News shocks in the data: Olympic Games and their macroeconomic effects

We examine the macroeconomic effects of bidding for the Olympic Games using panel data for 188 countries during the period 1950-2009. Our findings confirm that economies react to news shocks: investment, consumption, and output significantly increase nine to seven years before the actual event in bidding countries. Hosting countries also experience significant increases in investment, consumption and output five to two years before the hosting of the Games. The macroeconomic effects associated with hosting the Olympic Games occur well in advance before the actual event.


Introduction
Contemporary economists have no clear answer on whether the organization of the modern Olympic Games is a matter of honor or a matter of money. Although the acquirement of international prestige is obvious, the economic advantages for hosting the Games are not that clear. The economic benefits of hosting the Games are dubious for most academics who have conducted independent research on the issue (see, for example, Owen (2005)). In particular, the current consensus in the literature is that the contemporaneous and post-hosting effects of the Olympic Games on output and aggregate demand are limited. From a macroeconomists point of view, this is a puzzle because total expenditures on these mega events are large: for example, the total cost of the Summer Olympic Games that were held in China in 2008 is estimated to be in the range of 40-60 billion USD. 1 The starting point of our analysis is that the bidding for a mega event such as the Olympic Games is associated with important anticipation effects. One of the main results in modern macroeconomic theory, summarized in Lorenzoni (2011), is that the economy should react to this news already before the event takes place: consumers and firms may become more optimistic about future economic prospects; they, thus, spend more and aggregate private consumption and investment increase at the time of optimism. If the optimism is justified, the economy converges to a higher long-run path; if it is not, output returns to its original trend.
Identifying empirically news shocks is a tough task because identification problems can be severe. For example, Beaudry and Portier (2006) use a bivariate VAR that includes a measure of TFP and the quarterly S&P 500 index and identify news shocks as a shock that does not affect productivity in the short run but affects productivity in the long run. The identified news shocks lead to positive conditional comovements among macroeconomic aggregates and, moreover, are found to be an important source of business cycle fluctuations. On the other hand, Barsky and Sims (2012) identify news shocks as shocks that have the most predictive power in explaining future TFP but are orthogonal to the current TFP innovation. They find that output, investment and hours fall slightly after a news shock while consumption increases. Blanchard et al. (2010) show that the use of structural identification assumptions in a VAR setting is sensitive to the informational assumptions made and that SVAR shocks may be perfectly anticipated by the consumers. These authors point to the adoption of more structural models for the identification of news shocks. Schmitt-Grohe and Uribe (forthcoming) present a structural model in which agents receive anticipated news. They find that such shocks account for about half of predicted aggregate fluctuations.
The literature has also looked at the effects of data revisions as a source of noise in private sector forecasts. For example, Frenkel (1981), Engel and Frankel (1984) and Hardouvelis (1987) have looked at the response of interest rates to the news embodied in the weekly money supply announcements. Hardouvelis (1988) and Gürkaynak et al. (2005a and2005b) examine how exchange rates, interest rates, and long-term forward rates respond to monetary news and to monthly macroeconomic news. Such news coming from the unemployment rate, the industrial production index, personal income, and orders of durable goods series provide information about the state of the business cycle and are closely watched by economic forecasters. Bartolini et al. (2008) explore how the release of new economic data affects asset prices in the stock, bond, and foreign exchange markets.
All these studies employ time series information and do not exploit specific announcementssuch as the organization of the Olympic Games -to measure news shocks in the data. Because of the particular nature of the bidding and planning process, the Olympic Games provide a unique opportunity to test for anticipation effects in macroeconomic data: about ten to eight years before the actual hosting of the Olympic Games the bidding process begins; then, after a technical evaluation of the original bids by the International Olympic Committee (IOC), the top five bids are shortlisted --and a host city is selected seven years before the actual organization of the Olympic Games. The election of the host city is made by the assembled active IOC members, each possessing one vote. Members from countries that have a city taking part in the bidding for the Olympic Games are unable to vote.
