The Relaionship Beween Consumer Senimen and Sock Prices by Kevin P. Chris Assisan Professor of Economics, Rose-Hulman Insiue of Technology and Dale S. Bremmer Professor of Economics, Rose-Hulman Insiue of Technology To be presened a he Financial Economics Session of he 78 h Annual Conference of he Wesern Economics Associaion Inernaional in Denver, Colorado, on July 15, 2003
The Relaionship Beween Consumer Senimen and Sock Prices I. INTRODUCTON Tesable hypoheses abou he relaionship beween sock prices and consumer confidence seem o appear in he business press whenever new consumer survey daa is released. Headlines like Rise in Consumer Senimen Sends Share Prices Higher 1 and Socks Tumble, Spurred by Dive in Consumer Confidence 2 sugges ha sock prices respond direcly o measures of consumer confidence. Previous research ino he maer, however, suggess ha he headline wriers have i backwards ha he direcion of influence runs only one way, from sock prices o consumer confidence, and ha any sock price response o new informaion abou consumer confidence is ephemeral. This paper focuses on he shor-run and long-run relaionship beween sock indices and measures of consumer senimen, and i presens hree key empirical resuls. Firs, coinegraion es confirm ha here is no long-run relaionship beween differen sock indices and he Universiy of Michigan s measure of consumer confidence. Second, regarding he shor-run relaionship beween sock indices and consumer confidence, Granger-causaliy ess indicae ha sock prices affec consumer confidence, bu consumer confidence does no affec sock prices. Third, while unexpeced changes in consumer confidence have no saisically significan effec on sock prices, expeced changes in consumer confidence are direcly relaed o changes in sock prices. Following his inroducion, he second secion of he paper presens a brief review of he lieraure abou he relaionship beween measures of consumer confidence and oher economic variables. Uni roo ess, coinegraion ess, and Granger causaliy ess are discussed in he 1 New York Times, December 22, 2001. 2 New York Times, Ocober 31, 2001. Page 1
hird secion. The fourh secion of paper presens a model o predic consumer confidence. These predicions and heir errors are used o explain sock prices. The final secion of he paper summarizes he resuls and offers conclusions. II. LITERATURE REVIEW As ineres in he forunes of he sock marke have increased, so oo has ineres in he links beween sock marke indexes and oher indicaors of economic aciviy. Recenly, researchers have urned heir aenion o poenial links beween sock indexes and measures of consumer confidence. Generally, sock indexes and measures of consumer confidence appear o be conemporaneously correlaed, wih he direcion of influence running from sock price movemens o consumer confidence, bu no he oher way (Ooo, 1999; Jansen and Nahuis, 2002). Despie his empirical evidence of causaion, here are reasonable heoreical links beween sock indexes and consumer confidence. Moreover, i is possible ha he direcion of causaion simulaneously works in boh direcions, hereby making deecion of causal relaionships exremely difficul. The wo published measures of consumer confidence in he Unied Saes are he Conference Board Consumer Confidence Index and he Universiy of Michigan s Index of Consumer Senimen (ICS). 3 Of he wo indices, he Universiy of Michigan s index has a longer ime series, and mos academic research has focused on his measure. Because of his longer daa series, he ICS is also he focus of his paper. The ICS is based on a sample of 500 people responding o five differen quesions. Two quesions survey respondens abou heir curren financial siuaion and how hey feel i will change over he nex welve monhs. Two oher quesions ask paricipans abou heir views of fuure business condiions for he counry as a 3 For a comparison of hese wo measures of consumer confidence, see he discussion by Bram and Ludvigson (1998). Page 2
