Lecture 2: Consumption (Continued) Wrapping Up from Last Time: Non Separabilities My belief: U(C,N) can be written as u(C) + v(N) However – we do not measure C directly: C = f(x,h) where h is directly related to N (through time budget constraint). We measure X and N in the data. X = f-1(C,h(N)) Implication: U(X,N) cannot be written as U(X) + V(N). Take Away Non Separabilities between X and N (expenditure and labor supply) are important. When is it important to implicitly model the home production sector? When changes to home production technology are important! When care about cross good predictions. When have actual consumption (intake) measures. For most applications, a reduced form assumption that X and N are nonseparable can be important. Wrapping Up from Last Time: Synthetic Cohorts From last time, we estimate: ln(Citk ) 0 age Ageit cCohortit t Dt fs Familyit itk What is the intuition of this regression? (I went through it fast last time). Underlying the estimation is repeated cross section of regressions. Hard to identify lifecycle or time series effects from cross sectional data. Use the repeated cross sections to create “synthetic cohorts” based on observables. Examples (Done in class) Another Data Set: Survey of Consumer Finances (SCF) Detailed data on household balance sheets. Cross sectional in design (small panel 1983 – 1989). Data: 1983, 1986, 1989, 1992, 1995, 1998, 2001, 2004, and 2007 Sample Size: ~5,000 households per wave Quality of Data: Assets (general balance sheet, housing, some pension) – very good Demographics and Income – very good Some data imputed (need to account for the imputations – each observation has 5 “records” given the imputations). Other Wealth Surveys: Health and Retirement Survey • Surveys “Older Households” (not too old – over the age of 50) • Panel data (same households are tracked) • Years: • Detailed wealth and pension data (along with very good income, health and demographic data). • Can apply to get the social security earnings records of participants! • Have full income data and detailed wealth data on the eve of retirement. Can explore retirement saving adequacy in detail. Every two years starting in 1992. Scholz, Seshadri, and Khitatrakun (JPE 2006) “Are Americans Saving ‘Optimally’ For Retirement?” • Great use of HRS data (I love this paper) • Writes down an individual optimization problem (with stochastic income, stochastic length of life, imperfect capital and insurance markets, realistic government programs, and a bequest motive). • Solves the optimal saving rule for each household given their actual income (from their social security records), health trajectories, and expected length of life (from life tables based on observables). • Assumes everyone has the same preferences and preference parameters. • Computes the optimal amount of wealth they should have (on the eve of retirement) and compares that to the households actual wealth. Scholz, Seshadri, and Khitatrakun: Key Findings Other Wealth Surveys: PSID • Panel Study of Income Dynamics (PSID) – Discussed in last class • Panel data (same households are tracked) • Years: • Detailed wealth data for broad asset classes “stocks”, “checking accounts”, “debt”, etc. Until recently, pension data is not that good. • Housing wealth (and mortgage debt) asked every year. • Very good income and demographic data. • See description in Hurst, Luoh, and Stafford (1996 – Brookings Papers on Economic Activity). 1984, 1989, 1994, 1999, 2001, 2003, 2005, 2007 and 2009 Topics for Today (May Extend Into Next Week) Part 1. Estimating preference parameters using consumption data Part 2. Discuss parts 1 and 3 of homework Part 3. Discuss how consumption data can be used to learn about the income process households are facing. Part 4. Discuss CEX data (part 2 of homework) and link to measures of changing consumption inequality. Part 5. Discuss risk sharing and consumption Part 6. Discuss my favorite of my papers (which empirically documents the importance of “status” in determining household consumption decisions). Part 1: Estimating Household Preferences Part 1: Estimating Preferences • Intertemporal elasticity of substitution (I.E.S.) • Risk Aversion • Time discount rates Note: Risk aversion = (1/I.E.S.) with CES preferences Note: Using notation from last week: (1/ρ) = I.E.S. δ = time discount rate Why is the I.E.S. important? • The intertemporal elasticity of substitution determines how levels of consumption respond over time to changes in the price of consumption over time (which is the real interest rate – or more broadly – the real return on assets). • This parameter is important for many macro applications. • Economics: Raising interest rates lowers consumption today (substitution effect) Raising interest rates raises consumption today (income effect – if net saver) Consumption tomorrow unambiguously rises Estimating I.E.S. T t 1 max Et j 0 1 t j 1 (Ct j ) 1 C Et t 1 (1 rt 1 ) 1 Ct ln Ct 1 t 1 1 ln(1 rt 1 ) t 1 Graphical Illustration – No Substitution Effect C High interest rate ΔC2 = X ΔC1 = X 1 Low interest rate 2 period With only an income effect – consumption growth rate will not respond to interest rate changes. Estimate of (1/ρ) = 0. Graphical Illustration – With Substitution Effect C High interest rate ΔC2 > X Low interest rate ΔC1 < X 1 2 period As the substitution effect gets stronger, the growth rate of consumption increases more as interest rates increase. Estimate of (1/ρ) > 0. Issues With Estimating I.E.S. ln Ct 1 t 1 1 ln(1 rt 1 ) t 1 • Use of data source (micro or aggregate) • Forecast of future interest rates? • Correlation of forecast of interest rate with error term (things that make interest rates go up could be news about permanent income – which effect consumption). Hall 1988 “Intertemporal Substitution in Consumption” (JPE) Uses aggregate data (National Accounts) Attempts to deal with time aggregation Uses various measures of interest rates (stock market return, t-bill, etc.) Instruments interest rate with lag interest rates and lags of consumption. Estimate: 1/ρ ≈ 0.00 Attanasio and Weber 1993 “Consumption Growth, the Interest Rate and Aggregation” (ReStud) Uses micro data (cohort data – British Family Expenditure Survey) - Aggregate the micro data appropriately to aggregate data Use