13
This article was downloaded by: [USC University of Southern California] On: 09 November 2014, At: 22:39 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK Applied Financial Economics Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/rafe20 Consumer confidence announcements: do they matter? O. David Gulley & Jahangir Sultan Published online: 07 Oct 2010. To cite this article: O. David Gulley & Jahangir Sultan (1998) Consumer confidence announcements: do they matter?, Applied Financial Economics, 8:2, 155-166, DOI: 10.1080/096031098333131 To link to this article: http://dx.doi.org/10.1080/096031098333131 PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions

Consumer confidence announcements: do they matter?

Embed Size (px)

Citation preview

Page 1: Consumer confidence announcements: do they matter?

This article was downloaded by: [USC University of Southern California]On: 09 November 2014, At: 22:39Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: MortimerHouse, 37-41 Mortimer Street, London W1T 3JH, UK

Applied Financial EconomicsPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/rafe20

Consumer confidence announcements: do theymatter?O. David Gulley & Jahangir SultanPublished online: 07 Oct 2010.

To cite this article: O. David Gulley & Jahangir Sultan (1998) Consumer confidence announcements: do they matter?,Applied Financial Economics, 8:2, 155-166, DOI: 10.1080/096031098333131

To link to this article: http://dx.doi.org/10.1080/096031098333131

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purposeof the Content. Any opinions and views expressed in this publication are the opinions and views of theauthors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should notbe relied upon and should be independently verified with primary sources of information. Taylor and Francisshall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, andother liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relationto or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Page 2: Consumer confidence announcements: do they matter?

Applied Financial Economics, 1998, 8, 155 Ð 166

Consumer con® dence announcements:do they matter?

O. DAVID GULLEY* and JAHANGIR SULTAN §

* Department of Economics and § Department of Finance, Bentley College, Waltham,MA 02154, USA

This paper examines the response of ® nancial markets to consumer con® denceannouncements during 1980 Ð 93. Several hypotheses are tested to examine the impactof consumer con® dence announcements on the conditional mean and the conditionalvolatility of stock, bond and foreign exchange prices. Despite a plethora of causalempiricism in the popular press, consumer con® dence appears to in¯ uence only theDow Jones Industrial Average, and not bond or other stock indexes. However,changes in the consumer con® dence index are found to have asymmetric e� ects on thedollar exchange rates of ® ve major currencies. Finally, we ® nd that the impact of theconsumer con® dence index on the conditional volatility is not uniform across ® vemajor currencies.

I . INTRODUCTION

There is growing interest among academics and the ® nan-cial press on the impact of the Consumer Con® dence Indexand the Consumer Sentiment Index on various economicand ® nancial variables. Carroll et al. (1994) ® nd that pastvalues of the University of Michigan Consumer SentimentIndex can help explain current changes in consumer spend-ing. Using the same index, Throop (1992) ® nds thatexogenous changes in the index (caused by the Gulf War, forexample) can help forecast expenditures on consumerdurables. Garner (1991) reaches a similar conclusion withregard to the Conference Board’s Consumer Con® denceIndex (CCI). The latter two authors argue that underordinary economic circumstances, consumer con® dence/sentiment measures add little information to forecasts ofconsumer durables expenditures and none at all to forecastsof other consumer expenditure categories. Fuhrer (1993)employs both indexes and ® nds that their addition to vectorautoregression (VAR) models of various consumption com-ponents adds little to the forecast accuracy of the models.

1 The Wall Street Journal, 2-26-1993, p. C19.2 The Wall Street Journal, 2-26-1992, p. C6.3 The Wall Street Journal, 7-29-92, p. C6.4 New Y ork Times, 7-29-92, p. D1.

The reasoning behind these preceding ® ndings is thatunder ordinary economic circumstances, consumer’s viewsof future economic conditions are highly correlated withcurrent economic conditions, which are well measured byvariables other than the CCI, such as the unemploymentrate. Huth et al. (1994) compare the predictive powers of theUniversity of Michigan Consumer Sentiment Index and theCCI in a VAR framework. They ® nd that while the Senti-ment Index outperforms the CCI in predicting durablesexpenditures, the CCI does a better job predicting move-ments in variables such as the unemployment rate and theDow Jones Industrial Average.

The media’s infatuation with the consumer con® dence isevident in headlines that portray a causal association be-tween the CCI and ® nancial market responses. For instance,headlines such as: Bonds rally after reports shows con® -dence of consumers fell to 17-year low this month,1 Nasdaqindex declines 0.56% in wake of negative consumer-con® -dence data,2 Nasdaq index surges 1.2% as report onconsumer con® dence sparks rally,3 Markets jump asconsumer con® dence declines,4 and Bonds fall as economy

0960 Ð 3107 Ó 1998 Routledge 155

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a] a

t 22:

39 0

9 N

ovem

ber

2014

Page 3: Consumer confidence announcements: do they matter?

strengthens,5 imply that the CCI a� ects ® nancial marketparticipants’ expectations of the future course of the eco-nomy. In an intertemporal equilibrium model, suchexpectations should a� ect the investor’s choice betweenconsumption and investment.

However, the assertion made by the popular press thatconsumer con® dence a� ects stock, bond, and foreign ex-change prices has not been subjected to rigorous empiricalveri® cation. No researcher has examined the impact of CCIannouncements on asset prices. This paper uses a GARCHmodel to assess the extent to which positive and negativechanges in consumer con® dence a� ect both the level andvolatility of stock, bond and foreign exchange prices.GARCH techniques are suited for modelling high frequency® nancial market data that are non-normal and lepto-kurtotic. In the present context, CCI announcements arehypothesized to be leading causes for such non-linear returndistributions.

Financial market response to the CCI o� ers a usefulframework of analysis for several reasons. First, marketreaction to both positive and negative news about the eco-nomy may be indicative of how investors revise their esti-mates of future returns. Such revisions may be the reason forchanges in the risk of assets sensitive to announcements ofthe CCI. In the context of a time varying asset pricingmodel, it is possible to examine if changes in asset prices andrisk are due to the way market participants react to news.Second, given the extent of increasing capital market inte-gration among countries, it is important to know whether ornot the e� ects of US consumer con® dence are transmitted tothe global ® nancial market. If the CCI is viewed as a world-wide factor, then the risk due to general economic condi-tions should be also priced in the global stock index as wellas foreign exchange rates of the US dollar. With respect tothe global stock index, the evidence of country-speci ® c riskin the absence of barriers to capital ¯ ows is consistent withthe notion that when country-speci ® c risk becomes a part ofglobal ® nancial risk, such risk cannot be completely diversi-® ed away.