Given the nature of the bidding process, agents in candidate countries receive a signal for possible macroeconomic changes ten to eight years before hosting. The bidding signal gives typically a 1/5 probability to the country for holding the event, while the winning signal, seven years before the organization, produces certain news for investment planning in the following years. This particular information structure allows us to test whether such news affect economic behavior and explicitly examine the role of expectations and uncertainty realization in shaping macroeconomic outcomes. To the best of our knowledge our study is the first in the literature to investigate if agents react in anticipation of future events with macroeconomic data using a widespread signal such as the organization of the Olympics.
To analyze the effects of Olympic "news" we use a panel of 188 countries spanning the period 1950-2009. We construct two separate Olympic Games indicator variables. One for countries that bid and were selected to host the Games; and another for countries that bid and were not selected to host the Games. We employ panel fixed effects estimation techniques that allow for contemporaneous, future, as well as lagged effects of the Olympic Games. Within this panel fixed effects estimation framework, we examine how variations in countries' per capita GDP growth, private consumption, investment, government consumption expenditures, the price level and the exchange rate are related to the hosting and bidding for the Olympic Games.
Our empirical analysis provides support for the identification scheme adopted in Beaudry and Portier (2006): in the bidding countries output growth, investment, and private consumption significantly increase about nine to seven years before the Olympic Games are hosted. According to the theoretical predictions, after the announcement of the hosting country, investment decreases in the bidding countries (that unsuccessfully bid to host the Games) and there are no significant long-run effects associated with bidding for the Olympic Games on output, investment, or private consumption.
In contrast to the bidding countries, there is a significant positive effect on macroeconomic variables in hosting countries that occurs five to two years before the hosting of the Games. GDP growth increases significantly during the previous five years before hosting, and the peak response occurs four years before the actual hosting. This increase in GDP per capita growth is due to a significant positive and quantitatively large increase in investment and private consumption.
Government consumption spending also increases significantly four years in advance of the actual event. Our empirical analysis of the effects in hosting countries thus implies that hosting the Olympic Games entails more than an "Olive Wreath": it has real macroeconomic effects that are positive as well as long-lasting; and these effects show up in the data well in advance of the actual hosting of the event.
It is also important to note that the responses of key macroeconomic variables are significantly different between bidding and hosting countries after the announcement of which country gets to host the Olympic Games (for the years preceding the announcement the differences in effects between bidding and hosting countries are all insignificant).
Recently, Rose and Spiegel (2011) show that hosting and bidding for the Olympic Games has a positive impact on international trade. Using a variety of trade models, the authors document that this effect is statistically robust, permanent, and large. Our paper should be viewed as complementary to their study. First, the focus of our paper is on a broad set of macroeconomic variables such as growth in GDP per capita, consumption, and investment. Second, Rose and Spiegel claim that what matters for the increase in international trade is the signal countries send to international markets when bidding for the Olympics. We show that the bidding and hosting of the Olympics entails more than that since these events are associated with significant anticipation effects that induce positive output, investment, and consumption responses well in advance before the actual hosting of the Games. Our finding of significant anticipation effects is of substantive interest, both to the field of international macroeconomics as well as to the policy debate on the costs and benefits of hosting the Olympic Games. Third, by focusing on the anticipation effects associated with the hosting and bidding for the Olympic Games, we can provide an answer to a number of questions left open in Rose and Spiegel such as, for example, why open economies should bid to host the Olympic Games and why countries bid repeatedly for the organization of such events.
The remainder of the paper is organized as follows. Section 2 describes the data and estimation approach. Section 3 presents the main results. Robustness checks are discussed in Section 4 and Section 5 concludes.