aggregate data (from National Accounts) Uses building society deposit rate as measure of interest rate Instruments interest rate with lag interest rates. Estimate: 1/ρ ≈ 0.35 (National Accounts) ≈ 0.60 (FES Data - aggregating) ≈ 0.75 (FES Data – micro data) Vissing-Jorgensen (2002) “Limited Asset Market Participation and the Elasticity of Intertemporal Substitution” (JPE) Data: CEX Innovation: Split sample to those who are “saving” in financial markets Bond returns should only apply to bond holders Stock returns should only apply to stock holders Others are not on the margin because of fixed costs of participating. Estimate: 1/ρ ≈ 0.8 (Bond holders) ≈ 0.3 (Stock holders) Gourinchas and Parker (2002) “Consumption Over the Lifecycle” (Econometrica) You should read this paper. Estimates lifecycle consumption profiles in the presence of realistic labor income uncertainty (via calibration). Use CEX data on consumption (synthetic cohorts). Estimates the riskiness of income profiles (from the Panel Study of Income Dynamics) and feeds those into the model. Use the model and the observed pattern of lifecycle profiles of expenditure to estimate preference parameters (risk aversion and the discount rate). Gourinchas and Parker Structure N t N 1 max E u (Ct , ) VN 1 (WN 1 ) t 1 Wt 1 (1 r )(Wt Yt Ct ) C 1 u (C , Z ) v() 1 Yt PV t t Pt Gt Pt 1 N t Methodology Find in the income process (use different education and occupation groupings) Using PSID • Computed “G” from the data (mean growth rate of income over the lifecycle). • Estimated the variances from the data. Using CEX • Compute lifecycle profiles of consumption • Compute lifecycle profile of wealth/income (at beginning of life) Intuition No Uncertainty: No “Buffer Stock Behavior” Consumption growth determined by Rβ (where β = 1/(1+δ)) With Income Uncertainty Buffer stock behavior takes place (household reduce consumption and increase saving to insure against future income shocks). Consumption will track income if households are sufficiently “impatient” Sufficiently Impatient with Uncertainty: RβE[(GN)-ρ] < 1 Results Estimates (Base Specification): δ = 4.2% - 4.7% (higher than chosen r = 3.6%) ρ = 0.5 – 1.4 (1/ρ = 0.6 – 2.0) Interpretation Early in the lifecycle, households act as “buffer stock households”. As income growth is “high”, consumption tracks income (do not want to accumulate too much debt to smooth consumption because of income risk) In the later part of the lifecycle, consumption falls because households are sufficiently impatient such that δ > r. Barsky, Juster, Kimball, and Shapiro (1997) Preference Parameters and Behavior Heterogeneity: An Experimental Approach in the Health and Retirement Survey (QJE) “Suppose that you are the only income earner in your family, and you have a good job guaranteed to give you (and your current (family)) income every year for life. You are given the opportunity to take a new and equally good job, with a 50-50 chance it will double your (family) income and a 40-40 chance that it will cut your (family) income by a third. Would you take the new job?” If answer yes to base question, give a new question changing “third” to “half”. If answer no to base question, give a new question changing “third” to “20 percent”. Barsky, Juster, Kimball, and Shapiro (1997) Have four sets of answers: No – No ‘Low Risk Tolerance’ No – Yes ‘Medium Low Risk Tolerance’ Yes – No ‘Medium High Risk Tolerance’ Yes – Yes ‘High Risk Tolerance’ Use Survey Evidence to Measure Risk Parameters Risk Grouping Percent Low Tolerance “reject all gambles” 64.6% Medium Low Tolerance 11.6% Medium High Tolerance 10.9% High Tolerance “accept 50-50 gamble” 12.8% Using some structure (on distributions and preferences), estimate the coefficient of relative risk aversion (ρ) to be about 4.0 (standard error of 5 or so). Implication: (1/ρ) = 0.25 (lower than other estimates) Summary of Estimated I.E.S and Risk Aversion For those that ignore non-separabilities (or labor supply broadly), researchers usually use CES utility such that: T t 1 max Et j 0 1 ρ = 1.5 – 2.0 t j (Ct j )1 1 (1/ρ = 0.5 – 0.66) δ = r = 3.0 – 3.5% (sometimes δ > r ) We will talk about preferences with non-separable leisure in a few weeks. A Separate Question: The Importance of Precautionary/Buffer Stock Savings • How much of total wealth accumulation can be attributed to a precautionary motive? • Carroll and Samwick “How Important is Precautionary Saving?” (ReStat, 1998) • Hurst et al “The Importance of Business Owners in Assessing the Size of Precautionary Savings” (ReStat, forthcoming). • Use panel data from the PSID and estimate: ln(Wit ) 0 1 itpermy 2 ittransy 3 ln( yit ) Zit uit • Precautionary savings model predicts wealth will be higher the more risk that households face. The Importance of Precautionary Savings • Use income data to predict the transitory and permanent shocks to income by occupation and industry (specifically, we compute the variances for each individual and then instrument the two variances with income and occupation) • Identifying assumption: Occupation and Industry are independent of wealth aside from their effect on the variances of income. • Focus on households aged 26 – 50 (years 1984 and 1994) The Importance of Precautionary Savings Permanent variance Transitory variance Percent of sample Total sample 0.0162 (0.0023) 0.0513 (0.0040) 100 Professional and technical workers 0.0135 (0.0042) 0.0404 (0.0069) 23.74 Managers (non self-employed) 0.0171 (0.0048) 0.0305 (0.0083) 14.60 Managers (self-employed) 0.0272 (0.0163) 0.0866 (0.0270) 5.27 Clerical and sales workers 0.0192 (0.0075) 0.0541 (0.0128) 13.25 Craftsmen 0.0129 (0.0043) 0.0524 (0.0079) 20.10 Operatives and laborers 0.0199 (0.0055) 0.0592 (0.0094) 15.35 Farmers and farm laborers 0.0079 (0.0209) 0.1414 (0.05) 2.01 Service workers 0.0126 (0.0096) 0.0547 (0.0184) 5.69 Group Results Variables Pooled Pooled Variance of Permanent Income Shocks (α1) 15.91 (2.98) -1.57 (4.35) Variance of Transitory Income Shocks (α2) 7.52 (1.48) -0.27 (1.87) Percentage of Net Worth Explained by Precautionary Savings 47.5% 13.3% Dependent Variable (Log) Total Net Worth Total Net Worth Permanent Income Measure (Averaged) Non-capital Income Non-capital Income 2,144 1,729 Sample Size • • Carroll/Samwick results (our replication) in column I Our results (controlling for business owners) in column II • Our results ranged from 0.0 – 14% of total wealth. An Aside • Here are some good notes from Chris Carroll on the underpinnings of the “Buffer Stock Saving Model” http://econ.jhu.edu/people/ccarroll/public/lecturenotes/Consumption/Tract ableBufferStock/ They can be found on Chris Carroll’s Johns Hopkins web site. Things I am Interested In: Heterogeneity of Preferences • “Grasshoppers, Ants, and Pre-Retirement Wealth” (Erik Hurst ; permanent working paper) – my dissertation “It was wintertime, the ants’ store of grain had got wet and they were laying it out to dry. A hungry grasshopper asked them to give it something to eat. ‘Why did you not store food in the summer like us?’ the ants asked. ‘I hadn’t time’, it replied. ‘I was too busy making sweet music.’ The ants laughed at the grasshopper. ‘Very well’, they said. ‘Since you piped in the summer, now dance in the winter’.” • Permanent income hypothesis (broadly defined) describes well roughly 80% of the population. Roughly 20% appear “rule of thumb” or “time inconsistent”. Discussion of My Dissertation (including origins) • Discuss in class Things I am Interested In: Stability of Preferences • “The Correlation of Wealth Across Generations” (Kerwin Charles and Erik Hurst ; JPE 2002) • Do high saving parents have high saving kids? (Not the question I was originally interested in). • Real question of interest: “Can shocks to “preferences” today have long lasting effects on economic decisions?” “If we disenfranchise a group (Blacks) in the past – and then stop – how long will differences between two groups persist” Estimating Parent-Child Correlations Use data from the Panel Study of Income Dynamics and estimate: Wk = a + d1 Wp + a 1k Agek + a 2k Agek2 + a 1 p Age p + a 2 p Age2p + ek Wk 2Wp 1k Agek 2k Agek2 1 p Agep 2 p Age2p k Zk p Z p uk δ1 ≈ 0.40 δ2 ≈ 0.20 (where Z vectors include permanent income, direct transfers, education, etc.) δ2 can be interpreted as the correlation in saving rates (conditional on income, how similar are parent and child wealths) Can be do to “active” component or “passive” component. Wealth Persistence Parental Age-Adjusted Log Wealth Quintile (1984-1989) Child Age-Adjusted Log Wealth Quintile (1999) 1 2 3 4 5 1 36 26 16 15 11 2 29 24 21 13 16 3 16 24 25 20 14 4 12 15 24 26 24 5 7 12 15 26 36 Total 100 100 100 100 100 Persistence in Portfolio Persistence Parent Owns Stock I II III Child Owns Stock? A B C Child Owns Business? A B C Child Owns Home? A B C 0.133 0.057 0.058 (0.039) (0.041) (0.041) Parental Owns Business 0.110 0.081 0.065 (0.033) (0.034) (0.034) Parental Owns Home Parent and Child Age Controls a Parent and Child Income Controls b Parent and Child Risk Tolerance Controls c Adjusted R-Squared 0.245 0.145 0.147 (0.073) (0.072) (0.073) Yes No No Yes Yes No Yes Yes Yes Yes No No Yes Yes No Yes Yes Yes Yes No No Yes Yes No Yes Yes Yes 0.030 0.115 0.138 0.029 0.062 0.072 0.087 0.180 0.181 Persistence in Portfolio Persistence Regressors Very Low A B Child’s Risk Tolerance Measure Low Medium A B A B High A B Parental Risk Tolerance Low Risk Tolerance 0.059 (0.065) 0.064 (0.066) 0.008 (0.051) -0.021 (0.052) -0.054 (0.054) -0.042 (0.054) -0.012 (0.057) -0.001 (0.058) Medium Risk Tolerance -0.117 (0.079) -0.125 (0.083) 0.072 (0.062) 0.039 (0.065) 0.081 (0.065) 0.107 (0.068) -0.037 (0.069) -0.021 (0.072) High Risk Tolerance -0.138 (0.057) -0.098 (0.057) -0.005 (0.045) -0.013 (0.047) -0.010 (0.047) -0.012 (0.049) 0.154 (0.050) 0.123 (0.053) Part 1: Thoughts/Conclusions • Use consumption data to estimate preference parameters • Precautionary savings is an important feature in modern macro models • The importance of precautionary saving depends on household risk aversion, their impatience, and the risk they face. • Empirically, the importance of “precautionary savings” in explaining aggregate wealth holdings is mixed. Recent evidence suggests that it is small. • Preference heterogeneity seems to exist in the data. How important is it? • Are preferences stable? Part 2: Homework Part 1 and Part 3 Discussion of “The Age of Reason: Financial Decisions Over the Lifecycle” Erik Hurst University of Chicago, GSB Paper Synopsis • The main findings – – – • Emphasized interpretation – • Focus on a cross section of households Within the cross section, interest rates paid (fees, inverse of financial sophistication) is U-shaped Holds in a wide variety of settings Financial learning and declining cognitive ability Other interpretations offered (differing risk, opportunity cost of time, medical expenses, sample selection, cohort effects, etc.) My comments • I will focus on the “old” vs. “middle age” results (the upward sloping portion of the U-shaped profiles). I am going to ignore the young. • Comment 1: The importance of selection? Use data from existing nationally representative surveys to show that selection issues are very important (the 60-70 year olds that are borrowing are not random 60-70 year olds). • Comment 2: Are these magnitudes big? Maybe….Aggregating across all different debt types, difference in rates/fees paid by 55 year olds relative to 75 year olds is about $175 per year (~$3.50/week). Two (related) Questions • Why do people hold debt? (people do not hold debt randomly) – – – • Smooth out consumption over their lifecycle Borrowing will be peak when household income profiles are “low” or household consumption needs are “high” Who borrowers when they are 20? when they are 50? when they are 70? What interest rate will borrowers pay? (interest rates not charged randomly) – – – – Function of default probabilities Function of collateral amounts Function of borrower search (opportunity cost of time and value of lower interest rate) Function of financial sophistication Issue 1: Examining Selection • Use data from 2003 PSID • • Nationally representative Cross sectional comparisons (just like in this paper) • Focus on 25 – 75 year olds. • Look at: Lifecycle profile of credit card debt and mortgage debt Differences in the types of people (based on observables) that hold debt over the lifecycle Proportion of households with positive debt levels 80% 70% Any Mortgage 60% 50% Credit Card 40% 30% 20% 10% 0% 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 