Empirical evidence presented below suggests that suchannouncements have asymmetric e� ects on the daily returnand conditional volatility of the Dow Jones Industrial Aver-age, but not on bond or other broader stock indexes. Fur-ther, while we ® nd that decreases in the CCI are associatedwith the depreciation of the dollar against various curren-cies, increases in the CCI are not associated with any statist-ically signi® cant change in key exchange rates. Finally,increases in the CCI are associated with a decrease in the

5 The Wall Street Journal, 12-1-93, p. C1. The article refers to a strong increase in the CCI, among other factors, in¯ uencing bond prices.6 An interesting question arises: why should consumer con® dence announcements a� ect the prices of ® nancial assets when there is littleevidence that the CCI can add to predictions of economic variables? The answer may be in the timing of the release of the information onconsumer con® dence. For example, the September 1993 value of the CCI was released on 28 September. However, information on otherSeptember variables appears later Ð the September 1993 unemployment rate was released on 8 October. To the extent that CCIannouncements provide advance information about other economic variables, such announcements can a� ect asset prices.7 Hardouvelis (1988) suggests that trade balances are autocorrelated. If so, current balances provide information on future balances.

conditional volatility of the $/DM exchange rate, whiledecreases in the CCI are associated with a decrease in theconditional volatility of the $/BP rate.

Overall, the link between CCI announcements and ® nan-cial markets is important for increasing our understandingof the global economic and ® nancial integration currently inthe making.

II . REVIEW OF THE LITERATURE ANDMOTIVATION

While the ® nancial press o� ers anecdotal evidence thatconsumer con® dence announcements a� ect prices in ® nan-cial markets, a more solid theoretical basis for such e� ects isneeded. Garner (1991) and Throop (1992) describe howchanges in consumer psychology, as measured by changesin indexes of consumer sentiment/con® dence, should berelated to economic variables, most notably consumerspending on durables.6 As such spending is an importantcomponent of GDP, the level and volatility of stock pricescould be in¯ uenced by information about future consump-tion spending. Further, to the extent that consumer con® -dence is able to predict spending, such information wouldbe relevant to predicting both the real interest rate andexpected in¯ ation components of nominal interest rates.Since the CCI is a leading economic indicator, it re¯ ectsconsumers’ sentiments about the state of the economy. If theCCI registers a drop, this implies that consumers’ expecta-tions about the economy are worsening. This, in turn, im-plies falling interest rates, which may be lowered further bythe Federal Reserve in an attempt to induce an economicrecovery. Thus, one should expect to see a negative relation-ship between the CCI and the bond market. The link be-tween the CCI and stock prices is less clear. On the onehand, the CCI and stock prices could be positively corre-lated if the CCI is a predictor of future economic activity.However, to the extent that bonds and stocks are sub-stitutes, lower interest rates (associated with sluggisheconomic activity and low levels of the CCI) may induceinvestors to move out of bonds and into stocks.

Further, if consumer con® dence provides informationabout future economic activity, then it should also provideinformation about future trade balances.7 In turn, informa-tion on future trade balances yields information about thefuture (and therefore, current) foreign exchange value of thedollar. Thus, CCI announcements may in¯ uence the spotforeign exchange rate of the dollar. Speci® cally, an increase

156 O. D. Gulley and J. SultanD

ownl

oade

d by

[U

SC U

nive

rsity

of

Sout

hern

Cal

ifor

nia]

at 2

2:39

09

Nov

embe

r 20

14

Page 4: Consumer confidence announcements: do they matter?

in the CCI is expected to be associated with a dollar ap-preciation, which is consistent with the notion that strongeconomies tend to have a strong currency.

While our study is cast in the context of previous studiesthat relate news to asset prices,8 it makes several contribu-tions to the literature. First the extant literature does notexamine how CCI announcements a� ect prices of ® nancialassets. Given the attention that such announcements havereceived in the popular press, it is worthwhile to examine theextent, if any, to which consumer con® dence in¯ uences assetprices.

Second, this study examines the impact of news on boththe mean (® rst moment) and the variance (second moment)of asset price changes. The evidence of a link betweeninformation and the variance of an asset’s price is alsohelpful for making inferences on the impact of news on® nancial markets. We employ the GARCH methodologybecause this technique is especially suited for measuringhow announcements of economic data in¯ uence boththe level and volatility of asset returns. Standard regressiontechniques are ill-equipped to evaluate how announcementsa� ect the level and volatility of asset prices.

Finally, because several previous studies have foundthat news can have asymmetric e� ects on asset prices,we investigate the extent to which changes in the CCIhave asymmetric e� ects on the level and volatility of assetprices.

II I . DATA AND DIAGNOSTIC TESTS

The Conference Board’s monthly Consumer Con® denceIndex (1985 = 100) is used as a measure of informationwhich is expected to a� ect ® nancial markets. The CCI iscompiled from the responses of approximately 5000 house-holds that receive a questionnaire each month. The ques-tions ask about the household’s view of current and futureeconomic situations.9 Figure 1 presents the plot of the CCIagainst time. Note the large ¯ uctuations around the early1980s (associated with the beginning of the economic expan-sion of the 1980s) and the early 1990s (associated with theGulf War). While such ¯ uctuations may be re¯ ections ofeconomic conditions, if they are indicators of future eco-nomic conditions, changes in the CCI may impact the stock,bond and foreign exchange markets.

8 For example, Hakkio and Pierce (1985) ® nd that unanticipated money announcements a� ect the dollar exchange rate after the October1979 regime change. In¯ ation and real activity information are found not to a� ect exchange rates. Dwyer and Hafer (1989) ® nd thatunanticipated money announcements a� ect interest rates over the 1970Ð 87 period, while information about in¯ ation, real activity, andtrade balances do not a� ect interest rates. However, Cook and Korn (1991) ® nd that the monthly employment report a� ects interest rates.Ederington and Lee (1993) ® nd that announcements of variables (such as the employment report and durable goods orders) a� ect thevolatility of interest rates and the dollar Ð deutschmark exchange rate.9 See Nathanson (1984) for a description of the survey instrument and for a discussion of the similarities and di� erences between the CCIand the University of Michigan Consumer Sentiment Index.1 0 The Wall Street Journal sometimes reports the CCI prior to the public release date.1 1 The data were obtained from the Futures Industry Institute.

Fig. 1. Consumer con® dence (1985 = 100)

An important point to note is that prior to public release,the CCI is released to paying customers in the corporateand ® nancial sectors. These customers (most importantlyfor our purposes, stock, bond and foreign exchange traders)can then act on this information as they see ® t. Further, theinformation can be leaked to other non-paying customers.1 0

For our measure of stock prices, we use the daily closingprices of the Dow Jones Industrial Average (DJIA), the S&P500 Index (SP500), and the Financial Times World Index(FTWI). The FTWI consists of approximately 2200 stocksfrom 24 countries and is employed to measure the extent towhich the CCI is a global factor in asset pricing. Ourmeasure of bond price is the daily closing value of the DowJones 20 Bond Index (DJ20BNX), consisting of 10 industrialand 10 utility bonds.

Exchange rates are the daily spot closing prices (noonbuying) of the BP, CD, DM, JY and SF.1 1 For the DJIA andDJ20BNX, we have 3470 observations and for the foreignexchange market we have 3475 observations coveringthe period January 1980 to September 1993. The slightdi� erence in the number of observations is due to variousholidays. For the SP500, we have 3379 observations fromMarch 1980 to September 1993, and for the FTWI we have1913 observations from January 1986 to September 1993.

Consumer con® dence announcements 157D

ownl

oade

d by

[U

SC U

nive

rsity

of

Sout

hern

Cal

ifor

nia]

at 2

2:39

09

Nov

embe

r 20

14

Page 5: Consumer confidence announcements: do they matter?

