Environmental news and stock markets performance: The need for further evidence in developing countries

More and more firms tend nowadays to adopt environment-friendly attitudes. Their motivation originates in the extent to which revenue and costs are affected by consumers and investors´ valuation of pro-environment initiatives. There is a well-established literature capturing the impact on stock prices of environmental information releases through the event study methodology. The latter is based on information diffusion within environmental regulations or on media coverage of environmental news. In this manuscript, we focus specifically on Argentina. We find that positive environmental news have no impact, while negative news do have an effect on average rates of return a few days following its appearance. Our results are robust to different estimation periods and models, and are subject to parametric and also non parametric tests. However, we find in all cases that abnormal returns are of a much smaller magnitude than other studies for developing countries. We believe that is reasonable because there seem to be no strong reason why the level of abnormal returns (not its volatility) should be larger for environmental news in developing countries than in developed ones. JEL: Q58, G14 Keywords: developing countries, environmental management, environmental news, capital markets, event study

1

Environmental news and stock markets formance: The need for further evidence in developing countries I.

Introduction

The environmental economics literature recognizes three “waves” of environmental regulation: “Command and Control” (CAC) approach consisting mainly of establishing standards (i.e., on emissions or effluents on environmental quality or, indirectly, on inputs or goods), “Incentives-based” regulations (also called “market based instruments”) as the establishments of taxes, tradable permits or deposit-refund systems, and information-based regulations (see Tietenberg, 1998).1 The latter implies making public information regarding firms´ environmental behavior, relying on community and investors for the monitoring and enforcement of environment-friendly behavior. The cooperation of those two additional actors is not a minor point, especially for governments with weak regulatory power and scarce funding for control as those of developing countries. There are numerous examples in the world of explicit establishment of regulations based on information diffusion. The main ones are the Toxics Release Inventory (TRI) in the United States (that provides information to the public on releases of toxic substances to the environment) and the Canada´s National Release Inventory. In developing countries, there are two leading cases: the one of Indonesia with PROPER (Program for Pollution Control, Evaluation and Rating) and the one of Philippines with ECOWATCH (see World Bank, 2000).2 These last two programs consist in releasing not concrete information of firms´ emissions (as in the TRI), but rather a rating assigning different colors to firms according to their compliance with standards. Other (less direct, but with the same objective) policies of this kind are the “voluntary agreements” between firms and regulator and the creation of environmental quality awards given to the cleanest industries. The idea behind regulating via information dissemination (or voluntarily promoting a pro environment image) is that consumers would punish a polluter by buying less of its products as a way of sanctioning it and investors would reduce their interest in that firm (because they weigh future losses from expected regulatory penalties, liability settlements, and eventually cleaning up costs, as well as lower revenues from consumers´ behaviour). Both (pressure by consumers and by investors) could give polluting firms the incentives to improve their behavior (see Telle 2006 or Reinhardt, Stavins and Vietor 2008 for further discussion on firms´ incentives to be green). Argentina´s environmental regulation is mainly of the “first wave” type (i.e., CAC), and there is no explicit policy in the line of regulator releases of firms´ environmental behavior. Moreover, although it is known that data on firms´ environmental behavior is available (for example, for hazardous waste generation), authorities are generally reluctant to provide it. Environmental information is usually not supplied on government web sites. It can be requested specifically under the national law 25 831 of 2004. While local newspapers and television networks assign some place to environmental news, there is no measurement of the existence and the extent of the implicit effect. However, even in this context, Argentina´s firms do generally have some environmental directives, and, as shown by Chudnovsky, López and Freylejer (1997), the bigger and the more open to the rest of the world are the firms, they tend to undertake more environmentally-friendly actions. This finding could imply that the most important local firms respond to investors and consumers pressure and to

1

In fact, this “last wave” is also “market-based” since it allows informed citizens to decide by themselves which firm to buy from and invest in. 2 There are other experiences throughout the world. For example, the UK´s Pollutant Inventory, Australia´s National Pollutant Inventory, Mexico´s Registro de Emisiones y Transferencia de Contaminantes, Czech Republic´s Pollutant Release and Transfer Register, and more recent ones as the China´s Greenwatch Program. Several of these programs are undertaken under World Bank Projects (see www.worldbank.org/nipr).

2

foreign regulations (which in turn, at least in part, correspond to foreign consumers´ demands). At the international level, there is a well-established literature on capturing the impact of environmental information releases based on information regulation and on media coverage of general environmental news. A rather large number of empirical studies have attempted to identify the stock market reaction to news on environmental performance. The research based on stock markets data is more abundant than that based on sales or other less public data. Within that empirical literature, there are three well-known approaches: a) portfolio analyses, b) long-term studies using regression analysis and c) event (short-term) studies and (see Ambec and Lanoie 2008 for an in depth review). In the first type of studies, a comparison is made between the economic performance of portfolios consisting of companies with a higher environmental and portfolios of companies that do not pursue that kind of objective. In long-term studies, the relationship between certain characteristics of companies (including their environmental performance), and their economic performance is analyzed using regressions. Finally, short-term studies use the “event study” methodology.3 In this manuscript, we focus on the latter type of studies for being the only ones for which some reference for Argentina is available. In general, the event studies literature finds significant impacts of environmental news in stock prices (both though information releases by a regulatory program and through any environmental news in the media), though their magnitude depends on the type of news. These papers differ in the database they deal with, but also on some technical aspects. Those are: the model selected for the event study (i.e., constant mean model, market model, CAPM, etc.), the window size utilized, the type of events compared (i.e., lawsuits filing versus settlements, foreign versus local firms, etc.) and the test performed to assess the significance of the results (i.e, parametric or non parametric ones). With respect to the type of database used, there are two lines of studies. On the first hand, based on information regulation, Hamilton (1995) studies the impact due to news appearing in Nexis database and Wall Street journal based on media coverage of the Toxic Release Inventory information, Lanoie, Laplante and Roy (1998) examine the effects of announcements of the list of “complying” and “of concern” polluters in Canada, Gupta and Goldar (2005) analyze the impact of the announcement of the “green leaf rating” in India while Dasgupta et al (2006) study the impact of the disclosure of environmental violations (MVR Program) in Korea. On the other hand, based on general media coverage of environmental news, Muoghalu, Robinson and Glascock (1990) examine the capital market impacts of hazardous waste mismanagement lawsuits filing and settlements announced in the Wall Street Journal, Lanoie and Laplante (1994) assess the impact of different type of environmental news appearing in the Financial Post and Globe and Mail of Canada, and Klassen and McLaughlin (1996) report the stock prices effect of U.S. environmental media coverage on the Nexis database. Finally, within this latter line of research (i.e., that based on media coverage), a direct precedent of analyzing how environmental news impact on Argentinean firms stock prices is the paper by Dasgupta, Laplante and Mamingi (2001). That paper studies how environmental news alter asset returns of tradable firms in Argentina, Chile, Mexico, and Philippines from 1990 to 1994 and conclude that: “markets react positively (increase in firms´ market value) to the announcement of rewards and explicit recognition by the government of superior environmental performance”, and “capital markets react negatively (decrease in firms´ value) to citizens´ complaints targeted at specific firms.”. Dasgupta, Laplante and Mamingi (DLM), Gupta and Goldar (2005), Dasgupta et al (2006) are taken as the main references that 3

Another related line in the literature is that of management with, for example, Hendricks and Singhal (1996) that investigate the impact of quality awards for manufacturing firms in the US, Przasnyski and Tai (2002) that examine the impact on stocks of the Malcolm Baldrige National Quality Award in the U.S. and Beirão and Sarsfield Cabral (2002) which study the effect of ISO 9000 certification on Portuguese stock market. There are also some papers on the impact on stock prices of major environmental accidents (see citations in Hong and Hwang 2001).

3

releasing information on environmental behavior could also be effective in the developing world. Moreover, all three studies find significant impacts much larger than those in developed countries. In this manuscript, we select two relatively stable periods in the Argentinean economy (1995 to 2001 and 2003 to 2008), we take information based purely on local sources for asset (and market) returns, we use an internet systematic searcher for environmental news, we perform various sensitivity analysis which show that our results are robust to different estimation periods and models, and the significance of our results is tested using the usual parametric tests but also non-parametric ones (i.e., sign and Wilcoxon signed rank tests).4 This paper is organized as follows. In the next Section, we describe the methodology and the data we used to capture the impact of environmental news on the capital market. Our results are presented in Section III. And, Section IV summarizes our conclusions.

II.