whole. Finally, he las quesion asks respondens wheher i is good ime o purchase a major household appliance. The individual survey quesions and he acual calculaion of he index are described in he paper s appendix. There are wo channels hrough which sock movemens may influence consumer confidence. The firs linkage is he radiional wealh effec, in which movemens in sock indexes ranslae ino changes in curren wealh, hereby influencing consumer senimen direcly. The second channel is he "leading indicaor" linkage, in which consumers inerpre curren changes in sock indexes as reliable indicaors of fuure income changes (Poerba and Samwick, 1995; Morck, Shleifer and Vishny, 1990). Eiher scenario -- changes in curren wealh or anicipaed changes in fuure income --may reasonably be expeced o direcly influence consumer senimen. Using individual observaions from he Universiy of Michigan Consumer Senimen survey and daa from he Wilshire 5000 sock index, Ooo (1999) argues ha he empirical resuls are more consisen wih he view ha households use changes in equiy prices as a leading indicaor. Jansen and Nahuis (2002) exend Ooo s analysis o eleven European counries. Wih few excepions, hey find ha sock reurns and changes in consumer confidence are posiively correlaed. Like Ooo, hey find ha sock prices Granger-cause consumer confidence, bu consumer confidence does no Granger-cause sock prices. Their empirical resuls confirm Ooo s finding ha higher sock prices are a leading indicaor ha increases consumer confidence. Jansen and Nahius characerize his leading indicaor link as he confidence channel, ha is independen of he radiional wealh effec. The empirical resuls of boh Ooo and Jansen and Nahius sugges ha he confidence channel is a separae ransmission mechanism ha is no par of he convenional wealh effec. Page 3
Explanaions of possible causal relaionships beween consumer confidence and equiy prices ha work in he oher direcion are heoreically reasonable, bu empirically unsuppored. Again, here are wo channels of poenial influence. The firs channel is he link beween consumer spending and corporae profis. Several sudies show ha measures of lagged consumer senimen are saisically significan explanaory variables in explaining he behavior of curren household spending. 4 If his is so, hen here should be an indirec link beween consumer senimen and expeced corporae profis, hus providing a link (albei enuous) beween consumer senimen and sock prices. However, he relaionship beween consumer senimen and measures of oupu differs considerably across counries and across he differen measures of consumer confidence. In erms of predicing fuure oupu, measures of consumer confidence have less explanaory power han measures of business confidence. 5 The second poenial channel of influence is he so-called "publicaion effec" (Jansen and Nahuis, 2002), whereby publicaion of consumer survey daa exers a psychological effec on he marke. Such an effec, if i were o occur, is mos likely highly ransiory. 6 III. UNIT ROOT, COINTEGRATION, AND GRANGER CAUSALITY TESTS The daa Three differen sock indices were used o explore he relaionship beween sock prices and measures of consumer senimen. While Ooo s sudy only used he Wilshire 5000 index, his sudy uses hree sock indices ha are regularly repored in he news: he Dow Jones Indusrials, he S&P 500, and he NASDAQ. 4 See Carroll, Fuhrer, and Wilcox (1994), Bram and Ludvigson (1998), and Souleles (2001). 5 See Sanero and Weserland (1996). 6 On Ocober 30, 2002, he Financial Times, in a sory on consumer confidence reaching a 9-year low, repored ha he Dow Jones Indusrial Average iniially fell abou 1% afer he survey daa was released, bu "made up he los ground by he close." Page 4