Diagnostic tests

The results of diagnostic tests on the distributions of thevariables under consideration are reported below. All vari-ables are logged and subjected to Augmented Dickey Ð Fuller(ADF) unit root tests. All are found to be non-stationary .We then subject the log-di� erences of the variables to theADF test. As can be seen in Table 1, the log-di� erences of allvariables are stationary, as is required by the GARCH model.

Table 2 reports descriptive statistics for the variablesboth in log-level and log-di� erenced forms. The uncondi-tional distributions of D ln DJIA, D ln SP500, D ln FTWI,D ln DJ20BNX, D ln CCI and the foreign currency deprecia-tion rates are non-normal, as indicated by skewness, kurto-sis, and the Bera Ð Jarque statistics. Skewness ranges from thelow of - 3.87 (D ln DJIA) to a high of 0.55 (D ln DJ20BNX).Notice that the sample period covers the 1987 and 1989stock market crashes, either or both of which could beresponsible for the skewness, which would imply that the

Table 1. Augmented DickeyÐ Fuller unit root tests

Variable tu

D ln CCI - 3.16*

D ln DJIA - 16.87*D ln SP500 - 17.09*D ln FTWI - 11.10*

D ln DJ20BNX - 13.26*

D ln BP - 16.63*D ln CD - 16.13*D ln DM - 16.09*D ln JY - 14.60*D ln SF - 16.11*

* Signi® cant at the 5% level.The test involves estimating the following regression

D ln Xt = l 0 + l 1 D ln Xt ± 1 + l 2 D 2 ln Xt ± 1 + e t

where D ln Xt refers to the variable under consideration and D isthe ® rst di� erence operator. The second-di� erence term(s) is usedto whiten the residuals. The null hypothesis is that the coe� cient ofD ln Xt ± 1 is one (i.e. the series is non-stationary). tu is calculated as(l 1 - 1)/std. error of l 1 . 2.86 is the critical value at the 5% level.A rejection of the null hypothesis suggests that the series is station-ary. Preliminary Dickey Ð Fuller unit root tests indicated that thelog-levels (ln Xt) of the variables were non-stationary. Thus, thevariables were di� erenced and the tests reported above were car-ried out.

CCI is the Conference Board’s Consumer Con® dence Index(1985 = 100).DJIA is the Dow Jones Industrial AverageSP500 is the S&P 500 IndexFTWI is the Financial Times World IndexDJ20BNX is the Dow Jones Twenty Bond Index.

BP, CD, DM, JY, and SF are the spot exchange rates for theBritish pound, Canadian dollar, Deutsche-Mark, the Japanese yen,and the Swiss franc.data have asymmetric distributions. Kurtosis values suggest

that the distributions of the variables are leptokurtic: highkurtosis for stock and bond market returns are consistentwith two major crashes the markets had experienced. Fi-nally, based on the Bera Ð Jarque statistics, the hypothesis ofnormality is rejected for all variables.

Ljung Ð Box Q statistics are computed for the ® rst 24 lagsof each variable to measure the degree of autocorrelation.The results are reported in the last column of Table 2. TheQ (24) test shows that the null hypothesis of no autocorrela-tion cannot be rejected except for D lnDJIA, D lnSP500,D lnFTWI, D lnDJ20BNX, D lnDM and D lnSF. Q (24) stat-istics for these variables exceed the critical x 2 .

Finally, Engle’s ® rst-order LM test for ARCH residualsfound evidence of time-varying volatility for stock, bond andforeign exchange data. The tests are carried out using log-dif-ferenced data and require estimating the following regression

D ln Xt = a 0 + e t (1)

where X is the variable under consideration. A secondregression is estimated using the residuals from Equation 1

e 2t = b 0 + b 1 e 2

t ± 1 + n t (2)

Based on this regression, the statistic TR2 is calculated,where T is the number of observations. Table 3 presents theresults of the ARCH tests. As can be seen, there is strongevidence of time varying volatility for stock, bond andforeign exchange market data. The evidence of ARCH-typeerrors suggests that GARCH models are suitable for model-ling such types of distributions.

Overall, the diagnostic statistics provide convincing evi-dence that daily data are not normally distributed and thatvariance/covariances are changing over time. The GARCHmodel, which permits time varying volatility, is suitable formodelling such non-linear distributions of asset returns.

IV. THE IMPACT OF CONSUMERCONFIDENCE ANNOUNCEMENTS ONSTOCK, BOND AND FOREIGNEXCHANGE RETURNS:BENCHMARK ESTIMATES

GARCH models are estimated using the Berndt et al. (1974)(BHHH) procedure

100 * D ln Yt = b 0 + b 1 CRASH + + u k D ln CCIt ± k + e t

k = 0, 1, 6 (3)

e t | c t ± 1 ~ N(m , s 21 ) (4)

s 2t = V +

q

+i = 1

a i e2t ± l +

p

+j = 1

b j s t ± j + + G k D ln CCIt ± k

k = 0, 1, 6 (5)

where Y t in the mean Equation 3 is the daily return cal-culated using the closing price of the variable, and

158 O. D. Gulley and J. SultanD

ownl

oade

d by

[U

SC U

nive

rsity

of

Sout

hern

Cal

ifor

nia]

at 2

2:39

09

Nov

embe

r 20

14

Page 6: Consumer confidence announcements: do they matter?

Table 2. Descriptive statistics

Variable Mean Std. Dev. Skewness Kurtosis Bera Ð Jarque Q(24)

CCI 87.1056 20.9664 - 0.1632 - 1.2693 11.5940*D ln CCI - 0.0016 0.0792 - 0.2670 3.2993* 75.3991* 31.6000DJIA 1932.2124 874.0432 0.2980 - 1.2936 293.4038*D ln DJIA 0.0423 1.0795 - 3.8737* 96.7997* 1363843.8500* 40.3900*

SP500 250.9787 105.1629 0.2939 1.7818 257.5811*D ln SP500 0.0448 1.0341 - 3.3805* 78.9676* 818552.2500* 48.0142*

FTWI 121.8629 16.0842 - 0.6944 2.9537* 153.9095*D ln FTWI 0.0326 0.8100 - 1.5588 34.4951* 79799.0241* 103.1272*

DJ20BNX 82.6605 14.1672 - 0.2037 - 1.0623 187.2106*D ln DJ20BNX 0.0113 0.3191 0.5475 9.2344* 12506.3264* 425.0800*

D ln BP - 0.0001 0.0075 0.0665 3.0968* 1391.1327* 31.9600D ln CD - 0.0001 0.0027 - 0.2390 4.0655* 2426.2593* 29.9300D ln DM 0.0001 0.0079 0.2580 4.6896* 3222.8205* 39.5700*D ln JY 0.0002 0.0068 0.4591 3.5547* 1951.0248* 34.5700D ln SF 0.0001 0.0090 0.0424 7.7096* 8607.2151* 48.3900*

* Signi® cant at the 5% level.The critical values at the 5% level for the skewness and kurtosis tests are + / - 1.96 because the distributions of both statistics areasymptotically normal. The null hypotheses are no skewness or no kurtosis.The Bera Ð Jarque statistic is a test for normality and is distributed as chi-square and has a critical value of 5.99 at the 5% level. Q (24) is theLjungÐ Box Q-test for autocorrelation. The critical value of Q (24) at the 5% level is 36.42.CCI is the Conference Board’s Consumer Con® dence Index (1985 = 100).DJIA is the Dow Jones Industrial AverageSP500 is the S&P 500 Stock IndexFTWI is the Financial Times World Index.DJ20BNX is the Dow Jones Twenty Bond Index.

