RESEARCH NOTE: CONTRARY EVIDENCE ON THE ECONOMIC IMPACT OF THE SUPER BOWL ON THE VICTORIOUS CITY

RESEARCH NOTE: CONTRARY EVIDENCE ON THE ECONOMIC IMPACT OF THE SUPER BOWL ON THE VICTORIOUS CITY Draft: 3/15/2003 The results presented n ths paper should be seen as prelmnary n nature. Please do not cte or quote wthout the author s permsson Vctor A. Matheson Department of Economcs Wllams College Fernald House Wllamstown, MA 01267 (413) 597-2144 (phone) (413) 597-4045 (fax) Vctor.A.Matheson@wllams.edu Abstract: Prevous research has ndcated a surprsng statstcally sgnfcant mpact on the local economy n the cty that wns the Super Bowl. An analyss of personal ncome growth n vctorous Super Bowl ctes from 1972-2000 cannot further confrm these results, fndng no statstcally sgnfcant effect on the local economes of these ctes. JEL Classfcaton Codes: L83 - Sports; Gamblng; Recreaton; Toursm; R53 - Publc Faclty Locaton Analyss; Publc Investment and Captal Stock

Introducton Professonal sports are bg busness n the Unted States. If the popular meda s to be beleved, the champonshp games of the major professonal leagues such as Major League Soccer s MLS Cup, the Natonal Basketball Assocaton (NBA) and Natonal Hockey League (NHL) Fnals, Major League Baseball s (MLB) World Seres, and the Natonal Football League s (NFL) Super Bowl generate huge profts for the metropoltan areas lucky enough to host these events. Numerous scholars have attempted to estmate the mpact of large sportng events and league champonshps on host ctes. Baade and Matheson (2001) use employment and taxable sales data to fnd the effect of MLB s All-Star Game metropoltan areas. Porter (1999) uses taxable sales data determne the mplcatons of hostng the Super Bowl for host ctes. Baade and Matheson (2003a, 2003b) use metropoltan area personal ncome data to estmate the mpacts of MLB s post-season and the Super Bowl on local economes. In all cases, the economc consequences of hostng these mega-events are statstcally nsgnfcant and certanly much smaller than the fgures quoted by league and team boosters. The prevalng opnon among economsts s that whle these sportng events may be large n a gross sense, because of crowdng out, leakages, and substtuton effects, the net nfluence on the host cty s small. For further dscusson see Segfred and Zmbalst (2000) among others. The one excepton to ths s rule s Coates and Humphreys (2002). Ther examnaton of post-season play n the NFL, NBA, and MLB, smlar to all of the prevous studes, fnds that the ctes hostng post-season play experence no sgnfcant ncrease n real per capta personal ncome. In a very surprsng dscovery, however, they found that over the tme perod of ther sample, 1969-1997, the cty wnnng the Super Bowl experenced a statstcally sgnfcant ncrease of roughly 1

$140 n per capta ncome. Ths result s partcularly surprsng consderng that the Super Bowl, unlke the champonshps n the other major professonal sports, hockey, basketball, and baseball, s held at a pre-determned neutral ste rather than at one of the partcpants home felds. Therefore, whle one mght predct that the economes n the ctes of the other sports champons wll be nfluenced by the economc actvty surroundng the actual game(s), n the case of the Super Bowl, the wnner s home town receves no drect revenue from the team s bg vctory snce the wn wll lkely take place thousands of mles away. In fact, no Super Bowl champon has ever won the bg game n ther own home stadum. Furthermore, because of the sngle-game elmnaton playoff system, t s qute possble that the wnnng team may never have played even a sngle post-season game at home. Coates and Humphreys offer several reasons for ther fndng. They offer that the result could smply be an anomaly or the result of model ms-specfcaton or an omtted varable problem. Indeed, wth 28 ndependent varables n ther matrx of sports envronment varables, one mght expect at least one case of spurous correlaton just based on the law of averages. They argue, however, that ther fndng s reasonably robust to the ncluson of a varety of addtonal explanatory varables and alternatve functonal forms. Instead they propose that the ncrease s possbly the result of labor productvty ncreases n the wnnng cty. If wnnng the Super Bowl has a stmulatng effect on the productvty of the fans of the wnnng team, then the value of margnal product of these workers would ncrease as would the wage bll and ncome of these workers. Ths could possbly lead to an ncrease n real per capta ncome n a cty for a short perod of tme. (Coates and Humphreys, 2002, pg. 298) It s curous that only football, and not professonal basketball or baseball leads to ths 2