Average debt levels, conditional on having positive debt 160,000 32,000 140,000 28,000 120,000 24,000 Mortgage Levels (Left Axis) 100,000 20,000 80,000 16,000 60,000 12,000 Credit Card (Right axis) 40,000 8,000 20,000 4,000 0 0 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 Are the people that hold debt at older ages representative? • Not Really - For example, sizeable differences by race: Black Head Have Mortgage debt Age Non-Borrower Borrower Difference 50s 0.150 0.082 -0.068 60s 0.134 0.100 -0.034 0.035 * Have Credit Card debt Age Non Borrower Borrower Difference 50s 0.122 0.100 -0.021 60s 0.102 0.146 0.044 0.065* * Indicates significance at the 1% level Are the people that hold debt at older ages representative? • Why is the racial composition important? Blacks are found to pay higher interest rates than Whites in many markets adjusting for a full vector of demographics (including age). Charles, Hurst and Stephens (2008) – presented in a AEA session earlier today. Blacks pay higher rates for car loans than otherwise comparable whites (using SCF data). The effect is entirely due to the type of establishments frequented by blacks. Consistent with a plethora of recent lawsuits against vehicle finance service companies (GMAC, Ford Credit, etc.) Discrimination or financial sophistication? Are the people that hold debt at older ages representative? • Not Really - Health Differences: Report Health Deterioration in Prior Two Years Have Mortgage debt Age Non-Borrower Borrower Difference 50s 0.246 0.141 -0.105 60s 0.223 0.203 -0.020 0.085 * Have Credit Card debt Age Non Borrower Borrower Difference 50s 0.179 0.190 0.011 60s 0.186 0.259 0.073 0.062* Are the people that hold debt at older ages representative? • Additional differences by age (conditional on borrowing): - Self reported health much worse (mortgage and credit card) - Hospitalization more likely in the prior two years (credit card) - Gross wealth much lower (mortgage and credit card) • No difference by education (interesting) Summary: Is selection important? --- Yes • Probability of holding debt (and conditional levels of debt) diminish rapidly with age. • Who holds debt among older households? Much more likely to be Black. Much more likely to have received an adverse health shock (even if health spending is not put on the credit card, health shocks have occurred). Poorer individuals. Issue 2 – Examining Magnitudes • Cost of interest burden between 55 and 75 year olds: Annual Cost* Home equity loan interest gap: (~25 basis points) $100 Home equity line interest gap: (~30 basis points) $180 Credit card interest gap: (~5 basis points) $4 Auto interest rate gap: (~5 basis points) $2 Mortgage interest rate gap: (~12 basis points) $53 Total $350/year * All costs valued at the mean level of debt (as reported in the Appendix) Magnitudes? • Numbers on the previous page are likely way overstated! • As seen above, the amount of debt holdings seem to fall with age by roughly 50% (so my estimated costs should fall by 50%). • Suggestion: Why not compute exact dollar differences by different age ranges using data from SCF, PSID, HRS, AHEAD which tells the amount of debt of each type held at each age. • Prediction: For those holding debt, my guess is that the annual difference in expenditures is going to be less than $175/year (between 55 and 75 year olds). (About 50 minutes a month valued at pre-retirement wages) • Note: This number again would still be biased upwards if borrower composition is changing between 55 and 75. Conclusion • The policy prescription (particularly for the aged) depend on the reasons for the upward sloping interest rate profile. Are the old unable to process complex interest rate tasks (relative to their young selves)? I am not sure. • Selection seems to be important – much more work can be done on this (the data sets to address this are readily available). Moreover, interest rate data exists in some of these other data sets. • • Race and health composition changes over the lifecycle! The magnitudes are pretty small (not zero – just small). Would a cost benefit analysis recommend a policy intervention (again – particularly for the old)? A table of costs would be a great addition to the paper. Thoughts on “Depression Babies” Why Has The U.S. Saving Rate Declined? Part 3: Consumption and Income Consumption and Income Shocks 1 s t 1 2 max Et C C s t 1 2 Bt 1 ( Bt Yt Ct )(1 r ) where C is bliss point consumption, B is beginning of period wealth, and Y is labor income. Note: Assumption of "log utility" For simplicity: Assume r. Income Shocks and Consumption Growth • Given above preferences, consumption is a random walk such that: Ct 1 t 1 , Et [ t n ] 0 n 0 • Suppose, income process is as follows: Pt 1 Pt t 1 Yt 1 Pt 1 t 1 Et [t n ] Et [ t n ] 0 • Optimal Consumption Growth: r Ct 1 t 1 t 1 1 r 64 Deaton and Paxson (1994) “Intertemporal Choice and Inequality” (JPE) Hypotheses: PIH implies that for any cohort of people born at the same time, inequality in both consumption and income should grow with age. How much consumption inequality grows informs researchers about: o o Data: Lifecycle shocks to permanent income Insurance mechanisms available to households. U.S., Great Britain, and Taiwan 65 Deaton and Paxson Methodology (U.S. Application) • Variance of Residual Variation k k ln Citk 0 age Ageit cohort Cohortit tk Dt fsk Familyit itk • Compute variance of εkit at each age and cohort • Regress variance of εkit on age and cohort dummies (equation (2)) • Plot age coefficients (deviation from 25 year olds) Note: This is my application of the Deaton/Paxson Methodology (very similar in spirit to theirs). 