BP, CD, DM, JY, and SF are the spot exchange rates for the British pound, Canadian dollar, Deutsche-Mark, the Japanese yen, and theSwiss franc.

D ln CCI are current and lagged announcements of the CCIthat are included in the model to account for the autoregres-sive nature of the CCI. Similar to Cumby and Glen (1990),we account for the October 1987 stock market crash byincluding a dummy variable, CRASH. CRASH takesa value of one during the 1987 crash and zero otherwise.C t ± 1 is the information set and it is assumed that the returndistribution is conditional normal. The variance Equation5 models the conditional variances as a GARCH(p, q) pro-cess where p and q denote the lag lengths. V is the interceptterm, a i are ARCH terms, and b j are GARCH terms.1 2 Thea and the b terms are expected to be positive and signi® cantdeterminants of the conditional variance of returns. Thevariance equation also includes current and lagged values ofD ln CCI as exogenous variables, which permits the CCI toa� ect also the time varying volatility of asset prices. Thebenchmark estimates obtained in this section will be com-pared with the results of separating changes in the CCI intopositive and negative changes.

The robustness of the GARCH models depends largelyon post-estimation diagnostics using the standardized resid-uals (e ij , t/hij , t ). As Bollerslev and Wooldridge (1992) note,

1 2 If (a 1 + b 1 ) exceeds 1, the model is an IGARCH (integrated in variance) model, implying that shocks to the variance persist over time.

statistical inferences based on the student-t may be mislead-ing if the standard errors are not normal. Since preliminaryestimations suggest excess kurtosis in the standardizedresiduals, we estimate GARCH models by maximizingthe following log-likelihood function, assuming a condi-tional bivariate elliptical t-distribution (Muirhead, 1982,pp. 48 Ð 49)

f (e t) =G 1 (n + v)

2 2G 1 v2 2 (Ï p (v - 2))n

| Ht |± Æ 3 1 +

1v - 2

e 9t H± 1t e t 4

±n+ v

2

(6)

where n is the number of variables, v is degrees of freedom,and Ht is s 2

t obtained from Equation 5. Maximizing Equa-tion 6 improves statistical inferences because the resultingstandard errors of the coe� cients are robust to excesskurtosis.

In Table 4, univariate GARCH results are reported usingthe daily stock and bond returns. The results for the meanequations do not indicate that current changes in the CCI

Consumer con® dence announcements 159D

ownl

oade

d by

[U

SC U

nive

rsity

of

Sout

hern

Cal

ifor

nia]

at 2

2:39

09

Nov

embe

r 20

14

Page 7: Consumer confidence announcements: do they matter?

Table 3. Engle’s LM Test

Variable TR2

D ln DJIA 98.43*D ln SP500 36.23*D ln FTWI 470.07*

D ln DJ20BNX 197.22*

D ln BP 90.57*D ln CD 158.23*D ln DM 289.05*D ln JY 75.62*D ln SF 444.85*

* Signi® cant at the 5% level.Engle’s LM tests are carried out using log-di� erenced data. Thetest requires estimating the following regression

D ln Xt = a 0 + e t

where D ln Xt is the variable under consideration. An auxiliaryregression is then estimated using the residuals from above

e 2t = b 0 + b 1 e 2

t ± 1 + n t

Based on the second regression, the test statistic TR2 is calculated.T is the number of observations. It is distributed as chi-square andthe critical value at the 5% level is 3.84. Rejection of the nullhypothesis indicates the presence of ARCH-type errors.

DJIA is the Dow Jones Industrial Average.SP500 is the S&P 500 Stock Index.FTWI is the Financial Times World Index.DJ20BNX is the Dow Jones Twenty Bond Index.

BP, CD, DM, JY, and SF are the spot exchange rates for theBritish pound, Canadian dollar, Deutsche-Mark, the Japanese yen,and the Swiss franc.

are associated with changes in stock and bond marketreturns. This is in sharp contrast to the alleged link reportedin the ® nancial press. The lack of a relationship between theCCI and equity and debt returns could be for one of severalreasons. First, the data used in this study are daily data thatcannot measure the instantaneous impact of economicnews. Obviously, a solution would be to use intraday data.Second, prior to public release, the CCI is released to corpo-rate clients, so our dating of the announcements could be o�by several days. However, conversations with ConferenceBoard o� cials do not suggest that dating is a major prob-

1 3 We estimated an ARIMA model of the CCI and used the residuals as the unanticipated component of the CCI. The results arequantitatively similar to those presented in this paper.1 4 The BHHH estimation procedure requires several stages. First, a benchmark model is estimated without exogenous variables in theconditional variance equation. Next, exogenous variables are included in the conditional variance equation. The kurtosis of thestandardized residuals (e t/s

2t ) is used to solve for theoretical DF as follows: Kurtosis = 6 + (3/(6 - DF ± 1 )). In the ® nal stage, DF is

computed and the model is estimated. Because the DF term is now ® xed, the GARCH model does not estimate a t-statistic for it. Thelog-likelihood ratio statistics (LR) are calculated as 2(L - L 9 ) where L is the log-likelihood value from the benchmark model and the L 9 isthe log-likelihood from the ® nal model. The LR is distributed as chi-square with three degrees of freedom.1 5 However, experimentation indicates that higher ordered GARCH terms did not contribute to any signi® cant improvement in the results.1 6 This is solved as follows. Set (Et - Et ± 1 )/Et ± 1 = - 0.018785, where E is the spot exchange rate ($/£). From this, solve for Et . Next,substitute the value for Et in the following equation X = (Et ± 1 - Et)/Et , where X yields the amount of appreciation of the US dollarvis-a -vis the British pound.

lem. Third, ® nancial theory predicts that the announcementof the unexpected component of the CCI will in¯ uence assetprices. Our ® ndings could be the result of using the actualCCI, which is made up of both the anticipated and theunacticipated components.1 3 Finally, it is possible that theCCI has very little in¯ uence on the way traders revise theirexpectations of future returns. If so, our results are contra-dictory to reports in the popular press.

The variance equations in Table 4 indicate that while CCIannouncements do not a� ect the daily volatility of theSP500, FTWI or bond market, the announcements havea positive relationship with the volatility of the DJIA, as isindicated by the large and statistically signi® cant coe� cienton D ln CCIt . Thus, while the CCI does not a� ect the ® rstmoment of DJIA returns, it a� ects the second moment.

