+ All Categories
Home > Documents > Factors Determining Consumer Sentiment - Evidence From Household Survey Data

Factors Determining Consumer Sentiment - Evidence From Household Survey Data

Date post: 14-Apr-2018
Category:
Upload: yasumasa-sugai
View: 218 times
Download: 0 times
Share this document with a friend

of 41

Transcript
  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    1/41

    1

    For

    Te 64th Annual Meeting (2011) of the New York State Economics Association

    Rochester Institute of echnology, Rochester, NY.

    Factors Determining Consumer Sentiment

    Evidence from Household Survey Data

    1

    Kajal Lahiri2, Yongchen Zhao3

    September 2011

    1his paper is still a work in progress. Please do not quote or redistribute without permission from author. Comments

    are welcome.2 Department of Economics, University at Albany, SUNY, [email protected] Department of Economics, University at Albany, SUNY, [email protected] or [email protected]

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    2/41

    2

    Abstract

    Measures of consumer sentiment have long been used by policy makersand researchers in making monetary policies and forecasting consumptionexpenditures. Justification for such use centers on the intrinsic informationcontained in sentiment measures, i.e. what consumer sentiment reflects.

    Unfortunately, the literature has not been able to give a solid conclusion.In this paper, we study the five components of University of Michigansconsumer sentiment index using individual data from household survey forthe period from January 1978 to March 2009 in a nonparametric orderedchoice model. We find that the most important set of factors affectingconsumer sentiment is their own perceptions and expectations on the eco-nomic conditions, both overall and of their own, as well as on theeffectiveness of government policies and on information obtained fromnew media. After this set of factors is controlled for, consumers demo-graphic characteristics, actual macroeconomic conditions, and professionalforecasts of future conditions account for little in addition.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    3/41

    3

    1.IntroductionLooking back at the most recent recession, one may find it surprising that the Index of

    Consumer Sentiment started to fell before April 2007, while the Stock and Watsonsmonthly GDP estimates started to fell notably from about May 2008. However, this ismerely another time when changes in the Index of Consumer Sentiment lead similarchanges in the real economy. Such important relationship between consumer sentimentand the real economy, proved repeatedly by the history of our economy, justifies the im-portant role consumer sentiment measures play in business and economic research.However, as is pointed out in Ludvigson (2004), weather consumer confidence surveyscontain meaningful independent information about the economy is still largely undeter-mined, even though work has been done (for example, Fuhrer, 1993) to determine thedeterminants of consumer sentiment.

    In addition, there has been a prolonged interest in the economic literature in thepower of consumer confidence to predict future changes in aggregate consumptiongrowth, but the econometric evidence is mixed on the question of whether sentimentaffects consumer expenditures once other measured economic factors are allowed for. Ear-ly research showed that expectations variables, including the University of MichiganIndex of Consumer Sentiment (ICS), usually increases the explanatory power of automo-bile and consumer durables demand equations ( Juster and Wachtel, 1972 and Mishkin,1978). Te sharp decline in U.S. consumer sentiment in July 1990 and its coincidencewith the beginning of the most recent U.S. recession and he Gulf War revived a new,worldwide interest in the confidence consumers have regarding the prospects of the do-mestic economy and its relationship to consumers spending decisions. However, the

    evidence of the power of sentiment to predict household consumption growth has re-mained mixed. Troop (1992) finds contemporaneous and lagged consumer sentimentstatistically significant when added to equations for consumer durables and for total con-sumption. Fuhrer (1993) and Carroll, et al. (1994) again show small but statisticallysignificant effect of sentiment on consumption growth even when other variables, such asincome growth, are accounted for. Bram and Ludvigson (1998) also find that bothmeasures of U.S. consumer sentiment the University of Michigans Index of ConsumerSentiment and the Conference Boards Index of Consumer Confidence have significantpredictive power for several categories of consumer spending even after economic funda-mentals are controlled for.

    In other studies however, sentiment proved redundant in the presence of variableslike income, interest rates, and financial assets and liabilities. Mishkin (178) points outthat the presence of financial asset and liability measures typically reduces consumer sen-timent effects to insignificance. Garner (1991) also reports that, regardless of what othervariables are included, spending on consumer durables is not affected by the MichiganICS.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    4/41

    4

    Te international evidence is equally mixed. Berg and Bergstrom (1996), Acemogluand Scott (1994), and Helfenstein and Wolter (1997) present evidence of significant pre-dictive power of consumer confidence measures in models of consumption for Sweden,the U.K. and Switzerland respectively. However, Fan and Wong (1998) apply the meth-

    odology in Carroll, et al. (1994) and find the effects of sentiment on Hong Kongconsumption statistically insignificant. More recently, Easaw, et al. (2005) find that U.K.consumer confidence indices does predict durable good consumption but in predictingconsumption growth, consumer sentiment performs better in models using U.K. datathen U.S. data.

    As mainstream economic theory does not explicitly incorporate consumer sentimentin models of consumption, different explanations have been suggested for this relation-ship. At the core of these explanations is the issue of what is reflected in consumersentiment. Competing views of sentiment can be summarized as follows (Fuhrer, 1993):(1) Sentiment independently causes economic fluctuations. (2) Sentiment accurately fore-

    casts economic fluctuations but does not cause them. (3) Sentiment captures consumerspessimism and hence reflects consumers forecasts of economic fluctuations, inaccurate asthese forecasts may be. (4) Sentiment is a reflection of personal, respondent-specific con-ditions. (5) Sentiment reflects only current, widely-known economic conditions. (6)Sentiment measures consumers perceptions of uncertainty and risk, associated with thelikelihood of job/income loss.

    Te issue of what is captured by consumer confidence measures should be illuminat-ed by an investigation of the factors, which determine sentiment. However, the literatureprovides a limited discussion of these factors, and offers conflicting evidence. For example,Praet and Vuchelen (1989) and Jennings and McGrath (1994) show that increases in the

    value of the U.S. dollar affects consumer confidence in the U.S., Germany and France, butnot in the U.K. and Italy. Tey find changes in the U.S. stock market index to affect posi-tively German confidence, but negatively sentiment in the U.K. and to have no effect inFrance and Italy. Other studies attempt to explain sentiment with general macroeconomicindicators (Lovell, 1975 and Garner 1981), financial assets (Mishkin, 1978) and liabilitiesor political events (Vuchelen, 1985 and Blood and Phillips, 1995). Troop (1992) showsthat a structural model for the U.S. ICS including inflation rate, unemployment rate andshort-term interest rates as explanatory variables explains sentiment fairly well in times ofusual political and economic activity but collapses around points of extraordinary eventssuch as the Persian Gulf War. Estelami, et al. (2001) focus on consumer price knowledgeand find that macroeconomic factors explain considerable proportion of variations. In a

    more recent study, Ludvigson (2004) investigates the relationship between consumer sen-timent measures and consumer spending and suggests that it is possible that there aremore complex, possibly nonlinear, interactions between consumer confidence and eco-nomic variables . Vuchelen (2004) shows that a few well-chosen variables explainsentiment of Belgian consumers quite well.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    5/41

    5

    In light of these inclusive findings, this paper begins by examining the evidence onthe explanatory power of sentiment in models for consumption using monthly data onthe University of Michigans ICS and on personal consumption expenditures, dis-aggregated into durable, nondurables, and services for the period 1978:01 to 2009:03. We

    base our nonparametric estimation on a household level survey data and look very closelyat the overall explanatory power of our model as well as the incremental explanatory pow-er of each category of independent variables. Unlike previous studies of this issue, whichgenerally uses quarterly data, we use high-frequency monthly data. Since the original ICSis reported monthly, quarterly data will lose much of the variation in sentiment throughaveraging.

    Te focus of this paper is the issue of what determines consumer sentiment. Allstudies on this subject use time series analysis to explain an aggregate measure of con-sumer sentiment. It should be noted, however, that measures of sentiment are derivedfrom individual survey data, which contain information about consumer expectations and

    personal characteristics unavailable at the aggregate level. Tese data can help fully answerthe question of what groups of factors affect sentiment, and how much of it is explainableat all. In Section 2, we examine the data used in this study. Section 3 discusses the modelsand variables, which are grouped into four broad categories. Section 4 focuses on differentaspects of the explanatory power of our models. Section 5 shifts attentions to individualfactors and identifies the ones that are important and interesting. Section 6 briefly dis-cusses the specification test for our nonparametric model and Section 7 concludes.

    2.Survey o consumersTe index of consumer sentiment (ICS) was developed by George Katona (1951) as ameasure of consumers changing attitudes about the business conditions and job pro-spects, and the Survey Research Center at the University of Michigan has continued toproduce it on a regular basis to measure consumer confidence in the United States. TeICS has generally led the U.S. business cycle and the 1990-1991 recession has been wide-ly attributed to a drop in the ICS in the wake of Iraqs invasion of Kuwait. Te SRCssurvey of Consumer Attitudes and Behavior on which the ICS is based, started in 1952,and has been conducted quarterly since 1960 and monthly since 1978. Te five questionson which the ICS is based are given below, with their respective range of answers andFigure 1 shows the index itself as well as the five components of the ICS, correspondingto the following questions.

    PerFin_Current: We are interested in how people are getting along financially thesedays. Would you say that you (and your family living there) are better off or worse off fi-nancially than you were a year ago? Possible responses are: better off; same; worse off.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    6/41

    6

    PerFin_Expected: Now looking ahead do you think that a year from now you (andyour family living there) will be better off financially or worse off, or just about the sameas now? Possible responses are: better off; same; worse off.

    BusCond_12m: Now turning to business conditions in the country as a whole doyou think that during the next 12 months well have good times financially, or bad time,or what? Possible responses are: good times; good with qualifications; pro-con; bad withqualifications; bad times4.