66 Figure 1b: With and With Out Housing Services 0.24 Log Deviation From Age 25 0.20 0.16 0.12 0.08 0.04 0.00 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 Figure 1b: With and With Out Housing Services 0.24 Cross Sectional Variance of Total Nondurables for 25 Year Olds = 0.16 0.20 Log Deviation From Age 25 0.16 0.12 0.08 0.04 0.00 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 -0.04 Figure 1b: With and With Out Housing Services 0.24 Cross Sectional Variance of Total Nondurables for 25 Year Olds = 0.16 0.20 Log Deviation From Age 25 0.16 0.12 0.08 0.04 0.00 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 -0.04 More Aguiar/Hurst (2009) • Examine lifecycle profile of cross sectional inequality by category • Goods which have expenditures that increase with market work (due to home production or complementarity) should experience increasing dispersion when the dispersion of work increases. • Portion of lifecycle profile of cross sectional inequality due to these goods does NOT inform researchers about: o o Lifecycle profile of shocks to permanent income Insurance mechanisms available to households 70 Dispersion of Propensity to Work Over Life Cycle 0.60 0.50 0.40 0.30 0.20 0.10 0.00 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 Age Cross Sectional Lifecycle Dispersion: Entertainment Difference in Variance From Age 25 0.4 0.2 0.0 -0.2 -0.4 -0.6 -0.8 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 Non Increasing Dispersion Categories 0.5 Difference in Variance From Age 25 0.0 -0.5 -1.0 -1.5 -2.0 -2.5 -3.0 -3.5 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 Entertainment Utilities Housing Services Food At Home Other Non Durable Where is the Increase in Dispersion Coming From? 4.0 3.5 Difference in Variance From Age 25 3.0 2.5 2.0 1.5 1.0 0.5 0.0 -0.5 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 -1.0 Clothing Transportation Domestic Services Food Away From Home Lifecycle Variation in Standard Deviation Consumption Category Variance at Age 25 Change 25 - 44 Change 45 - 59 Change 59 - 68 Change 25 - 75 Increasing Transportation Clothing/P. Care Food Away Alcohol /Tobacco Domestic Services 0.70 0.63 1.54 5.80 6.82 -0.14 0.18 0.00 1.53 0.84 0.11 0.53 1.29 2.62 1.15 0.04 0.09 0.42 1.05 0.47 0.38 0.91 1.91 4.82 2.85 Non Increasing Housing Services Utilities Entertainment Other Non-Durable Food at Home 0.41 0.89 1.29 9.57 0.41 -0.07 -0.56 -0.31 -0.71 -0.05 -0.12 -0.09 -0.10 -0.91 0.02 -0.07 -0.05 -0.17 -0.27 0.01 -0.27 -0.76 -0.69 -2.39 -0.02 0.50 Cross Sectional Dispersion Over Lifecycle Percentage Point Deviation From Age 25 0.40 0.30 0.20 0.10 0.00 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 -0.10 -0.20 Core Nondurable 0.50 Cross Sectional Dispersion Over Lifecycle Percentage Point Deviation From Age 25 0.40 0.30 0.20 0.10 0.00 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 -0.10 -0.20 Core Nondurable Work Related Food At Home Cross Sectional Dispersion Over Lifecycle: Figure 6b 0.50 Percentage Point Deviation From Age 25 0.40 0.30 Total 0.20 0.10 0.00 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 Core -0.10 -0.20 Cross Sectional Dispersion Over Lifecycle: Figure 6b 0.50 Percentage Point Deviation From Age 25 0.40 0.30 Total 0.20 0.10 0.00 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 Core -0.10 -0.20 What Does it Mean? • Aguiar and Hurst (2009) Write down a model where households maximize utility with three consumption goods (and leisure) with the following constraints: one good (food) is amenable to home production one good (transport, clothes) are complements to market work there is a time budget constraint Assumptions: o o o conditional on work, income process is uncertain take the lifecycle process of work as exogenous assume that individual receives no utility for the lifecycle component of work related expenses. Implications When use disaggregated consumption data to match moments of model, get: • The estimated uninsurable/unanticipated permanent income volatility gets reduced by more than half (increases transitory volatility) Reason: The consumption volatility of “core nondurables” increases by roughly 50% less than “total nondurables” • The estimated importance of precautionary savings due to income fluctuations in explaining the wealth holdings of individuals is reduced. Reason: Permanent income volatility is lower • Agents are estimated to be significantly more patient Reason: Mean spending on core nondurables does not fall over the back half of the lifecycle. Conclusions • Beckerian model of consumption is important for explaining not only lifecycle profile of mean expenditures but also lifecycle profile of cross sectional dispersion in expenditures. - Explains decline in mean during back half of the lifecycle. - Explains increase in cross sectional dispersion post middle age. • The assumption that consumption (expenditure) and leisure are nonseparable is not a valid assumption. Part 4: Homework Part 2 and Time Series of Consumption Inequality Average Consumption in CEX 9.8500 9.8000 9.7500 9.7000 9.6500 9.6000 9.5500 84 Percent Change in Consumption in CEX (from 1981) 0 -0.02 -0.04 -0.06 -0.08 -0.1 -0.12 -0.14 -0.16 85 Income and Consumption Inequality • Large literature documenting the increase in income inequality within the U.S. during the last 30 years (Katz and Autor, 1999) • Consumption is a better measure of well being than income (utility is U(C) not U(Y)). • Does income inequality imply consumption inequality? Depends on whether income inequality is “permanent” Depends on insurance mechanisms available to households Depends on other margins of substitution (home production, female labor supply, etc.). • Topic taken up by Attanasio and Davis (1994, JPE), Krueger and Perri (2006, ReStud), and Attanasio, Battistin, and Ichimura (2004, orazio’s web page). 86 Kevin Murphy’s Web Page 87 Kevin Murphy’s Web Page 88 Consumption Inequality (Time Series) 0.5600 0.5500 0.5400 0.5300 0.5200 0.5100 0.5000 0.4900 89 Consumption Inequality: Adjusting For Family Size 0.06 0.05 0.04 0.03 0.02 0.01 0 -0.01 -0.02 No Controls Family Size Dummies Family Size Adjustment 90 Trends in CEX Consumption (Attanasio et al, 2004) “What really happened to consumption inequality in the US?” 91 Trends in CEX Consumption Inequality (Attanasio et al, 2004) “What really happened to consumption inequality in the US?” 92 Aguiar and Hurst (2009) Change in the Cross Sectional Variance of Log Expenditure Over Different Time Ranges Log Expenditure Measure 19811990 I. 19902003 19811990 II. 19902003 19811990 III. 19902003 19812003 19812003 19812003 Log Total Non Durable Expenditures 0.055 0.039 0.094 0.042 0.045 0.087 0.036 0.037 0.073 Log Core Non Durable Expenditures 0.063 0.027 0.090 0.037 0.031 0.068 0.034 0.023 0.058 Log Work Related Expenditures 0.104 0.041 0.145 0.076 0.038 0.114 0.052 0.020 0.072 Log Food at Home Expenditures -0.047 -0.020 -0.066 -0.005 0.003 -0.002 -0.004 0.000 -0.003 First Stage Controls None None None Full Full Full Full Full Full Second Stage Controls None None None None None None Age Age Age 93 Part 5: Consumption and Insurance Consumption Insurance • The Broad Question of Interest: Are households “insured” against “shocks” to their income process? • The Problem at Hand: How does one measure a “shock” from the household’s perspective? What we (the econometricians) label as a shock may be anticipated from the household’s perspective. Given that, households may react little to our identified “shocks”. • The Methodology: Use the joint distributions of income and consumption (and sometimes expectations) to analyze the extent of consumption insurance. 