Overall, diagnostic results suggest that GARCH(1,1)models ® t the data well. In the mean equation, the GARCHmodelling of the bond market returns required AR (auto-regressive) and MA (moving average) terms. Both variablesare signi® cant at the 1% level. In the variance equation, theARCH and the GARCH coe� cients are signi® cant for bothmodels. The DF term is the inverse of the degrees of freedomand the low value suggests that the assumption of condi-tional normality is not appropriate.1 4 Finally, diagnosticsusing the standardized residuals (e t/ s

2t ) suggest that the

stock market return distribution is modelled adequatelywith a GARCH(1, 1) model. However, serial correlation inthe squared residuals detected by the Q2 (24) test suggeststhat higher ordered GARCH models may be needed tocorrect for residual heteroscedasticity.1 5

In contrast to the stock and bond market data, estimatesreported in Table 5 suggest a signi® cant link between theCCI and foreign currency prices. Increases (decreases) inD ln CCIt are associated with a spot dollar appreciation(depreciation) vis-a -vis the British pound, German mark,and Swiss franc. For example, a 1% increase in the CCIfrom the previous month is associated with a 1.8785%depreciation of the British pound (or a 1.9145% appreci-ation of the dollar).1 6 The highest amount of dollar appreci-ation is noticed vis-a -vis the German mark, followed by theSwiss franc and the British pound.

The results also show the presence of a link between theCCI and the conditional volatility of the Canadian dollar

160 O. D. Gulley and J. SultanD

ownl

oade

d by

[U

SC U

nive

rsity

of

Sout

hern

Cal

ifor

nia]

at 2

2:39

09

Nov

embe

r 20

14

Page 8: Consumer confidence announcements: do they matter?

Table 4. The impact of consumer con® dence announcements on equity and bond returns: univariate GARCH(1,1) results

Mean equation DJIA DJ20BNX SP500 FTWI

Constant 0.0422 0.0022 0.0470 0.0589(3.1485)* (1.9479)** (3.3877)* (3.9856)*

Crash - 26.9168 - 0.6372 - 14.1674 - 5.1762( - 19.9382)* ( - 3.9870)* ( - 10.6812)* ( - 5.7722)*

D ln CCIt - 0.7776 - 0.1189 - 0.2868 0.1266( - 0.8987) ( - 0.8125) ( - 0.3542) (0.1623)

D ln CCIt ± 1 - 0.4995 - 0.0375 - 0.8027 0.3252( - 0.4691) ( - 0.2326) ( - 0.9107) (0.4041)

D ln CCIt ± 6 0.0872 - 0.0145 0.3046 0.0806(0.1022) ( - 0.0874) (0.3793) (0.1304)

AR 0.8159(26.8721)*

MA - 0.7022 0.0433 0.2507( - 18.5625)* (2.5620)** (11.0642)*

Variance equationConstant 0.0104 0.0013 0.0102 0.04383

(3.00037)* (4.8597)* (3.2681)* (4.5203)*e 2

t ± 1 0.0379 0.0806 0.0348 0.1888(5.8616)* (8.3691)* (6.1393)* (5.7276)*

s 2t ± 1 0.9532 0.9099 0.9531 0.7449

(121.7152)* (97.3860)* (129.5684)* (20.4975)*D ln CCIt 0.8783 - 0.0062 0.3743 - 0.3449

(1.7574)*** ( - 0.1330) (0.8657) ( - 0.5883)D ln CCIt ± 1 0.1379 0.0309 - 0.2087 - 0.5677

(0.2934) (0.7677) ( - 0.4701) ( - 0.8587)D ln CCIt ± 6 - 0.2266 - 0.0362 - 0.8084 - 0.7217

( - 0.5405) ( - 1.0259) ( - 2.3968)** ( - 1.6451)***

1/DF 0.2075 0.2243 0.1792 0.2491L - 713.4532 3380.4773 - 200.7138 5.6253LR 6.9200 2.7200 6.5600 2.9700

Diagnostics with standardized residuals

Mean 0.0039 - 0.0085 - 0.0032 - 0.0269Variance 0.9596 0.9992 1.0060 0.9065Skewness - 0.2701 - 0.1477 - 0.3623 0.1236Kurtosis 7.8441 8.3018 6.7917 9.2201Q (24) 23.2473 25.7723 17.1969 10.1169Q2 (24) 8.8783 79.4717 18.4358 15.0941

* Signi® cant at the 1% level. ** Signi® cant at the 5% level. *** Signi® cant at the 10% level.The results for the mean and variance equations are from estimating Equations 3 and 5

D ln Y t = b 0 + b 1 CRASH + + u k D ln CCIt ± k + e t k = 0, 1, 6 (3)

s 2t = V +

q

+i = 1

a i e2t ± l +

p

+j = 1

b j s t ± j + + G k D ln CCIt ± k k = 0, 1, 6 (5)

CRASH is a dummy variable that is set to one during the 1987 crash and zero otherwise. CCI is the Consumer Con® denceIndex. AR and MA are autoregressive and moving average terms included in Equation 3 for the bond market. DF is the degreesof freedom. L is log-likelihood value from Equation 6. LR is the log-likelihood ratio statistic. See Footnote 14 for additionalinformation on DF, L and LR. The LR is distributed as chi-square with three degrees of freedom. The critical value is 7.81 (5%signi® cance level). See Table 2 for a discussion of the skewness, kurtosis and Q (24) statistics. Q2 (24) is a test for serial correlationin the squared residuals. It is distributed as chi-square and has a critical value at the 5% level of 36.42.

and the Japanese yen. It is surprising that the CCI lagged sixperiods is negatively associated with the conditional volati-lity of the $/Canadian dollar rate. Similarly, the CCI is alsonegatively linked to the conditional volatility of the $/yenrate. The negative link suggests that an improvement in the

US economy reduces the uncertainty associated with thevalue of the dollar.

Post estimation diagnostics reveal that GARCH models® t the exchange rates quite well. The log-likelihood ratiostatistics suggest that the null hypothesis that the CCI does

Consumer con® dence announcements 161D

ownl

oade

d by

[U

SC U

nive

rsity

of

Sout

hern

Cal

ifor

nia]

at 2

2:39

09

Nov

embe

r 20

14

Page 9: Consumer confidence announcements: do they matter?

Table 5. The impact of consumer con® dence announcements on exchange rates: univariate GARCH(1,1) results

BP CD DM JY SF

Mean equation

Constant - 0.0024 0.0044 - 0.0080 - 0.0019 - 0.0221( - 0.2179) (1.3025) ( - 0.7046) ( - 0.1992) ( - 1.8135)***

D ln CCIt - 1.8785 - 0.3209 - 2.1300 - 0.5875 - 1.9112( - 2.8568)* ( - 1.4115) ( - 2.8331)* ( - 0.9195) ( - 2.4773)**

D ln CCIt ± 1 0.1369 - 0.0215 - 0.6193 0.0139 - 0.3951(0.1969) ( - 0.0892) ( - 0.8775) (0.0231) ( - 0.4954)

D ln CCIt ± 6 - 0.0914 - 0.1388 - 0.006 0.2949 0.5292( - 0.1309) ( - 1.1470) ( - 0.0079) (0.4767) (0.7054)

Variance equation

Constant 0.0147 0.0013 0.0345 0.0278 0.0739(4.1147)* (3.9747)* (5.1941)* (5.0748)* (4.1576)*

e 2t ± 1 0.0627 0.1204 0.1000 0.1022 0.1016

(7.1638)* (8.8745)* (7.9994)* (7.6042)* (5.6906)*s 2

t ± 1 0.9091 0.8747 0.8414 0.8385 0.8370(71.9555)* (69.5182)* (44.1155)* (41.0964)* (29.6887)*