Methodology and data

It is broadly acknowledged that firms´ stock prices reflect their future cash flow. The idea behind the methodology of event studies is that, given rationality in the marketplace, any event affecting a specific firm is reflected immediately on its asset price. There are numerous cases in which this method has been applied: to capture the effect of releasing macroeconomic information, to value mergers and acquisitions, etc. (see Campbell, Lo and Mac Kinlay 1997 for more references). While the use of this method began with a 1933 paper, the more well-known methodological works are those of Ball and Brown (1968) and Fama et al (1969). The steps to perform this type of studies are: 1) Define what is an event (and its “window”), 2) Select the model for estimation of “Normal” (or expected) returns, 3) Select the criteria to include firms. Search their “Actual” returns, and 4) Follow testing procedures on “Abnormal” returns (i.e., the difference between “Actual” and “Normal” returns). In this section, we will describe with some detail the steps we followed and the data we used in each stage.

II. 1.

Definition of event

It is important to differentiate the number of newsclips from the number of events since an “event” is an “environmental new” providing novel information. As it is well-known, Argentina´s economy is far from being stable. In what concerns the last two decades, relative “stability” followed the “convertibility of one peso = one dollar” period (on April 1st 1991, convertibility was established by law 23.928) and ended some time before that convertibility was abandoned by the devaluation on of the peso (on January 6th 2002 by law 25.561).5 We chose two periods for our analysis: 1995 to 2001 and 2003 to 2008, which means that we avoid dealing with data for the period that goes from the resignation of President De la Rúa in December to the beginning of Néstor Kirchner presidency in May 2003. Those were very unstable years, during which four transitory chiefs of state lead the country. We chose news appearing on the second newspaper in terms of daily circulation in Argentina: la Nación. In fact, for the period 1996-2000 la Nación had a daily circulation of approximately 160,000 from Monday to Saturday and 250,000 for Sundays (INDEC, 2001). Our data collection was made using a web search mechanism for older editions called “La Nación online” (www.lanacion.com.ar). We did not choose Clarín (the top newspaper in 4

Non parametric tests are not usual in the environmental literature, being the exception Klassen and McLaughlin (1996), who report results based on the Wilcoxon signed rank test. However, they are usual in finance. 5 It does not follow that capital markets were quiet during that period. For example, in 1998 the Argentine stock recorded a significant decline, a trend recorded since October 1997, at the time of the outbreak of the crisis in Asia. The Russian default in August 1998 gave it a further downward push, and although a recovery began subsequently, in January 1999 the devaluation in Brazil led to renewed volatility in the prices of Argentine stocks.

4

terms of circulation) because the Clarín web search is newer than that of La Nación (and so covers contain a shorter period), and because we believe environmental news appearing in those newspapers can not be so distinct among both newspapers.

II. 2.

Model selected for estimation

There are three main statistical models available for estimation in event-studies: the constant mean model, the market model and the factor model. The simplest one is the “Constant-Mean Return Model” (CMM), which relates linearly the return of any given security to a constant and a disturbance term. More precisely, expected returns are estimated from:

Rit = µ i + ζ it ,

(1)

with E (ζ it ) = 0 and Var (ζ it ) = σ ζ2i . Then, the so-called “Market Model” (MM), is potentially an improvement over the constant mean return model since it relates linearly the return of any given security to the return of the market portfolio. So, since it removes a portion of the return that is related to the market portfolio return, the variance of the abnormal return is reduced. More precisely, for any asset i,

Rit = α i + β i ⋅ Rmt + ε it

(2)

where Rit and Rmt are the period t returns on security i and the market portfolio, respectively, and εit is a disturbance term with mean and variance given by:

E [ε it ] = 0

Var [ε it ] = σ ε2i

Finally, another possibility is to add other factors to the right hand-side of equation (2) beyond the market return. For example, use as independent variables the rate of return on a risk-free asset in addition to that of the market in order to estimate the Capital Asset Pricing Model (CAPM), or add other factor that may determine asset returns. However, as stated in Campbell, Lo and Mac Kinlay (1997), there are limited gains in practice of using multifactor models in event studies. The related literature generally employs the constant mean and market models (when data on market returns is available), except for example, for Lanoie and Laplante (1994) and Lanoie, Laplante and Roy (1998) who utilize the CAPM model. Hence, we run here estimations based on the CMM and the MM. Coming back to the methodology, using any of the described models, it is possible to estimate, over the period previous to the event window, the expected return for each event window for each firm. The commonly used estimation method is Ordinary Least Squares. The estimation period includes generally between 120 and 210 trading days (Campbell, Lo and MacKinlay, 1997). Here, we report results for an “intermediate” estimation window of 165 working days (and we run sensitivity analysis for 120 and 210 working days). Hence, having estimated the expected returns, it is straightforward to predict a “Normal” return, during the days covered by the event window. This period was defined as 5 days before the announcement day (day 0) to 5 days after the announcement day, allowing possible information leakage as well as delays in processing it. Graphically, the time line can be represented as in Figure 1:

5

Figure 1. Event window

ARi ,−5

ARi , 0

…..

pre-event -5

…..

0

window

ARi ,5

+5

post-event window

(120 working days, 165 working days, 210 working days)

Event window (11 working days)

The difference between the “Actual” and the “Normal” return during the event window is the so- called “Abnormal” return and can be depicted by the following equations ((1´) and (2´), for the CMM and MM respectively):

= ζˆit

(1´)

AR = Rit − αˆ i + βˆ i ⋅ Rmt = εˆit {it { 14 4244 3

(2´)

ARit {

= Rit − {

" Abnormal"

" Actual"

"Normal"= µi

[

]

" Abnormal "

" Actual "

Rit {

" Normal "

The idea of the MM is that while stock returns tend to move with the market, unexpected firm-specific events also affect the returns. Hence, the market valuation and significance of an event can be estimated by measuring any abnormal change in the stock return. The same is true from the CMM, but expected returns are not related with the market return but rather with the mean of the return in the period previous to the event. Those “Abnormal” returns (ARit) are calculated for each event (for each firm) at each point in time within the event window.6 However, in order to derive conclusions about the effect of events in capital markets in a broader sense, it is important to analyze three extra concepts: “Average Abnormal” returns (AARt, across events for the same moment in time), “Cumulated Abnormal” returns (CARi, along time within the event window of a single event), and “Average Cumulative Abnormal” returns (CAAR, along time and across events). Their respective formulas are:

AARt =

1 N ⋅ ∑ ARit N i =1

(3)

where N is the number of events of similar nature in a full set of event. Similarly, for CAR and CAAR: t

CARi = ∑ ARit

(4)

t =t

6

We leave the discussion of their probability distribution for Subsection II. 4.

6

where t and t are the lower and upper limits within the event window (i.e., they can be the limits of the window themselves or some other days within the window), and,

CAAR =

II. 3.

1 N ⋅ ∑ CARi N i =1

(5)

Criteria to define events (firms) and their returns

Coming back to environmental news and the underlying events, it remains to be checked which of the events found in the La Nación database are related to firms trading their assets in the market during the time around the event (those firms have to be included in the index chosen as the reference for the market, since we perform estimations based on the CMM but also on the MM). And, there should be also sufficient data available (pre event) for the estimation of “Normal” returns ( Rˆ it ). In this case, market returns ( Rmt ) are based on the MERVAL (Mercado de Valores) index. There are five market indices in the Buenos Aires stock market (MERVAL and MERVAL 25, Burcap, Indice Bolsa de Comercio de Buenos Aires and the new MERVAL Argentina –M.AR-). The M.AR. and MERVAL 25 are discarded because they began in year 2000 and 2003 respectively, while the Indice Bolsa de Comercio de Buenos Aires is a too broad index since it represents the evolution of all traded stocks. MERVAL weighting are based on traded volume and its base is 1986, while BURCAP weightings are based on market capitalization. We chose the MERVAL for being the most publicly known (and older) index for the market.7 The stocks included in MERVAL are up to 80% of the participants in the market and the weights are updated every 3 months. Finally, we used the software EconomaticaTM to search stock prices in order to construct “Actual” (daily) returns (Rit) of the firms selected as Ri ,t = Pi ,t − Pi ,t −1 ⋅ 100 Pi ,t −1 ,

(

)

where Pi ,t is the i-stock closing price on trading day t and Pi ,t −1 is the stock closing price on trading day t-1, all adjusted for capital changes as stock splits and stock dividends.8

II. 4.

Testing procedures We test the significance of our results using both parametric and no parametric tests.

II. 4. 1.

Parametric tests

We begin by testing abnormal returns significance for each event at each moment in time within the event window. To do so, according to the fact that abnormal returns are the disturbance term of the CMM and MM we rely on (1´) and (2´). Hence, under the null hypothesis, abnormal returns will be distributed for MM (and similarly for CMM) as

N (0, σ 2 ( ARit )) , where (see Campbell, Lo and MacKinlay 1997 for a complete derivation) 7

However, we have also estimated equation (2) based on BURCAP and we find similar results. In particular, those events for which βs are not significant under the MM using MERVAL are the same that are not relevant using BURCAP. 8 The EconomaticaTM database covers over 5,000 companies in Latin America (countries included are: Argentina, Brazil, Chile, Colombia, Mexico, Peru and Venezuela). The data starts as early as 1986, depending on the country. It includes quarterly company balance sheets and daily market data (stock prices, ADRs, indexes, currency exchange rates, inflation rates, net asset value per share for mutual funds, etc), financial and trading ratios.

