On the Rationality of Institutional Investors: The Case of Major Industrial Accidents

Abstract

To determine if institutional investors act more rationally than individual investors, we examine the stock market price drop following 173 major industrial accidents from 1959 to 2017. In most cases, this drop, generally considered excessive, rapidly reverses. Using meta-analysis techniques, we distinguish events likely to trigger a large negative mid-term market reaction (potentially destructive events) from other events. For firms with large institutional holdings, we show that the market drop following potentially destructive events (potentially nondestructive events) is significantly larger (smaller). This is consistent with the hypothesis that investors’ potential irrational reaction depends on their sophistication.

Keywords:

• Industrial accident
• Institutional investor
• Rationality
• Event study
• Stock market

JEL classification:

• G14
• G23
• G41
• Q53

Notes

1 We restrict the survey to the subsample of studies that analyze post-announcement abnormal returns to study the impact of the events on the equity value of the firms directly involved in industrial accidents, excluding 1) those exclusively devoted to the spillover effects on industrial sectors or on the market, as in the study by Kaplanski and Levy (2010) and 2) those analyzing indirect effects like the reaction of airplane manufacturers to air crashes (Krieger and Chen 2015). We also exclude the studies of a single event, for which a list is provided by Capelle-Blancard and Laguna (2010). To our knowledge, we survey all the published studies fulfilling the above criteria.

2 Racine et al. (2020) classify 78 out of their 223 firms in sectors other than the classical sectors defined by Fama-French, which cover all sectors excluding airline.

3 The excess return of Phillips Petroleum from the day before the Pasadena explosion to the anniversary of the blast is about 29% (Knight and Pretty 1996). Union Carbide obtained an excess return of 8.9% during the year following the accident, starting before the announcement of the Bhopal disaster (Salinger 1992).

4 In June 2008, the US Supreme Court reduced what had once been a $5 billion punitive damages award against ExxonMobil to about $500 million, related to the Exxon Valdez spill in Alaska in 1989. Exxon reported total earnings of about $30 billion in 2010. In March 2010 the Paris Court of Appeal handed down its judgment on the Erika wreckage and pollution, which occurred in 1999. The Erika was last chartered by Total-Fina-Elf. Total was found to be imprudent in implementing its vessel vetting process and was ordered to pay a fine of €375,000. Total reported net income of about €14 billion in 2010. Following the Bhopal accident in 1984, legal procedures are ongoing in 2019. The Supreme Court of India ordered a final settlement of all Bhopal litigation of $470 million, and Union Carbide paid $420 million in 1989. However, 35 years after the disaster, the former Union Carbide factory remains abandoned, with more than 350 tons of untreated toxic waste inside, and victim compensation is still pending. According to United Nations Special Rapporteur Baskut Tuncak, the merge of Dow with DuPont in 2017 could obstruct justice for victims in Bhopal https://www.theweek.co.uk/88401/dow-dupont-merger-may-scupper-bhopal-disaster-compensation-fears-un-official.

5 Federal Reserve Board, Flow of Funds Account of The United States, Table 213, available at https://www.federalreserve.gov/releases/z1/release-dates.htm. From 1965 to 2017, these data are compiled by the Securities Industry and Financial Markets Association (SIFMA); SIFMA (2010) and SIFMA (2019).

6 The FACT database reports about 23,000 accidents in the chemical sector worldwide between 1960 and 2014 http://www.factsonline.nl/. In the US airline industry, Walker et al. (2014) report 173 accidents between 1950 and 2009, and the US Department of Transportation reports 65,534 accidents in the railway industry between 1990 and 2015, inducing 361 fatalities and 9,181 injuries https://www.bts.gov/content/train-fatalities-injuries-and-accidents-type-accidenta.

7 Developments in media technologies over the past three decades have changed the way and pace that people get the news. However, institutions have been precursors in the use of wired media technology like Bloomberg Terminal, Thomson Reuters or Dow Jones Newswire, since the early ‘80s. Moreover, the pace of news diffusion has only a slight impact on our analysis, because we use daily returns and the event time as the starting point for our estimation.

8 To detect simultaneous events that can influence the results, we check the information published by Value Line, Moody’s, Factiva and the New York Times for the months surrounding each accident. We find one case of simultaneous accidents, and restricted the analysis of the first accident to the period of time that elapsed until the next accident. This eliminates the influence of the second accident on the estimation of abnormal returns.

9 We use ADR returns, when available, for non-US companies whose stocks are not traded in the US.

10 For fiscal years ending in December, we use the current fiscal year if the crisis occurs before June 30, and the next fiscal year for any crisis occurring between July 1 and the end of December.

11 The single accident we get during the ‘50s occurs in 1959 and is included in the ‘60s decade.

12 Estimating the global media coverage of major accidents is out of reach for a large sample. For example, Smith (1993) located 1,156 stories containing the words “Exxon Valdez” and “oil spill” during the 6 months following this accident.

13 First, meta-analyses are constrained by the limits of the studies analyzed. In our case, some variables are evocated in one or two studies only and their effects cannot be estimated at the meta-level. Second, the samples differ between the studies. Because of the large overlap between the samples used in various studies of industrial accidents, the problem is probably not major in our analysis. Third, the outcome of a meta-analysis depends on the studies analyzed; we are subject to the publication bias by omitting unpublished research. Another limit of meta-analysis comes from the difference in the results of individual studies. The summary effect possibly ignores important differences across studies. Lastly, the use of meta-analysis for rare and adverse events can be problematic.

14 Because they are available in numerous publications and textbooks, we do not present the details of the estimation of the size effects (Wolf 1986; Shelby and Vaske 2008; Stanley and Doucouliagos 2012).

15 There is no statistically significant difference for the other variables, based on Wilcoxon tests of the difference between the mean (median) of the distributions for firms depending on their institutional holdings, except for Denv and Decade.

16 We get similar coefficients and levels of significance for other windows. For the sake of brevity, we do not report the results for the other windows.