D ln CCIt 0.4332 - 0.0014 0.0010 - 0.5746 - 0.1586(1.4514) ( - 0.314) (0.0020) ( - 1.6801)*** ( - 0.1606)

D ln CCIt ± 1 - 0.3779 - 0.0156 0.0011 - 0.4647 - 0.5990( - 1.2080) ( - 0.3266) (0.0024) ( - 1.3012) ( - 0.6196)

D ln CCIt ± 6 - 0.2457 - 0.0550 0.0010 - 0.2704 - 1.0156( - 0.8286) ( - 2.5530)** (0.0021) ( - 0.7566) ( - 1.0833)

1/DF 0.1272 0.1918 0.1452 0.1906 0.2780L 1401.0770 4020.1330 948.1288 716.8547 - 1429.0414LR 2.4400 5.0600 0.0040 6.9200 1.6400

Diagnostics with standardized residuals

Mean - 0.0156 - 0.0397 0.0062 0.0320 0.0209Variance 1.0254 1.0068 1.0344 1.0453 0.8540Skewness - 0.1542 - 0.6816 0.0609 0.5188 - 0.1541Kurtosis 4.7857 7.2430 5.4570 7.2029 10.2218Q (24) 29.6758 22.6851 31.7504 34.6488 30.9378Q2 (24) 24.8145 15.2550 32.8362 26.3083 22.4641

* Signi® cant at the 1% level. ** Signi® cant at the 5% level. *** Signi® cant at the 10% level.BP, CD, DM, JY, and SF are the spot exchange rates for the British pound, Canadian dollar, Deutsche-Mark, the Japanese yen,and the Swiss franc.

See Table 4 for a description of the estimation technique.

not belong in the conditional variance equation can berejected for all but the Japanese yen. The standardizedresiduals are quite close to being white noise. Finally,Ljung Ð Box Q statistics show that the standardized residualsare not serially correlated in levels or in squares.1 7

V. ASYM METRIC IMPACT OF THE CCI ONSTOCK, BOND AND FOREIGNEXCHANGE RETURNS

Recent evidence suggests that news can have asymmetrice� ects on asset prices. Hafer (1985) ® nds that unanticipated

1 7 To capture how the Gulf-War induced focus on the CCI may have altered how ® nancial markets responded to the CCI, we createda dummy variable equal to zero prior to August 1990, and one thereafter. An interaction variable consisting of this dummy variable andCCI was statistically insigni® cant.

weekly money supply increases are associated with lowerstock prices, but that unanticipated decreases in the moneysupply had no a� ect on stock prices. Engle and Ng (1993)® nd that negative news shocks produce more volatility inJapanese stock prices than do positive news shocks. Finally,Sultan (1994) shows that unanticipated increases in themonthly US trade de® cit are more likely to be associatedwith dollar depreciation than are smaller than expectedtrade de® cits. We now investigate the extent to which CCIannouncements have asymmetric e� ects on asset prices.

Two new variables are created by decomposing D ln CCIinto D ln CCI > 0 and D ln CCI< 0. D ln CCI > 0 takes on

162 O. D. Gulley and J. SultanD

ownl

oade

d by

[U

SC U

nive

rsity

of

Sout

hern

Cal

ifor

nia]

at 2

2:39

09

Nov

embe

r 20

14

Page 10: Consumer confidence announcements: do they matter?

the value of D ln CCI when D ln CCI is positive and zerootherwise. D ln CCI< 0 is de® ned in an analogous fashion.Equations 3 Ð 5 are re-estimated and the results are presentedin Tables 6 and 7. In Table 6, the mean equations indicatethat the impact of CCI announcements is limited to theDJIA. Recall that the benchmark estimates in Table 4 foundno statistically signi® cant a� ect of the CCI on the DJIA.Table 6 further shows that the contemporaneous impact ofnegative changes in the CCI is larger than for positivechanges in the CCI. Thus, negative news seems to havea larger a� ect on the DJIA than does positive news.

Table 6. Asymmetric impact of consumer con® dence announcements on equity and bond returns: univariate GARCH(1,1)results

DJIA DJ20BNX SP500 FTWI

Mean equation

Constant 0.0366 - 0.0002 0.0385 0.0541(2.7242)* ( - 0.1661) (1.8349)*** (3.6855)*

Crash - 26.0911 0.4860 - 23.5561 - 10.0793( - 7.7221)* (2.3637)* ( - 17.0596)* ( - 3.9601)*

D ln CCIt > 0 2.4595 0.0294 2.3528 1.1410(1.9724)*** (0.1436) (1.0608) (0.8020)

D ln CCIt ± 1 > 0 - 1.3969 0.2976 1.5566 - 0.7449( - 0.8981) (1.1573) (0.7047) ( - 0.4344)

D ln CCIt ± 6 > 0 0.6495 0.2088 1.3479 0.7351(0.5458) (0.8251) (0.7308) (0.5656)

D ln CCIt < 0 - 3.0122 - 0.3982 - 2.2537 - 0.6858( - 2.2157)** ( - 2.0253) ( - 1.1390) ( - 0.6090)

D ln CCIt ± 1 < 0 0.3185 - 0.2717 - 2.9494 - 0.3300(0.1873) ( - 1.5209) ( - 1.1470) ( - 0.2827)

D ln CCIt ± 6 < 0 - 0.5639 - 0.1687 2.9109 - 1.9965( - 0.3631) ( - 0.7242) (1.2041) ( - 1.3666)

AR 0.8657(37.5109)*

MA - 0.7675 0.901 0.2531( - 25.4998)* (3.8608)* (11.1422)*

Variance equation

Constant 0.0135 0.0012 0.0076 0.0489(3.5085)* (4.4573)* (1.8933)*** (4.1350)*

e 2t ± 1 0.0353 0.0742 0.0276 0.2111

(5.8505)* (8.3653)* (3.9994)* (5.3020)*s 2

t ± 1 0.9544 0.9181 0.9642 0.7578(127.2106)* (107.0947)* (96.6464)* (20.3910)*

D ln CCIt > 0 - 0.2181 0.0001 - 0.9055 0.0100( - 0.2759) (0.0015) ( - 0.6962) (0.0086)

D ln CCIt ± 1 > 0 0.2449 0.0010 1.4061 0.0030(0.2861) (0.0152) (1.0802) (0.0015)

D ln CCIt ± 6 > 0 - 0.5773 0.0020 - 1.1564 0.0004( - 1.1554) (0.0312) ( - 1.6701)*** (0.0004)

D ln CCIt < 0 1.7731 0.0030 1.7646 0.0030(2.8045)* (0.0463) (1.5878) (0.0020)

D ln CCIt ± 1 < 0 0.1078 0.0469 - 1.4025 0.0006(0.1662) (0.9911) ( - 1.2031) (0.0004)

D ln CCIt ± 6 < 0 0.1326 0.0425 - 0.5871 0.0020(0.1765) (0.7346) ( - 0.5959) (0.0010)

1/DF 0.1994 0.2200 0.1230 0.2900L - 587.3406 3450.4676 357.6723 - 311.1984

In the variance equation, however, only negative changesin the CCI a� ect the conditional volatility of the DJIA. Thecoe� cient estimate indicates that decreases in the CCI areassociated with decreases in the conditional volatility of theDJIA. This result may be due to the market’s interpretationof a decrease in the CCI as a resolution of uncertainty,which will act to reduce volatility. The results in Table 6contrast sharply to the benchmark estimates in Table 4,which indicate that changes in the CCI have symmetrice� ects on the DJIA. Bond and other broader stock marketindexes seem una� ected by increases and decreases in the CCI.