7

σ 2 ( ARit ) = σ ε2 + i

2 1  (Rmt − Rm )  ⋅ 1 +  L  σ m2 

(6)

This variance has an additional component (in brackets) that is the sampling error in the estimation of the two parameters based on regression (2). But, if L (the estimation period) is large (here, at least 120 trading days), then, the second term vanishes. Having calculated cumulative abnormal returns as in (4), we test for their significance using the relatively easy to derive fact that CAR are distributed N (0, σ 2 (CARi )) , where

σ 2 (CARi ) = (τ − τ + 1) ⋅ σ ε2

(7)

i

Note that the parenthesis indicates the number of days returns are accumulated. We also calculate average abnormal returns, and test them to infer significance of abnormal returns of similar events. To do so, it is also straightforward (from (3) and (6) and for L large) that AAR are distributed N (0, σ 2 ( AARt )) , where

σ 2 ( AARt ) =

1 N 2 ⋅ ∑σ ε N 2 i =1 i

(8)

Finally, based on (5), tests on the cumulative average abnormal returns can be derived from the fact that CAAR are distributed N (0, σ 2 (CARi )) , where

σ 2 (CAAR ) =

1

N

⋅ (τ − τ + 1) ⋅ σ ε2i 2 ∑ 144244 3 N i =1

(9)

σ 2 (CARi )

However, parametric tests embody detailed assumptions about the probable distribution of returns (i.e., that they follow a normal distribution). Hence, to check for robustness of our conclusions, we also run non parametric tests.

II. 4. 2.

Non parametric tests

We perform the two main non-parametric tests used in the finance literature (the sign test and the Wilcoxon signed rank test). The former judges the proportion of positive and negative abnormal returns against and assumes fifty percent split under the null hypothesis of no reaction to the event. The latter is a more complete non-parametric test that considers both, the sign and magnitude of the abnormal performance.

The Sign Test One of the simplest non-parametric tests for examining the hypothesis of no abnormal return is the sign test. The basis of this test is that under the null hypothesis of no market reaction to events, it is equally probable that the AR' s at time t will be positive or negative. Hence, the expected proportion of positive (negative) abnormal residuals equals ½. For example, for any given subset of positive (negative) events, we wish to test the null hypothesis that there is no positive (negative) abnormal performance at time t against the alternative hypothesis that there is a positive (negative) abnormal return. These hypotheses can be expressed in the following form: For positive (negative) events

8

H 0 : p ≤ 1 / 2  H 1 : p > 1/ 2 where p denote the probability of positive (negative) abnormal return on day t . Assuming that abnormal returns are cross sectional independent, the number of cases where ARit > 0 ( ARit < 0 ) will have a binomial distribution with parameters N and p .9 Let the test statistic S N be the number of successes in the N independent observations. The null distribution of S N is therefore the binomial distribution corresponding to N observations with success probability ½. So, given the binomial distribution:

 N  N* N −N*  ⋅ p ⋅ q P( S N = N *) =  for N*=0,1,….,N  N *

(10)

where p is the probability of success and q = 1 − p The null distribution is given by:

N  1  ⋅ N P ( S N = N *) =   N * 2

for N* = 0,1,….,N

(11)

where N* is the number of positive (negative) ARit in a day within the event window according to the side of the test. So, in the sign test, the number of cases for which ARit is positive (negative) is counted and the decision to reject or not H 0 is then based solely on the number of positive (negative) abnormal returns. It is interesting to mention here that, for large N, the distribution of S N is approximately

normal.

With p = 1 / 2 ,

the

mean

and

the

variance

are

given

by: E ( S N ) = N ⋅ p = 1 / 2 ⋅ N and σ ( S N ) = N ⋅ p ⋅ q = 1 / 4 ⋅ N . But, as we show below, since in our sample N is small, we calculate the exact distribution of the statistic. 2

Wilcoxon signed rank test The idea behind the sign test is that under the null hypothesis, the proportion above and below the true median should be ½. In fact, the null hypothesis of no abnormal performance at a given time t is equivalent to the statement that Pr( ARit > 0) = Pr( ARit < 0) = 1 / 2 . This statement is equivalent to the statement that the median of the abnormal returns ( θ ) equals zero at each instant t . Similarly, the alternative hypothesis H 1 that there is positive (negative) abnormal performance is equivalent to the

statement that the median of ARit , θ > 0 (θ < 0) . Thus, these hypotheses can be expressed as:

9

A binomial experiment has the following characteristics: is composed of N identical trials; there is only two possible results for each trial -success or failure-; the probability of having success is constant in each trial and equals p. The probability of a failure equals (1-p) =q; trials are independent; and the random variable under study is the number of observed success in the N trials.

9

For positive events

For negative events

H 0 : θ = 0  H 1 : θ > 0

H 0 : θ = 0  H 1 : θ < 0

However, if the magnitude of the ARit can be measured, it is useful to apply a test procedure which not only considers the sign but also recognizes the fact that a large ARit is more important than a small one. We shall now describe a procedure that considers both the sign and the magnitude of the abnormal performance, known as the Wilcoxon signed rank test. This test begins by transforming each ARit into its absolute value, then these values are ranked from lowest to highest10 and finally the positive or negative sign that was removed in the first step is reattached to each rank. The main effect of replacing the original measures with ranks is that it brings us to focus only on the “ordinal” relationships among measures. The Wilcoxon signed rank test is based on the idea that under the null hypothesis, the sum of the ranks above and below the median should be similar. The statistic is defined as: N

T * = ∑ ri*

(12)

i =1

where, according to the nature of the events under observation, ri* is the positive (negative) rank of the absolute value of abnormal returns on a day of the event window. Now we consider the null distribution of the statistic T * . Let # (v;N) denote the number of sign combination of the ranks 1,…,N for which the sum of positive (negative) signed ranks is equal to v. Since under H 0 the probability of any particular sign combination is

1 , it 2N

follows that:

P (T * = v) =

# (v; N ) . 2N

(13)

It can be proved that the significance probability is higher when using the sign test than when using the Wilcoxon signed rank test. Since our rejection region is on the right side of the null distribution, the sign test will reject the null hypothesis more often than the Wilcoxon test. Finally, for large N we can use the normal approximation, where for T * = T + ( T + is the sum of positive ranks) the mean and the variance under the null hypothesis are

10

It is assumed that no two of these absolute values are equal, and that each is different from 0.

10

E (T + ) = N ⋅ ( N + 1) / 4 and σ 2 (T + ) = N ⋅ ( N + 1) ⋅ (2 N + 1) / 24 .11 As stated above, the number of observations is relatively small, so, we use a table of the null distribution of the statistic (Lehmann, 1998).

III.

Results

In this particular case, as can been seen in Table 1, for the first period (1995-2001), we have 61 environmental news by publicly traded companies. Table 1.

Environmental Newsclips in Argentina: 1995-2001

Name of firm

Sector of activity

ACINDAR ALUAR ASTRA ATANOR BAESA CELULOSA INDUPA PEREZ COMPANC SIDERAR SIDERCA TELECOM TELEFÓNICA YPF Total

Metal Metal Oil Chemical Food Pulp and paper Chemical Oil Metal Metal Communication Communication Oil

Nature and number of newsclips Positive

Negative 4 1 0 1 1 0 0 2 2 1 1 1 3 17

0 0 1 5 0 1 5 6 0 0 0 0 26 44

Source: own elaboration based on La Nación online.

Some of the positive events (for example, Baesa 07/25/1999) were discarded for the analysis because there were no traded prices around the event window. Similarly, environmental news that are mere follow-ups or repetitions of previous news cannot be selected as “events”. Hence, after “cleaning up” the dataset for these two types of issues, there were 15 and 10 positive events and 17 and 6 negative events for the first and second period respectively. The final lists of events are detailed in Table 2 and Table 3.12

11

Defining the sum of the N unsigned ranks as

T=

N ( N + 1) 2

assigned a positive or negative sign is ½ we have the following:

and taking into account that the probability of being

E (T + ) =

1 N 1  N ( N + 1)  N ( N + 1) ri = ⋅  ∑  = 2{ 2  2 4 i =1 { p

;

T

1 1 1  N ( N + 1)(2 N + 1)  N ( N + 1)(2 N + 1) 2 + and σ (T ) = ⋅ ⋅ ∑ ri 2 = . .  = 2 { 2 1 4 6 24  i =1 { 23 p (1− p ) N

σ 2 (T )

12

When events appear in newspapers on days where the stock market is close (i.e., Saturdays, Sundays, or public holidays), the immediate following day of trading is used as day 0.

11

Table 2.