The index of consumer confidence is generaed by he Survey Research Cener of he Universiy of Michigan. The unemploymen rae is used as an explanaory variable o describe he behavior of consumer senimen. The regression model explaining he behavior of sock prices includes he yield on 10-year U.S. governmen bond rae as an explanaory variable. All of he daa consis of monhly ime series. Wih he excepion of he NASDAQ sock index, he daa sample is 1978:01 2003:01. However, he NASDAQ sock index is only available for he 1984:10 2003:01 sample. 7 Uni roo ess To avoid regressions wih spurious resuls, each ime series is esed for a uni roo. Firs, all he variables are expressed in heir naural logs. Table 1 repors he augmened Dickey-Fuller ess (Dickey and Fuller, 1979) for boh levels daa and firs-differenced daa. Daa plos indicae ha he underlying regressions used o generae he Dickey-Fuller es saisics for level daa should include boh an inercep and a rend variable. On he oher hand, daa plos of he firs differences indicae ha neiher an inercep nor a rend variable is necessary in he regressions ha generae he Dickey-Fuller es for he firs-differenced daa. Each es saisic is derived by including he number of lagged dependen variables ha minimizes he Akaike Informaion Crieria (Akaike, 1973) for each specificaion. Referring o Table 1, he augmened Dickey-Fuller ess indicae ha he null hypohesis of a zero roo canno be rejeced for measures of consumer confidence, he hree sock indices, and he unemploymen rae. The null hypohesis of zero roos is rejeced a he five percen level for he ineres rae on he en-year U.S. governmen bond. Using he firs-differenced daa, he 7 Consumer senimen and he yield on U.S. en-year governmen bonds were downloaded from he Federal Reserve Bank of S. Louis FRED a hp://research.slouisfed.org/fred/. Monhly Dow Jones, S&P 500, and NASDAQ daa were downloaded from hp://finance.yahoo.com/. The Bureau of Labor Saisics web page, hp://www.bls.gov, was he source of he monhly unemploymen rae. Page 5
null hypohesis of a zero roo is rejeced a he one-percen level for every variable. In heir sudy of he relaionship beween consumer confidence and sock indices in Europe, Jansen and Nahuis found ha heir ime series also exhibied zero roos. Coinegraion es Given ha he consumer confidence and sock indices have zero roos, coinegraion ess (Johansen and Juselius, 1990) are preformed o deermine wheher a long-run relaionship exiss beween consumer confidence and each of he sock indices. If consumer confidence and a sock index are coinegraed, ha implies a long-run relaionship beween he wo variables exiss. Three independen saisical ess were performed o deermine wheher consumer confidence and he Dow Jones Indusrial, consumer confidence and he S&P 500, and consumer confidence and he NASDAQ were pair wise coinegraed. The resuls of he coinegraion ess are repored in Table 2. The naural log of he consumer senimen and sock index variables were used in deriving he saisics in Table 2. The es saisics repored in Table 2 are based on he null hypohesis ha a coinegraing vecor beween consumer confidence and a given sock index does no exis. In oher words, he null hypohesis is here is no long-run relaionship beween a measure of consumer senimen and a given index of equiy prices. The race es indicaes ha such a long-run relaionship does no exis. In each of he hree cases, he es saisic was less han he criical value associaed wih a five-percen level of significance. Hence, he null hypohesis of no long-run relaionship canno be rejeced. Likewise, in all hree cases, he max-eigenvalue es saisic was less han he fivepercen criical value. This adds addiional evidence ha he null hypohesis of no long-run relaionship canno be rejeced. The oucomes of hese coinegraion ess are similar o he resuls ha Jansen and Nahuis found wih heir European daa. Page 6
Granger-causaliy ess Given ha here is no long-run saisical relaionship beween consumer confidence and he sock indices, he naure of he shor-run relaionship was explored. Granger-causaliy ess (Granger, 1969) were performed by esimaing a wo-equaion, vecor auoregressive sysem. Le he naural log of he index of consumer confidence in monh be denoed by c, while he naural log of a given sock index in monh is denoed by s. There are wo equaions in he VAR. The firs equaion in he wo-equaion sysem of seemingly unrelaed equaions is N c = δ + δ c + γ s + u 0 i - 1 i - 1 i = 1 i = 1 N, (1) and he second equaion of he VAR is N s = θ + θ c + φ s + e 0 i - 1 i - 1 i = 1 i = 1 N. (2) Noice ha boh equaions (1) and (2) have he same lag srucure, ha is, here are N lagged explanaory variables for boh he sock index in quesion and he consumer senimen. Referring o equaion (1), if