Consumer con® dence announcements 163D

ownl

oade

d by

[U

SC U

nive

rsity

of

Sout

hern

Cal

ifor

nia]

at 2

2:39

09

Nov

embe

r 20

14

Page 11: Consumer confidence announcements: do they matter?

Table 6. (Continued )

DJIA DJ20BNX SP500 FTWI

Diagnostics with standardized residuals

Mean 0.0035 - 0.0121 0.0064 - 0.0265Variance 0.9831 1.0134 1.024 0.8227Skewness - 0.2493 - 0.3217 - 0.0203 - 0.0748Kurtosis 7.5108 8.8969 4.6544 11.7608Q (24) 23.0582 17.9234 14.6717 12.4899Q2 (24) 9.4918 91.6067 16.1465 11.0968

* Signi® cant at the 1% level. ** Signi® cant at the 5% level. *** Signi® cant at the 10% level.The results for the mean and variance equations are from estimating Equations 3 and 5

D ln yt = b 0 + b 1 CRASH + S u kD ln CCI > 0t ± k + S l k D ln CCI < 0t ± k + e t k = 0, 1, 6 (39 )

s 2t = V +

q

+i = 1

a i e2t ± l +

p

+j = 1

b j s t ± j + + G k D ln CCI > 0t ± k + + d k D ln CCI < 0t ± k k = 0, 1, 6 (59 )

CRASH is a dummy variable that is set to one during the 1987 crash and zero otherwise. CCI is the ConsumerCon® dence Index. D ln CCIt > 0 is set to D ln CCIt when D ln CCIt is positive and zero otherwise. D ln CCIt < 0 is de® nedin analogous fashion. AR and MA are autoregressive and moving average terms included in Equation 3 for the bondmarket. DF is the degrees of freedom. L is log-likelihood value from Equation 6. LR is the log-likelihood ratio statistic.See Footnote 14 for additional information on DF, L and LR. The LR is distributed as chi-square with six degrees offreedom. The critical value is 12.59 (5% signi® cance level). See Table 2 for a discussion of the skewness, kurtosis and Q (24)statistics. Q2 (24) is a test for serial correlation in the squared residuals. It is distributed as chi-square and has a criticalvalue at the 5% level of 36.42.

The asymmetric e� ects of changes in the CCI are evenmore pronounced for the foreign exchange rates in Table 7.As the mean equations show, only negative changes in theCCI a� ect the exchange rates, with the exception of the JY.The sign of the coe� cients indicates that the dollar tends todepreciate with decreases in the CCI. Increases in the CCIdo not seem to a� ect the dollar price of any of the foreigncurrencies investigated. Again, the results here are in con-trast to the benchmark estimates in Table 5, which indicatethat changes in the CCI have symmetric e� ects on thevarious exchange rates. As for the variance equations, in-creases in the CCI appear to decrease the volatility of the

Table 7. Asymmetric impact of consumer con® dence announcements on exchange rates univariate GARCH(1,1) results

BP CD DM JY SF

Mean equation

Constant - 0.0061 0.0043 - 0.0125 - 0.0014 - 0.0290( - 0.5481) (1.2366) ( - 1.1023) ( - 0.1455) ( - 2.3655)**

D ln CCIt > 0 - 1.9879 0.0560 - 2.6295 - 0.4272 - 1.2731( - 1.4246) (0.1438) ( - 2.2571)** ( - 0.3582) ( - 0.9131)

D ln CCIt ± 1 > 0 1.5567 - 0.4638 0.7675 0.1940 1.6129(1.1519) ( - 1.1055) (0.5871) (0.2016) (1.1249)

D ln CCIt ± 6 > 0 0.0471 0.0451 0.1859 0.2151 0.8793(0.0439) (0.1188) (0.1487) (0.2181) (0.8241)

D ln CCIt < 0 - 2.3175 - 0.6213 - 2.4180 - 0.6444 - 2.7825( - 3.3559)* ( - 1.8381)*** ( - 3.4532)* ( - 0.6479) ( - 2.2094)**

D ln CCIt ± 1 < 0 - 1.0478 0.3570 - 1.4007 - 0.3485 - 1.3808( - 0.8820) (0.9096) ( - 1.1745) ( - 0.3085) ( - 0.8149)

D ln CCIt ± 6 < 0 - 0.0001 - 0.2161 - 0.0001 - 0.0010 0.5914( - 0.0001) ( - 0.6314) ( - 0.0001) ( - 0.0010) (0.4532)

$/DM rate, while decreases in the CCI appear to decreasethe volatility of the $/BP rate. The reductions in the condi-tional volatilities could be due to the resolution of uncer-tainty from the announcement of the CCI.

Thus, the CCI appears to a� ect the level of all but the JY,and the volatility of the DM and BP. It is important to notethat the e� ects are not uniform across currencies. Suchinformation is of use to market participants in evaluatingthe impact of news on asset prices.

As in the benchmark estimates, post estimation diagnos-tics here reveal that GARCH models ® t the exchange ratesquite well. The log-likelihood ratio statistics suggest that the

164 O. D. Gulley and J. SultanD

ownl

oade

d by

[U

SC U

nive

rsity

of

Sout

hern

Cal

ifor

nia]

at 2

2:39

09

Nov

embe

r 20

14

Page 12: Consumer confidence announcements: do they matter?

Table 7. (Continued )

BP CD DM JY SF

Variance equation

Constant 0.0155 0.0013 0.0344 0.0273 0.0749(4.1337)* (3.5663)* (5.3215)* (5.0849)* (4.1771)*

e 2t ± 1 0.0632 0.1236 0.0997 0.1010 0.1024

(7.2345)* (8.9052)* (8.0958)* (7.6669)* (5.7336)*s 2

t ± 1 0.9065 0.8717 0.8416 0.8403 0.8345(70.4800)* (66.8377)* (45.3106)* (42.0430)* (29.2005)*

D ln CCIt > 0 - 0.0669 0.0001 - 1.2724 - 0.4196 - 0.2758( - 0.1188) (0.0011) ( - 1.7616)*** ( - 0.8785) ( - 0.1730)

D ln CCIt ± 1 > 0 - 0.0746 0.0000 0.3378 - 0.6334 - 0.3042( - 0.1155) (0.0001) (0.4073) ( - 1.2461) ( - 0.1792)

D ln CCIt ± 6 > 0 - 0.1495 0.0001 0.2683 - 0.5405 - 0.8395( - 0.3079) (0.002) (0.3441) ( - 1.0421) ( - 0.5616)

D ln CCIt < 0 0.6070 0.0001 0.5622 - 0.6952 - 0.1680(1.6455)*** (0.0016) (1.0214) ( - 0.8807) ( - 0.1071)

