Red Zero, Black Zero: The Economic Impact of Business Cycle Reporting Andrew C. Eggers { London School of Economics Alexander B. Fouirnaies { London School of Economics First version: November 2013 This version: November 20130 Abstract By convention, a recession is said to take place when two successive quarters of negative economic growth occur. We exploit the arbitrary cutoff implied by this convention to show that the announcement of a recession reduces consumer confidence and private consumption in OECD countries, conditional on actual economic fundamentals. We find that the effect is concentrated in countries with smaller social safety nets, which suggests that social spending reduces output volatility in part by making consumer expectations less pro-cyclical. Our findings highlight the fact that economic actors are only partially attentive to economic conditions and indicate that these information constraints play an important role in the transmission of economic shocks. We thank Ethan Ilzetzki, Daniel Sturm, Simon Hix, Dominik Hangartner, and the seminar audience at DIW Berlin for useful comments, and we thank Federico Bruni, Maoko Ishikawa, Yoojin Ma, and Miguel Rivi`erefor excellent research assistance. 1. Introduction In November of 2001, German newspapers announced that the country had entered a reces- sion. The German Federal Statistics Agency's new estimates for GDP growth in the second and third quarters of 2001 were -.03% and -.1%, respectively; this qualified as a recession according to the conventional definition, which requires consecutive quarters of negative growth. The statistics agency downplayed the news, noting that the -.03% could be rounded to 0, but in newspaper coverage it was pointed out that because the original number was negative this was a \red zero" and thus a recession was indeed taking place.1 In this paper, we ask how the announcement of a recession affects subsequent economic and political outcomes. As the episode of the \red zero" in Germany in 2001 illustrates, in some situations essentially arbitrary factors determine whether a recession is announced; chance could easily have turned the \red zero" that provoked alarming headlines into an apparently innocuous \black zero." We use a regression discontinuity design to compare these situations as a way of identifying the effect of media on economic perceptions, as well as the effect of sentiment and economic expectations on economic and political outcomes. To briefly preview our main results, what we find is that the announcement of a recession reduces both consumer confidence and growth in private consumption (the latter by as much as 1 percentage point) in the quarter during which the recession is announced.2 We find that these effects are concentrated in countries with weaker social safety nets (i.e. \liberal market economies" as categorized by the literature on welfare states), which suggests that recession announcements reduce confidence and spending by increasing the perceived risk of negative income shocks. We do not find effects on outcomes that we do not expect to be affected by recession announcements, such as government spending or lagged outcomes. These findings speak to at least four questions of broad interest. The first of these in- volves the role of expectations and sentiment in explaining the business cycle. An important strain of economic thinking going back at least to Pigou(1929) and Keynes(1936) holds that the sentiments of economic actors (\animal spirits" in Keynes' classic phrase) help account for swings in output and consumption. These ideas have received some attention from a 1For examples, see \Deutschland befindet sich in einer Rezession", S¨uddeutscheZeitung 21 Nov. 2001; \REZESSION. Alle f¨uhrendaw R-Wort im Mund", S¨uddeutscheZeitung 23 Nov. 2001. 2That is, we find lower confidence and spending growth in quarter t when growth is barely negative in quarters t − 2 and t − 1 (provoking a recession announcement in quarter t), than in cases when growth is barely positive in one of those preceding quarters, controlling for previous growth and other covariates. 