γ i = 0 for every i, hen one could conclude ha he sock index does no Granger cause consumer confidence. Likewise, if θ i = 0 for every i, hen consumer confidence does no Granger cause he sock index. The resuls of he Granger causaliy ess are repored in Table 3. The regressions are preformed on firs-differences of he naural logs of he variables o avoid he problems of spurious correlaions caused when regressing ime series wih uni roos on each oher. The lengh of he lag was chosen by picking ha value of N ha minimized he Akaike Informaion Crierion for each VAR. Regardless he sock index used, he Dow Jones, he S&P 500, or he NASDAQ, he resuls were he same. In all hree cases, he null hypohesis ha a given sock index did no Granger cause consumer senimen was rejeced a he one-percen level of Page 7
significance. Likewise, in each of he hree cases, he null hypohesis ha measures of consumer confidence do no Granger cause measures of he sock index could no be rejeced. The finding ha sock indices affec consumer confidence, bu no he reverse, is consisen wih he findings of Ooo and Jansen and Nahius. However, neiher of hese sudies invesigae wheher he impac of expeced changes in consumer confidence and unexpeced changes in consumer confidence differ. If he sock marke is efficien, hen sock prices should reflec only expeced changes in consumer confidence IV. EFFECT OF EXPECTED AND UNEXPECTED CONSUMER CONFIDENCE The ypical approach o esimaing he relaionship beween movemens in sock prices and movemens in measures of consumer confidence is o regress changes in he logs of sock indexes on changes in he logs of consumer confidence indexes. A formal model is s = a 0 + a1 c + a c 2-1 + ε (3) where, as before, s is he naural log of a sock index for he curren period, and c and c - 1 are he naural logs of he conemporaneous and previous period's consumer confidence index. Such regressions usually yield lile predicive value, wih R 2 s of less han 0.10, and saisical significance for only he conemporaneous consumer confidence variable. Our own esimaions of such regressions yielded saisically significan resuls for he coefficien on he conemporaneous consumer confidence variable, and R 2 s ranging beween 0.04 and 0.08. Our alernaive approach is o disinguish beween expeced changes in he consumer confidence index and unexpeced changes. In an efficien marke, movemens in sock prices should only reflec expeced changes in consumer confidence. To invesigae he proposiion ha only expeced changes in he consumer confidence are relaed o sock indexes, i is necessary o firs sele upon a forecasing model for consumer Page 8
confidence. A naive approach would be o assume ha he curren period index is a funcion of is mos recen value. The hird column of Table 4 presens he resuls of esimaing c = α 0 + α1c - 1 + υ. (4) As a baseline forecas, such a model performs well, as indicaed by he high R 2 and low Roo Mean Square Error (RMSE). A more sophisicaed approach would be o assume ha an index of consumer confidence is influenced by economic indicaors ha consumers use o formulae heir personal views regarding he healh of he economy. The las hree columns of Table 4 presen he resuls of such a regression using differen sock indexes: c = α 0 + α2s + α3s - 1 + α4u - 1 + µ (5) where u 1 is he naural log of he previous period's unemploymen rae. As indicaed by he high R 2 s and low RMSEs, hese forecasing models also perform well. The prediced values from regressions such as equaion (4) or equaion (5) are used o forecas expeced changes in consumer confidence, while he residuals of hese regressions serve as proxy for unexpeced changes in consumer confidence. The inuiion behind his approach is ha markes should fully anicipae expeced changes in consumer senimen. Table 5 presens he esimaion resuls of wo differen specificaions. The firs specificaion, he basic model, assumes he change in sock prices is a funcion lagged changes in ineres raes and he acual change in consumer confidence, wih no aemp o disinguish beween expeced and unexpeced changes in consumer confidence. This specificaion is s = a+ 0 a i 1-1 + a2 c + ω (6) where i is he naural log of he yield on he en-year U.S. governmen bond. The second specificaion, he revised model, disaggregaes he change in consumer confidence ino wo Page 9