Data and Estimation Approach
We obtain data on the Olympic Games bidding and hosting countries from www.olympic.org. Data Appendix Table 1 lists the bidding and hosting countries in our sample. Data on real per capita GDP, private consumption, investment, government consumption expenditures, the consumer price level, and the nominal exchange rate (defined as the ratio of home to US currency) are from the Penn World Table, version 7.0 (Heston et al., 2011). The data on exports and imports are from the World Development Indicators (World Bank, 2011).
We estimate the effects of the Olympic Games using a panel of 188 countries spanning the period 1950-2009. 2 This is the largest possible sample given the availability of data. The econometric model is:

Y it = a 0 Host it + A(L)Host it +B(F)Host it +c 0 Bid it + C(L)Bid it +D(F)Bid it +α i +β t +e it
(1) where Host it is an indicator variable that is unity in country i and year t if the country hosted in year t the Olympic Games. Bid it is an indicator variable that is unity in country i and year t if the country was bidding to host the Olympic Games that were held in year t. Because over the 1950-2009 period bidding to host the Olympic Games took place about eight to ten years before the actual hosting of the Games, we include up to ten leads of the bidding and hosting country indicator variable on the right- Note that as control variables in equation (1)  variables, but it rejects the hypothesis of a unit root for the first difference. Therefore, we use the first difference of these logged variables in the estimating equation. We note that we cluster the error term, e it , at the country level so that it may be arbitrarily serially correlated within countries over time.
Our key identifying assumption in the estimation of equation (1) is that, conditional on fixed country characteristics, the hosting and bidding for the Olympic Games is randomly assigned. Of course, a certain level of economic development is required to host the Games. However, whether countries have this level of economic development is captured by the country dummies, α i . In other words, these country dummies take into account that some countries have the capacity to host the Games while other countries do not have this capacity. Our assumption of random assignment is, thus, that conditional on having the capacity to host the Olympic Games the timing of when countries bid and host the Games is random. In this regard, it is important to note that the bidding for the Olympic Games takes place ten to eight years prior to the celebration of the Games. The incidence of the Olympic Games is therefore a predetermined event. The fact that bidding countries are excluded from voting when the International Olympic Committee convenes to decide which country will be selected to host the Olympic Games also implies that bidding countries have no influence on which country is chosen to host the Games. We will come back to the assumption of random assignment in Section 4.1 where we discuss and investigate it further.

Anticipation Effects
We begin the discussion of the results by focusing on the bidding countries, and thus, on anticipation effects. Beaudry and Portier (2006) show using a SVAR approach that news shocks lead to positive comovements in aggregate variables that mimic demand shocks. From a partial equilibrium point of view it is easy to model an economy in which news increase output, investment and consumption.
Agents receive news about future developments that change their expectations about the future, which makes them spend and invest more. Both the permanent income hypothesis and the neoclassical investment theory would support this view. The literature has provided mechanisms (see Lorenzoni (2010) for an extensive review) that deliver increases in consumption, investment, output and hours after a news shock in general equilibrium models. The news shock we consider can have many interpretations: it can be thought of as news about future government investment that enhances private productivity and results in increases in aggregate demand during the bidding phase; it can be thought of as a shock to expected demand from tourism; or as a structural reform that increases the openness of the economy (see, e.g., Rose and Spiegel (2011)).
Our results in Table 1 are consistent with the finding in Beaudry and Portier (2006): output growth, investment, and private consumption significantly increase during the period that countries bid for the Olympics. There is a significant positive output growth, investment, and private consumption response in the bidding countries about seven to nine years before the actual hosting of the Games.
Investment responds first, while output and private consumption react with a one-year lag relative to investment. Government consumption spending, the price level and the exchange rate do not react significantly to the news.
