اثر ماندگار یک شوک خبری دروغ

In September 2008, a six-year-old article about the 2002 bankruptcy of United Airlines' parent company resurfaced on the Internet and was mistakenly believed to be reporting a new bankruptcy filing by the company. This episode caused the company's stock price to drop by as much as 76% in just a few minutes, before NASDAQ halted trading. After the “news” had been identified as false, the stock price rebounded, but still ended the day 11.2% below the previous close. We explore this natural experiment by using a simple asset-pricing model to study the aftermath of this false news shock. We find that, after three trading sessions, the company's stock was still trading below the two-standard-deviation band implied by the model and that it returned to within one standard deviation only during the sixth trading session. On the seventh day after the episode, the stock was trading at the level predicted by the asset-pricing model. We investigate several potential explanations for this finding, but fail to find empirical evidence supporting any of them. We also document that the false news shock had a persistent negative effect on the stock prices of other major airline companies. This is consistent with the view that contagion effects would have dominated competitive effects had the bankruptcy actually taken place.

A central question of financial economics is whether markets are efficient. Among other things, market efficiency requires that asset prices react to news about fundamentals, as opposed to noise. However, in most circumstances relevant information and noise arise simultaneously, and cannot be easily separated. Agents have to make inference about fundamentals from possibly noisy pieces of information, and thus the noise component usually affects agents' investment decisions. In this paper we explore a natural experiment that allows us to study a stock market's reaction to an information release for which the noise component can be singled out very cleanly. On September 8, 2008, an old article about the 2002 bankruptcy of United Airlines' parent company (henceforth UA) resurfaced on the Internet and was mistakenly believed to be reporting a new bankruptcy filing by the company.1 This caused the company's stock price to drop by as much as 76% in just a few minutes, before NASDAQ halted trading. After the false news had been identified as such, the stock price rebounded, but still ended the day 11.2% below the previous close. Trading volumes skyrocketed during these extreme price movements. The episode can be thought of as comprising two pieces of information: the “news” that UA had filed for bankruptcy protection again, and the subsequent statements by UA and the media companies involved in the article's release clarifying that it pertained to the 2002 bankruptcy filing. The clarification statements were widely circulated shortly after the large price drop, and were publicly available when trading resumed. Moreover, the false news appears to have made its way to the main sources of financial information by sheer accident. This justifies our assumption that the episode provides a natural experiment to study the effects of what we refer to as a false news shock: two pieces of information that cancel each other. Given this shock, we are left with the task of trying to make sense of the 11.2% drop of UA's stock price on that day and its slow recovery on subsequent days. In order to study the impact of the false news shock on UA's stock price, we need a so-called “counterfactual”: the path that the stock price would likely have followed in the absence of the false news. In Section 3 we construct such a counterfactual path using a simple factor pricing model for UA's stock return. In particular, we postulate that the excess return on UA stock depends linearly on the excess returns of three factors: the “market” (as proxied by the S&P 500), the “airline industry” (as proxied by Bloomberg's World Airline Index), and crude oil. We estimate the asset-pricing model using data until the day before the false news impacted the market. The model captures the dynamics of UA excess returns quite well, explaining about 40% of its variation at both daily and intraday frequencies. We use our model to construct point estimates and standard-error bands for UA's stock price given the evolution of the three pricing factors on the day of the false news event, and over subsequent trading sessions. We find that after three trading sessions UA shares were still trading below the two-standard-deviation band implied by the model, and only returned to within one standard deviation of the model-implied price on the sixth trading session after the event. On the seventh day after the episode — and for quite a few days thereafter — the shares traded essentially at the level predicted by the asset-pricing model. These findings are robust to different specifications of the factor model. Throughout our analysis we maintain the assumption that the two pieces of information that comprise the false news shock exactly cancel each other, in the sense that after the clarification statements investors fully understood that the article was six years old, and that UA had not filed for bankruptcy protection again. However, it is possible that the false news shock had indirect asset-pricing effects not captured by our factor model — e.g., by affecting the liquidity of UA shares or investors' views about the quality of information about UA's fundamentals. We explore these possibilities in Section 5. However, we fail to find empirical evidence that is supportive of the theory-based explanations that we entertain. In that section we also investigate a more idiosyncratic potential explanation, motivated by the special circumstances in which the episode took place — namely, in