componens: he expeced change in consumer confidence, consumer senimen, u c. This specificaion becomes e c, and he unexpeced change in s = a+ a i + a c + a c + η. (7) e u 0 1-1 3 4 In each case, he basic model yields resuls ha are similar o hose found in previous research: he coefficien on he change in consumer confidence is saisically significan, bu he overall model yields lile explanaory value as exhibied by very low R 2 s. The revised model has considerably more explanaory power, and he imporance of he expeced change in consumer confidence is en imes greaer han ha for he acual change in consumer confidence in he basic model. Forecased changes in consumer confidence are priced ino he marke. Surprise changes in consumer confidence are no. This is precisely wha one would expec from an efficien marke. V. CONCLUSIONS Similar o he resuls of Jansen and Nahuis, his paper finds ha measures of consumer confidence and sock indices exhibi uni roos. The finding ha no long-run relaionship beween U.S. measures of consumer confidence and sock indices exiss also agrees wih findings of Jansen and Nahius. In erms of he shor-run relaionship beween hese variables, Granger-causaliy es indicae ha measures of he Dow Jones, S&P 500, and he NADAQ affec consumer confidence, bu he opposie does no hold. Jansen and Nahuis found he same resul using European daa, while Ooo found a similar relaionship in he U.S. using he Wilshire 5000 sock index. This paper exends Ooo s findings o he more commonly used indices of he Dow Jones, he S&P 500, and he NASDAQ. However, in a deparure from he sudies of Jansen and Nahius and Ooo, his paper finds ha sock prices reflec expeced changes in consumer confidence. There is no saisically Page 10
significan correlaion beween unexpeced changes in consumer confidence and sock prices. This finding complemens he heoreical conclusions of he efficien markes lieraure. Forecass of expeced changes in consumer confidence are based on commonly available daa ha are also incorporaed in he deerminaion of sock prices. Expeced increases in consumer confidence lead o increases in he demand for sock and higher equiy prices. Our empirical resuls are consisen wih he noion ha unexpeced changes in consumer confidence exhibi no susained correlaion wih sock prices. Page 11
References Akaike, H., Informaion Theory and an Exension of he Maximum Likelihood Principle, 2 nd Inernaional Symposium on Informaion Theory, B. N. Perov and F. Csaki (eds.),. Budapes: Akadémiai Kiadó, 1973. Bram, Jason and Sydney Ludvigsun, Does Consumer Confidence Forecas Household Expendiure? A Senimen Index Horse Race, Federal Reserve Bank of New York Economic Policy Review, June 1998, 4(2), 59-78. Carroll, Chrisopher, Jeffrey Fuhrer, and David Wilcox, Does Consumer Senimen Forecas Household Spending? If So, Why? American Economic Review, December 1994, 84(5), 1397-1408. Dickey, David and Wayne Fuller, Disribuion of he Esimaors for Auoregressive Time Series wih a Uni Roo, Journal of he American Saisical Associaion, June 1979, 74(366), 427-431. Fama, Eugene, Efficien Capial Markes: A Review of Theory and Empirical Work, The Journal of Finance, May 1970, 25(2), 383-417. Granger, C. W. J., Invesigaing Causal Relaions by Economeric Mehods and Cross-Specral Mehods, Economerica, July 1969, 37(3), 424-438. Jansen, W. Jos and Niek J. Nahuis, The Sock Marke and Consumer Confidence: European Evidence, Moneary and Economic Policy Deparmen, De Nederlandsche Bank, July 2002, hp://www.dnb.nl/moneair_beleid/pdf/serie2002-11.pdf. Johansen, Soren and Kaarina Juselius, Maximum Likelihood Esimaion and Inference on Coinegraion - - Wih Applicaions o he Demand for Money, Oxford Bullein of Economics and Saisics, May1990, 52(2), 169-210. Morck, Randall, Andrei Shleifer, and Rober Vishny, The Sock Marke and Invesmen: Is he Marke a Sideshow? Brookings Papers on Economic Aciviy, 1990, 2, 157-202. Ooo, Maria Ward, Consumer Senimen and he Sock Marke, Board of Governors of he Federal Reserve Sysem, November 1999, hp://www.federalreserve.gov/pubs/feds/1999/199960/199960pap.pdf. Poerba, James M. and Andrew A Samwick, Sock Ownership Paerns, Sock Marke Flucuaions, and Consumpion, Brookings Papers on Economic Aciviy, 1995, 2, 295-357. Sanero, Teresa and Niels Weserlund, Confidence Indicaors and Their Relaionship o Changes in Economic Aciviy, Organizaion for Economic Cooperaion and Developmen, Economic Deparmen, Working Papers, No. 170, 1996, hp://www.oecd.org/pdf/m00001000/m00001194.pdf Page 12