ncrease n productvty, but ths s explaned by the authors by the fact the t s truly football and not any of the other sports that truly captures the heart and soul of a cty and, ndeed, the world. Some mght even argue that wnnng champonshps mght reduce productvty snce the resultng parades and celebratons often result n busness closures and because of the unfortunate crcumstance that fan rotng after sportng champonshps has become ncreasng commonplace n the Unted States. (Baade and Matheson, 2003c) The questons remans, however, how to explan ths curous result. In the next secton, an alternatve model to answer ths rddle s proposed. The paper ends wth results and conclusons. Data and Methodology The economc actvty generated by or related to any sportng event s lkely to be small relatve to the overall economy, and solatng the event s mpact, therefore, s not a trval task. The approach used here s an ex post examnaton of personal ncome growth n ndvdual metropoltan statstcal areas (MSA) and s dentcal to that used by Baade and Matheson (2003a, 2003b). Explanatory varables dentfed from past models are used to help establsh what personal ncome growth would have been n the absence of wnnng the Super Bowl, and then these predctons are compared to actual ncome growth rates to assess the contrbuton of the wn to the local economy. The success of ths approach, of course, depends on the ablty to dentfy those varables that explan the majorty of observed varaton n growth n personal ncome n those ctes that have won the Super Bowl. Equaton (1) represents the model used to predct changes n ncome for wnnng ctes. n t β β β β t-1 t = 0+ Y Y 1 + 2 Y t-1 + 3 + β t β t β t β n 4W + 5T + 6TR + OTt + εt (1) =1 nt Y t-1 / n = 1 3 Y 7

For each tme perod t, Y t s the real ncome and Y t s the change n real personal ncome n the th MSA, n s the number of ctes n the sample, W t s the nomnal wages n the th MSA as a percentage of the average for all ctes n the sample, T t s the state and local taxes n the th MSA as a percentage of the average for all ctes n the sample, TR t s an annual trend varable, and ε t s the stochastc error. OT t s a dummy varable used n certan ctes regresson equatons to specfy cty-specfc events such the sgnfcant economc nfluence of Hurrcane Andrew on the economy of Mam and the effect of the ol boom and bust on ol patch ctes such as Denver and Dallas. The cohort of ctes used n the sample ncludes the seventy-three largest MSAs n the Unted States by populaton over the tme perod 1969-2000. Personal ncome data from 1969-2000 were obtaned from the Regonal Economc Informaton System at the Unversty of Vrgna, whch derves ts data from the Department of Commerce statstcs. Data regardng state and local taxes as a percentage of state GDP were avalable for all ctes from 1970 to 2000 and were obtaned from the Tax Foundaton n Washngton, D.C. Manufacturng wage data from the Bureau of Labor Statstcs Current Employment Statstcs Survey were avalable for all ctes over varyng tme perods. A complete descrpton of the data s avalable from the author upon request. For the purposes of ths analyss, the functonal form s lnear n all the varables ncluded n equaton (1). The equaton was estmated for 14 dfferent metropoltan areas representng all of the ctes that have won at least one Super Bowl snce 1969. Not every varable specfed n equaton (1) emerged as statstcally sgnfcant for every cty. The decson of whether to nclude an ndependent varable known to be a good predctor n general but falng to be 4