Moreover, responses should be similar for all bidding countries between the bidding dates and the decision date. Panel A of Table 4 reports the p-values on the null hypothesis that the ex-ante and contemporaneous coefficients on the hosting and bidding indicator variables (that we obtained from equation (1)) are the same. The main finding is that the coefficients ten to six years prior to the event are not significantly different in the bidding and hosting countries; this can be seen from the p-values which are all larger than 0.1 for the null hypothesis that the coefficients are equal. Hence during the time of the bidding phase, there is no evidence that the macroeconomic effects differ for unsuccessful bidding countries and hosting countries. Given that there is no evidence of a significant difference in the effects, we report in Table 2 estimates from an alternative regression where for the bidding phase, unsuccessful bidding countries are pooled with hosting countries. The main result is that this alternative regression confirms the presence of anticipation effects: private investment, consumption and output react significantly eight to nine years before the organization of the Games. We now turn to discussing our estimates of the effects of the actual hosting of the Olympic Games.

Contemporaneous and Ex-Ante Effects
We report in Table 3 estimates of the contemporaneous and ex-ante effects of hosting the Olympic Games. Column (1) shows that countries which hosted the Olympic Games experienced a significant increase in their real per capita GDP growth up to five years before the organization of the event. The peak effect occurs about four years prior to hosting. The estimated coefficients imply that at peak real per capita GDP growth increases by up to two and a half percentage points. Notice that in the year when the Olympic Games are hosted the effect on GDP per capita growth is positive but insignificant, which suggests that the main effects on GDP per capita growth occurred already well in advance before the actual hosting of the Games.
It is noteworthy that the results in column (1) of Table 3 are compatible with the patterns of actual investment since many countries followed the 2-4-1 investment plan before the Games: 2 years of planning, 4 years of constructions and 1 year of testing (see, e.g., Preuss (2004)). Indeed column (2) of Table 3 shows that the reason for the significant increase in GDP per capita growth is a significant positive and quantitatively large increase in investment. The peak investment effect coincides with the peak output effect. Investment responses are still significantly different from zero three years before actual hosting. Investment reacts faster than government consumption spending (see column (4) of Table 3). The strongest reaction of government consumption spending occurs on average two years before hosting. Column (3) shows that the private consumption response follows a similar pattern as output; the exception is that, in contrast to output, the contemporaneous response of private consumption to the hosting of the Olympic Games is positive and significant at the 5 percent significance level.
We find also a significant positive ex-ante effect of the hosting of the Olympic Games on the consumer price index and the nominal exchange rate. The estimates in columns (5) and (6) of Table 3 show that there is already a significant positive response in consumer prices five years before the actual hosting of the Olympic Games; and in the exchange rate about seven years before the actual hosting, that is, at the time when the announcement of the hosting country is made.
In Figure 1 we present our findings differently by plotting the cumulative effects (and 90% confidence bands) after the announcement of the Olympic Games hosting country. The figure is clear about the importance of the ex-ante effects: all variables, except government spending, increase significantly prior to hosting with exchange rates reacting more significantly immediately after the announcement.
Comparing the maximum output responses between the bidders (Table 1) and the winners (  (1) of Table 3 imply a peak multiplier (at four years before the hosting of the Olympic Games) of about six.
According to the expectations hypothesis responses should be similar for all bidding countries between the bidding dates and the decision date. However, once the International Olympic Committee has announced which country will host the Games, responses should differ between hosting and unsuccessful bidding countries. The Olympics bidding can also be interpreted as a noise shock in the spirit of Angeletos and La'O (2010). In unsuccessful bidding countries the agents' optimism turns out to be unjustified and, as a result, the economy returns to its original trend. Panel A of Table 4 shows that indeed this is the case. The coefficients that capture the effects of the Olympic Games on GDP growth two to five years before the celebration of the Games in the host countries are significantly different from the coefficients in the bidding countries. Hence, pooling unsuccessful bidding countries with hosting countries as we did in Table 2 is a valid approach for testing anticipation effects (since the p-values in Panel A of Table 4 for ten to six years prior to the hosting of the Games indicate no significant difference between unsuccessful bidding and hosting countries), but it is not a valid approach for testing ex-ante effects that may be present up to five years before the actual hosting of the Games.  The results in Table 5 might justify the findings of many studies (see Owen (2005)) that indicate that the economic benefits of organizing mega events such as the Olympic Games are quantitatively and economically small. Our analysis stresses that the benefits from organizing such events are not contemporaneous, nor lagged. We find that the positive effects of the Olympic Games have to do with the ex-ante effects they induce.