Souleles, Nicholas, Consumer Senimen: Is raionaliy and Usefulness in Forecasing Expendiure - - Evidence from he Michigan Micro Daa, Naional Bureau of Economic research Working Paper Series, Working paper 8410, Augus 2001, hp://www.nber.org/papers/w8410. Page 13
Table 1 Augmened Dickey-Fuller Tess for Uni Roos Levels Daa: Regressions Include Inercep and Trend Variable (in naural log) Tes Saisic Lags Sample Consumer Confidence -3.12 0 1978:02 2003:01 Dow Jones Indusrial -2.16 0 1978:02 2003:01 S&P 500-1.36 0 1978:02 2003:01 NASDAQ -1.32 1 1984:12 2003:01 Unemploymen Rae -2.91 6 1978:08 2003:01 10-Year Governmen Bond -3.66** 3 1978:05 2003:01 Firs-Differenced Daa: Regressions Don Include Inercep or Trend Variable (in naural log) Tes Saisic Lags Sample Consumer Confidence -9.06*** 5 1978:08 2003:01 Dow Jones Indusrial -16.91*** 0 1978:03 2203:01 S&P 500-6.67*** 4 1978:07 2003:01 NASDAQ -13.12*** 0 1984:12 2003:01 Unemploymen Rae -4.35*** 5 1978:08 2003:01 10-Year Governmen Bond -8.83*** 2 1978:05 2003:01 *** indicaes he null hypohesis ha he ime series has a uni roo is rejeced a he 1% level, while ** indicaes he same null hypohesis is rejeced a he 5% level. Page 14
Table 2 The Long-Run relaionship beween Consumer Confidence and a Sock Index : Johansen Coinegraion Tes Resuls Null Hypohesis: A coinegraing equaion for he following wo variables does no exis. Pair-Wise Combinaion of Consumer Confidence and Sock Index (1) Consumer Confidence And Dow Jones (2) Consumer Confidence And S&P 500 (3) Consumer Confidence And NASDAQ Sample 1978:04-2003:01 1978:04-2003:01 1985:02-2003:01 Eigenvalue 0.03 0.04 0.06 Number of lags 2 2 3 Akaike Informaion Crieria -6.71-6.74-6.04 Trace Tes Tes Saisic 11.40 13.13 15.74 5% Criical Value 15.41 15.41 25.32 1% Criical Value 20.04 20.04 30.45 Max-Eigenvalue Tes Tes Saisic 10.60 11.28 13.71 5% Criical Value 14.07 14.07 18.96 1% Criical Value 18.63 18.63 23.65 Page 15
Table 3 Pair-wise Granger Causaliy Tess Model A: Model B: N c = δ + δ c + γ s + u 0 i - 1 i - 1 i = 1 i = 1 N N s = θ + θ c + φ s + e 0 i - 1 i - 1 i = 1 i = 1 N All dependen and explanaory variables are firs differences of naural logs. Pair of Variables: Consumer Confidence ( c ) and Dow Jones Indusrial ( s ) Null Hypohesis Tes Saisic Lags Sample Dow Jones Does No Granger Cause Consumer Confidence 15.39*** 2 1978:04 2003:01 Consumer Confidence Does No Granger Cause Dow Jones 0.29 Pair of Variables: Consumer Confidence ( c ) and S&P 500 ( s ) Null Hypohesis Tes Saisic Lags Sample S&P 500 Does No Granger Cause Consumer Confidence 15.38*** 2 1978:04 2003:01 Consumer Confidence Does No Granger Cause S&P 500 0.17 Pair of Variables: Consumer Confidence ( c ) and NASDAQ ( s ) Null Hypohesis Tes Saisic Lags Sample NASDAQ does no Granger Cause Consumer Confidence 14.32*** 2 1985:01 2003:01 Consumer Confidence Does No Granger Cause NASDAQ 0.18 *** indicaes he null hypohesis can be rejeced a he 1% level. Page 16
Table 4 Forecasing models for consumer confidence Regression Parameer Associaed Inercep Or Explanaory Variable Naïve Forecas (1) 1978:02 2003:01 α 0 Inercep 0.198** (0.762) α 1 Lagged dependen variable, c 0.956*** (0.017) α 2 Complex Forecas (2) 1978:03 2003:01 0.977 (0.890) Conemporaneous Dow Jones Index, s 0.181** (0.056) α 3 Lagged Dow Jones Index, s 0.253*** 1 (0.056) α 2 Complex Forecas (3) 1978:03 2003:01 2.082*** (0.639) Conemporaneous S&P 500 Index, s 0.188*** (0.057) α 3 Lagged S&P 500 Index, s 0.232*** 1 (0.057) α 2 Complex Forecas (4) 1984:12 2003:01 3.452*** (0.387) Conemporaneous NASDAQ Index, s 0.074** (0.035) α 3 Lagged