95 The Conceptual Issue: Uncertain Income Income Time t T 96 Uncertain Income Income Shock (as identified by econometrician) Time t T Suppose, from individual perspective, the household truly did receive an unexpected permanent shock to income. 97 No Insurance Consumption (dotted line) Income Time t T Household consumption responds completely to permanent shock to income. 98 Complete Insurance Income Consumption Time t T Household consumption will not respond to the permanent income shock. 99 The Conceptual Issue: Deterministic Income Income Time t T Suppose, from individual perspective, the income process is completely deterministic. 100 The Conceptual Issue: Deterministic Income Income Shock (as identified by econometrician only) Time t T Suppose, from individual perspective, the income process is completely deterministic. 101 The Conceptual Issue: Deterministic Income Income Consumption Time t T Forward looking consumers will incorporate the expected change in income into their current consumption decisions. 102 The Conceptual Issue: Deterministic Income Income Consumption No change in consumption growth Time t T Forward looking consumers will incorporate the expected change in income into their current consumption decisions. 103 The Conceptual Issue: Deterministic Income Income Consumption No change in consumption growth Time t T From the perspective of the econometrician, households appear to be completely insured against permanent “shocks” to income. 104 The Conceptual Issue: Deterministic Income Income Consumption No change in consumption growth Time t T From the perspective of the econometrician, households appear to be completely insured against permanent “shocks” to income. Results from not properly identifying unanticipated changes in income. 105 Blundell, Pistaferri, and Preston (AER, 2008) • Write down and estimate an econometric model to uncover the extent to which households are insured against both transitory and permanent income shocks. • They use data on actual income and consumption data. • Using data on only observed income and consumption does not allow the econometrician to separately identifying unanticipated changes in income from anticipated changes in income. (Akin to the simplified example above). • Blundell, Pistaferri and Preston made the implicit assumption that variance of anticipated permanent changes in income and the variance of the anticipated transitory changes in income were zero (i.e., there was no uncertainty over the anticipated changes in income). • As seen above, if that assumption fails to hold, the estimated extent of household insurance would be over stated (change in consumption understated). 106 Kaufmann and Pistaferri (AER P&P, 2009) • Use data on: Actual income realizations Actual consumption data Expected income changes • Use the moments of these three series to identify how consumption responds to the unexpected permanent and transitory innovations in income. • The key is using data on expected income changes to better isolate income “shocks” from the perspective of the household. 107 Some More Preliminaries • Data is from the Italian Survey of Household Income and Wealth • Survey questions on individual expectations of future income. • With a tad bit of structure, can compute the expected future income for all households who report answers to the survey questions. • Strong correlation between expected income and actual income (~0.5). 108 Key Results • As theory predicts, the amount of insurance is OVERSTATED with respect to permanent income shocks when econometricians ignore the fact that individuals have superior information about their own income process. - • Some of our identified “shocks” are expected by the household resulting in a muted consumption response. Key results from these paper: Response to Transitory Income Shocks Response to Permanent Income Shocks BPP 0.14 (0.05) KP 0.31 (0.43) 0.69 (0.27) 0.94 (0.51) 109 Benefits of Risk Sharing • An important implication of complete markets, full insurance model is that allows the construction of a “representative” consumer. • Good for aggregating individuals • Aggregate consumption moves as if it were determined by a representative consumer who only responds to aggregate risk (no need to worry about idiosyncratic risk). Formalize the test: ln(Cti ) k vt yti ti where full risk sharing implies that = 0 110 Important Earlier Empirical Papers Testing Full Risk Sharing • Townsend (1994) “Risk and Insurance in Village India” (Econometrica) • Cochrane (1991) “A Simple Test of Consumption Insurance” (JPE) • Attanasio and Davis (1996) “Relative Wage Movements and the Distribution of Consumption” (JPE) All papers reject perfect risk sharing. Some limited evidence of partial risk sharing (government transfers, self insurance for transitory shocks, family transfers). 111 Something You Should Read Job Market Paper from Greg Kaplan (out of NYU – now at Penn Economics Department) “Moving Back Home: Insurance Against Labor Market Risk” Had offers from Booth, Wharton, Penn Econ, Berkeley Econ, Sloan, Michigan, and 6 others. Dissertation looked at the role families play (particularly the ability to move back home) in insuring labor market risk for young low educated workers. http://homepages.nyu.edu/~gwk210/Greg_Kaplan/Home.html All of you could have written this dissertation. 112 Conclusions on Risk Sharing • There is some risk sharing (within families). • However, we are far from perfect risk sharing. • Permanent idiosyncratic shocks have permanent effects on household consumption. 