statstcally sgnfcant n a partcular cty s case s largely an arbtrary one. The ncluson of theoretcally valuable varables that are dosyncratcally nsgnfcant wll mprove some measures of ft such as R-squared but may reduce other measures such as adjusted R-squared or the standard error of the estmate. Snce the purpose of equaton (1) s to produce predctve rather than explanatory results, varables were ncluded n the regresson equaton as long as they mproved predctve success. Table 1 presents the regresson results for all ctes wth the combnaton of varables that mnmzes the standard error of the estmate (SEE). For about half of the ctes, autocorrelaton was dentfed as a sgnfcant problem, and, therefore, the Cochrane-Orcutt method was used for ctes where ts use agan reduced the SEE. Results The model dentfed n Table 1 for each cty s used to estmate ncome growth for each cty for each year that data are avalable, 1969-2000. Cty-specfc wage data are not avalable for all ctes pror to 1972, and therefore only the Super Bowl wnners snce 1972 are examned. Ths leaves 12 ctes and 29 champonshps n the data set. Once ncome growth s estmated by the model, the predcted ncome growth s then compared to the actual ncome growth that each MSA experenced durng the year(s) n whch t won the Super Bowl. If t s assumed that any dfference between actual and predcted ncome can be accounted for by wnnng the Super Bowl, ths method allows for an estmate of the mpact of the game on vctorous ctes. Table 2 shows the wnnng cty, predcted growth, estmated growth, the dfference between predcted and actual growth (the resdual), and the standardzed resdual. In an effort to compare these results n a meanngful way to the results of 5

Coates and Humprheys, the real per capta ncome (n 2000 dollars) and the change n the per capta ncome above or below predctons are shown. It s mportant to note that ths model was not desgned to drectly calculate changes n per capta personal ncome. Instead changes n per capta ncome are nferred from the change n total personal ncome, and so t must be assumed that wnnng the Super Bowl has no effect on populaton growth n an MSA for these fgures to have valdty. If a wn n the Super Bowl spurs mgraton nto the cty, then the fgures n Table 2 actually overstate the ncrease n real per capta ncome. Only f the Super Bowl wn nduces resdents to flee the cty over the course of the year, a hghly unlkely scenaro, do the calculatons understate the true per capta ncome gans. The statstcs recorded n Table 2 suggest two thngs worth notng. Frst, the dollar dfferences recorded n the fnal column vary substantally wth some ctes exhbtng per capta ncome gans nearly $1,000 n excess of model predctons, and other ctes showng a large negatve mpact. Second, actual and predcted growth on average are almost exactly the same wth actual ncome growth exceedng predcted growth by 0.169 percent. Agan assumng no effect from the Super Bowl on populaton growth, real per capta ncome rose by just under $50 on average, a fgure roughly one-thrd that estmated by Coates and Humphreys. The magntude of the varaton of the estmates at frst blush appears hgh. The explanaton for ths range of estmates s smply that the models do not explan all the varaton n estmated ncome, and, therefore, not all the varaton can be attrbutable to wnnng the Super Bowl. In short, there are omtted varables. Whle the model ft statstcs for the ndvdual cty regressons dsplay moderately hgh R-squared numbers, the standard error of the estmate for the typcal cty s roughly one percent meanng that one would expect the models to predct 6

actual economc growth for the ctes wthn one percentage pont about two-thrds of the tme. Gven the sze of these large, dverse economes, the effect of even a large event wth hundreds of mllons of dollars of potental mpact s lkely to be obscured by natural, unexplaned varatons n the economy. Indeed, none of the standardzed resduals are statstcally sgnfcant at the 5% level. Whle t s unlkely that the models for any ndvdual cty wll capture the effects of even a potentally large occason lke wnnng the Super Bowl, one would expect that across a large number of ctes and years, any event that produces a bg mpact would emerge on average as statstcally sgnfcant. Usng the seemngly unrelated regressons approach, one can compare the standardzed resduals for the 29 observatons wth resduals beng normally dstrbuted wth a standard devaton of 1. A test on the null hypothess that the average standard resdual s dfferent than zero provdes a p-value of 39.5 percent, well outsde any range of statstcal sgnfcance. As detaled n Baade and Matheson (2003a, 2003b), the seemngly unrelated regressons approach can be taken one step further by ncorporatng assumed economc mpacts nto the model predctons. Usng these technques, t s found that per capta ncome gans of $123.35, $145.34, and $188.60 can be rejected at the 10%, 5%, and 1% sgnfcance level respectvely. Conclusons An analyss of personal ncome growth n the year followng a wn n the Super Bowl n the wnnng team s home cty shows a slght ncrease n personal ncome growth but one that s not statstcally dfferent than zero. Ths contradcts recent fndngs by Coates and Humphreys 7