NASDAQ Index, s 0.132*** 1 (0.035) α 4 Lagged Unemploymen Rae, u -0.191* 1 (0.094) ρ Auoregressive parameer 0.991*** (0.008) -0.191** (0.094) 0.987*** (0.009) -0.224** (0.098) 0.964*** (0.017) R 2 0.913 0.921 0.921 0.884 DW 1.967 2.139 2.140 2.183 F-es 3,136.56 861.99 857.13 404.59 Forecas RMSE 0.046 0.044 0.044 0.038 Dependen variable: c = naural log of he Universiy of Michigan Consumer Senimen Index. The "naive" model is a simple auoregressive funcion: c= α 0+ α1c -1+ υ. The "complex" models incorporae sock indexes and he unemploymen rae: c= α 0 + α 2s+ α 3s -1+ α 4u -1+ µ. All variables are expressed in naural logs. Sandard errors are in parenheses. *, **, and *** indicae ha he null hypohesis ha he slope coefficien is equal o zero may be rejeced a he 10%, 5%, and 1% level of significance, respecively. These -ess are wo-ail ess. indicaes ha he null hypohesis ha all he slope coefficiens are simulaneously equal zero is rejeced a he 1% level. Page 17
Table 5 Relaionship Beween Expeced and Unexpeced Changes in Consumer Confidence s = change in he naural log of he Dow Jones Indusrial Index Revised model: Regression Basic model Parameer using acual c a 0 0.007** (0.002) a 1-0.221*** (0.062) c and c e u generaed from forecas (2) in Table 4 0.008*** (0.002) -0.123*** (0.043) s = change in he naural log of he S&P 500 Index Revised model Basic model using acual c 0.007** (0.002) -0.186*** (0.062) c and c e u generaed from forecas (3) in Table 4 0.008*** (0.002) -0.090** (0.043) s = change in he naural log of he NASDAQ Composie Index Revised model Basic model using acual c 0.001 (0.005) -0.148 (0.131) c and c e u generaed from forecas (4) in Table 4 0.008** (0.004) -0.150 (0.105) a 2 0.190*** (0.057) 0.195*** (0.056) 0.316** (0.127) a 3 1.820*** (0.125) 1.910*** (0.131) 2.936*** (0.336) a 4-0.017 (0.045) -0.012 (0.045) 0.034 (0.117) ρ -0.076 (0.060) -0.414*** (0.059) -0.072 (0.060) -0.373*** (0.060) 0.045 (0.071) -0.259*** (0.077) R 2 0.076 0.417 0.067 0.422 0.042 0.252 DW 2.002 2.158 2.012 2.130 1.992 2.054 F-es 8.04 52.29 7.02 53.37 3.09 17.79 Sample 1978:04 2003:01 1978:04 2003:01 1978:04 2003:01 1978:04 2003:01 1894:12 2003:01 1894:12 2003:01 e u The basic regression model is s =a+ 0 a i 1-1+ a2 c + ω. The revised regression model is s = a+ 0 a i 1-1+ a3 c + a c 4 + η. *, **, and *** indicae ha he null hypohesis ha he slope coefficien is equal o zero may be rejeced a he 10%, 5%, and 1% level of significance, respecively. These -ess are wo-ail ess. indicaes ha he null hypohesis ha all he slope coefficiens are simulaneously equal zero is rejeced a he 1% level. Page 18
Appendix The Index of Consumer Senimen (ICS) Calculaed by he Survey Research Cener, Universiy of Michigan x 1 x 2 x 3 x 4 x 5 The ICS is derived from he responses o he following five quesions. We are ineresed in how people are geing along financially hese days. Would you say ha you (and your family living here) are beer off or worse off financially han you were a year ago? Now looking ahead - - do you hink ha a year from now you (and your family living here) will be beer off financially, or worse off, or jus abou he same as now? Now urning o business condiions in he counry as a whole - - do you hink ha during he nex welve monhs we ll have good imes financially, or bad imes, or wha? Looking ahead, which would you say is more likely - - ha in he counry as a whole we ll have coninuous good imes during he nex five years or so, or ha we will have periods of widespread unemploymen or depression, or wha? Abou he big hings people buy for heir homes - - such as furniure, a refrigeraor, sove, elevision, and hings like ha. Generally speaking, do you hink now is a good or bad ime for people o buy major household iems? x i is equal o he percen of favorable replies minus percen of unfavorable replies plus 100, rounded o neares whole number. x 1 + x 2 + x 3 + x 4 + x5 ICS = + 2.0 6.7558 The denominaor, 6.7558, is he 1966 base period oal, and he addiion of 2.0 correcs for survey design changes ha occurred in he 1950s. Source: Survey Research Cener, Universiy of Michigan, hp://www.sca.isr.umich.edu. Page 19