    BusCond_5y: looking ahead, which would you say is more likely that in the coun-try as a whole well have continuous good times during the next five years or so, or thatwell have periods of widespread unemployment or depression, or what? Possible re-sponses are: good times; good with qualifications; pro-con; bad with qualifications; badtimes.

    BuyCond: About the big things people buy for their homes such as furniture, refrig-

    erator, stove, television, and things like that. Generally speaking do you thin know is agood or a bad time to buy major household items? Possible responses are: good; pro-con;bad.

    Based on the responses to the above five questions, we can compute the balance sta-tistic as follows

    % of respondents who answer better or good % of respondents who answer worse or bad) + 100Ten simple average of the five balance statistics is computed to obtain index of

    consumer sentiment. Te ICS is reported relative the base month February 1960 = 100.

    For our time series analysis we use monthly data on the ICS dating from January 1978 toMarch 2009 (375 months). Tis is the most recent and complete data available to thepublic at the time this study began.

    Te design of the survey allows us to extract additional information about individu-als expectations and perceptions of general and personal economic conditions, as well aspersonal characteristics, which is not contained in the five index questions alone. Specifi-cally, the survey asks questions regarding the individuals perception of how the economyhas done over the past one year, how the government is doing its job at present, andwhether the individual has recently heard any good or bad news about the general eco-nomic and political situation in the country. Tere are also questions about the individuals

    expectations about general economic conditions over the next year, about prices, real in-come, unemployment, and interest rate on borrowing, all with answers in the three-response framework better/same/worse or up/same/down. Tere is also information aboutthe respondents employment status, education, marital status, gender, race, age, region of

    4 Responses good time / good with qualifications and bad with qualifications / bad times are grouped together respec-tively for index construction. Same for the next question.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    7/41

    7

    residence, household size, and annual income. In addition to the response to the five indexquestions, we use information on more than 50 other questions to explain response toeach of the component questions at the individual level. We used raw data sets frommonthly survey, which amounts to a total of 375 months. We have a final sample contains

    156,070 observations.

    3.Model and variablesAs outlined previously, we utilize the information from over 50 questions from the surveyof consumers asked consistently throughout the sample period 1978:01 2009:03 to ana-lyze the factors, which determine individual responses to each of the five ICS componentquestions in an ordered response model. Te variables we define are explained in able 1.

    In this study we are specifically interested in (1) how much of the variation in re-

    sponses to each question can be explained by expectations and perceptions of generaleconomic conditions; how much is due to personal idiosyncrasies; and how much is animmediate reflection of widely known economic conditions ; and (2) what are the maindeterminants of the responses to each question; are they the same for all of the five com-ponents; and what effect do they have on the components of sentiment measure.Answering these questions will enable us to better discriminate among the existing hy-potheses about the informational content of the ICS.

    As stated before, we combine the response, when necessary, so that for all of the fivecomponents, we have three different kinds of responses: good / same / bad or up / same /down. Tis enables us to consider the questions raised before in a unified latent-variableframework. Let

    = + where the index corresponds to individual interviewed at time , with lower values of denoting higher optimism, i.e., better/up = 2, same = 1, worse/down = 0, and

    = , , , )is a vector of variables affecting an individuals level of optimism, in which is individu-als expectation and perception, is a set of variables explaining individual idiosyncrasies, is a set of macroeconomic variables and is a set of macroeconomic forecasts. Te da-ta for

    comes from monthly mean of Blue Chip forecasts. Te last column of able 1shows which variable belongs to which category. Note that a set of monthly dummy vari-

    ables are included in the category and a dummy variable for whether the economy is ina recession is included in the category. More specifically, for each category

    Expectations/Perceptions () category contains individuals perception on overallnews, news specifically related to inflation, news specifically related to unemployment,

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    8/41

    8

    news specifically related to war, current business condition compared with that from oneyear ago and the performance of government economic policies, as well as individuals ex-pectation of future business conditions, future interest rate, future price level, future realincome and future unemployment.

    Personal characteristics/idiosyncrasies (

    ) category contains individuals age, real per

    capita family income, racial origin, marital status, education level, gender and location ofresidence.

    Macroeconomics conditions () category contains the levels and percentage changesfrom previous month of the index of coincident indicators, index of leading indicators,industrial production, S&P 500 index, personal income and purchasing managers index,as well as the levels of inflation rate, 3-month treasury bill rate, 5-year government bondrate, unemployment rate, along with a dummy variable indicating NBER recession andthe standard deviation of the S&P 500 index in the previous month.

    Macroeconomic forecasts(

    ) category contains monthly mean of Blue Chip forecasts,

    with one year horizon, of inflation rate, real GDP and unemployment rate.

    Given the latent variable , the observation rule is =

    0 if < 11 if 1 < 22 if 2

    where is an independently distributed error term with distribution function ). Teprobability of observing response 0, 1, 2) is given by

    Pr[ = 0] = 1 )

    Pr[ = 1] = 2 ) 1 )Pr[ = 2] = 1 2 )

    For identification, we assume that ) is known and there is no constant in . So themodel can be estimated by maximum likelihood. Te likelihood function is given by

    ln)= Pr =)

    =1

    =1

    For the error distribution F, we consider two types of specification here. Firstly we

    consider a usual Probit5 specification, where F = is a normal distribution function. Sec-5 A logit specification is also considered and proved to produce very similar results as that produced by a probit specifi-cation. hus, the result from logit model is omitted from this paper but is available from the authors upon request.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    9/41

    9

    ondly, we consider a nonparametric specification (Stewart, 2003) where the density of theerror distribution is approximated by a Hermite form

    )=1 (

    =0

    2

    )

    where ) is standard normal density and = (

    =0

    2 ) In the density, we normalize 0 = 0 so that the density is invariant to multiplication ofthe s. o achieve identification, we also normalize the first threshold to be the same asthe first threshold estimated using a normal distribution. (Gabler et al., 1993; Melenbergand Soest, 1996) Based on this density function, the distribution function of the errorterm is thus given by

    )= =0 2) =0 2)

    which is a family of distributions given parameter . As increases, the approximationbecomes finer and the computational cost increases accordingly. Based on the willingnessto consider normal distribution as an alternative to this more general form of nonpara-metric distribution, we are accepting the assumption that the true error distribution is notwildly oscillatory, which means the above nonparametric distribution is able to approxi-

    mate the true error distribution well. Under regularity conditions, the parameters in thenonparametric model can be estimated consistently given that is increasing with sam-ple size (Gallant and Nychka, 1987). In this study, is chosen for each of the fivecomponents separately according to the BIC statistic.

    o facilitate the comparison of the effects of different variables on components ofsentiment index, alongside the estimated coefficients we show the marginal effects foreach possible value of the dependent variable. Te marginal effect of at = measures the change in the predicted probability that will take on the value resultingfrom a unit increase in . Te standard methodology for computing marginal effects isto take derivative of the predicted probability of = with respect to . For the non-parametric model in this study, the predicted probabilities are

    Pr[ = 0] = 1 )Pr[ = 1] = 2 1 )

    Pr[ = 2] = 1 2 )

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    10/41

    10

    where the hatted values denote estimates. And the marginal effects are given by

    = 0 ) = 1 ) |=

    = 1 ) = [2 1 )] |=

    = 2 ) = [1 2 )] |= where the derivatives are evaluated at the sample means and the maximum likelihoodestimates , 1 and 2 . Tis computation is appropriate when is a continuous variable,but for a dummy variable , we use the following alternative method to calculate themarginal effect (Caudill and Jackson, 1989),

    = ) = P r = | = 1 , , , 1, 2 Pr = | = 0 , , , 1, 2 where 0, 1, 2).4.Explanatory powerTe first question we address here is how much variation in each of the five sentimentcomponents can be explained. o measure the explanatory power of our model, we use thepseudo- statistic proposed by McKelvey and Zavoina (1975), modified to take into ac-

    count the feature of the nonparametric distribution, which is given by

    2 = 2=1=1 2=1=1 +)where

    2 = = Te pseudo- measures the proportion of variation in the underlying latent variable explained by the model. able 2 shows the proportion of variation explained by the full

    model, and by each group of factors separately and the explanatory power of the modelfor the full sample and for the last 10 years separately. Te incremental reported in a-ble 2 refers to the change in pseudo- resulting from the addition of each group offactors to the rest of the variables.

    Overall explanatory power: From the table we can see that in terms of overall explan-atory power, the model explains from 17% to 56% of the total variation. BusCond_12m

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    11/41

    11

    the one-year ahead expectation of the overall business condition is best explained with apseudo- of 55.86% in probit model and 54.50% in nonparametric model. Te least wellexplained variables are PerFin_Current personal financial condition compared to a yearago and BuyCond buying condition for large household items, with a pseudo- of

    around 17-18% for both the probit model and the nonparametric model. Consideringthat BusCond_5y expectation of five year ahead business condition is the second bestexplained variable with a pseudo- about 40% and PerFin_Expected expectation ofone year ahead personal financial situation is the third best explained variable with apseudo- of about 30-40%, we come to the conclusion that the model explains expecta-tions about the overall economic condition better that it explains expectations aboutpersonal situation. Tis observation is further confirmed by that fact that macroeconomicvariables accounts for a larger proportion of variation in explaining overall business condi-tions. Another finding we get from able 2 is that in terms of overall explanatory power,the probit model provides a result that is highly consistent with what the nonparametric

    model provides. Te differences in pseudo-

    for four of the five variables are within 1 to2%, except for PerFin_Expected. While the probit model explains about 30% of the totalvariation of this variable, nonparametric model explains about 50%. Tis obvious increasein explanatory power may be due to the non-normality of the error distribution for thisvariable.