113 Part 6 “Conspicuous Consumption and Race” Charles, Hurst, and Roussanov QJE 2009 Racial Differences in Economic Outcomes • Large literature documenting differences in wealth holdings, savings rates, and portfolio allocation between Blacks and Whites. (e.g., Barsky et al. (2002), Hurst et al. (1998), Charles and Hurst (2001), etc.) Question: Why do Blacks save less (hold less wealth) than otherwise similar Whites? • Likewise, there is some work documenting racial differences in individual consumption categories such as education and health insurance. Question: Why do Blacks spend less on health insurance and education than similar Whites? • Related Question: • Question: What are Blacks spending more on? Can racial differences in spending patterns on these goods explain (at least partially) racial differences in savings rates or racial differences in education or health spending? Conspicuous (Visible) Consumption • Veblen (1899) : Consumption communicates information about economic status. “Consumption is evidence of wealth, and thus becomes honorific, and…failure to consume a mark of demerit.” o The argument does not necessarily apply to “total consumption” – only the portion of consumption that is observable by others. • Theoretically, models of conspicuous consumption have been explored by many. • Empirically, the signaling value of consumption is relatively unexplored in economics. Some Preliminaries: An Overview of Main Data Set • Use data from Consumer Expenditure Survey (CEX) o o o o Use data from 1986 – 2002 (pooled). Include one observation per household (collapse multiple observations throughout the year into a single observation). Restrict the primary analysis sample to households with a head aged 18 to 49 (inclusive). Include households with a head being either Black, Hispanic, or White (we also look at Asians in some cuts of the data). Sample includes roughly 37,300 Whites; 6,800 Blacks; 5,300 Hispanics Will use other data (PSID) to confirm the CEX findings An Overview of the Data (continued) • Summary: We define visible goods to include expenditures on: o o o • Treat housing separately o o • Clothing and Jewelry Personal Care Spending on vehicles (excluding maintenance) Hard to separate the quantity from the price effect. Evidence of discriminatory practices. Note: Racial differences in visible spending get slightly LARGER if we include housing as a visible good. Some Descriptive Statistics (Tables 1 and A2) All White Black Hispanic Total Annual Income (Conditional Inc > 0 ) 57,800 63,800 38,400 39,800 Total Expenditure (Quarterly) 10,700 11,600 7,700 8,400 Visible Expenditures (Quarterly) 2,029 2,176 1,538 1,681 Vis Expend/Total Expend 0.12 0.12 0.12 0.12 All in 2005 dollars Part 1: Documenting the Facts Estimate: ln(Visible Exp) = βo + β1 Black + β2 Hispanic + φ Permanent Income + θ X + η Additional Controls (X): o o o o o o Year dummies ; Sex dummy ; Quadratic in age; Family structure dummies (number of adults, number of children, married) ; Location dummies (urban dummy, MSA dummy, census region dummies, city size dummies (post-1996)) ; Wealth controls (in some specifications) Measuring Permanent Income Approach 1: Use current income controls (current income, education dummies, and occupation dummies) to proxy for permanent income. CEX current income data is notoriously bad (27% of sample had missing income – no imputations). Racial gaps in income using CEX data do not match the racial gaps in income using CPS data (although the CEX expenditure gaps match the CPS income gaps). Approach 2: Use CEX total expenditure as a proxy for permanent income. Potential Issues with Approach 2 Potential problems with using total expenditure as a proxy for permanent income: 1. Total expenditure is not exogenous (expenditure components are jointly determined). 2. Measurement error in visible expenditure will cause a correlation between visible expenditures and total expenditures. Solution: Instrument total expenditure with our current income controls (either current income or current income, education and occupation dummies). Verify our results in the PSID where we can use panel aspect to create a better measure of permanent income. Preferred Specification Estimate: ln(Visible Exp) = βo + β1 Black + β2 Hispanic + φ ln(Total Exp) + θ X + η Notes: Instrument Total Expenditure with: a dummy for whether current income was zero, a cubic in current income (or the log of current income) if income was positive, education and occupation dummies. Included non-linear total expenditure controls as a robustness. Similar to standard “consumption demand system” model. Will estimate separately by race and plot the visible expenditure Engel curves. Table 2: Base Regression Results Regression Controls Included Black Coefficient Hispanic Coefficient 1. No Additional Controls -0.38 (0.04) -0.23 (0.04) 2. Specification 1 plus current income controls -0.03 (0.03) 0.14 (0.04) 3. Specification 1 plus ln(Total Expenditure) 0.31 (0.03) 0.26 (0.06) 4. IV Regression of Specification 3 0.23 (0.03) 0.20 (0.05) 5. Specification 4 plus time dummies 0.24 (0.03) 0.21 (0.05) 6. 0.26 (0.02) 0.23 (0.05) Specification 5 plus rest of X vector Magnitudes • Blacks Hispanics spend roughly 26% (23%) more on visible consumption than comparable whites. • Average household in sample spends roughly $2,100 per quarter on visible consumption. • Blacks (Hispanics) spend roughly $2,200 ($1,900) a year more on visible goods than comparable Whites. • The level is likely an under estimate (research shows that the CEX under reports total expenditures relative to NIPA). • Mean total pre-tax family income for Blacks (Hispanics) during the 1990s (March CPS): $42,500 ($48,300) 2 4 6 8 Estimated Engel Curves (Figure 1) 7 8 9 Log Quarterly Total Expenditure Black Estimated Difference at sample mean income ~ 0.3 White 10 Separately Analyzing Visible Components (Table 3) I. Full Sample Visible Consumption SubCategory Clothing/Jewelry Personal Care Cars (Limited) Cars (Expanded) II. Positive Car Spending Black Dummy Hispanic Dummy Black Dummy Hispanic Dummy 0.38 0.41 0.36 0.37 (0.03) (0.03) (0.04) (0.02) 0.73 0.43 0.81 0.42 (0.05) (0.03) (0.06) (0.05) -0.43 -0.29 0.12 0.09 (0.07) (0.10) (0.04) (0.06) -0.46 -0.34 0.09 0.04 (0.10) (0.17) (0.03) (0.05) Separately