Cumulative Effects
The hosting of the Olympic Games may be associated with a long-term legacy effect. According to the supporters of the Olympic Games infrastructure investments lead to improvements in overall production conditions for domestic and foreign enterprises, making investment more attractive and increasing GDP per capita in the long run. We can examine whether the hosting and bidding for the Olympic Games had a long-run effect on the level of GDP per capita and the other variables of interest.
We do this by summing up the estimated coefficients, and test whether their sum is significantly different from zero.
Panel A of Table 6 shows that for the hosting countries the sum of the estimated coefficients for the GDP per capita and private consumption response are positive and significantly different from zero at the 90 percent level. For the hosting countries the Olympic Games were hence associated with permanently higher levels of GDP per capita and private consumption. On the other hand, Panel B shows that for the bidding countries there were no significant long-run effects. The sum of the estimated coefficients is quantitatively small and statistically insignificant at any conventional confidence level. In Panel C we test for the existence of significant differences between the long-run effects of the Olympic Games in the host countries and the bidding countries. For GDP per capita and consumption we can reject with over 95 percent confidence that the long-run effects are the same in the host and the bidding countries. 3 In Figure 2 we plot the cumulative effects for the host countries (continuous line) and the bidding countries (dashed line) for the set of variables we consider. The cumulative responses of all the variables for the periods before the announcement of the hosting country are similar and we cannot reject the hypothesis that both hosting and bidding countries behave the same. However, after the 3 In Appendix Table 1 we document that the estimates on the lagged Olympic Games bidding indicators are either significantly negative or insignificant. In Appendix Table 2 we document that the findings on the long-run estimates continue to hold when controlling on the right-hand side of the estimating equation for lags of the dependent variable. In that case, also the long-run investment response is significantly different from zero in the hosting countries, and significantly different from the long-run investment response of the bidding countries.
announcement of the host, per capita GDP continues to increase in the hosting countries while in the countries that unsuccessfully bid for the Olympic Games per capita GDP decreases. This is apparent in the top left-hand panel of Figure 2 where we plot the cumulative effects on output growth. Output increases significantly two periods before the announcement of the winner, but output growth starts immediately to fall after the bad "news" of having lost the bid are revealed. On the other hand, for the hosting countries, output growth increases with a faster pace, and, particularly so, for the four years before the celebration of the Games. For the components of output which are displayed in the next three panels the pattern is similar. The cumulative effects on inflation and the exchange rate, displayed in the bottom panels of Figure 2, are also different for hosting and bidding countries.
Results are informative about the nature of the Olympic 'news' signal. As mentioned earlier, Olympics bidding might be associated with news about changes in infrastructure; changes in foreign demand through tourism; or structural changes in the degree of openness of the economy. The results confirm the nature of the news have to do with future changes in aggregate supply relative to demand since the effect in the hosting countries turns out to be permanent and quantitatively large.
Also, results provide support for the identification of news shocks proposed by Beaudry and Portier (2006). Blanchard et al. (2010) show that if attention is limited in models in which consumers can perfectly anticipate future shocks the SVAR identification scheme can recover the true news shocks. If we think of the Olympic games bidding as a signal that reveals information about future shocks, it is not surprising to find that the responses of the macroeconomy to Olympics' news resembles the ones of the economy with news identified through a SVAR using the restrictions of Beaudry and Portier (2006), even though the identification method to recover the shocks is very different.