1 theoretical literature exploring multiple equilibria and self-fulfilling prophesies (Howitt and McAfee, 1992; Farmer and Guo, 1994; Benhabib and Farmer, 1999; Farmer, 1999) and par- ticular downturns have been attributed to consumer sentiment (Blanchard, 1993; Hall, 1993), but overall economists have taken a skeptical view of the idea that sentiment substantially affects the macroeconomy. It is accepted that measures of confidence (e.g. from surveys or stock market movements) predict future economic outcomes (Carroll, Fuhrer and Wilcox, 1994), but this relationship is mostly attributed not to sentiment per se but rather to in- formation about current or future economic fundamentals that is contained in confidence measures (Beaudry and Portier, 2006; Barsky and Sims, 2012). The challenge faced by this entire literature is that it is difficult to distinguish sentiment from information. Existing approaches rely heavily on applying structural restrictions to time series data, but the as- sumptions underlying these efforts are difficult to test and produce dismayingly contradictory results (Lorenzoni, 2011). Our contribution to this literature is to adopt a completely dif- ferent identification strategy that allows us to measure the effect of an expectational shock that (by construction) contains no information about fundamentals. The fact that we find a strong effect of this expectational shock on aggregate consumption provides clearer evidence that changes in sentiment have macroeconomic consequences.3 This paper also sheds light on the importance of information imperfections in under- standing macroeconomic outcomes and the transmission of economic shocks. In a classical rational expectations view of macroeconomics, agents are assumed to possess correct beliefs that incorporate all available information, which of course is unrealistic. In recent years, macroeconomists have shown how models that incorporate noisy, costly, or delayed informa- tion can provide alternative accounts of core phenomena such as unemployment, the Phillips curve, and aggregate volatility across the business cycle (e.g. Akerlof, 2002; Mankiw and Reis, 2002; Sims, 2003; Veldkamp, 2011; De Grauwe, 2011; Ma´ckowiak and Wiederholt, 2012). Al- though these approaches are gaining wider acceptance, it remains unclear how important it is for macroeconomists to incorporate information imperfections into their models and which imperfections in particular deserve attention (Coibion and Gorodnichenko, 2012). This pa- per contributes to this literature by providing firm evidence of consumers' inattentiveness to easily available information about economic fundamentals, and by providing evidence that this inattentiveness has consequences for the transmission of economic shocks. If economic 3Our study cannot of course resolve the question of whether sentiment or information about fundamentals explains the predictive power of e.g. consumer confidence. 2 agents understood the definition of recession and were even vaguely aware of recent quarterly growth estimates, the announcement of a recession would not affect consumer confidence in the way we document here. The fact that it does, and that consumer spending notice- ably responds, suggests that routine inattention to economic fundamentals not only smooths consumption (as shown by Reis(2006)) but in some circumstances may exacerbate volatility. Third, our findings contribute to a literature (much of it outside of economics) attempting to measure the effects of media on economic perceptions and, by extension, economic and po- litical outcomes. Studies in this literature (see, for example, Behr and Iyengar(1985); MacK- uen, Erikson and Stimson(1992); Blanchard(1993); Haller and Norpoth(1994); Nadeau et al. (1999); Wu et al.(2002); Hester and Gibson(2003); De Boef and Kellstedt(2004); Doms and Morin(2004); Soroka(2006); Starr(2012)) typically use time series methods to argue that media coverage has an effect on economic behavior, controlling for economic funda- mentals. They all leave open serious questions of endogeneity, however: given the complex relationships among news, attitudes, and economic fundamentals, it is difficult to rule out the possibility that the correlation these studies find between news shocks and contempora- neous (or even subsequent) economic outcomes is due not to the actual impact of news on economic outcomes but rather to functional form misspecification or omitted variable bias.4 In short, the problem this literature faces is the same as the problem faced by scholars trying to measure the effect of sentiment on economic outcomes. Again, this paper contributes by adopting an identification strategy that disentangles the effect of business cycle news from underlying economic fundamentals. Finally, we contribute to research examining the relationship between social spending and output volatility. A substantial literature highlights the role of progressive taxation and social transfers as \automatic stabilizers" that tend to reduce business cycle fluctuations by producing lower effective tax rates and higher social spending during recessions and the reverse during booms (DeLong and Summers, 1986; Gal´ı, 1994; Auerbach and Feenberg, 2000). We find that the announcement of a recession has larger effects on consumer confi- dence and private spending in liberal market economies like the U.S. that provide a weaker social safety net. Not only does this provide a robustness check for the