callahan, and N Chipalklatti, 2002, Earnings predictability, information asymmetry and market liquidity, Journal of Accounting Research pp. 501–17. Barberis, Nicholas, Ming. Huang, and Tanos Santos, 2001, Prospect theory and asset prices, Quarterly Journal of Economics 116, 1–53. Barberis, Nicholas, Andrei Shleifer, and Robert Vishny, 1998, A model of investor sentiment, Journal of Financial Economics 49, 307–43. Bhattacharya, Utpal., Hazem Daouk, and Michael Welker, 2003, The world price of earnings opacity, Accounting Review 78. Bhushan, Ravi, 1994, An informational efficiency perspective on the post– earnings–announcement–drift, Journal of Accounting & Economics 18, 45– 65. Brown, Larry D., 1997, Earnings surprise research: Sythesis and pespectives, Financial Analysts Journal 53, 13–19.

22

Brown, Stephen J., William N. Goetzmann, and Stephen A. Ross, 1995, Survival, Journal of Finance 50, 853–873. Caramanis, C, 1997, The enigma of the greek the greek auditing profession: Some preliminary results concerning the impact of liberalisationon auditor behaviour, European Accounting Review 6, 85–106. , 1999, International accounting firms versus indigenous auditors: intra-professional conflict in the greek auditing profession 1990-93, Critical Perspectives on Accounting 10, 153–196. Coutts, Andrew., Chistos. Kaplandis, and Jennifer. Roberts, 2000, Security price anomalies in an emerging market: the case of the athens exchange, Applied Financial Economics. Daniel, Kent D., David Hirshleifer, and Avanidhar Subrahmanyam, 1998, Investor psychology and security market under- and over-reactions, Journal of Finance 53, 1839–1886. DeBondt, Werner, 2005, The value and beliefs of european investors, Driehaus Center of Behavioural Finance, DePaul University, Chicago. DellaVigna, Stephano, and Joshua Pollet, 2005, Investor attention, firm reaction and friday earnings announcement, NBER Working Paper 11683. Diacogiannis, George., Nikolaos. Patsalis, Nickolaos Tsangarakis, and Emmanuel Tsiritakis, 2005, Price limits and overreaction in the athens stock exchange, Applied Financial Economics 15, 53–61.

23

Dokery, E., and M. Kavassanos, 1995, Testing the efficient market hypothesis, in G. Karathanasis, and M. Glezakos, ed.: The Mechanism of prices in stock market: Empirical studies on the Athens Stock Exchange (Papasisi: Athens). Glosten, Lawrence R., and Paul R. Milgrom, 1985, Bid, ask and transaction prices in a specialist market with heterogeneously informed traders, Journal of Financial Economics 14, 71–100. Grinblatt, Mark, and Bing Han, 2005, Prospect theory. mental accounting and momentum, Journal of Financial Economics 78, 311–339. Hirshleifer, David, Avanidhar Subrahmanyam, and Sheridan Titman, 1994, Security analysis and trading patterns when some investors receive information before others, Journal of Finance 49, 1665–1698. Hong, Harrison, and Jeremy Stein, 1999, A unified theory of underreaction, momentum trading and overreaction in asset markets, Journal of Finance. Jackson, Andrew, and Timothy Johnson, 2006, Analysing underreaction anomalies, Journal of Business 7, 75–114. Kothari, S.P, 2001, Capital markets research in accounting, Journal of Accounting and Economics 31, 105–231. Krinksy, Itzhak, and Jason Lee, 1996, Earning announcements and the components of the bid-ask spread, Journal of Finance 51, 1523–35.

24

Kyriazis, Dimitris., and George. Diacogiannis, 2004, Testing the performance of value strategies in the athens stock exchange, SSRN Working Paper http://papers.ssrn.com/sol3/papers.cfm?abstracti d = 499662. Leuz, Chhristian., Dhananjay. Nanda, and Peter. Wysocki, 2003, Earnings management and investor protection: an international comparison, Journal of Financial Economics 69, 505–527. Mendenhall, Richard, 1991, Evidence of possible underweighting of earningsrelated information, Journal of Accounting Research 29, 170–180. Phylaktis, K., M. Kavussanos, and M. Manalis, 1999, Price limits and stock market volatility in the greek exchange, European Financial Management 5, 69–84. Sadka, Ronnie, 2006, Momentum and post-earnings-announcement drift anomalies: The role of liquidity risk, Journal of Financial Economics 80, 309–349. Strong, Norman., and Martin Walker, 1989, Information and Capital Markets (Basil Blackwell: Oxford and New York).