    Explanatory power of each category: In terms of the explanatory power of each catego-ry of variables, the two models again give highly consistent results. Generally, expectationsand perceptions category has the highest incremental explanatory power, ranging fromabout 10% for BuyCond and PerFin_Current to 52% for BusCond_12m. In the rest ofthe categories, personal idiosyncrasies category plays an important role in explaining Per-Fin_Current and PerFin_Expected, providing pseudo- of 8-10% for the former and 5-8% for the later. Macroeconomic variables and macroeconomic forecasts play equally im-portant role in explaining business conditions and buying condition. And not surprisingly,these two categories are more important in explaining BusCond_12m than the rest twovariables. Note that while overall explanatory power for BusCond_5y is much higher thanthat for BuyCond, macroeconomic variables and macroeconomic forecasts are much lessimportant in BusCond_5y. Tis shows that uncertainty of long-term expectation of busi-ness condition is so large that the variation caused by idiosyncratic factors is much morethan variations caused by the variation in the common information set current and pro-fessionally forecasted economic conditions.

    Incremental explanatory power: As for incremental explanatory power, while there are

    more notable differences in the results produced by the probit model and the nonpara-metric model, the general conclusions we can draw from able 2 in still clear. Once wecontrol for the rest of the variables, addition of macroeconomic variables and macroeco-nomic forecasts provide the least additional benefit, i.e., explanatory power. Tis isespecially obvious when looking at the results provided by the probit model. Te nonpar-ametric model shows that for PerFin_Expected, adding macroeconomic variables or

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    12/41

    12

    macroeconomic forecasts to the model with other variables already in it brings another18% pseudo- , which is quite notable considering that the benefit of doing so for othercomponents of sentiment index is usually quite negligible 6% for PerFin_Current andless than 1% for business conditions and buying condition.

    ime variation in explanatory power: Another interesting perspective from which wecan look at the results is the time variation in explanatory power. We first estimated theprobit model for the last 10 years (2000 to 2009). In terms of overall explanatory power,there is marginal improvement for four of the five components. For BuyCond, the ex-planatory power for the last 10 years is about 1% less than that for the entire sampleperiod from 1978 to 2009. In terms of the incremental explanatory power of each com-ponent, we find that despite the increase in overall explanatory power if we restrict thesample to the last 10 years, the contribution of macroeconomic variables and macroeco-nomic forecasts are generally even less. o show how the explanatory power evolves overthe years, we presented a year by year pseudo- plot for each of the five components in

    Figure 2. From the figure we can see that explanatory power for BusCond_12m andBusCond_5y basically follow the same trend, except that we see a dramatic increase inexplanatory power for BusCond_12m since 2008 while that for BusCond_5y remainedlow for the same period. Explanatory power for PerFin_Current and PerFin_Expected al-so share almost the same trend in the earlier years. But since 2007, explanatory power forPerFin_Current dropped sharply while that for PerFin_Expected remained the same. Wethink this is partly because the most recent recession is so strong that short term uncer-tainty spiked causing a decrease in explanatory power for personal variables but anincrease for economic conditions. In other words, during the recent recession, while con-sumers are more uncertain about their own conditions, they are less able to form anindependent view regarding the overall business conditions, resulting in an increased reli-ance on macroeconomic variables and forecasts in forming such view.

    5.Main determinantsWe have discussed the explanatory power of the model and its variation across the fivecomponents of the sentiment index and variation over the entire sample period. In thissection we focus on the individual factors in each category that has interesting and im-portant impact on the components of sentiment. able 3 to 7 give the regression result ofthe nonparametric model for each of the sentiment components where marginal effectsare reported alongside the estimated coefficients.

    Expectations and perceptions: We find that perception of government economic policyhas an important role in explaining PerFin_Current, BusCond_12m, and BusCond_5y.On average, the marginal effect of perception of government performance is about 2 to3%; meaning that other things being equal, the likelihood of people have a good senti-ment is 2-3% higher if they think the government is doing a good job. Interestingly, theeffect of this on PerFin_Expected and BuyCond is much smaller. A possible explanation

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    13/41

    13

    is that while consumers do think that government economic policy is important to theoverall economy and to the change in their own financial situation, they tend not to thinkof it as an important factor in forming expectation about their own future and makingtheir own purchasing decisions. Meanwhile, we find that the effect on the sentiment is

    bigger if consumers think the government is doing a poor job than the effect if consumersthink the government is doing a good job. We also find that whether consumers haveheard any good or bad news mainly affects BusCond_12m but not the rest of the compo-nents, in particular, BusCond_5y. Tis may be because of the time-sensitiveness nature ofnews stories they mostly focus on whats happening now instead of on whats going tohappen in a relatively long term. In addition, our result shows that good news has a biggereffect than bad news, somehow contradicting to traditional wisdom that consumers aremore sensitive to bad news. Te marginal effect of good news on BusCond_12m is about3% while that of bad news is less than 1%. In the rest of the variables in this category,consumers perception and expectation of overall economic condition, price level, and real

    income are the most important factors in terms of magnitude of marginal effects.Personal idiosyncrasies: In terms of racial origin, we find that white consumers gener-

    ally have higher sentiment, in all five sentiment components, then black consumers, thenHispanics. For example, white consumers are about 2-3% more likely to think that theirown financial situations as well as the general economic conditions have been improvingand will continue to improve in the future. Tis number is generally less than 1% (or notstatistically significant) for black consumers and Hispanics. In terms of gender, male con-sumers tend to have higher sentiment, especially for long term expectations BusCond_5y, where mail consumers are 4% more likely to have higher sentiment or un-likely to have lower sentiment. For other components of sentiment, the marginal effect ofbeing male is generally 2-3%. Te level of education also has a significant effect on con-sumer sentiment, though the effect is not so clear for some components of sentiment.Our result shows that consumers with higher education level have significantly highersentiment in PerFin_Current and BusCond_5y consumers with a college degree orhigher are about 1.5% more likely to have higher sentiment than consumers without highschool diploma. For the rest of the sentiment components, the effect of education is notclear. As for residence location, we find that for current personal financial situation Per-Fin_Current, consumers residing in the south have higher sentiment then consumersresiding in the west and the consumers residing in the northeast have lowest sentiment.For expected personal financial situation PerFin_Expected and 12-month ahead expecta-tion of business conditions BusCond_12m, consumers from the west have higher

    sentiment than consumers from northeast. For the rest of the sentiment components, theeffect of location of residence is not clear.

    Macroeconomic variables and forecasts: Macroeconomic variables and macroeconomicforecasts are generally not so important in determining personal financial situations, bothpast and expected, PerFin_Current and PerFin_Expected, though an increase in unem-ployment increases consumers sentiment in PerFin_Current. Te rationality behind this

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    14/41

    14

    seemingly unreasonable result may be that the worse the condition for their peers, thebetter the consumers feel for themselves, provided that most of them are still employed.As for business conditions, changes in industrial production, forecasted inflation, 5-yeargovernment bond rate and unemployment rate has strong effects on BusCond_12m,

    while the index of coincident indicators and forecasted unemployment rate strongly affectBusCond_5y. For example, a 1% increase in industrial production makes consumers1.13% more likely to think that the overall business condition will be better in one year; a1% increase in the coincident indicator makes consumers 1.27% more likely to think thatbusiness condition will be better in five years. Unemployment also has a strong effect onBuyCond.

    Information content of sentiment: While the current setting of the model does notpermit a statistical test on the information content of consumer sentiment, the aboveanalysis and findings do provide some solid clues. In general, we find that sentiment cap-tures consumers perception and expectation of economic fluctuations, based on their day-

    to-day experience (hypothesis 3 and 4). Firstly, variation in the sentiment components canbe largely explained by variation in consumers perception and expectations of their ownsituation, the information they get from news media as well as the condition of the econ-omy, at least the part of the economy they experience. If sentiment was an unmediatedexperience, and reflects nothing more than official statistical reports, professional forecastsand noise, as in hypothesis 5, we would expect actual lagged macroeconomic variables andmacroeconomic forecasts to be able to explain a large portion of the variation in senti-ment on the individual level. However, we find contrary. Moreover, macroeconomicvariables and macroeconomic forecasts, by themselves, do not provide sufficiently largeexplanatory power, as can be seen f rom able 2. Secondly, sentiment is clearly not a directreflection of political and business news, otherwise, our news variables would be expectedto capture a lot more variations then what they do now. Not to mention that even they docapture a lot more variation, they are still the perception of the consumers, i.e., differentconsumers may interpret the same piece of news differently, especially when such inter-pretation is used to form expectation of the future. Our results show that new variablesgenerally have less than 2-3% marginal effect, about the same as all other important vari-ables. In particular, news of war, unemployment and inflation do not have any importanteffects on sentiment components, which make us believe that consumers tend not tooverreact to particular word(s) that appear in news report. So it is highly unlikely that onenews story containing a particular bad word, e.g., the r word recession would trigger asudden fall in consumer sentiment. Tis is supported by the result that good news has

    larger effect on sentiment compared with bad news. Te weak performance of news varia-bles as determinants of sentiment directly contradicts the results of Blood and Phillips(1995), who find that recession related headline news have a significant negative prior in-fluence on ICS, even after accounting for the actual state of the economy. Our findingssupport the view of Linden (1982) and Blendon, et al. (1997) that consumers expecta-tions are formed in conversations between neighbors over the backyard fence and arenot a direct reflection of media coverage or published statistics (Garner, 1991). Finally, we

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    15/41

    15

    find a clear difference between consumers sentiment regarding their own personal situa-tions and the conditions of the overall economy. Teir personal situations are more likelyto be affected by their demographic characteristics and other idiosyncrasies rather thanmacroeconomic variables or their forecasts. But macroeconomic variable do greatly affect

    consumers sentiment on business conditions. Expected business conditions have marginaleffects about 4-6% for BusCond_12m and BusCond_5y. Expected prices, expected in-come and expected unemployment situation generally have marginal effects about 1-2%for BusCond_12m and BusCond_5y. Expected business condition and expected employ-ment are also important in BuyCond, with marginal effects more than 1%.