(2002) that seemed to show a sgnfcant gan n per capta ncomes n wnnng ctes. Ther plausble explanatons for the gan ncluded both the possblty that the fndng was spurous and a purported productvty ncrease due to jublant fans n the cty. The fndng n ths paper supports ther frst concluson that the fndng was purely an anomaly. Obvously, addtonal research wll be requred to determne whch set of results truly reflects the real mpact of wnnng sports champonshps on local economes. Based on ths study, however, the mpact appears to be small or non-exstent. Whle Caesar may thought that the way to keep ctzens happy and productve was to provde bread and crcus, wnnng the Super Bowl does not seem to be ths magc tcket to rches. 8

REFERENCES Baade, Robert A. and Vctor A. Matheson. (2001). Home Run or Wld Ptch? Assessng the Economc Impact of Major League Baseball s All-Star Game. Journal of Sports Economcs, 2(4), 307-327. Baade, Robert A. and Vctor A. Matheson. (2003a). Super Bowl or Super (Hyper) Bole? Assessng the Economc Impact of Amerca s Premer Sportng Event. Workng paper. Baade, Robert A. and Vctor A. Matheson. (2003b). Assessng the Impact of Major League Baseball s Post-Season. Workng paper. Baade, Robert A. and Vctor A. Matheson. (2003c) The Paradox of Champonshps. Workng paper. Coates, Denns and Brad Humphreys. (2002). The Economc Impact of Post-Season Play n Professonal Sports. Journal of Sports Economcs, 3(3), 291-299. Porter, Phlp. (1999). Mega-Sports Events as Muncpal Investments: A Crtque of Impact Analyss. In Fzel, J., Gustafson, E. & Hadley, L. Sports Economcs: Current Research. Westport, CT: Praeger Press. Segfred, John and Andrew Zmbalst. (2000). The Economcs of Sports Facltes and Ther Communtes, Journal of Economc Perspectves, Summer 2000, 95-114. 9

TABLE 1 Regresson results for Equaton 1 all varables ncluded that mnmze SEE. (t-stats n parentheses) MSA Cons. Avg. Y t Y t-1 Inc. Wages Taxes Tme Other Ft Chcago.343 (4.94).961 (12.00) - -.073 (-1.13).068 (1.32) -.348 (-4.90) - - Adj. R 2 =.8954 SEE = 0.7958% Dallas -2.449 (-2.68).980 (8.47) - - - -.280 (-3.34).0014 (2.81) -.0151 (-1.79) Adj. R 2 =.7730 SEE = 1.1578% ** Denver 1.947 (1.44).918 (-1.32).152 (1.25) -.122 (-1.29) - -.225 (-1.95) -.0008 (-1.32) -.0377 (-4.69) Adj. R 2 =.7549 SEE = 1.3175% Green Bay.447 (2.79).847 (9.68).053 (0.60) -.482 (-2.78) - - - ** Adj. R 2 =.7769 SEE = 0.9159% Los Angeles 10.81 (2.55) 1.032 (8.67) -.071 (-0.78) -.530 (-2.46) - - -.0052 (-2.54) ** Adj. R 2 =.7831 SEE = 1.2470% Mam 10.44 (2.65).809 (5.12).213 (2.40) -.576 (-2.55).331 (2.16) - -.0051 (-2.67) -.0871 (-6.02) Adj. R 2 =.8646 SEE = 1.3471%.0845 (3.35) New York Cty -7.601 (-4.88) Oakland -.210 (-1.36) Pttsburgh 2.078 (3.08) Sant Lous San Francsco Wash., D.C..613 (3.32) -0.486 (-5.27) -1.401 (-0.33) 1.018 (8.07).882 (7.68).613 (7.10).973 (17.16).853 (5.40).653 (6.11) -.249 (-2.39) -.421 (-3.69) -.361 (2.80).235 (2.34).248 (2.16) - -.509 (-2.87) -.338 (5.29).265 (1.87) -.202 (-4.06) -.169 (-2.57) - - -.111 (-1.90) - - ** Adj. R 2 =.7374 SEE = 1.265%.049 (1.34) -.150 (-1.84) - -.138 (-2.88) - - Adj. R 2 =.7678 SEE = 1.1782% -.0010 (-2.89) - Adj. R 2 =.7461 SEE = 0.8575% - ** Adj. R 2 =.9134 SEE = 0.5552% - - - - Adj. R 2 =.6691 SEE = 1.714%.049 (1.41).0007 (1.04) ** Adj. R 2 =.7244 SEE = 0.9581% OLS regresson used n all cases except those noted by **. The Cochrane-Orcutt method was used n these cases where the elmnaton of seral correlaton mproved model ft as measured by the SEE. 10