5

Tables

25

Nature of Signal Confirmation (ERR & REV same sign) Negation (ERR & REV of opposing sign)

Information Environment Public Private A B C

D

Table 1: Revising the M91 framework to allow for the importance of private signals about earnings

26

Industry # companies/years % of sample Food & household 74 17.70 Construction 60 14.35 Textiles 45 10.77 Metals Non-ferrous 29 6.94 Financial 23 5.50 Miscelaneous Matierials 21 5.02 Banking 19 4.55 Building 19 4.55 Merchandising 14 3.35 Metals 12 2.87 Electricals 8 1.91 Health Care 8 1.91 Industrial equipment 8 1.91 Electrical equipment 7 1.67 Insurance 7 1.67 Recreation 7 1.67 Multi-Product 6 1.44 Leisure & Tourism 6 1.44 Transport 6 1.44 Beverages 5 1.20 Broadcasting 5 1.20 Real Estate 5 1.20 Business publishing 4 0.96 Telecommunications 4 0.96 Appliances 3 0.72 Chemicals 3 0.72 Data processing 2 0.48 Forest Products 2 0.48 Automobiles 1 0.24 EAFE CND MULTI 1 0.24 EAFE CSV MULTI 1 0.24 Energy sources 1 0.24 Technical Multi Products 1 0.24 Utiilties Multi-product 1 0.24 Total # of companies/years 223 Table 2: Industrial breakdown of sample companies 27

Liquidity Good Bad Liquid 15 (12.4%) 14 (11.6%) Illiquid 54 (44.6%) 38 (31.4%) Table 3: Decomposition of sample by earnings news and liquidity NB Liquid stocks are defined as those with above the median volume of trade ¯ and below median bid-ask spreads as a proportion of total price.

28

F orecastError P rice F orecastrevision P rice

Mean -2.56 -1.81

Median SD 0.38 26.85 -0.46 4.31

Max 36.83 2.92

Min Skew Kurtosis N -294.85 -7.56 71.27 223 -34.41 -4.59 25.69 223

Table 4: Summary statistics for sample forecast errors used in test Mendenhall Propositions

29

Variable Mean Median SD Min Max N Volume 27435.23 9485.55 60668.96 0 655112.81 223 Spread 0.14 0.02 0.24 0 1.02 223 price Table 5: Summary statistics for liquidity data

30

Constant Lag error -0.7 0.55 (-0.52) (13.43)

R2 0.44

Hetero N 7.31* 223

ERR2,n = α0 + α1 ERR1,n + errorn Table 6: Test of Hypothesis H1M : Persistence of forecast errors

31

Constant Revision -0.004 (-1.52)

-0.001 (2)

Error Error R2 (Confirm) Negate 0.001 0.001 0.02 (0.22) (0.95)

Hetero 4.13

CARn (F2 −1, F2 +3) = γ0 +γ1 REVn +γ2s ERR1,n +γ2d (1−Sn )ERR1,n +errorn Table 7: Test of Hypothesis H2M : Post forecast revision drift

32

Constant Forecast Error Forecast Revision R2 -0.004 0.0005 -0.0004 0.03 (-0.77) (1.97) (-0.33)

Hetero 4.72*

CAR(A2 − 1, A) = β0 + β1 ERR1,n + β2 REVn + errorn Table 8: Test of Mendenhall Hypothesis 3: The information content of forecast revisions for predicting PEAD

33

Month 1 2 3 4 5 6 Confirmatory 1.01 1.00 0.99 1.00 1.00 1.01 Negatory 1.02 1.01 1.01 1.01 1.01 1.01 Table 9: Dispersion of earnings forecasts for next year’s earnings (normalised on the mean forecast) in the year following an earnings announcement for firms subject to confirmatory and negatory earnings announcements NB. A confirmatory earnings announcement is one where the consensus forecast revision over the last three months has the same sign as the forecast error. The forecast error is calculated as the difference between the outstanding consensus forecast in the month before the announcement and announced earnings. A negatory earnings announcement is one where the forecast revision and forecast error have opposing signs. Regime Mean SD Confirmatory-Liquid (A) 0.0055 0.036 Confirmatory-Illiquid(B) -0.0053 0.046 Negatory-Liquid (C) 0.0020 0.039 Negatory-Illiquid (D) -0.0006 0.070

Min Max N -0.072 0.124 27 -0.132 0.133 89 -0.176 0.319 59 -0.097 0.064 23

Table 10: Data descriptives for 3-day cumulative returns for typology of firms described by DHS98 model

34

Constant 0.0005 (-2.06)

Error Error Revision (Confirm) (negate) (Confirm) 0.002 0.005 -0.007 (1.72) (2.29) (-0.38)

Revision R2 (Negate) 0.12 10.09* (-2.03)

Hetero -0.012

Table 11: Test of Mendenhall Hypothesis 3 with confirmatory & negatory signals CARn (A2 − 1, A2 ) = γ0 + γ1s Sn REVn + γ1d (1 − Sn )REVn + γ2s Sn ERR1,n + γ2d (1 − Sn )ERR1,n + errorn (

Sn =

1, If REVn & ERR1,n are the same sign and, 0, If REVn & ERR1,n are of different sign

35