the week before the bankruptcy of Lehman Brothers. Specifically, we consider the possibility that UA's financial conditions around that time made it particularly susceptible to changes in market perceptions about the health of the U.S. financial sector, due to high borrowing needs in a context of tightening borrowing constraints and lending standards. We augment the asset-pricing model with a factor that captures the market's assessment of U.S. banks' health, and repeat our counterfactual analysis.2 While the financial factor comes out as extremely statistically significant, it does not affect any of our conclusions, as the changes in the estimated counterfactual and error bands are negligible. In Section 6 we analyze the evolution of the stock prices of other major U.S. airlines during the episode (American Airlines, Continental Airlines, Delta Airlines and U.S. Airways). We find a very similar, although attenuated, pattern. On September 8, 2008, their share prices experienced maximum drops in the range of − 25.6% to − 31.8%, and ended the day between − 2.5% to − 5.3% relative to the previous closing price. The timing of the sharp price moves coincides with UA's. Employing the same type of factor pricing model as for UA, we construct a counterfactual path for the stock price of each of these four companies and find that the effects of the false news shock originated from the article on UA were also persistent. Finally, we document that intraday trading volumes for all five stocks spiked up considerably during the sharp price movements. We discuss our findings in the context of the literature on the “contagion and competitive effects of bankruptcy” (e.g. Lang and Stulz, 1992). Our paper adds to the available evidence on systematic deviations from informationally frictionless and efficient markets. Huberman and Regev (2001) document that a front-page New York Times article about an old scientific discovery had a huge impact on the stock price of the company responsible for it (EntreMed), even though the scientific findings had been published in Nature and covered by a not-so-prominent New York Times article more than five months before. The prominent article also had spillover effects on the stock prices of other biotechnology companies. The authors conclude that “enthusiastic public attention” may induce important movements in stock prices in response to old news that may have been overlooked by a large fraction of market participants. Like Huberman and Regev (2001), our paper provides very clean evidence on the importance of media vehicles in transmitting information to market participants and affecting how they perceive the world. While it is usually taken for granted that people receive and act on information transmitted by various media outlets, most models have no role for them — information is simply “received” (or inferred) by agents without any reference to concrete communication channels. There is, however, a growing body of literature that aims at estimating the asset-pricing impact of news identified through application of linguistic tools to newspaper articles. Tetlock (2007) constructs a media-based measure of “sentiment” towards stock markets from a linguistic analysis of the Wall Street Journal's “Abreast of the Market” column. He finds that high negative sentiment predicts lower returns for the Dow-Jones index over the next few days followed by a reversion, and that unusually high or low pessimism predicts high trading volume. Sinha (2009) uses a sentiment score from Thomson-Reuters to measure the tone of news articles and constructs portfolios based on past sentiment. He finds that a portfolio long in positive- and short in negative-sentiment firms is positively correlated with a long-short momentum portfolio and generates positive returns. Tetlock et al. (2008) use the fraction of negative words in firm-specific news articles to predict future earnings and stock returns. They find that negative language predicts negative earnings, even when they control for analysts' forecasts and historical accounting data. They also document that stock prices respond with a one-day delay to negative language in the firm-specific news. The papers summarized in the previous paragraph share the feature that they measure sentiment in news articles without assessing whether the information contained in the articles is actually novel. Accordingly, they are only loosely related to Huberman and Regev (2001), who document the asset-pricing impact of an article that contained old news about a company. Tetlock (2009) is somewhat closer to Huberman and Regev's work. He constructs empirical proxies to capture the degree to which a news story about a company is stale, such as the presence of another news story in the prior week, the presence of an extreme abnormal stock return in the prior week, or high media coverage in the past month. Tetlock (2009) sorts individual stocks into calendar-time portfolios based on the firms' recent public news. He documents that return reversals after news events are stronger when these events have a higher content of stale information according to his empirical proxies of news staleness. While our paper is close to Huberman and Regev, 2001 and Tetlock, 2009 in some dimensions, there is a crucial distinction, associated with the difference between the concept of stale (or old) news, and that of false news. The old news in the EntreMed case studied by Huberman and Regev (2001), and likely in many of the stale news identified by Tetlock (2009), refer to factors which at the time were still potentially important for the future profitability of the firm. In contrast, in the United Airlines event that we study, the “news” that produced a significant and persistent deviation of the company's stock price from its fundamental value was simply