    6.Specifcation testSince both the probit model and the nonparametric model are estimated by maximumlikelihood method, we can test the validity of the nonparametric distribution against

    normal distribution using a standard likelihood ratio test. Let

    0be the log-likelihood

    value of the nonparametric model and 1 be the log likelihood value of the probit model,with 0 the degree of freedom of the nonparametric model and 1 the degree of freedomfor the probit model, the test statistic is

    = 21 0)which is approximately2 with 0 1) degrees of freedom.

    Te result of the test is given in able 8. As we can see from the table, nonparamet-ric specification is accepted for all five components of the index of consumer sentiment.Tough we have to note here that while nonparametric specification does indeed producedifferent result than the probit specification, the basic conclusions regarding the overallexplanatory power and the importance of variables one can draw from the two set of re-sults are the same. From Panel IV of able 2 we can see that nonparametric modelprovides stronger explanatory power than the probit model when only variables from eachcategory are included. We also note here that nonparametric estimation used in this studyis highly computationally costly and probit model is a good alternative in this sense.

    7.SummaryIn this study, we focus on the informational content of the Index of Consumer Sentiment,

    in particular, its five components. Models based on household level survey data are esti-mated assuming normal error distribution as well as assuming nonparametric distribution.Our results provide some solid clue on how much we can explain variations in the com-ponents of consumer sentiment, what are the important factors affecting components ofconsumer sentiment, and what information consumer sentiment index reflects.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    16/41

    16

    We find that up to 48% of variations in personal sentiment, and up to 56% of varia-tions in sentiment on business conditions can be explained by variables includingconsumers perception and expectation of social economic fundamentals, macroeconomicvariables and their professional forecasts, and consumers demographic characteristics such

    as gender, racial origin, education level, age, and region of residence. Among them, con-sumers perception and expectation are the most important category of variables,accounting for 11% to 53% of the total variation in components of sentiment. Macroeco-nomic variables and forecasts generally provide much lower explanatory power around 5%to 20%, especially for personal sentiment components where they only provide about 3%to 5% explanatory power.

    Te most important and influential factors affecting components of sentiment at theindividual and household level are consumers expectation of future business conditionsand unemployment level, as well as consumers perception of the performance of govern-ment economic policies and news stories. Consumers response to general questions

    regarding, for example, the overall business condition contains complex information fromall sources and balance statistics constructed for this kind of questions are not easy to in-terpret. Our results thus suggest the incorporation of the statistics constructed from directquestions about expected unemployment or family income in the sentiment index.

    Te above finding suggests that the informational content of consumer sentiment iscomplex yet tractable. Te information contained in sentiment measure is neither purelymacroeconomic fluctuation nor the tone of news reports delivered by mass media. Rather,different information takes up the role of the principle driving force for different compo-nent of consumer sentiment. It is more informative if components of the sentiment indexare treated separately in economic and econometric models. When the components are

    combined, information of the driving force of changes in sentiment may be obscured.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    17/41

    17

    Reerences

    Acemoglu D, Scott A. Consumer confidence and rational expectations: are agents beliefs con-

    sistent with the theory? Te Economic Journal. 1994;104(422):119.

    Berg L, Bergstrm R. Consumer Confidence and Consumption in Sweden. 1996.

    Blendon RJ, Benson JM, Brodie M, et al. Bridging the Gap between the Publics and Economists'

    Views of the Economy. Te Journal of Economic Perspectives. 1997;11(3):105118.

    Blood DJ, Phillips PCB. Recession Headline News, Consumer Sentiment, Te State of the Econ-

    omy and Presidential Popularity: A ime Series Analysis 1989-1993. International Journal of

    Public Opinion Research. 1995;7(1):2.

    Bram J, Ludvigson S, York FRBN. Does consumer confidence forecast household expenditure?: A

    sentiment index horse race. Federal Reserve Bank of New York; 1997.

    BRESLAW J, McIntosh J. Simulated latent variable estimation of models with ordered categori-

    cal data.Journal of Econometrics. 1998;87(1):25-47.

    Carroll CD, Fuhrer JC, Wilcox DW. Does consumer sentiment forecast household spending? If

    so, why? Te American Economic Review. 1994;84(5):13971408.

    Caudill SB, Jackson JD. Measuring marginal effects in limited dependent variable models. Te

    Statistician. 1989;38(3):203206.

    Easaw JZ, Garratt D, Heravi SM. Does consumer sentiment accurately forecast UK household

    consumption? Are there any comparisons to be made with the US? Journal of Macroeconomics.

    2005;27(3):517532.

    Estelami H, Lehmann DR, Holden AC. Macro-economic determinants of consumer price

    knowledge: A meta-analysis of four decades of research.International Journal of Research in Mar-

    keting. 2001;18(4):341355.

    Fan CS, Wong P. Does consumer sentiment forecast household spending?:: Te Hong Kong case.

    Economics Letters. 1998;58(1):7784.

    Fuhrer JC. What role does consumer sentiment play in the US macroeconomy? New England

    Economic Review. 1993;(Jan):3244.

    Gabler S, Laisney F, Lechner M. Seminonparametric Estimation of Binary-Choice Models with

    an Application to Labor-Force Participation. Journal of Business and Economic Statistics.

    1993;11(1):61-80.

    Gallant AR, Nychka DW. Semi-nonparametric Maximum Likelihood Estimation.Econometrica.

    1987;55(2):363-390.

    Garner CA. Economic determinants of consumer sentiment. Journal of Business Research.

    1981;9(2):205220.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    18/41

    18

    Garner A. Forecasting consumer spending: should economists pay attention to consumer confi-

    dence surveys?Economic. 1991:57.

    Helfenstein R, Wolter SC. Unemployment, consumer confidence and private consumption in

    Switzerland. In: Business and Economics for the 21st Century, Volume 1.; 1997.Jennings EJ, McGrath P. Consumer sentiment and international conditions. Te Journal of Eco-

    nomics. 1994;XX:17-21.

    Juster F, Wachtel P. Anticipatory and objective models of durable goods demand. Te American

    Economic Review. 1972;62(4):564579.

    Katona G. oward a macropsychology.American Psychologist. 1979;34(2):118126.

    Katona G. Psychological analysis of economic behavior. McGraw-Hill New York; 1951.

    Linden F. Te consumer as forecaster. Public Opinion Quarterly. 1982;46(3):353360.

    Lovell MC. Why Was the Consumer Feeling So Sad? Brookings Papers on Economic Activity.

    1975:473479.

    Ludvigson SC. Consumer confidence and consumer spending. Journal of Economic Perspectives.

    2004;18(2):2950.

    MacKuen MB, Erikson RS, Stimson JA. Peasants or bankers? Te American electorate and the

    US economy. Te American Political Science Review. 1992;86(3):597611.

    Maddala GS. Limited-dependent and qualitative variables in econometrics. Cambridge Univ Pr;

    1986.

    McKelvey RD, Zavoina W. A statistical model for the analysis of ordinal level dependent varia-

    bles. Te Journal of Mathematical Sociology. 1975;4(1):103120.Melenberg B, Soest AHOV. Measuring the costs of children: Parametric and semiparametric es-

    timators. Statistica Neerlandica. 1996;50:171-192.

    Mishkin FS, Hall R, Shoven J, Juster , Lovell M. Consumer sentiment and spending on durable

    goods. Brookings Papers on Economic Activity. 1978:217232.

    Praet P, Vuchelen J. Te contribution of consumer confidence indexes in forecasting the effects of

    oil prices on private consumption.International Journal of Forecasting. 1989;5(3):393397.

    Stewart MB. Semi-nonparametric Estimation of Extended Ordered Probit Models. Te Stata

    Journal. 2003;4(1):27-39.

    Troop AW. Consumer sentiment: Its causes and effects. Federal Reserve Bank of San FranciscoEconomic Review. 1992;1:3559.

    Vuchelen J. Consumer sentiment and macroeconomic forecasts. Journal of Economic Psychology.

    2004;25(4):493506.

    Vuchelen J. Political events and consumer confidence in Belgium. Journal of economic psychology.