TABLE 2 Year SB Wnner Real per Capta Income Predcted Growth Actual Growth Dfference Standard Resdual Per Capta +/- SEE 1972 Dallas $ 20,420 7.614% 7.669% 0.055% 0.047 $ 11.20 1.158% 1973 Mam $ 22,989 6.852% 8.358% 1.506% 1.118 $ 346.23 1.347% 1974 Mam $ 22,309 0.683% -0.267% -0.950% -0.705 $ (211.98) 1.347% 1975 Pttsburgh $ 20,281-0.372% 0.425% 0.797% 0.929 $ 161.65 0.858% 1976 Pttsburgh $ 21,127 3.304% 3.897% 0.594% 0.692 $ 125.44 0.858% 1977 Oakland $ 25,547 5.124% 3.915% -1.209% -1.026 $ (308.85) 1.178% 1978 Dallas $ 24,316 7.309% 9.219% 1.910% 1.650 $ 464.49 1.158% 1979 Pttsburgh $ 22,779 0.042% -0.137% -0.179% -0.209 $ (40.75) 0.858% 1980 Pttsburgh $ 22,254-2.051% -2.791% -0.741% -0.864 $ (164.83) 0.858% 1981 Oakland $ 26,751 1.766% 1.753% -0.013% -0.011 $ (3.50) 1.178% 1982 San Francsco $ 33,652 1.380% 0.702% -0.678% -0.396 $ (228.09) 1.713% 1983 Washngton $ 28,386 4.670% 5.261% 0.591% 0.617 $ 167.78 0.958% 1984 Los Angeles $ 26,263 6.005% 6.366% 0.361% 0.290 $ 94.84 1.247% 1985 San Francsco $ 36,935 4.750% 2.450% -2.300% -1.342 $ (849.57) 1.713% 1986 Chcago $ 27,972 4.315% 3.951% -0.364% -0.458 $ (101.85) 0.796% 1987 New York $ 30,979 3.181% 3.820% 0.639% 0.505 $ 197.99 1.265% 1988 Washngton $ 34,775 5.876% 6.290% 0.414% 0.433 $ 144.12 0.958% 1989 San Francsco $ 40,650 2.854% 2.320% -0.534% -0.312 $ (217.20) 1.713% 1990 San Francsco $ 41,426 1.104% 2.102% 0.998% 0.583 $ 413.56 1.713% 1991 New York $ 33,253-1.395% -1.636% -0.241% -0.191 $ (80.19) 1.265% 1992 Washngton $ 34,957 2.593% 2.251% -0.342% -0.357 $ (119.44) 0.958% 1993 Dallas $ 28,708 2.567% 2.920% 0.353% 0.305 $ 101.43 1.158% 1994 Dallas $ 29,247 3.972% 4.414% 0.442% 0.382 $ 129.37 1.158% 1995 San Francsco $ 41,839 2.526% 4.772% 2.246% 1.311 $ 939.71 1.713% 1996 Dallas $ 30,468 4.390% 5.770% 1.380% 1.192 $ 420.46 1.158% 1997 Green Bay $ 27,882 3.901% 4.953% 1.052% 1.148 $ 293.21 0.916% 1998 Denver $ 34,355 8.067% 7.266% -0.801% -0.608 $ (275.19) 1.318% 1999 Denver $ 35,408 5.938% 5.877% -0.061% -0.046 $ (21.46) 1.318% 2000 St. Lous $ 31,313 2.004% 1.990% -0.014% -0.025 $ (4.38) 0.555% Average 3.413% 3.582% 0.169% 0.160 $ 47.73 T-stat = 0.864 P-value = 39.5% 11