Description of events: 1995-2001

(a) Positive events Name of firm ACINDAR

ALUAR ATANOR PEREZ C. SIDERAR SIDERCA TELECOM TELEFÓNICA YPF

Date

Nature of event

12/27/98 6/5/99 11/18/01 9/23/98 9/16/97 12/9/98 9/6/99 10/22/99 3/13/99 5/23/00 4/12/97 4/12/97 4/4/98 6/5/99 12/10/99 9

Company ISO for environmental performance Company ISO for environmental performance Company reward for environmental performance by US representative in Argentina Investment in plant expansion Shutdown removal by government Court action against Greenpeace for complaint Company ISO for environmental performance Company reward for environmental performance by an environmental group Agreement with Loma Negra to produce ecology cement Announcement: agreement to construct the first waste hard metal treatment platform Investment in technology to preserve environment Agreement with Aguas Argentinas, Edenor y Edesur on environmental protection work Investment in construction of two sulphur treatment plants Company ISO for environmental performance Waste water treatment plant inauguration (

Date 3/25/00 8/14/97 12/9/97 12/17/98 8/25/00 12/14/00 3/9/97 6/25/98 10/17/96 3/9/97 5/25/97 12/13/97 8/12/98 9/14/99 12/19/00 3/5/01 6/25/01

Nature of event Dock Sud plant ordered to shut down Temporary shutdown Greenpeace complaint Greenpeace complaint in Bahía Blanca Shutdown: chlorine escape Citizens complaint about air pollution in Ingeniero White Accidental oil spill in Neuquén Mapuches complaint against Mega project Accident: hydrocarbon spill Accidental oil spill in Neuquén Court action against the company Accident: oil tank got burnt Mapuches complain against the construction of gas pipes Partial shutdown: toxic emissions suspicion Greenpeace complaint for canals contamination Mapuches complaint against environmental policies Government intimation

b) Negative events Name of firm ASTRA ATANOR CELULOSA INDUPA

PEREZ C. YPF

Source: Own elaboration based on La Nación online.

For the second period, 2003-2008, we find 10 positive and 6 negative relevant events (see Table 3). Table 3.

Description of events: 2003-2008 Name of Firm

(a) Positive events Acindar Aluar Repsol YPF

Banco Frances Ledesma Tenaris Transportadora de Gas del Sur (b) Negative events Ledesma Molinos Rio del Plata Repsol YPF Indupa Solvay

Date

Sector

Nature of Event

04/01/2004 Metal 21/02/2004 Metal 19/12/2003 Metal 27/04/2005 Oil 17/12/2003 Oil 27/08/2008 Oil 09/04/2003 Banking 17/12/2003 Paper 17/12/2003 Metal 17/12/2003 Gas

Environmental Prize Environmental Innovation in product Environmental Prize Product development with environmental technology Agreement with ONU for better environmental policy Investment in green efficient technology First in ranking of social responsibility of environment Agreement with ONU for better environmental policy Agreement with ONU for better environmental policy Agreement with ONU for better environmental policy

03/07/2003 Paper 21/06/2006 Food 05/12/2005 Oil 27/09/2006 Chemical 17/04/2003 Chemical 21/06/2006 Chemical

Citizens complaint because of deforestation Court action against company Government action because of oil spill Court investigation against company Research showing contamination by company Court action against company

12

On one side, we find that positive environmental events appearing in the media are related to two kind of news: announcements (or inaugurations) of investments and ISO norms approval and other “voluntary rewards”. And, on the other side, we find that negative events are mainly associated with two types of news: environmental accidents, and court/government rulings or citizens/ONG complaints. Another characteristic of our dataset is that the proportion of negative versus positive events is almost 50% (23 and 25 events respectively). In terms of the estimation of the “Normal” returns, as is shown in Appendix A, the β coefficients (in the case of the market model) have the expected positive sign in all cases and are significant except for a few events, which are then omitted from the analysis.13 The resulting “Abnormal” returns (AR) are in all cases of the expected sign. Using the mean model and a 165 days estimation period, we show in Appendix B that 5 of the 15 positive events (i.e., 33%) and for 12 of 17 negative events (i.e., 70%) turned out to be significant in the 1995-2001 period, while 7 out of 10 positive news (i.e., 70%) and 3 of 6 negative news (i.e., 50%) were significant from 2003 to 2008. All significant ARs have the expected signs (a positive sign for positive news and a negative sign for negative ones). Similar percentage of significant events occurs with the alternative market model and alternative estimation periods. With respect to the estimated CAR (from day –5 to each relevant day in the window), they are significant (and are of the expected sign) in 2 of the 15 positive events, and in 6 of the 17 negative events for the first period and in 3 over 10 and 1 over 6 events in the second period of our dataset (CAR results are also reported in Appendix B). As the literature in the field, we follow our analysis to a comparison between events to better interpret the reason and average magnitude of the impacts. First of all, we aggregate positive and negative events for each of the periods and use parametric tests. We find that positive news have no significant effect, while negative events have an impact on average stock returns on days +1 and +3 (see Table 4 for CMM and n=165). The same pattern is maintained for the alternative MM and estimation periods (see in that respect Table 5), with day +3 being consistently significant for negative news while positive news have no significance. In addition, the consistency of results extends to the magnitude of AAR, which is in all cases around a 1 % impact. Note that CAAR are never significant. Table 4.

AAR Z-stat

CAAR Z-stat

AAR Z-stat

Positive versus negative events: parametric tests 1995-2001

Day -5

Day -4

Day -3

Day -2

Day -1

Day 0

Day 1

Day 2

0,9470

-0,1117

0,2918

-0,2527

Positive events -0,3075 -0,3375 -0,0289 -0,1031 -0,1237 -0,0427 -1,2044

0,8535

-0,1006

0,2630

-0,2277

-0,2771

-0,3041

-0,0260

-0,0929

0,8794

0,7757

1,0466

0,8120

0,5265

0,1890

0,1622

0,0665

0,8190

0,5109

0,5628

0,3781

0,2193

0,0719

0,0571

0,0219

-0,4385

0,4822

-0,1750

0,8950

-0,7959

0,8751

-0,3176

1,6244

Z-stat

-0,1115

0,0002

0,1054

-1,9194

-0,2427

-0,4127

0,0127

-0,1519

0,7430

0,7431

-0,7567

0,0165

-0,1608

0,6811

0,6093

Day 4

-0,0384

Day 5

-1,0854

-0,0572 -0,0999 -1,3043 -0,0178

Negative events 0,0001 0,0581 -1,0575 -0,1337 -1,2372 **

CAAR

Day 3

-2,2456

-0,0294

-0,3663

0,4943 -0,5541 0,8971

-1,0057

***

0,8012 -0,1941 -0,3121 -1,4765 -1,0113 -1,5329 0,5997

-0,1345

-0,2023

-0,9024

-0,5863

-0,8474

Note: Results are for CMM and n= 165. ***, ** and * denote significance at 1%, 5% and 10% respectively (one-tail test).

13

No significant coefficients appear in the last YPF negative news, varying according to the length of the estimation period. Hence, for n = 120, we discarded the last 3 YPF news, for n= 165 we rejected YPF 03/05/01, and for n = 210 we took off the sample YPF 03/05/01 and 06/25/01.

13

When looking at AAR but testing through non parametric tests, we find for positive events that, for all the alternative model and estimation periods, and with the sole exception of day –5 (the lower border of our event window), our results are robust (see Table 6). Regarding negative events, the significance of day + 3 is confirmed in all cases, while it holds in general for day + 1 (see Table 7). Note we can say Wilcoxon is the more “complete” of the two non parametric tests.14 Table 5. Positive versus negative events with alternative models and estimation periods: parametric tests 1995-2001 Model*