D ln CCIt ± 1 < 0 - 0.2542 0.0000 - 0.1989 - 0.3947 - 0.0782( - 0.6474) (0.0001) ( - 0.3398) ( - 0.5833) ( - 0.0536)

D ln CCIt ± 6 < 0 - 0.1322) 0.0001 - 0.0793 - 0.2188 - 0.3528( - 0.2800) (0.0015) ( - 0.1099) ( - 0.4060) ( - 0.2206)

1/DF 0.1272 0.1918 0.1451 0.1906 0.2780L 1405.2949 4018.5897 953.5504 717.1609 - 1425.2610LR 2.3080 0.0120 4.0200 7.4400 1.2400

Diagnostics with standardized residuals

Mean - 0.0160 - 0.0398 0.0087 0.0294 0.0218Variance 1.0476 1.0064 1.0524 1.0544 0.8627Skewness - 0.1586 - 0.6745 0.0641 0.5190 - 0.1533Kurtosis 4.7952 7.1754 5.4346 7.1789 10.1657Q (24) 29.3225 22.5879 30.4837 34.1653 30.2384Q2 (24) 23.6689 14.5734 31.8001 26.3718 22.7005

* Signi® cant at the 1% level. ** Signi® cant at the 5% level. *** Signi® cant at the 10% level.BP, CD, DM, JY, and SF are the spot exchange rates for the British pound, Canadian dollar, Deutsche-Mark, the Japanese yen,and the Swiss franc.See Table 6 for a description of the estimation technique.

null hypothesis that the CCI does not belong in the condi-tional variance equation can be rejected for all but theJapanese yen. The standardized residuals are quite close tobeing white noise. Finally, Ljung Ð Box Q statistics show thatthe standardized residuals are not serially correlated inlevels or in squares.1 8

VI . SUMMARY AND CONCLUSIONS

Despite a plethora of news clippings suggesting that the CCIa� ects stock and bond markets, we ® nd evidence of sucha link only for the DJIA, a relatively narrow index of stocks.Other broader stock market indexes, as well as the bondmarket, do not appear to be a� ected by changes in the CCI.

1 8 Similar to the procedure reported in Footnote 17, we found that an interaction variable consisting of a Gulf-War dummy variable andCCI was statistically insigni® cant. Thus, it appears that the Gulf War may not have altered how ® nancial market participants view theinformation content of the CCI.

However, foreign exchange market data suggest that theCCI is an important variable for market participants.The results also show that changes in the CCI haveasymmetric e� ects on foreign exchange rates, with negativechanges in the CCI being associated with depreciation ofthe dollar against all currencies but the Japanese yen.Furthermore, increases in the CCI are associated with adecrease in the conditional volatility of the $/DM rate,while decreases in the CCI are associated with decreasedconditional volatility of the $/BP rate. These results areimportant for understanding the way information a� ectsmarket participants and asset prices. Our results suggestthat foreign exchange markets are more reactive to newsabout the US economy and that such responses are globallytransmitted.

Consumer con® dence announcements 165

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a] a

t 22:

39 0

9 N

ovem

ber

2014

Page 13: Consumer confidence announcements: do they matter?

The literature on the e� ects of news announcements on® nancial markets is not complete. Avenues for future re-search are evident. First, we have noted that the CCI hasquite di� erent e� ects on various asset prices. While we havespeculated as to the reasons for this, more rigorous invest-igation is called for. Second, the use of intraday data mayallow more precise estimates of the e� ects of news on ® nan-cial markets. Ederington and Lee (1993) is one of a fewpapers that use such data. Third, news other than the CCIhas been shown to a� ect the level and volatility of assetprices. GARCH methodology may allow ® nancial econo-mists to gauge more accurately the impact of news on assetprices.

ACKNOWLEDGEMENTS

The authors thank James Zeitler and Tim Wolski for pro-viding excellent research assistance and an anonymousreferee for very helpful comments. We are responsible for allremaining errors.

REFERENCES

Berndt, E. K., Hall, B. H., Hall, R. H. and Hausman, J. A. (1974)Estimation and inference in nonlinear structural models, An-nals of Economic and Social Measurement, 4, 653 Ð 65.

Bollerslev, T. and Wooldridge, J. (1992) Quasi-maximum likeli-hood estimation and inference in dynamic models with time-varying covariance, Econometric Reviews, 11, 143 Ð 72.

Carroll, C., Fuhrer, J. and Wilcox, D. (1994) Does consumersentiment predict household spending? If so, why?, AmericanEconomic Review, 84, 1397 Ð 408.

Cook, T. and Korn, S. (1991) The reaction of interest rates to theemployment report: the role of policy anticipations, FederalReserve Bank of Richmond Economic Review 77 (5): 3 Ð 12.

Cumby, R. E. and Glen, J. D. (1990) Evaluating the performance ofinternational mutual funds, Journal of Finance, 45, 497 Ð 521.

Dwyer, G. P., Jr. and Hafer, R. W. (1989) Interest rates andeconomic announcements, Federal Reserve Bank of St. L ouisReview 71 (2), 34Ð 46.

Ederington, L. H. and Lee, J. H. (1993) How markets processinformation: news releases and volatility, Journal of Finance,48, 1161 Ð 91.

Engle, R. F. and Ng, V. K. (1993) Measuring and testing the impactof news on volatility, The Journal of Finance, 48, 1749 Ð 78.

Fuhrer, J. C. (1993) What role does consumer sentiment play inthe U.S. macroeconomy? New England Economic Review,Jan/Feb, 32 Ð 44.

Garner, C. A. (1991) Forecasting consumer spending: should eco-nomists pay attention to consumer con® dence surveys? Feder-al Reserve Bank of Kansas City Economic Review, May/June,57 Ð 71.

Hafer, R. W. (1985) The response of stock prices to changes inweekly money and the discount rate. Federal Reserve Bank ofSt. L ouis Review, 67, 5Ð 16.

Hakkio, C. and Pierce, D. (1985) The reaction of exchange rates toeconomic news, Economic Inquiry, 23, 621 Ð 35.

Hardouvelis, G. (1988) Economic news, exchange rates and interestrates, Journal of International Money and Finance, 7, 23 Ð 35.

Huth, W. L., Eppright, D. R. and Taube, P. M. (1994) The indexesof consumer sentiment and con® dence: leading or misleadingguides to future buyer behavior, Journal of Business Research,29, 199 Ð 206.

Muirhead, R. J. (1982) Aspects of Multivariate Statistical Theory,John Wiley and Sons, New York.

Nathanson, P. N. (1984) Consumer con® dence. In The Handbookof Economic and Financial Measures, Frank J. Fabozzi andHarry I. Green® eld (eds) Dow Jones-Irwin, Homewood, IL.

Sultan, J. (1994) Trade de® cit announcements and exchange ratevolatility: evidence from the spot and futures markets, TheJournal of Futures Markets, 14, 379 Ð 404.

Throop, A. W. (1992) Consumer sentiment: its causes and e� ects,Federal Reserve Bank of San Francisco Economic Review, 1,35 Ð 59.

166 O. D. Gulley and J. Sultan

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a] a

t 22:

39 0

9 N

ovem

ber

2014