    1995;16(4):563579.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    19/41

    19

    Table 1 - Variable Description

    Variable Name Descri tion GrouYYYYMM Year and month of the observation

    BusCond_12m Expected overall business conditions in 1 yearBusCond_5y Expected overall business conditions in 5 yearsBuyCond Whether now is a good time to buy major household itemsPerFin_Current Personal financial situation compared to 1 year agoPerFin_Expected Expected personal financial situation in 1 yearBadNewsOnly rue if GoodNews = 0 and BadNews = 1 EBNInfl Bad news about inflation has been heard in past several months EBNUnemp Bad news about unemployment has been heard in past several months EBNWar Bad news about war/defense has been heard in past several months EBusCond_Better Business condition better than one year ago EBusCond_Worse Business condition worse than one year ago EEBusCond_Better Expect business conditions to be better in one year EEBusCond_Worse Expect business conditions to be worse in one year EEInt_Rate_Down Expect interest rate to go down in one year E

    EInt_Rate_Up Expect interest rate to go up in one year EEPrices_Down Expect prices to go down in one year EEPrices_Up Expect prices to go up in one year EEReal_Income_down Expect income to increase less than price in one year EEReal_Income_Up Expect income to increase more than price in one year EEUnempl_Less Expect lower unemployment rate in one year EEUnempl_More Expect higher unemployment rate in one year EGNInfl Good news about inflation has been heard in past several months EGNUnemp Good news about unemployment has been heard in past several months EGNWar Good news about war/defense has been heard in past several months EGoodGovt Government is considered doing a good job EGoodNewsOnly rue if GoodNews = 1 and BadNews = 0 EPoorGovt Government is considered doing a poor job EAge Age IAgeSq Square of age IApr rue if interviewed in Apr IAug rue if interviewed in Aug IBlack Black IDivorced Divorced IFeb rue if interviewed in Feb IGrade_9_11 Highest education level is grade 9 to 11 IHigh_School Highest education level is high school IHispanic Hispanic IIncome otal real annual family income adjusted for size of family IJan rue if interviewed in Jan IJul rue if interviewed in Jul IJun rue if interviewed in Jun IMale Male IMar rue if interviewed in Mar IMarried Married IMay rue if interviewed in May INorth_Central Lives in north central INortheast Lives in northeast INov rue if interviewed in Nov IOct rue if interviewed in Oct ISep rue if interviewed in Sep I

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    20/41

    20

    Some_College Highest education level is college but no degree ISouth Lives in south IWhite White IWidowed Widowed IC_ICI Change in ICI from previous month M

    C_ILI Change in index of leading indicators MC_Indpro Change in industrial production from previous month MC_PI Change in personal income from previous month MC_PMI Change in PMI from previous month MC_SP500 Change in SP500 index from previous month MICI Index of composite indicators MILI Index of leading indicators MInfl Inflation rate MIrate_3m 3 month -bill rate MIrate_5y 5 year government bond rate MPMI Purchasing managers' index MRecession rue if the economy is in recession when interview conducted MStd_SP500 Standard deviation of SP500 index MUnemp Unemployment rate M

    F_Infl Forecasted inflation rate F F_RGDP Forecasted real GDP F F_Unemp Forecasted unemployment rate F

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    21/41

    21

    Table 2 Summary Statistics

    Variable Mean Std. Dev. Variable Mean Std. Dev.YYYYMM 199271 898.335 Hispanic 0.0457 0.2088

    BusCond_12m 1.0565 0.9811 Income 8.5866 9.5923BusCond_5y 0.9320 0.9494 Jan 0.0848 0.2785BuyCond 1.5103 0.8350 Jul 0.0848 0.2786PerFin_Current 1.1493 0.8478 Jun 0.0849 0.2788PerFin_Expected 1.2698 0.6610 Male 0.4576 0.4982BadNewsOnly 0.1822 0.3860 Mar 0.0795 0.2705BNInfl 0.0526 0.2232 Married 0.6053 0.4888BNUnemp 0.1162 0.3205 May 0.0810 0.2728BNWar 0.0120 0.1088 North_Central 0.2738 0.4459BusCond_Better 0.4150 0.4927 Northeast 0.1931 0.3948BusCond_Worse 0.4687 0.4990 Nov 0.0851 0.2791EBusCond_Better 0.2874 0.4526 Oct 0.0842 0.2777EBusCond_Worse 0.1971 0.3978 Sep 0.0845 0.2781EInt_Rate_Down 0.1885 0.3911 Some_College 0.2289 0.4201

    EInt_Rate_Up 0.5075 0.4999 South 0.3328 0.4712EPrices_Down 0.0361 0.1866 White 0.8436 0.3632EPrices_Up 0.7962 0.4028 Widowed 0.0760 0.2649EReal_Income_Down 0.3523 0.4777 C_ICI_L1 0.1486 0.3210EReal_Income_Up 0.2136 0.4099 C_ILI_L1 0.1812 0.6855EUnempl_Less 0.1582 0.3649 C_Indpro_L1 0.1564 0.6898EUnempl_More 0.3583 0.4795 C_PI_L1 0.5342 0.5381GNInfl 0.0157 0.1241 C_PMI_L1 -0.0745 4.9047GNUnemp 0.0383 0.1919 C_SP500_L1 0.7440 4.3604GNWar 0.0021 0.0462 ICI_L1 80.5226 15.9750GoodGovt 0.2265 0.4185 ILI_L1 70.3767 19.6007GoodNewsOnly 0.5008 0.5000 Infl_L1 4.2227 4.3971PoorGovt 0.2635 0.4405 Irate_3m_L1 6.1599 3.4187Age 43.8424 16.342 Irate_5y_L1 7.4914 3.2435AgeSq 2189.24 1599.0 PMI_L1 50.9964 6.1926Apr 0.0812 0.2731 Recession 0.1694 0.3751Aug 0.0843 0.2779 Std_SP500_L1 10.2174 11.2467Black 0.0866 0.2813 C_SP500_L1 0.7440 4.3604Divorced 0.1399 0.3469 Unemp_L1 6.2351 1.5001Feb 0.0815 0.2736 F_Infl_L1 4.2195 2.5306Grade_9_11 0.1021 0.3028 F_RGDP_L1 2.6334 1.5390High_School 0.3163 0.4650 F_Unemp_L1 6.2893 1.3942

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    22/41

    22

    Table 3 Regression Result Current Financial Situation

    Variable Coe. Std. Err. z P>z ME(y=1) ME(y=3)GoodGovt 0.1777 0.0084 21.050 0.0000 -0.0059 0.0765

    PoorGovt -0.2289 0.0074 -31.060 0.0000 0.0373 -0.1026BusCond_Better 0.2746 0.0108 25.380 0.0000 -0.0006 0.1158BusCond_Worse -0.1594 0.0091 -17.510 0.0000 0.0226 -0.0713Married 0.1183 0.0100 11.810 0.0000 -0.0152 0.0529Divorced -0.0028 0.0125 -0.220 0.8240 0.0003 -0.0012Widowed -0.0055 0.0140 -0.390 0.6970 0.0005 -0.0024Grade_9_11 -0.1997 0.0115 -17.310 0.0000 0.0308 -0.0895High_School -0.1349 0.0081 -16.630 0.0000 0.0181 -0.0603Some_College -0.0921 0.0085 -10.810 0.0000 0.0111 -0.0411White 0.0756 0.0210 3.590 0.0000 -0.0087 0.0337Black 0.0718 0.0239 3.000 0.0030 -0.0049 0.0315Hispanic 0.0158 0.0253 0.620 0.5340 -0.0014 0.0070North_Central -0.0149 0.0089 -1.680 0.0930 0.0014 -0.0066Northeast -0.0610 0.0095 -6.400 0.0000 0.0068 -0.0272

    South 0.0311 0.0087 3.580 0.0000 -0.0025 0.0137Male 0.0400 0.0063 6.350 0.0000 -0.0031 0.0176Age -0.0427 0.0011 -40.290 0.0000 0.0017 -0.0083AgeSq 0.0003 0.0000 26.770 0.0000 NA NAIncome 0.0151 0.0009 17.470 0.0000 -0.0014 0.0067Jan 0.0327 0.0145 2.250 0.0240 -0.0033 0.0145Feb 0.0281 0.0149 1.890 0.0590 -0.0023 0.0124Mar 0.0276 0.0148 1.870 0.0610 -0.0023 0.0122Apr 0.0293 0.0147 2.000 0.0460 -0.0024 0.0129May 0.0094 0.0145 0.650 0.5160 -0.0008 0.0042Jun 0.0111 0.0146 0.760 0.4460 -0.0010 0.0049Jul 0.0510 0.0149 3.440 0.0010 -0.0038 0.0224Aug 0.0456 0.0146 3.110 0.0020 -0.0035 0.0201Sep 0.0182 0.0145 1.250 0.2100 -0.0016 0.0080Oct -0.0121 0.0145 -0.840 0.4030 0.0011 -0.0054

    Nov 0.0040 0.0145 0.280 0.7830 -0.0004 0.0018Std_SP500_L1 -0.0015 0.0004 -3.930 0.0000 0.0001 -0.0007Unemp_L1 -0.0670 0.0037 -18.070 0.0000 0.0061 -0.0296Irate_3m_L1 -0.0263 0.0040 -6.550 0.0000 0.0024 -0.0117Infl_L1 -0.0033 0.0011 -3.180 0.0010 0.0003 -0.0015Infl_L2 -0.0018 0.0010 -1.750 0.0800 0.0002 -0.0008Irate_5y_L1 0.0312 0.0046 6.810 0.0000 -0.0028 0.0138C_12m_SP500 0.0009 0.0002 3.740 0.0000 -0.0001 0.0004C_12m_Indpro 0.0016 0.0014 1.130 0.2580 -0.0001 0.0007C_ILI_L1 0.0181 0.0056 3.250 0.0010 -0.0016 0.0080

    Tresholds SNP Coefs Est. Error dist.1 -1.9007 1 0.2642 Variance 1.27112 -1.1448 2 -0.0465 Skewness -0.5636

    3 -0.1017 Kurtosis 4.4083

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    23/41

    23

    Table 4 Regression Result Expected Financial Situation

    Variable Coe. Std. Err. z P>z ME(y=1) ME(y=3)GoodNewsOnly -0.0139 0.0107 -1.3000 0.195 -0.0027 -0.0034