Estimation period

Mean

120 days

AAR Z-stat

210 days

AAR Z-stat

Market

120 days

AAR Z-stat

165 days

AAR Z-stat

210 days

AAR Z-stat

Mean

120 days

AAR Z-stat

Day -5

Day -4

Day -3

Day -2

Day -1

0.9399

-0.0994

0.3040

-0.2404

Positive events -0.2953 -0.3360 -0.0167 -0.0908 -0.1223 -0.0412 -1.2029

0.7895

-0.0835

0.2554

-0.2020

0.9782

-0.0941

0.3094

-0.2351

0.9143

-0.0880

0.2891

-0.2197

-0.2710

0.8869

0.1222

0.0291

-0.4434

0.1986

0.9228

0.1272

0.0303

-0.4613

0.2066

0.8457

0.1324

0.0380

-0.3986

0.2125

0.9917

0.1553

0.0445

-0.4675

0.2492

0.8358

0.1001

0.0346

-0.3082

0.1965

1.1113

0.1331

0.0459

-0.4097

0.2613

-0.4289

0.4974

-0.1657

0.8994

-0.7983

0.9259

-0.3084

1.6741

-0.2480

Day 0

-0.2822

Day 1

-0.0140

Day 2

-0.0763

Day 3

-0.1027

Day 4

-0.0346

Day 5

-1.0105

-0.2899 -0.3055 -0.0113 -0.0855 -0.0918 -0.0107 -1.1724 -0.2856

-0.0106

-0.0799

-1.0209 0.0621

0.0913

-1.0622

0.0646

0.0950

-0.9991 -0.0330 0.0950 -1.1716

-0.0387

0.1114

-0.9078 -0.1116 0.0323 -1.2070

-0.1484

0.0430

-0.0858

0.1163

-1.9505

-0.2396

-1.0959

-0.3857 -0.2510 -1.1224 -0.4013

-0.2611

-1.1678

-0.3369 -0.2456 -1.0718 -0.3951

-0.2880

-1.2569

-0.3338 -0.2782 -1.0721 -0.4438

Negative events 0.0045 0.0625 -1.0479 -0.1287 -1.2375 0.0084

-0.0100

-2.3036

*** 0.0509 -1.0574 -0.1442 -1.2549

-0.3699

-1.4254

0.5039 -0.5445 0.9380

-1.0135

**

210 days

AAR Z-stat

-0.4384

0.4783

-0.1864

0.8878

-0.0071

-0.8140

0.8882

-0.3461

1.6486

-0.0132

0.0945

-1.9635

-0.2678

-2.3302

*** -0.1209 -0.5611 -0.1330 -1.1845

0.4944 -0.5540 0.9181

-1.0287

**

Market

120 days

AAR Z-stat

165 days

AAR Z-stat

210 days

AAR Z-stat

-0.3504

0.7893

0.5143

0.8842

0.5795

-0.7089

1.5970

1.0406

1.7889

1.1725

-0.3544

0.7053

0.4465

0.6492

0.4356

-0.7600

1.5122

0.9573

1.3920

0.9340

-0.4040

0.7320

0.4490

0.8067

0.4633

-0.9161

1.6599

1.0182

1.8293

1.0505

-0.2447

-1.1353

-0.2692

-2.3966

*** -0.1076 -0.5769 -0.1381 -1.1656 -0.2307

-1.2369

-0.2961

-2.4992

*** -0.1040 -0.4806 -0.0508 -1.0503 -0.2359

-1.0898

-0.1151

-2.3817

***

0.5514 -0.6176 1.1156

-1.2496

0.2222 -0.5330 0.4765

-1.1428

0.4451 -0.6174 1.0094

-1.3999

*

Note: ***, ** and * denote significance at 1%, 5% and 10% respectively (one-tail test).

14

To complete the analysis and to check if the impacts found on returns were due to environmental news, we also looked for other newsclips that appeared in newspaper within the event windows. We found that Pérez Companc and YPF were the companies with more news (positive and mainly negative). But, not all this information was relevant. However, regarding the window around the positive events of YPF, we found one relevant newsclip (about the decision to install a plant of methanol in Neuquén) a day after the event day (that is, on day 12/10/1999). Then, the positive impact found could be an anticipation of this event and not because the environmental new. On the other hand, when we focused on negative event windows, 60% of the news found were related to the firm YPF. So, because average impacts could have been caused by those news, we analyzed abnormal returns excluding the events related to that company. We found that positive events results do not change (no day within the event window is significant) and for negative events, day +1 is still significant but day +3 turned out to be not statistically significant.

14

Table 6. Estimation period

Results of non parametric tests for positive events 1995-2001 Model and Method

Day -5

Day -4

Day -3

Day -2

Day -1

Day 0

Day 1

Day 2

Day 3

Day 4

Day 5

14

14

14

14

14

-0.0994

0.3040

-0.2404

14

15

15

15

(0.7895) (-0.0835) (10)** (6) (71) (52)

(0.2554) (11)*** (66)

(-0.2012) (-0.2480) (-0.2822) (-0.0140) (-0.0763) (-0.1027) (-0.0346) (-1.0105) (10)** (6) (7) (7) (4) (8) (6) (7) (71) (38) (44) (43) (36) (54) (42) (42)

Positive events N

120 days

Mean

AAR 0.9399

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

Market

AAR 0.8869

Mean

0.1222

0.0291

-0.4434

0.1986

(0.1272) (10)** (70)

(0.0303) (6) (58)

(-0.4613) (6) (42)

(0.2066) (-1.0622) (0.0646) (0.0950) (-0.4013) (-0.2611) (-1.1678) (6) (4) (7) (8) (9) (6) (5) (51) (23) (48) (52) (64) (43) (29)

AAR 0.9470

-0.1117

0.2918

-0.2527

-0.3075 -0.3375 -0.0289 -0.1031 -0.1237 -0.0427 -1.2044

(0.8535) (-0.1006) (10)** (6) (73) (49)

(0.2629) (10)** (65)

(-0.2277) (-0.2771) (-0.3041) (-0.0260) (-0.0929) (-0.1115) (-0.0384) (-1.0854) (10)** (5) (5) (5) (5) (7) (6) (5) (71) (37) (42) (38) (38) (53) (45) (39)

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

Market

AAR 0.8457

Mean

0.0913

0.1324

0.0380

-0.3986

0.2125

(0.0445) (7) (59)

(-0.4675) (6) (41)

(0.2491) (-1.1716) (-0.0387) (0.1114) (-0.3951) (7) (4) (7) (8) (9) (54) (20) (47) (52) (67)

AAR 0.9782

-0.0941

0.3094

-0.2351

-0.2899 -0.3055 -0.0113 -0.0855 -0.0918 -0.2782 -1.0721

(0.9143) (-0.0880) (10)** (6) (73) (52)

(0.2891) (11)*** (66)

(-0.2197) (-0.2710) (-0.2856) (-0.0106) (-0.0799) (-0.0858) (-0.3699) (-1.4254) (10)** (5) (6) (6) (5) (7) (5) (5) (72) (37) (43) (43) (38) (57) (45) (29)

AAR 0.8358

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

(1.1113) (10)** (80)**

-0.9991 -0.0330 0.0950

-0.3857 -0.2510 -1.1224

(0.1553) (10)** (70)

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

Market

-1.0209 0.0621

(0.9917) (9)* (81)**

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

210 days

13

(0.9228) (10)** (83)**

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

165 days

14

-0.2953 -0.3360 -0.0167 -0.0908 -0.1223 -0.0412 -1.2029

-0.9078 -0.1116 0.0323

-0.3369 -0.0107 -1.1724 (-0.010) (-1.0959) (6) (5) (48) (39)

0.1001

0.0346

-0.3082

0.1965

(0.1331) (10)** (70)

(0.0459) (7) (57)

(-0.4097) (6) (43)

(0.2613) (-1.2070) (-0.1484) (0.0430) (-0.4438) (-0.3699) -1.4254 (8) (4) (6) (8) (8) (5) (5) (60) (21) (43) (51) (63) (45) (29)

-0.3338 -0.2782 -1.0721

Note: ***, ** and * denote significance at 1%, 5% and 10% respectively (one-tail test).

Table 7. Estimation period

Results of non parametric tests for negative events 1995-2001 Model and Method

Day -5

Day -4

Day -3

Day -2

Day -1

Day 0

Day 1

Day 2

Day 3

Day 4

Day 5

16

15

16

17

17

17

16

15

16

0.4974

-0.1657

0.8994

0.0045

0.0625

-1.0479

-0.1287

-1.2375

16

16

(-0.7983) (0.9259) (10) (6) (85) (50)

(-0.3084) (10) (75)

(1.6741) (7) (58)

(0.0084) (9) (90)

(0.1163) (-1.9505)** (-0.2396) (-2.3036)*** (0.9380) (-1.0135) (8) (12)** (7) (11)** (6) (7) (76) (114)*** (65) (116)*** (51) (75)

Negative events N

120 days

Mean

AAR -0.4289

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

Market

AAR -0.3504

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

165 days

Mean

Market

Mean

0.8842

0.5795

-0.1209

-0.5611

-0.1330

(1.7889) (8) (59)

(1.1725) (-0.2447) (9) (9) (81) (74)

(-1.1353) (11)*** (99)*

(-0.2692) (-2.3966)*** (1.1156) (-1.2496) (7) (12)*** (7) (11)** (60) (117)*** (58) (95)*

-1.0575

-0.1337

0.4822

-0.1750

0.8950

0.0001

0.0581

(-0.3176) (9) (75)

(1.6244) (7) (61)

(0.0002) (9) (90)

(0.1054) (-1.9194)** (-0.2427) (-2.2456)*** (0.8971) (-1.0057) (8) (12)*** (7) (11)** (6) (7) (77) (117)** (65) (116)** (51) (75)

0.7053

0.4465

0.6492

0.4356

-0.1076

-0.5769

-0.1381

(0.9573) (9) (74)

(1.3920) (8) (61)

(0.9340) (-0.2307) (10) (9) (86) (78)

(-1.2369) (10) (95)**

(-0.2961) (-2.4992)*** (0.4765) (-1.1428) (7) (12)** (8) (11)** (59) (115)*** (60) (93)**