Sensitivity Analysis
We have performed a number of sensitivity tests to ensure that our results are robust. In the paper we report only those estimates that we believe to be more relevant for providing robustness. The other estimates not reported here are reported in a supplementary online appendix that is available for download at http://sites.google.com/site/markusbrucknerresearch/research-papers.4

The Nature of the Experiment
The central assumption in our panel data regressions is that, conditional on fixed country characteristics, the hosting and bidding for the Olympic Games is randomly assigned. If this assignment is indeed random, then we have a natural experiment in hand which allows us to examine the causal effects that an exogenous, but anticipated, shock has on the macroeconomy. The International Olympic Committee (IOC) that is responsible for making the decision of which country will host the Olympic Games states: "only rich countries have the means to make a good return on such a large investment [the Olympic Games]." In the cross-section of countries, the random assignment assumption is thus indeed questionable since only countries which are sufficiently developed have the capacity of hosting the Games. We note, however, that all our regressions control for country fixed effects. Hence, whether countries are developed or not is captured by the country dummies. Indeed, when we estimate equation (1) using only OECD countries the results we obtain are very similar. Those results are reported in Table 3 of the Supplementary Online Appendix.
If the assignment of the Olympic Games is not random and depends on the economic growth performance of the bidding and hosting countries, then the ex-ante coefficients on GDP per capita growth should be different for the hosting and bidding countries. As we have already seen in Panel A of We can also directly test the randomness of the assignment by examining whether withincountry variations in GDP per capita growth have a significant effect on the likelihood of bidding for the Olympic Games. To do so, we report in columns (1) and (2) of Table 7 estimates that regress an indicator variable equal to unity during the year when countries bid to host the Olympic Games on the GDP per capita growth of that particular year. We report estimates from a linear probability model in column (1). In column (2), we report estimates from a conditional logit fixed effects model. We note that while the conditional logit fixed effects model takes into account the binary nature of the dependent variable the estimates produced by this model are not marginal effects. 5 The main finding in columns (1) and (2) is that both, the linear probability model and the conditional logit fixed effects model show that countries' GDP per capita growth has an insignificant effect on the likelihood of bidding for the Games. Also, it is noteworthy that in quantitative terms the estimates from the linear probability model imply that, at most, a one percentage point higher GDP per capita growth increases the likelihood of bidding for the Olympic Games by 0.007 percentage points. This is, indeed, a very small effect.
In columns (3) and (4) of Table 7 we repeat the exercise for an indicator variable that is equal to unity during the year when a country was selected to host the Olympic Games. In line with our identifying assumption that conditional on fixed country characteristics the Olympic Games are randomly assigned, we find that economic growth has no significant effect on the likelihood of being selected to host the Games. This result holds for both the linear probability model and the conditional logit fixed effects model. And, it holds both for the contemporaneous effect (Panel A) as well as for possible lagged effects (Panel B).

The Role of Country Characteristics
In Section 3 we found that the announcement of the Olympic Games hosting country generates positive output, investment, private consumption, and government consumption spending responses. An interesting question that we can examine with our panel data is whether the marginal effect of hosting the Olympic Games varies across countries as a function of important features; such as government size, trade openness, political institutions, or geography. These variables have been found to be significant determinants of economic growth in the cross-section of countries. It could, therefore, be also of interest to examine whether these variables induce significant heterogeneity in the marginal effect that hosting the Olympic Games has on output growth across countries. in open economies the size of local demand effects on output is typically smaller than in closed economies.

Summer vs. Winter Games
In the benchmark model we pool the summer and winter Olympic Games together. Since the winter games are a much smaller event constraining the coefficient estimates to be the same across those games might be misleading. 6 For that reason we re-estimated equation (1) separating the Olympics into winter and summer games. The results are presented in Tables 9 and 10. The main finding is that the winter games have a smaller effect than the summer games. Nevertheless, the effects of the winter games are still significantly positive and anticipation is confirmed in both events. Rose and Spiegel (2011) find that the winter games alone have no statistically significant impact on international trade levels. Our findings imply that there is much more than openness to trade to gain by hosting the Olympics.