    BadNewsOnly -0.0467 0.0135 -3.4500 0.001 -0.0088 -0.0114GNInfl 0.1026 0.0368 2.7900 0.005 0.0209 0.0255BNInfl -0.0533 0.0209 -2.5500 0.011 -0.0101 -0.0130GoodGovt 0.0526 0.0118 4.4500 0.000 0.0105 0.0130PoorGovt -0.1382 0.0114 -12.0800 0.000 -0.0249 -0.0332BusCond_Better 0.0646 0.0143 4.5200 0.000 0.0129 0.0160BusCond_Worse -0.0739 0.0146 -5.0700 0.000 -0.0138 -0.0179EBusCond_Better 0.4508 0.0152 29.6400 0.000 0.1028 0.1175EBusCond_Worse -0.4268 0.0139 -30.6900 0.000 -0.0629 -0.0988EPrices_Down 0.0159 0.0269 0.5900 0.555 0.0031 0.0039EPrices_Up -0.0337 0.0119 -2.8300 0.005 -0.0066 -0.0083EReal_Income_Up 1.0337 0.0291 35.5600 0.000 0.2487 0.2846EReal_Income_Down -0.5687 0.0153 -37.2300 0.000 -0.0741 -0.1292EUnempl_Less 0.1080 0.0140 7.7300 0.000 0.0220 0.0269

    EUnempl_More -0.1027 0.0108 -9.4700 0.000 -0.0189 -0.0248EInt_Rate_Up 0.0092 0.0101 0.9100 0.363 0.0018 0.0023EInt_Rate_Down 0.0379 0.0131 2.8800 0.004 0.0075 0.0093Married 0.0128 0.0138 0.9300 0.352 0.0025 0.0031Divorced 0.2538 0.0181 13.9800 0.000 0.0547 0.0644Widowed 0.1225 0.0213 5.7400 0.000 0.0251 0.0305Grade_9_11 -0.1599 0.0168 -9.5100 0.000 -0.0284 -0.0384High_School -0.0214 0.0113 -1.8900 0.058 -0.0041 -0.0052Some_College 0.0615 0.0123 5.0000 0.000 0.0123 0.0152White 0.1024 0.0297 3.4500 0.001 0.0188 0.0248Black 0.3827 0.0348 11.0000 0.000 0.0858 0.0989Hispanic 0.0938 0.0368 2.5500 0.011 0.0190 0.0233North_Central -0.1230 0.0134 -9.1900 0.000 -0.0223 -0.0297Northeast -0.1606 0.0145 -11.0700 0.000 -0.0285 -0.0385South -0.0230 0.0128 -1.8000 0.072 -0.0044 -0.0056

    Age -0.0018 0.0016 -1.1300 0.258 -0.0035 -0.0045AgeSq -0.0002 0.0000 -11.1600 0.000Income -0.0011 0.0005 -2.0200 0.044 -0.0002 -0.0003Recession -0.0322 0.0140 -2.3100 0.021 -0.0062 -0.0079C_SP500_L1 0.0022 0.0010 2.1600 0.030 0.0004 0.0006Infl_L1 -0.0032 0.0013 -2.4600 0.014 -0.0006 -0.0008ICI_L1 0.0002 0.0005 0.3800 0.701 0.0000 0.0000Irate_5y_L1 0.0093 0.0029 3.2200 0.001 0.0018 0.0023F_Infl_L1 -0.0322 0.0032 -10.0800 0.000 -0.0062 -0.0079F_Unemp_L1 -0.0712 0.0050 -14.3800 0.000 -0.0138 -0.0175C_ILI_L1 0.0167 0.0080 2.0900 0.036 0.0032 0.0041

    Tresholds SNP Coefs Est. Error dist.1 -2.0347 1 0.1982 Variance 2.01682 0.3957 2 -0.0115 Skewness -0.2507

    3 0.0461 Kurtosis 3.35604 0.0388

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    24/41

    24

    Table 5 Regression Result Business Condition 12m

    Variable Coe. Std. Err. z P>z ME(y=1) ME(y=3)GoodNewsOnly 0.1373 0.0116 11.8700 0.000 -0.0122 0.0559

    BadNewsOnly -0.1335 0.0143 -9.3200 0.000 0.0117 -0.0543GNWar -0.0566 0.0809 -0.7000 0.484 0.0041 -0.0228BNWar -0.1783 0.0379 -4.7000 0.000 0.0172 -0.0728GNUnemp 0.0817 0.0211 3.8600 0.000 -0.0036 0.0323BNUnemp -0.0167 0.0143 -1.1700 0.243 0.0011 -0.0067GNInfl 0.0858 0.0318 2.7000 0.007 -0.0037 0.0338BNInfl -0.0737 0.0208 -3.5500 0.000 0.0056 -0.0298GoodGovt 0.2762 0.0164 16.8400 0.000 -0.0001 0.1044PoorGovt -0.3306 0.0184 -17.9400 0.000 0.0407 -0.1359BusCond_Better 0.2910 0.0183 15.9400 0.000 0.0008 0.1095BusCond_Worse -0.5098 0.0253 -20.1900 0.000 0.0761 -0.2087EBusCond_Better 0.5419 0.0270 20.0600 0.000 0.0310 0.1884EBusCond_Worse -0.8572 0.0432 -19.8600 0.000 0.1557 -0.3363EPrices_Up -0.0537 0.0106 -5.0600 0.000 0.0027 -0.0213

    EPrices_Down -0.0212 0.0230 -0.9200 0.356 0.0014 -0.0085EReal_Income_Up 0.1129 0.0120 9.3900 0.000 -0.0041 0.0443EReal_Income_Down -0.2020 0.0129 -15.7100 0.000 0.0204 -0.0826EUnempl_Less 0.1653 0.0147 11.2700 0.000 -0.0042 0.0642EUnempl_More -0.4126 0.0202 -20.3800 0.000 0.0561 -0.1695EInt_Rate_Up -0.0946 0.0098 -9.7000 0.000 0.0038 -0.0373EInt_Rate_Down -0.0284 0.0117 -2.4200 0.015 0.0019 -0.0114Married 0.0474 0.0120 3.9700 0.000 -0.0034 0.0191Divorced 0.0028 0.0151 0.1800 0.854 -0.0002 0.0011Widowed -0.0486 0.0203 -2.4000 0.017 0.0034 -0.0196Grade_9_11 -0.0919 0.0159 -5.7700 0.000 0.0073 -0.0372High_School -0.0048 0.0100 -0.4700 0.635 0.0003 -0.0019Some_College 0.0144 0.0106 1.3600 0.173 -0.0008 0.0058White 0.0779 0.0261 2.9800 0.003 -0.0060 0.0315Black 0.0010 0.0289 0.0300 0.973 -0.0001 0.0004

    Hispanic -0.0441 0.0310 -1.4200 0.155 0.0031 -0.0178North_Central -0.0200 0.0115 -1.7400 0.082 0.0013 -0.0080Northeast -0.0586 0.0126 -4.6500 0.000 0.0043 -0.0236South 0.0088 0.0111 0.7900 0.429 -0.0005 0.0035Male 0.1576 0.0109 14.4700 0.000 -0.0042 0.0612Age -0.0078 0.0014 -5.4300 0.000 0.0001 -0.0004AgeSq 0.0001 0.0000 5.3700 0.000Income 0.0021 0.0004 4.8000 0.000 -0.0001 0.0008Jan 0.0757 0.0195 3.8700 0.000 -0.0058 0.0306Feb 0.0252 0.0195 1.2900 0.197 -0.0014 0.0100Mar 0.0580 0.0198 2.9300 0.003 -0.0028 0.0230Apr 0.0320 0.0192 1.6600 0.096 -0.0017 0.0128May 0.0373 0.0192 1.9400 0.052 -0.0020 0.0148Jun 0.0361 0.0192 1.8700 0.061 -0.0019 0.0143

    Jul 0.0275 0.0192 1.4300 0.151 -0.0015 0.0110Aug 0.0156 0.0190 0.8200 0.412 -0.0009 0.0062Sep 0.0428 0.0191 2.2400 0.025 -0.0022 0.0170Oct 0.0247 0.0192 1.2800 0.200 -0.0014 0.0098Nov 0.0333 0.0190 1.7500 0.080 -0.0018 0.0133Recession -0.1230 0.0159 -7.7300 0.000 0.0106 -0.0500C_SP500_L1 0.0090 0.0011 8.3700 0.000 -0.0006 0.0036C_SP500_L2 0.0102 0.0011 8.9300 0.000 -0.0006 0.0041C_SP500_L3 0.0068 0.0010 6.7100 0.000 -0.0004 0.0027

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    25/41

    25

    C_SP500_L4 0.0040 0.0010 4.0600 0.000 -0.0002 0.0016C_Indpro_L1 -0.0009 0.0077 -0.1100 0.909 0.0001 -0.0004C_Indpro_L2 0.0472 0.0073 6.4700 0.000 -0.0029 0.0189Unemp_L1 -0.1200 0.0067 -17.8100 0.000 0.0073 -0.0480Irate_3m_L1 -0.0207 0.0056 -3.6800 0.000 0.0013 -0.0083

    Irate_5y_L1 0.0613 0.0063 9.7700 0.000 -0.0037 0.0245Infl_L1 -0.0104 0.0015 -6.7500 0.000 0.0006 -0.0042Infl_L2 -0.0034 0.0015 -2.3100 0.021 0.0002 -0.0014C_ILI_L1 0.0360 0.0090 4.0000 0.000 -0.0022 0.0144F_RGDP_L1 0.0222 0.0038 5.9200 0.000 -0.0014 0.0089F_Infl_L1 -0.0515 0.0046 -11.1600 0.000 0.0031 -0.0206

    Tresholds SNP Coefs Est. Error dist.1 -1.0552 1 -0.0590 Variance 1.09412 -0.9172 2 -0.0496 Skewness 0.4289