-1.0574

-0.1442

0.0509

-1.1656

0.4943 -0.5541

(-0.7600) (1.5122) (9) (6) (85) (43)

-1.2549

0.2222 -0.5330

0.4783

-0.1864

0.8878

-0.0071

(-0.8140) (0.8882) (10) (6) (86) (51)

(-0.3461) (10) (77)

(1.6486) (7) (61)

(-0.0132) (0.0945) (-1.9635)** (-0.2678) (-2.3302)*** (0.9181) (-1.0287) (9) (8) (12)** (7) (11)** (6) (7) (90) (77) (118)*** (66) (116)*** (52) (75)

AAR -0.4040

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

-1.2372

0.5514 -0.6176

(-0.7959) (0.8751) (10) (6) (86) (50)

AAR -0.4384

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

Market

0.5143 (1.0406) (9) (74)

AAR -0.3544

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

210 days

0.7893

(-0.7089) (1.5970) (10) (6) (84) (42)

AAR -0.4385

Z-stat Sign (exact binomial stat) Wilcoxon signed rank (exact stat)

-1.1845

0.50393 -0.5445

-1.0503

0.49442 -0.5540

0.7320

0.4490

0.8067

0.4633

-0.1040

-0.4806

-0.0508

(-0.9160) (1.6599) (10) (6) (86) (44)

(1.0182) (9) (73)

(1.8293) (8) (62)

(1.050) (10) (86)

(-0.2359) (9) (80)

(-1.0898) (10) (92)

(-0.1151) (-2.3817)*** (1.0094) -1.3999* (8) (12)** (8) (11)** (60) (116)*** (60) (92)

0.44513 -0.6173

Note: ***, ** and * denote significance at 1%, 5% and 10% respectively (one-tail test).

Then, we deepen our analysis to see if we could make conclusions based on the type of news included in positive and negative aggregations. It might be the case that some kinds of positive news have an impact, even if on the average positive news have no effect. Hence,

15

we divide positive news in those linked to ISO certification issues, and those related with proenvironment investments (separated in announcements and inaugurations). We find that announcing having received ISO in the family of 14.000 has no impact whatsoever, and this is also the case for all news related to pro-environment investments (see Table 8). However, announcements of investments in favor of the environment do have an effect on day +3, while inaugurations of new plants with special provisions toward the environment are anticipated by the market and do have a significant positive correspondence on days –5 and –3. The magnitude of the effect is indeed larger than that of negative news (going from 1.95 to 3.47%). We perform a similar disaggregation for negative events. We find (as shown in Table 9) that day +1 and +3 negative impact is directly linked to court/government rulings or citizens/ONG complaints, following the same results as the aggregation of negative events. We also detect that oil companies (highly represented in the sample) are those for which there is particular impact.15 Here, again, the magnitude of the impacts are small (around 1%). We have also performed no parametric tests for the subsets of positive news and negative news, but we do not report them to avoid including more Tables. However, those tests confirm in almost all cases the results previously obtained. Table 8.

AAR Z-stat

CAAR Z-stat

AAR Z-stat

CAAR Z-stat

AAR Z-stat

CAAR Z-stat

AAR Z-stat

CAAR Z-stat

Disaggregation of positive events 1995-2001 Day -5

Day -4

Day -3

Day -2

1.1053

1.4455

0.4961

Positive events: ISO certification news 0.3875 0.6419 -1.3188 -0.3710 -0.4582 -1.0556 -0.9390 -0.3321

0.5771

0.7548

0.2590

0.2023

0.3352

-0.6886

-0.1937

-0.2392

-0.5512

1.1053

2.5509

3.0469

3.4344

4.0763

2.7576

2.3865

1.9283

0.8727

0.5771

0.9418

0.9185

0.8966

0.9519

0.5878

0.4710

0.3560

0.1519

1.3240

-1.1534

0.7989

-0.6960

0.3060

-0.7757

1.3240

0.3354

0.7700

-0.3319

0.7989

0.1431

0.2683

-0.1001

0.4650

-0.5879

0.3765

-0.4761

Day 0

Day 1

Day 2

Day 3

Positive events: related to pro-environment investments 0.5071 -1.2856 -1.2674 -0.5350 0.9875 1.2321 0.3531 -0.7648

-0.3228

0.5959

0.7435

-1.4183 -1.9533 -1.1069 -0.0508 -0.3827

-0.4812

-0.2524

-0.0108

0.2130

0.3023 0.0608

Day 4

-0.4903

Day 5

-0.1734

-0.0663 -0.3984 -0.0109

-0.0627

0.0491 -2.7282 0.0296

-1.6462

0.3514 -2.3768 0.0671

-0.4324

Positive events: announcements of pro-environment investments -0.2160 -1.9001 -1.0197 0.0604 1.0331 2.5148 0.2876 0.1540 -3.6739 -0.1749

-1.5386

0.4650

-0.0054

-0.1782

-1.6982

0.2054

-0.0017

-0.0454

-0.3751

3.4715

-2.2843

2.7345

-0.8257

0.8365

2.0363

0.2329

-2.5140 -2.4536 -1.6272

** 0.3846

0.6722

0.0601

0.0990

-0.4966

0.0489

-0.4425

-0.2717

0.1247

-2.9749

0.8263 -2.8477 0.1154

-0.3793

Positive events: inauguration pro-environment investments 1.9533 -0.0566 -1.7628 -2.0235 0.8964 -1.3332 0.5167 -0.2132 -0.3639

-1.7993

1.5386

-0.0446

*** 3.4715

1.1872

* 3.1405

3.0839

2.7345

0.6613

1.4282

1.2146

***

Day -1

-1.3886

-1.5939

1.3210 -0.7025 0.4654

-0.2259

0.7061

-1.0502

0.1939 -1.1393 0.0577

-0.3173

0.4070

-0.1679

-0.2866

-0.6226 -0.8358 -1.1996 -0.1635

-0.2082

-0.2849

*

Note: ***, ** and * denote significance at 1%, 5% and 10% respectively (one-tail tests).

15

We do separate rulings/government decisions from citizens/ONG complaints because a detailed analysis of the news shows that (at least in Argentina, and, in our data set) most regulatory decisions concerning the environment are taken ex - post citizens´ complaints. Hence, it would not be appropriate to combine news according to any of those criteria as if they were separate issues. However, we analyse positive versus negative events.

16

Table 9.

AAR Z-stat

CAAR Z-stat

AAR Z-stat

Disaggregation of negative events 1995-2001 Day -5

Day -4

Day -3

-0.5893

0.6324

0.2839

-0.8778

0.9420

0.4229

Z-stat

Day -1

Day 0

Day 1

2.0849

0.3421

-0.8691

-1.7637

0.1355 0.0713

0.2532

1.6527

1.8824

1.2990

-0.8103

-0.0093

0.2177

1.2310

1.2541

0.7900

0.1161

-0.6802

0.0330

-0.7249

-1.3425

0.0652

-1.4307

-0.5439 -0.0089

-1.2306

-0.8318

0.0655

-0.6803

1.3905

-1.4686

0.1554

*

-1.2604

-1.2273

-1.4363

-1.2111

*

-0.1244

Negative events: related to oil companies 0.0332 -0.3447 0.7045 -0.7441 0.0787

*

-0.6235 -0.5960

Day 2

Day 3

Negative events: citizen complaints and rulings 1.3996 0.2296 -0.5834 -1.1839 -0.0835 -1.1634 ** 0.2061

*

CAAR

Day -2

-1.7331

Day 4

Day 5

0.6163 -0.4858 0.9181

-0.7236

** -0.9384 -0.3695 -0.8179 -0.4660

-1.2110 -2.3901

-0.1741

-0.3674

0.2717 -0.5850 0.5363

-1.1547

***

-1.5720 -0.8674 -1.5495 -1.4773 -2.6883 -2.4392 -2.9755 -1.3875

-0.6989

-1.1559

*

-1.0309

-1.7686

-1.5224

-1.7707

**

*

**

Note: ***, ** and * denote significance at 1%, 5% and 10% respectively (one-tail tests).

Given the “stability” of our results for the first period, we only display our results of the second period for the constant mean model, the 165 days estimation period, and parametric tests. We find that positive between 2003 and 2008 do have an average positive significant impact (on the day of the event and after that), while average returns from negative news are not significant. This result (in terms of significance) is the opposite than the one found for the 1995-2001 period, where positive news do not significantly impact returns, while negative news do. One possible explanation for that result is that positive news in the second period are less related to own (firms) announcements and ISO and have more to do with voluntary agreements to international organizations and prizes, which could be considered more credible. Nevertheless, the significant ARs are very small. For negative news, court and government actions are less credible in the years 2000s than they were in the 90s. However, when averaging together positive and negative news for the two periods (all events), the result remain in line with that for the first period. Positive news have no significant impact, while negative ones do generate abnormal returns on days +1 and +3. Significant impacts are of the expected sign. The magnitude of the impacts for periods taken together are in the order of one digit. Table 10 summarizes our results for the second period and both periods taken together, when we average positive and negative events separately.