false: there was no new bankruptcy filing, and the reemergence of the six-year-old article should not matter for the company's profitability going forward. To the best of our knowledge, this particular nature of a false news event is unique to the episode that we document and study. Finally, when we first circulated our paper in February 2009 we were unaware of any other work on the UA episode. Since then we learned of two such papers. Lei and Li (2010) take a market microstructure approach, and use the episode to study how investors traded on September 8, 2008 to exploit their short-lived information. They find that during the sharp price movements associated with our false news event, investors used “intermarket sweep orders” — a liquidity-demanding type of limit orders — and traded aggressively. This led to a significant increase in the frequency of trades, and in the number of markets where trades are executed, but not in the size of trades. Marshall et al. (2010) is somewhat closer to our paper. They focus on the time it took for UA and other stocks to react to the false news shock as a way to test a specific theory, the so-called “gradual information diffusion hypothesis.” They find that other airline stocks and supplier firms recovered quickly,3 whereas UA's stock price took a few days to recover fully.4 They conclude that this is inconsistent with the gradual information diffusion hypothesis, and with the findings of Cohen and Frazzini (2008). Thus, their paper should be seen as complementary to ours, in that they analyze — and dismiss — a specific theory of persistent “mispricings” that we do not entertain in our analysis. Our paper is organized as follows. In the next section we provide a description of the episode. Section 3 describes our pricing model and the data used in the estimation. Section 4 presents our results, and discusses robustness issues. Section 5 provides an analysis of possible explanations for our findings. Section 6 documents the impact of the false news shock on the stock prices and trading volumes of other major airlines. The last section concludes.

We explore a natural experiment to study the impact of a false news shock on the stock price of United Airlines. We find that the shock had a persistent effect on the level of UA's stock price: it took six trading sessions for the stock to return to within one standard error of the model-implied counterfactual path. On the seventh trading session after the episode, and for quite a few days thereafter, UA's stock price was essentially trading at the counterfactual path implied by our factor model. We provide an in-depth analysis of two theories which could potentially rationalize our findings. According to the first, disrupted liquidity during a transition process might have resulted in the slow return of the share price back to its fundamental level. We do not find evidence of poor liquidity on the days following the false news shock. The second theory relies on ambiguity-averse traders as an explanation of a slow recovery back to the level implied by the asset-pricing model. We assess empirically whether the predictions of this theory are born out by the data and find little supportive evidence. We also consider the hypothesis that the slow recovery of UA's stock price can be explained by the firm's linkages with the U.S. financial sector which itself was in turmoil in September 2008. While we do find a significant correlation between UA's returns and the returns on financial sector stocks prior to the event, the counterfactual path implied by a model augmented with a financial factor does not alter our conclusion that it took unusually long for UA's share price to return to the level predicted by the model. A sudden bankruptcy of a firm may have contagious as well as competitive effects on other firms within the same industry (see e.g. Lang and Stulz, 1992). We study the behavior of other major U.S. airlines and find a similar pattern for their stock prices on the days after the false news event. This finding leads us to argue that contagion effects would have dominated the competitive effects — had the bankruptcy actually taken place. It is difficult to find other episodes that could be similarly characterized as a false news shock. There are a number of at first seemingly related cases that were subsequently shown to involve a fraud or hoax. In such cases, false news was deliberately produced to impact the stock price. This changes the nature of the trading environment, since some market participants trade with knowledge of the false news. It is reasonable to assume that the hoaxer takes advantage of the induced price movements by trading in the stock or its derivatives. One should thus expect a more complete reversal of the price movements produced by the false news, as the hoaxer trades to his or her advantage. Some prominent examples of false news due to fraud involved Pairgain Technologies (on April 7, 1999; the company later merged with ADC Telecommunications in 2000), and Emulex Corporation (on August 25, 2000). Curiously, despite reports that the frauds became apparent before the end of the respective trading sessions, in both cases the stock price still ended the day moving “in the direction” that the false information would have justified.19 Despite their different nature, we see these episodes as suggestive that our findings would generalize to other false news shocks. Finally, one may reasonably argue that one week is not a long enough spell for the misvaluation of stocks to have relevant economic effects — beyond gains and losses by traders and investors. However, this is how long it took for the effects of a pure false news shock to dissipate. In most circumstances, relevant information (“signal”) and noise arise simultaneously, and cannot be so easily separated.