    3 0.0291 Kurtosis 3.60174 0.0096

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    26/41

    26

    Table 6 Regression Result Business Condition 5y

    Variable Coe. Std. Err. z P>z ME(y=1) ME(y=3)GoodNewsOnly 0.0252 0.0086 2.9200 0.003 -0.0032 0.0100

    BadNewsOnly -0.0158 0.0111 -1.4200 0.155 0.0020 -0.0063GNWar 0.0788 0.0779 1.0100 0.311 -0.0087 0.0314BNWar -0.0649 0.0333 -1.9500 0.051 0.0086 -0.0256GNUnemp 0.0566 0.0179 3.1600 0.002 -0.0064 0.0225BNUnemp -0.0385 0.0125 -3.0800 0.002 0.0049 -0.0152GoodGovt 0.3327 0.0102 32.5700 0.000 -0.0221 0.1327PoorGovt -0.3045 0.0097 -31.2600 0.000 0.0495 -0.1164BusCond_Better 0.1050 0.0114 9.2100 0.000 -0.0111 0.0419BusCond_Worse -0.2086 0.0118 -17.7300 0.000 0.0316 -0.0810EBusCond_Better 0.4208 0.0100 42.1100 0.000 -0.0212 0.1671EBusCond_Worse -0.5734 0.0136 -42.2400 0.000 0.1073 -0.2073EPrices_Up -0.1446 0.0096 -15.0100 0.000 0.0143 -0.0578EPrices_Down -0.0762 0.0203 -3.7500 0.000 0.0102 -0.0300EReal_Income_Up 0.1250 0.0096 13.0800 0.000 -0.0128 0.0499

    EReal_Income_Down -0.2048 0.0085 -24.0000 0.000 0.0309 -0.0795EUnempl_Less 0.1991 0.0105 18.8800 0.000 -0.0179 0.0796EUnempl_More -0.4993 0.0107 -46.5900 0.000 0.0906 -0.1837EInt_Rate_Up -0.1220 0.0082 -14.8800 0.000 0.0125 -0.0487EInt_Rate_Down -0.0031 0.0103 -0.3000 0.767 0.0004 -0.0012Married 0.0359 0.0106 3.3800 0.001 -0.0046 0.0142Divorced -0.0233 0.0136 -1.7100 0.087 0.0029 -0.0092Widowed -0.0051 0.0183 -0.2800 0.779 0.0006 -0.0020Grade_9_11 -0.2979 0.0144 -20.7400 0.000 0.0482 -0.1140High_School -0.1495 0.0092 -16.3300 0.000 0.0215 -0.0585Some_College -0.0342 0.0094 -3.6300 0.000 0.0044 -0.0135White 0.0454 0.0233 1.9500 0.051 -0.0059 0.0179Black -0.1673 0.0261 -6.4100 0.000 0.0244 -0.0653Hispanic -0.1324 0.0284 -4.6700 0.000 0.0187 -0.0519Male 0.2444 0.0079 31.0800 0.000 -0.0200 0.0977

    Age -0.0021 0.0013 -1.5800 0.115 -0.0001 0.0004AgeSq 0.0000 0.0000 2.7200 0.007Income 0.0038 0.0005 8.0500 0.000 -0.0005 0.0015Recession 0.0544 0.0133 4.1000 0.000 -0.0062 0.0217Irate_5y_L1 0.0020 0.0020 1.0000 0.316 -0.0002 0.0008C_ICI_L1 -0.0289 0.0146 -1.9900 0.047 0.0035 -0.0115C_ICI_L2 -0.0164 0.0140 -1.1700 0.240 0.0020 -0.0065C_ICI_L3 -0.0229 0.0143 -1.6100 0.108 0.0028 -0.0091C_PMI_L1 -0.0014 0.0009 -1.5500 0.122 0.0002 -0.0006C_ILI_L1 0.0227 0.0071 3.1800 0.001 -0.0028 0.0090C_ILI_L2 0.0361 0.0071 5.0800 0.000 -0.0044 0.0143C_ILI_L3 0.0316 0.0068 4.6400 0.000 -0.0039 0.0125C_ILI_L4 0.0199 0.0061 3.2400 0.001 -0.0024 0.0079F_RGDP_L1 0.0170 0.0029 5.9100 0.000 -0.0021 0.0067

    F_Unemp_L1 -0.1221 0.0048 -25.5000 0.000 0.0149 -0.0484Tresholds SNP Coefs Est. Error dist.1 -1.0563 1 -0.0178 Variance 0.99262 -0.7398 2 -0.0023 Skewness 0.0786

    3 0.0066 Kurtosis 3.0089

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    27/41

    27

    Table 7 Regression Result Buying Condition

    Variable Coe. Std. Err. z P>z ME(y=1) ME(y=3)GoodNewsOnly 0.0257 0.0125 2.0500 0.041 0.0030 0.0051

    BadNewsOnly -0.0417 0.0149 -2.8000 0.005 -0.0049 -0.0084GNUnemp 0.0957 0.0280 3.4100 0.001 0.0119 0.0185BNUnemp -0.0427 0.0165 -2.5900 0.010 -0.0050 -0.0086GoodGovt 0.1297 0.0152 8.5100 0.000 0.0164 0.0248PoorGovt -0.2055 0.0155 -13.2800 0.000 -0.0222 -0.0430BusCond_Better 0.1953 0.0185 10.5400 0.000 0.0253 0.0365BusCond_Worse -0.2449 0.0241 -10.1600 0.000 -0.0259 -0.0518EBusCond_Better 0.0189 0.0124 1.5200 0.128 0.0023 0.0037EBusCond_Worse -0.1518 0.0156 -9.7400 0.000 -0.0169 -0.0314EPrices_Up 0.0286 0.0137 2.0900 0.037 0.0034 0.0057EPrices_Down -0.1355 0.0284 -4.7800 0.000 -0.0152 -0.0279EReal_Income_Up 0.0566 0.0141 4.0100 0.000 0.0069 0.0110EReal_Income_Down -0.1384 0.0132 -10.4900 0.000 -0.0155 -0.0285EUnempl_Less -0.0192 0.0153 -1.2500 0.210 -0.0023 -0.0038

    EUnempl_More -0.1880 0.0153 -12.2600 0.000 -0.0205 -0.0392EInt_Rate_Up 0.0200 0.0117 1.7100 0.087 0.0024 0.0040EInt_Rate_Down -0.1014 0.0150 -6.7600 0.000 -0.0116 -0.0207Married -0.0125 0.0153 -0.8200 0.414 -0.0015 -0.0025Divorced -0.0431 0.0193 -2.2400 0.025 -0.0051 -0.0086Widowed -0.1357 0.0264 -5.1400 0.000 -0.0152 -0.0279Grade_9_11 -0.2246 0.0219 -10.2500 0.000 -0.0240 -0.0473High_School -0.0009 0.0130 -0.0700 0.943 -0.0001 -0.0002Some_College 0.0314 0.0137 2.2900 0.022 0.0038 0.0062White 0.1595 0.0328 4.8600 0.000 0.0177 0.0330Black 0.1038 0.0361 2.8800 0.004 0.0130 0.0200Hispanic -0.0242 0.0390 -0.6200 0.535 -0.0029 -0.0048North_Central -0.0005 0.0146 -0.0400 0.970 -0.0001 -0.0001Northeast 0.0115 0.0157 0.7300 0.466 0.0014 0.0023South -0.0120 0.0140 -0.8600 0.390 -0.0014 -0.0024

    Male 0.1630 0.0128 12.7800 0.000 0.0209 0.0308Age -0.0080 0.0019 -4.2200 0.000 -0.0002 -0.0004AgeSq 0.0001 0.0000 3.5400 0.000Income 0.0039 0.0007 5.4500 0.000 0.0005 0.0008Jan 0.1823 0.0260 7.0200 0.000 0.0200 0.0380Feb 0.1313 0.0261 5.0300 0.000 0.0166 0.0251Mar 0.1257 0.0261 4.8200 0.000 0.0159 0.0240Apr 0.1138 0.0252 4.5100 0.000 0.0143 0.0218May 0.1338 0.0259 5.1700 0.000 0.0169 0.0255Jun 0.1234 0.0256 4.8200 0.000 0.0156 0.0236Jul 0.1761 0.0266 6.6100 0.000 0.0227 0.0332Aug 0.1285 0.0254 5.0700 0.000 0.0162 0.0245Sep 0.0680 0.0243 2.8000 0.005 0.0084 0.0132Oct 0.0571 0.0245 2.3400 0.019 0.0070 0.0111

    Nov 0.0320 0.0242 1.3200 0.185 0.0039 0.0063Recession -0.1969 0.0201 -9.7900 0.000 -0.0214 -0.0411C_SP500_L1 0.0076 0.0012 6.0700 0.000 0.0009 0.0015C_SP500_L2 0.0060 0.0013 4.6500 0.000 0.0007 0.0012C_SP500_L3 0.0041 0.0012 3.4200 0.001 0.0005 0.0008C_Indpro_L1 0.0285 0.0090 3.1600 0.002 0.0034 0.0057C_Indpro_L2 0.0202 0.0083 2.4400 0.015 0.0024 0.0040Unemp_L1 -0.1446 0.0090 -15.9800 0.000 -0.0173 -0.0287Irate_3m_L1 -0.0104 0.0070 -1.5000 0.135 -0.0012 -0.0021

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    28/41

    28

    Irate_5y_L1 0.0361 0.0074 4.8900 0.000 0.0043 0.0072Infl_L1 -0.0026 0.0018 -1.4500 0.146 -0.0003 -0.0005Infl_L2 -0.0012 0.0018 -0.6600 0.507 -0.0001 -0.0002C_ILI_L1 0.0099 0.0103 0.9600 0.337 0.0012 0.0020F_RGDP_L1 0.0104 0.0045 2.3200 0.020 0.0012 0.0021

    F_Infl_L1 -0.0523 0.0046 -11.4500 0.000 -0.0063 -0.0104Tresholds SNP Coefs Est. Error dist.1 -1.5568 1 0.1406 Variance 1.61062 -1.3572 2 0.1789 Skewness -0.0998

    3 0.0308 Kurtosis 2.6621

    Table 8 Result of Likelihood Ratio Test

    SentimentComponent

    Order ofSNP polynomial

    ()Likelihood Ratio

    est Statisticp-value

    PerFin_Current 3 1121.6211 0.00000

    PerFin_Expected 4 439.3399 0.00000

    BusCond_12m 4 296.4618 0.00000BusCond_5y 3 8.3058 0.00395

    BuyCond 3 14.8842 0.00011

    * Optimal K selected from K=3 to 6 using AIC and BIC.