17

Table 10. Positive versus negative events: parametric tests Day -5

Day -4

Day -3

Day -2

-0.0149

0.0061 0.0024 0.0023

Day -1

Day 0

Day 1

Day 2

Day 3

Day 4

Day 5

2003-2008 period AAR Z-stat

AAR

-1.9453

0.7976

0.3094

0.2962

-0.0109 -0.0119 0.0105 -0.0038 Z-stat

-1.1426

-1.2483

1.1052

-0.4011

Positive events 0.0025 0.0152 0.0166 -0.0405 0.0196 0.3286

1.9797

2.1699

** ** Negative events 0.0072 0.0131 0.0039 0.7522

1.3736

0.4079

-5.2880

0.0107

-0.0032

2.5627

1.3930

-0.4116

***

*

0.0075 -0.0043 0.7845

-0.4478

-0.0009

-0.0033

-0.0988

-0.3442

Together 1995-2001 and 2003-2008 news AAR

0.5288 Z-stat

AAR

0.8443

-0.3219 Z-stat

-0.7904

Positive events -0.0605 0.1660 -0.1418 -0.1727 -0.1905 -0.0091 -0.0759 -0.0640 -0.0965

0.2650

-0.2264

0.3410 -0.1244 0.6605 0.8373

-0.3054

1.6219

-0.2757

-0.3042

-0.0145

-0.1211

-0.1022

Negative events 0.0019 0.0464 -0.7680 -0.0934 -0.9010 0.0048

0.1138

-1.8859 **

-0.2293

-2.2124

-0.0204

-0.7039

-0.0326

-1.1238

0.3592

-0.4039

0.8821

-0.9918

**

Note: Results are for CMM and n= 165. ***, ** and * denote significance at 1%, 5% and 10% respectively (one-tail test).

IV.

Conclusions

Our conclusions can be stated on the base of the signs and the magnitude obtained, being the second issue the most innovative finding of this paper. With respect to the sign of the impact, we confirm that markets react negatively to court/government rulings or citizens/ONG complaints regarding firms´ environmental performance. But, we are not able to show that positive environmental news have impact. This result is intuitive to us since it is reasonable to state that Argentina´s society is generally rather skeptical of firm´s statements about their behavior. With respect to the magnitude of the impacts, previous literature has found a larger impact of environmental news on stock markets in developing countries in comparison to Canada and the United States. According to the authors, that difference could be due to a large volatility in developing countries´ stock markets. However, our results for Argentina do not confirm those findings. We find average (and average cumulative) abnormal returns of one digit, which are in line with results obtained in studies related to developed countries and not with those of two digits in Dasgupta, Laplante and Mamingi (2001) for four developing countries, Gupa and Goldar (2005) for India and Dasgupta et al (2006) for Korea. In our opinion, further studies are needed to confirm the existing evidence of a larger impact of environmental news in developing countries. More so, when the result (justified on a matter of volatility) happens in countries with low local government pressure and scarce environmental education, which makes the larger effect counterintuitive. We have shown that our results hold for different models and different periods used to estimate returns. And, that significance found through parametric tests was in most cases confirmed by non parametric tests.

References Ambec S. and P. Lanoie (2008), “When and why does it pay to be green?”, Academy of Management Perspectives, 23: 45-62. Ball R. and P. Brown (1968), “An Empirical Evaluation of Accounting Income Numbers”, Journal of Accounting Research, 159-178.

18

Beirão and Sarsfield Cabral (2002), “The reaction of the Portuguese stock market to ISO 9000 certification”, Total Quality Management, Vol.13, No.4, 465-474. Campbell J.Y., A.W. Lo, and A.C. MacKinlay (1997), The Econometrics of Financial Markets, Princeton University Press, Princeton, New Jersey. Chudnovsky D., A. López and V. Freylejer (1997), “La prevención de la contaminación en la gestión ambiental de la industria argentina", CENIT, Documento de Trabajo No. 24, October. Dasgupta S, JH Hong, B Laplante, N Mamingi “Disclosure of environmental violations and stock market in the Republic of Korea”, Ecological Economics, 58 (4): 759-777. Dasgupta S., B. Laplante and N. Mamingi (2001), “Pollution and Capital Markets in Developing Countries”, Journal of Environmental Economics and Management, Volume 42, Number 3, November, pp. 310-335(26). Fama E.F., L. Fisher, M. C. Jensen and R. Roll (1969), “The adjustment of stock prices to new information”, International Economic Review, 10: 1-21. Gupta S. and B. Goldar (2005), “Do stock markets penalize environment-unfriendly behaviour? Evidence from India”, Ecological Economics, 52: 81-95. Hamilton J.T. (1995), “Pollution as News: Media and Stock Market Reactions to the Toxics Release Data”, Journal of Environmental Economics and Management, 28(1):98-113. Hendricks K.B. and V.R. Singhal (1996), “Quality Awards and the Market Value of a Firm: an empirical investigation”, Management Science, 42: 415-436. Hong, J. H. and Hwang J. (2001), “Korean Major Environmental Accidents and Capital Market Responses”, Journal of Economic Research, May, v. 6, iss. 1, pp. 73-95. INDEC (2001), Circulación diaria neta de diarios y periódicos, sobre datos del Instituto Verificador de Circulaciones: www.indec.mecon.ar/nuevaweb/cuadros/9/a030601.xls, Instituto Nacional de Estadísticas y Censos de Argentina. Last access: January 29th 2010. Klassen R.D. and C.P. McLaughlin (1996), “The Impact of Environmental Management on Firm Performance”, Management Science, Vol. 42, No. 8: 1199-1214. Lanoie, P., and B. Laplante (1994), "The market response to environmental incidents in Canada: A theoretical and empirical analysis", Southern Economic Journal, 60, 65772. Lanoie, P., Laplante, B., and M. Roy (1998), "Can capital markets create incentives for pollution control?", Ecological Economics, 26, 31-41. Lehmann, E. L. Nonparametrics: Statistical Methods Based on Ranks. Prentice Hall, rev. 1st edition, 1998. Muoghalu, M., Robison, H.D., and J.L. Glascock (1990), "Hazardous waste lawsuits, stockholder returns, and deterrence", Southern Economic Journal, 7, 357-70. Przasnyski and Tai (2002), “Stock performance of Malcolm Baldrige National Quality Award winning companies”, Total Quality Management, Vol.13, No.4, 475-488. Reinhardt F.L., R. N. Stavins and R.H.K. Vietor (2008). "Corporate Social Responsibility Through an Economic Lens", Review of Environmental Economics and Policy, 2 (2) 219-239. Telle K. (2006), “It Pays to be Green” – A Premature Conclusion?, Environmental & Resource Economics, 35 (3): 195-220. Tietenberg T. (1998), “Disclosure Strategies for Pollution Control”, Environmental and Resource Economics, 11, 587-602. World Bank (2000), Greening Industry: New Roles for Communities, Markets, and Governments, World Bank Policy Research Report, Oxford University Press, New York.

19

Appendix A Table A.1. Results of regressions: events 1995-2001 Events Positive Acindar

Market Model Est. Alpha Est. Beta p-value

Date

n

12/27/98 6/5/99 11/18/01 9/23/98 9/16/97 12/9/98 9/6/99 10/22/99 3/13/99 5/23/00 4/12/97 4/12/97 4/4/98 6/5/99 12/10/99

165 165 165 113 139 122 141 165 165 165 165 165 165 165 165

-0.0073 -0.1153 -0.5129 0.9965 -0.0495 -0.1099 -0.2508 0.1197 0.1200 0.0012 0.0271 0.0249 0.1337 0.1811 0.0917

1.3182 1.2416 0.9872 0.8526 0.8359 0.9815 1.1234 0.8495 1.1589 1.2200 0.8223 0.9784 0.7762 0.7219 0.4684

0.0000

3/25/00 8/14/97 12/9/97 12/17/98 8/25/00 12/14/00 Perez Companc 3/9/97 6/25/98 YPF 10/17/96 3/9/97 5/25/97 12/13/97 8/12/98 09/14/99 12/19/00 3/5/01 6/25/01

165 138 135 165 165 162 165 165 165 165 165 165 165 165 154 146 127

0.1891 0.1322 -0.1969 -0.0446 -0.1362 -0.1679 0.0747 -0.0659 0.0235 0.0894 0.0891 0.1559 -0.0005 0.0834 -0.0529 -0.1535 -0.1055

0.5572 0.6093 0.4818 0.8439 0.6551 0.7595 0.7200 0.9059 0.7173 0.5671 0.5210 0.7007 0.7832 0.5482 0.1222 0.0617 0.0926

0.0000

Aluar Atanor Siderar Siderca Perez Companc Telecom Telefónica YPF

Negative Astra Atanor Celulosa Indupa

Std. Dev. Resid.