6 The fact that hosts for the winter games are not chosen at random because one needs a mountain with snow on it to qualify as a host (or even to be considered as a serious bidder) is not a problem in our panel fixed effects estimation. The reason is that during the time period we consider, these geographic characteristics of countries are time invariant variables and, therefore, are captured by the country fixed effects.

Further Robustness Checks
Many economists have shown that stock price movements reflect the market's expectation of future developments in the economy (see e.g. Fama (1990) and Schwert (1990)). Moreover, Beaudry and Portier (2006) use those stock market series to extract news shocks from the data. Given the nature of our exercise and its conclusions it is only natural to ask whether the anticipatory effects for the organization of the Olympic Games are reflected in the movements of the stock market. Using the IFS data on stock price indices, we can examine whether the Olympic bid has significant anticipatory effects on the evolution of stock prices of the hosting and bidding countries. The results that we present in Appendix Table 4 Appendix Table 9. The main finding is that the leads for years t+11 to t+15 are insignificant; both for the bidding and hosting indicator. Hence, this is reassuring evidence that, indeed, our estimation strategy correctly estimates the timing of the anticipatory effect of the Olympic Games.
As a further identification check we repeated the estimation of equation (1) splitting the Olympic Games bidding indicator into: (i) an indicator that is unity for bidding countries that are from the same continent as the host of the previous Olympic Games; (ii) an indicator that is unity for bidding countries that are not from the same continent as the host of the previous Olympic Games. Appendix Table 10 shows that the effects in these two groups of bidding countries are of similar magnitude. In particular, in both groups there is evidence of a significant effect on output growth eight to ten years before the hosting of the Olympic Games. It is therefore not the case that significant anticipatory effects are absent in bidding countries which are from the same continent as the host of the previous Olympic Games.
Our panel estimates reflect an average effect that the Olympic Games had on output growth over the 1950-2009 period. To examine whether the effects of hosting the Olympics have changed during the most recent decades, we re-estimated equation (1) interacting the hosting indicator variable with a dummy for the post-1990 period. One reason for why the effects could be different for the post-1990 period is the end of the Cold War which led to a significant change in international relations.
Other reasons, that could possibly go in opposite directions, include the decline in transportation costs that occurred in the 1980s and 1990s and the increase in competition for hosting the Olympics Games (see Preuss (2004)). Appendix Table 11 shows that the effects of the Olympic Games on output growth are not significantly different for the post-1990 period. Hence, this suggests that the output effects of the Olympics Games are not systematically different for the most recent decades.
In the empirical analysis we have used aggregate data to evaluate the effects of the Olympics.
Some of the countries in our sample are large and the organization of a mega-event in one region can have positive spillover effects to other regions in that country. Hence, our results reflect country-wide average effects. To examine also regional effects of hosting the Olympics we obtained data on total real gross per capita state product for the US from the Bureau of Economic Analysis.  Table 12, hosting the Olympic Games generates increases in GSP growth two years before the actual hosting of the event. GSP growth also increases significantly five years before the hosting of the Olympic Games in unsuccessful bidding countries, thus confirming the presence of anticipation effects (see Panel B). Appendix Table 13 also shows that the delayed effects of hosting the Olympics are considerable and last up to three years after the Games.
At the peak, US states that organized the Olympics saw their gross state product increase by more than three percentage points relative to states that did not undertake the organization of these events. Hence although some caution must be exercised because of the smaller sample, the findings of the baseline analysis are mostly confirmed by the regional data.
A remark regarding standard errors is worthwhile. In all our regressions we clustered standard errors at the country level. The purpose of clustering at the country level is that the computed standard errors are robust to arbitrary within-country serial correlation. To also ensure that our results are robust to arbitrary spatial correlation across countries within any given year (in addition to arbitrary withincountry serial correlation) we report in Appendix Table 14 results that cluster standard errors at the country and year level, using the Cameron et al. (2011) multi-cluster estimator. The main result is that using such a multi-cluster estimator produces standard errors that are very similar to those reported in our baseline regressions.