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    29/41

    29

    Table 9 Explanatory power of the model

    Panel I Explanatory power of the full model and each category of variables - Probit model

    Model PerFin_Current PerFin_Expected BusCond_12m BusCond_5y BuyCond

    Full Sample

    All Variables 18.085% 29.723% 55.942% 41.826% 16.404%

    Expectations 11.010% 24.938% 53.323% 38.298% 12.263%

    Personal Characteristics 8.558% 8.825% 3.230% 6.002% 2.323%

    Macroeconomic Variables 3.299% 1.411% 14.136% 2.851% 7.979%

    Forecasts 0.290% 1.555% 12.043% 5.743% 5.347%

    Macro and Forecasts 3.340% 2.290% 16.730% 6.086% 8.677%

    Last 10 Years

    All Variables 20.764% 30.695% 58.357% 46.959% 15.980%Expectations 13.520% 25.937% 57.038% 44.685% 13.330%

    Personal Characteristics 9.611% 9.611% 4.298% 6.754% 1.699%

    Macroeconomic Variables 4.291% 2.973% 14.042% 3.226% 7.035%

    Forecasts 0.353% 1.922% 10.364% 3.827% 6.024%

    Macro and Forecasts 4.356% 3.044% 15.296% 4.374% 7.798%

    Panel II Explanatory power of the full model and each category of variables - Nonparametric model

    Model PerFin_Current PerFin_Expected BusCond_12m BusCond_5y BuyCond

    Order of SNP polynomial 3 4 4 3 3All Variables 14.696% 30.019% 54.496% 41.986% 18.182%

    Expectations 5.631% 24.038% 52.366% 38.431% 10.686%

    Personal Characteristics 8.218% 3.439% 0.598% 3.238% 3.496%

    Macroeconomic Variables 3.553% 0.839% 10.253% 1.017% 11.414%

    Forecasts 0.248% 1.505% 10.274% 1.847% 8.795%

    Macro and Forecasts 3.564% 2.246% 10.304% 2.293% 5.599%

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    30/41

    30

    Panel III Incremental explanatory power of each category of variables - Probit model

    Model PerFin_Current PerFin_Expected BusCond_12m BusCond_5y BuyCond

    Full Sample

    Expectations 6.374% 18.662% 36.453% 30.249% 5.726%Personal Characteristics 6.540% 4.355% 0.432% 1.941% 1.057%

    Macroeconomic Variables 0.566% 0.019% 0.812% 0.020% 1.436%

    Forecasts 0.006% 0.140% 0.168% 0.534% 0.133%

    Macro and Forecasts 0.589% 0.625% 2.103% 1.229% 2.864%

    Last 10 Years

    Expectations 7.522% 19.247% 39.837% 36.107% 6.963%

    Personal Characteristics 6.687% 4.435% 0.303% 1.914% 0.812%

    Macroeconomic Variables 0.529% 0.052% 0.491% 0.135% 0.580%

    Forecasts 0.001% 0.014% 0.147% 0.023% 0.211%

    Macro and Forecasts 0.536% 0.140% 0.977% 0.331% 1.737%

    Panel IV Incremental explanatory power of each category of variables - Nonparametric model

    Model PerFin_Current PerFin_Expected BusCond_12m BusCond_5y BuyCond

    Expectations 6.125% 25.435% 42.416% 30.350% 9.206%

    Personal Characteristics 8.916% 5.825% 0.415% 2.077% 1.162%

    Macroeconomic Variables 0.547% -0.012% 0.932% 0.026% 2.057%

    Forecasts -0.010% -0.065% -0.035% 0.540% -0.613%

    Macro and Forecasts 0.567% 0.405% 1.598% 0.941% 3.429%

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    31/41

    31

    Figure 1 Index of consumer sentiment in Recent Recession

    12000

    12500

    13000

    13500

    50

    60

    70

    80

    90

    100

    Jun-04 Jul-05 Aug-06 Sep-07 Oct-08 Nov-09

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    32/41

    32

    Figure 2 Index of consumer sentiment and its components

    Panel I. Index of Consumer Sentiment

    Panel II. Components of Consumer Sentiment: PerFin_Current

    Panel III. Components of Consumer Senment: PerFin_Expected

    4000

    5000

    6000

    7000

    8000

    9000

    10000

    11000

    12000

    13000

    14000

    0

    20

    40

    60

    80

    100

    120

    140

    160

    180

    200

    Jan-78

    Feb-79

    Mar-80

    Apr-81

    May-82

    Jun-83

    Jul-84

    Aug-85

    Sep-86

    Oct-87

    Nov-88

    Dec-89

    Jan-91

    Feb-92

    Mar-93

    Apr-94

    May-95

    Jun-96

    Jul-97

    Aug-98

    Sep-99

    Oct-00

    Nov-01

    Dec-02

    Jan-04

    Feb-05

    Mar-06

    Apr-07

    May-08

    Jun-09

    4000

    5000

    6000

    7000

    8000

    9000

    10000

    11000

    12000

    13000

    14000

    0

    20

    40

    60

    80

    100

    120

    140

    160

    180

    200

    Jan-78

    Jun-79

    Nov-80

    Apr-82

    Sep-83

    Feb-85

    Jul-86

    Dec-87

    May-89

    Oct-90

    Mar-92

    Aug-93

    Jan-95

    Jun-96

    Nov-97

    Apr-99

    Sep-00

    Feb-02

    Jul-03

    Dec-04

    May-06

    Oct-07

    Mar-09

    4000

    5000

    6000

    7000

    8000

    9000

    10000

    11000

    12000

    13000

    14000

    0

    20

    40

    60

    80

    100

    120

    140

    160

    180

    200

    Jan-78

    Jun-79

    Nov-80

    Apr-82

    Sep-83

    Feb-85

    Jul-86

    Dec-87

    May-89

    Oct-90

    Mar-92

    Aug-93

    Jan-95

    Jun-96

    Nov-97

    Apr-99

    Sep-00

    Feb-02

    Jul-03

    Dec-04

    May-06

    Oct-07

    Mar-09

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    33/41

    33

    Panel IV. Components of Consumer Sentiment: BusCond_12m

    Panel V. Components of Consumer Sentiment: BusCond_5y

    Panel VI. Components of Consumer Sentiment: BuyCond

    * Note: In all above figures, the dotted line is Stock and Watson monthly real GDP estimates.

    4000

    5000

    6000

    7000

    8000

    9000

    10000

    11000

    12000

    13000

    14000

    0

    20

    40

    60

    80

    100

    120

    140

    160

    180

    200

    Jan-78

    Jun-79

    Nov-80

    Apr-82

    Sep-83

    Feb-85

    Jul-86

    Dec-87

    May-89

    Oct-90

    Mar-92

    Aug-93

    Jan-95

    Jun-96

    Nov-97

    Apr-99

    Sep-00

    Feb-02

    Jul-03

    Dec-04

    May-06

    Oct-07

    Mar-09

    4000

    5000

    6000

    7000

    8000

    9000

    10000

    11000

    12000

    13000

    14000

    0

    20

    40

    60

    80

    100

    120

    140

    160

    180

    200

    Jan-78

    Jun-79

    Nov-80

    Apr-82

    Sep-83

    Feb-85

    Jul-86

    Dec-87

    May-89

    Oct-90

    Mar-92

    Aug-93

    Jan-95

    Jun-96

    Nov-97

    Apr-99

    Sep-00

    Feb-02

    Jul-03

    Dec-04

    May-06

    Oct-07

    Mar-09

    4000

    5000

    6000

    7000

    8000

    9000

    10000

    11000

    12000

    13000

    14000

    0

    20

    40

    60

    80

    100

    120

    140

    160

    180

    200

    Jan-78

    Jun-79

    Nov-80

    Apr-82

    Sep-83

    Feb-85

    Jul-86

    Dec-87

    May-89

    Oct-90

    Mar-92

    Aug-93

    Jan-95

    Jun-96

    Nov-97

    Apr-99

    Sep-00

    Feb-02

    Jul-03

    Dec-04

    May-06

    Oct-07

    Mar-09

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    34/41

    Fig

    * Resul

    re 3 Exp

    ts in this table a

    lanatory p

    re from ordered

    wer over

    probit models.

    34

    ime

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    35/41

    Fig re 4 Bal nce statist cs for selec

    35

    ed binary ndepende t variable

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    36/41

    36

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    37/41

    37

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    38/41

    38

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    39/41

    39

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    40/41

    40

  • 7/27/2019 Factors Determining Consumer Sentiment - Evidence From Household Survey Data

    41/41

    Fig re 5 Estimated Err r Densiti s

    Position:

    PerFin_Curre

    BusCond_12

    BuyCond

    t PerFin_Ex

    BusCond_

    pected

    y


Recommended