0.0000 0.0000 0.0214 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

0.0000 0.0179 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.1010 0.4199 0.1101

1.7304 2.2108 3.1918 9.3041 2.3176 2.6779 2.7228 1.6139 1.5029 1.2979 1.1165 1.1796 1.3812 2.0682 1.6032 1.7702 2.5275 4.3338 2.3625 1.6093 1.8025 0.8917 1.0865 0.8965 0.9757 0.9364 1.1038 1.3538 1.7951 1.7885 1.6092 1.3722

Note: Results for n= 120 and 210 are omitted for space reasons, but are available upon request.

Table A.2. Results of regressions: events 2003-2008 (n=165) Events Positives Acindar Aluar Repsol YPF

Banco Francés Ledesma Tenaris TGS Negatives Ledesma Molinos Rio del Plata Repsol YPF Indupa Solvay

Est.Alpha

Est.Beta

p-value

Std.Dev.Resi d.

04/01/2004 0.00000246 21/02/2004 -0.0009851 19/12/2003 0.0003878 27/04/2005 0.001001 17/12/2003 0.0007669 27/08/2008 0.0012724 09/04/2003 0.001645 17/12/2003 -0.000812 17/12/2003 -0.0000275 17/12/2003 -0.001177

1.092.724 1.126.129 0.293348 0.165355 0.129335 0.789668 1.523.131 0.503489 0.501245 125.452

0.000 0.000 0.000 0.000 0.007 0.000 0.000 0.000 0.000 0.000

0.01573 0.01546 0.01216 0.01277 0.01212 0.02352 0.029372 0.015632 0.015663 0.021939

03/07/2003 21/06/2006 05/12/2005 27/09/2006 17/04/2003 21/06/2006

0.220364 0.411828 0.332296 0.506628 0.530592 0.4484

0.111 0.000 0.000 0.000 0.000 0.000

0.03795 0.0176 0.01473 0.01374 0.021522 0.01474

-0.0006342 -0.0020809 0.000903 -0.0010794 -0.0013016 -0.001994

Note: Results for n= 120 and 210 are omitted for space reasons, but are available upon request.

20

Appendix B Table B.1. AR and CAR for positive events 1995-2001: Mean Model n =165* Positive Events ACINDAR 12/27/98

AR

Day -5 Day -4 Day -3 Day -2 Day -1 Day 0 Day 1 Day 2 6.280

Z-stat

Day 4

Day 5

1.359

CAR

10.136 10.484

Z-stat

ACINDAR 06/05/99

Day 3

1.551

1.310

AR Z-stat

CAR Z-stat

ACINDAR 11/18/01

AR Z-stat

CAR Z-stat

ALUAR 09/23/98

AR Z-stat

CAR Z-stat

ATANOR 09/16/97

AR Z-stat

CAR Z-stat

ATANOR 12/9/98

AR

6.633

Z-stat

1.691

CAR Z-stat

PÉREZ C. 09/06/99

AR Z-stat

CAR Z-stat

PÉREZ C. 10/22/99

AR Z-stat

CAR Z-stat

SIDERAR 03/13/99

AR Z-stat

CAR Z-stat

SIDERCA 05/23/00

AR

5.417 6.869

Z-stat

2.488

3.154

CAR Z-stat

TELECOM 04/12/97

AR

2.580

Z-stat

1.516

CAR Z-stat

TELEFÓNICA 04/12/97

AR Z-stat

CAR Z-stat

YPF 04/04/98

AR Z-stat

CAR Z-stat

YPF 06/05/99

AR Z-stat

CAR Z-stat

YPF 12/10/99

AR

6.183

Z-stat

3.281

CAR 6.183 Z-stat

3.281

Note: We report here abnormal (and cumulative abnormal) return, which are significant at 10% or less. Results for n= 120 and 210 are omitted for space reasons, but are available upon request.

21

Table B.2. AR and CAR for negative events 1995-2001: Mean Model n =165 Negative Events ASTRA 03/25/00

AR

Day -5 Day -4 Day -3 Day -2 Day -1 Day 0 Day 1 Day 2 Day 3 -3.994 -2.983

Z-stat

ATANOR 08/14/97

-2.033

CAR

-4.333

Z-stat

-1.560

Day 5

-1.518

-8.284 -10.550 -12.453 -11.104 -1.491

AR

-1.790

-2.005

-3.769

Z-stat

-1.704

-4.699

-1.403

-1.749

CAR

-11.067 -12.565 -17.264

Z-stat

CELULOSA 12/09/97

Day 4

-1.373

AR

-1.479

-1.938

-6.447

Z-stat

-1.462

CAR Z-stat

INDUPA 12/17/98

AR

-7.929

Z-stat

-2.207

CAR

-12.177

Z-stat

INDUPA 08/25/00

AR Z-stat

-1.957

-3.040 -1.509

CAR Z-stat

INDUPA 12/14/00

AR

-3.100

Z-stat

-1.336

CAR Z-stat

PÉREZ C. 03/09/97

AR Z-stat

CAR Z-stat

PÉREZ C 06/25/98

AR Z-stat

CAR Z-stat

YPF 10/17/96

AR

-2.775

Z-stat

-1.899

CAR Z-stat

YPF 03/09/97

AR Z-stat

CAR Z-stat

YPF 05/25/97

AR

-1.832

Z-stat

-1.524

CAR Z-stat

YPF 12/13/97

AR Z-stat

CAR Z-stat

YPF 08/12/98

-5.077 -3.814 -2.663

-4.382

-2.001

-2.298

-7.184 -8.619 -1.884

-2.022

AR

-8.036

-1.319

-1.333

-3.267

Z-stat

YPF 09/14/99

-7.545

-6.962

-1.628

-3.470

CAR

-10.978

Z-stat

-1.650

AR Z-stat

CAR Z-stat

YPF 12/19/00

AR Z-stat

CAR Z-stat

YPF 03/05/01

AR

-2.526

Z-stat

-1.572

CAR Z-stat

YPF 06/25/01

AR Z-stat

CAR Z-stat

-1.648

-1.488 -1.110

-4.244

-2.758

-3.262

-2.859

-7.102

-8.400

-1.170 -1.465 -1.366 -2.853 -3.963

-6.721

-9.983

-9.013

-5.770

-8.130

-6.999

-1.506

-3.831 -1.688

-1.436

-2.777

-3.609

Note: We report here abnormal (and cumulative abnormal) return, which are significant at 10% or less. Results for n= 120 and 210 are omitted for space reasons, but are available upon request.

22

Table B.3. AR and CAR for events 2003-2008: Mean Model n =165* Day -5 Day -4 Day -3 Day -2 Day -1 Day 0 Day 1 Day 2 Day 3 Day 4 Day 5 Positive events ACINDAR 04/01/04

AR Z-stat

0.038

0.054

1.518

2.151

CAR Z-stat

ACINDAR 21/02/04

AR Z-stat

CAR Z-stat

ALUAR 19/12/03

AR

0.046 0.060 0.024

Z-stat

3.447

4.482

1.771

CAR Z-stat

BCO FRANCÉS 09/04/03

AR Z-stat

CAR Z-stat

LEDESMA 17/02/03

AR Z-stat

0.025

0.056

1.340

3.013

CAR Z-stat

REPSOL YPF 27/04/05

AR

0.017

Z-stat

0.018 0.024

1.284

1.409

1.814

CAR Z-stat

REPSOL YPF 17/12/03

AR Z-stat

0.018

0.018

0.026

0.026

1.471

1.441

2.127

2.087

CAR

0.057 0.047 0.072 0.080 0.088

Z-stat

REPSOL YPF 27/08/08

1.738

1.334

1.954

2.050

2.140

AR Z-stat

CAR Z-stat

TENARIS 17/02/03

AR

0.036

Z-stat

1.968

CAR

2.878

0.101 0.083 0.075

Z-stat

TGS 17/02/03

0.053

1.814

AR

0.048

Z-stat

1.434

CAR

1.409

1.229

0.046 1.391

0.170 0.178

Z-stat

1.626

1.617

Negative events

INDUPA SOLVAY 27/09/06

INDUPA SOLVAY 17/04/03

AR

-0.031

Z-stat

-1.932

CAR

-0.046

Z-stat

-1.452

AR

-0.060

Z-stat

-2.589

CAR Z-stat

INDUPA SOLVAY 21/06/06

AR Z-stat

CAR Z-stat

LEDESMA 03/07/03

AR Z-stat

CAR Z-stat

MOLINOS RLP 21/06/06

AR Z-stat

CAR Z-stat

REPSOL YPF 05/12/05

AR Z-stat

-0.021 -1.386

CAR Z-stat

Note: We report here abnormal (and cumulative abnormal) return, which are significant at 10% or less.

23