Finally, we have also repeated our exercise using GDP per worker instead of GDP per capita.
Results are very similar and are presented in Table 15 of the supplementary online appendix.

Conclusions
Starting from Pigou (1926) and Keynes (1936) until the recent work of Beaudry and Portier (2004, 2006, 2007, Jaimovich and Rebelo (2009) and Schmitt-Grohé and Uribe (2012), economists have stressed the importance of expectations in determining the evolution of aggregate macroeconomic data.
However, we are not aware of any empirical work that quantifies such effects directly in the data. In the sense that no study has shown so far how the uncertainty about news affects aggregate macroeconomic outcomes and the effects of its resolution. Most of the existing evidence for the presence of anticipation in macroeconomics is indirect. Anticipation effects present serious challenges to empirical research (see also the discussion in Blanchard et al., 2010). Recent studies on the identification of fiscal shocks have shown that anticipation effects might be crucial for determining the effects of such shocks in the macroecocomy (see e.g. Leeper et al. (2012), Mertens and Ravn (2010), or Ramey (2011)). Beaudry and Portier (2006) estimate the macroeconomic effects of shocks in the stock price index and showed that such shocks cause a boom in the economy. Barsky and Sims (2012), using a different identification strategy contradict their results, raising a debate in the literature on the role of news shocks in aggregate fluctuations. We use the bidding for the Olympic Games as an alternative way to identify news shocks in the data and we confirm the results of Beaudry and Portier (2006). By taking advantage of the time structure in the bidding for and hosting of the Olympic Games we are able to quantify the effects of anticipatory behavior on macroeconomic aggregates. We find that such effects are economically important and statistically significant. News about Olympic Games makes output and investment surge already at the time of the bidding. In unsuccessful bidding countries the agents' optimism turns out to be unjustified and, as a result, the economy returns to its original trend, while hosting economies enjoy quantitatively large and significant positive effects from hosting. The insignificant and quantitatively small cumulative effect for bidding countries that do not win the Games resembles the Pigou (1927) cycles analyzed in Beaudry and Portier (2004) --an initial period of overoptimism in response to submitting a potential winning bid is followed by a corresponding decrease in activity when the positive news does not materialize.
Many studies have claimed that hosting the Olympic Games has no measurable economic effects, but these studies concentrate on the delayed or contemporaneous effect. Our results indicate that the anticipatory and ex-ante effects of hosting the Olympic Games on economic growth are statistically significant and economically important. Hence, hosting of the Olympic Games involves more benefits than an "olive wreath", according to Herodotus.       Note: The method of estimation is least squares. Huber robust standard errors (reported in parentheses) are clustered at the country level. *Significantly different from zero at 90 percent confidence, ** 95 percent confidence, *** 99 percent confidence.  (1) and (3) is least squares; columns (2) and (4) maximum likelihood. Reported coefficients in columns (2) and (4) are from a conditional logit fixed effects regression, and do not represent marginal effects. The dependent variable in columns (1) and (2) is an indicator variable that is unity during the year when countries bid to host the Olympic Games. The dependent variable in columns (3) and (4) is an indicator variable that is unity during the year when the country was selected to host the Olympic Games. *Significantly different from zero at 90 percent confidence, ** 95 percent confidence, *** 99 percent confidence.  Note: The method of estimation is least squares. Huber robust standard errors (reported in parentheses) are clustered at the country level. **Significantly different from zero at 95 percent confidence, *** 99 percent confidence. Note: The method of estimation is least squares. Huber robust standard errors (reported in parentheses) are clustered at the country level. **Significantly different from zero at 95 percent confidence, *** 99 percent confidence.