Analyzing Visible Components (Table 3) I. Full Sample Visible Consumption SubCategory Clothing/Jewelry Personal Care Cars (Limited) Cars (Expanded) II. Positive Car Spending Black Dummy Hispanic Dummy Black Dummy Hispanic Dummy 0.38 0.41 0.36 0.37 (0.03) (0.03) (0.04) (0.02) 0.73 0.43 0.81 0.42 (0.05) (0.03) (0.06) (0.05) -0.43 -0.29 0.12 0.09 (0.07) (0.10) (0.04) (0.06) -0.46 -0.34 0.09 0.04 (0.10) (0.17) (0.03) (0.05) Table 4: Racial Differences in All Spending Categories Log Expenditure Housing Utilities Food Other Transport. Home Furnishings Education Black Hispanic 0.03 0.13 (0.02) (0.03) 0.09 -0.02 (0.03) (0.02) -0.06 0.06 (0.02) (0.02) -0.15 -0.02 (0.03) (0.04) -0.18 0.09 (0.04) (0.05) -0.16 -0.30 (0.10) (0.12) Log Expenditure Black Hispanic Entertain Services -0.29 -0.36 (0.03) (0.05) -0.35 -0.17 (0.05) (0.05) -0.51 -0.48 (0.05) (0.06) -1.04 -1.04 (0.05) (0.05) -0.08 -0.38 (0.04) (0.08) Entertain Durables Health Alc./Tobacco Other Table 5: Robustness Exercise Using PSID Log Expenditure Black Clothing Expenditures, No Controls -0.07 (0.07) Clothing Expenditures, Full Controls 0.24 (0.07) Price of Recent Car Purchase, Full Controls 0.12 (0.09) Food Expenditures, Full Controls -0.12 (0.03) Entertainment Expenditures, Full Controls -0.33 (0.08) Other Transportation, Full Controls -0.09 (0.06) Summary of the Facts • Large evidence that relative to economically similar Whites, both Blacks and Hispanics consume considerably more “visible” goods. o o o o The magnitudes are large: roughly 26% more which translates to about $2,100 more per year in visible spending for blacks. The findings are very robust – within different sub-groups of the population, across different time periods, across different specifications. The percentage differences are much smaller for older Black households (off a much smaller base). Aside from housing, all other consumption categories are lower for Blacks and Hispanics (including health spending and education) Part 2 – A Model of Conspicuous Consumption • Preference differences could explain the differences in consumption patterns across races. • Question 1: Is there any model that does not rely on differences in preferences between races that can explain the documented consumption patterns? • Question 2: If so, can the predictions of this model be distinguish from a model of preference differences? Part 2 – A Signaling Model of Conspicuous Consumption • Glazer and Konrad (1996) study the signaling value of observable charitable giving. • Other models include Mailath (1987) and Ireland (1994). • Similar in implications to the classic Spence model (1973) of job market signaling. • Our goal is to draw on the implications of these theoretical models. Part 2 – Model Components • Preferences (household i drawn from group k) ( yik cik ) u (cik ) w( sik ) where: ci is consumption of all visible goods yi is the total household income endowment y-c is consumption of all non-visible goods (static model) • • Income is not observable (only c is observable to others) Income is drawn from known distribution fk(y) with support [ykmin, ykmax] • Define: Status (sik) is society’s inference about i’s income based upon things observed about the person. sik E yik | cik * , k , where cik * is equilibrium visible consumption Part 2 – Model Components Notes: • • • All preferences are constant across all groups. v(.), u(.), and w(.) are each concave and twice continuously differentiable. We do not take a stand on the benefits of “status” . Focus on separating equilibrium such that: sik (cik * ( yik )) yik • Similar spirit to Glazer and Konrad (1996). Signaling Predictions 1. cik* is strictly increasing in yi (relationship can be concave or convex depending on the relative concavity of w(.) with u(.) and v(.)). 2. In equilibrium, the poorest individual in group k has no incentive to signal (cik* will be the same regardless of whether or not w(.) = 0). How does cik* relate with moments of the income distribution, f(.)? 3. The relationship between group income dispersion and cik* is theoretically ambiguous (holding own income constant). Depends on curvature of ∂c*/∂y 4. If poorer persons are added to the group such that the support of the group’s income distribution becomes [ymin – θ, ymax] and average group income falls, then cik* increases at every level of income. Comments • Framework is quite general. Reference groups k represent, in theory, any type of groupings into which the population can be sorted. • Depending on the situation, observers will know more or less about the distribution from which other individual’s unobserved income is drawn. • Key insight: Information about one’s reference group influences observer’s inferences about one’s income and thus interacts with the optimal choice of signaling expenditures. A leftward shift in the distribution of reference group income: cik* ↑ (holding yi constant) An increase in dispersion of reference group income: cik* ? (holding yi constant) 0 Density .00001 .00002 .00003 Black vs. White Permanent Income Distribution (Fig 2a) 0 20000 40000 60000 80000 100000 120000 140000 160000 Total Expenditure (Annual) White Black Permanent Income Measured by Total Expenditure (CEX data) 0 Density .00001 .00002 Black vs. White Permanent Income Distribution (Fig 2b) 0 25000 50000 75000 100000 125000 150000 175000 200000 Average Family Income White Black Permanent Income Measured by Average Income (PSID data) Relevant Questions at Hand • Are moments of the reference group income distribution (mean and variance) systematically related to visible consumption? Can we see such a relationship within a race? For example, do Whites from poorer reference groups consume more visible goods than otherwise similar Whites from richer reference groups? Note: • Use mean as proxy for the leftward shifting of the income distribution. Does controlling for moments of the reference group income distribution explain the racial differences in visible consumption? As seen above, the black distribution of income is, on average, to the left of the white income distribution. How Do We Define Reference Group Income Distribution • Main approach (when assessing CEX data) Define reference group at the state/race level States is the lowest level of geographic location available in the CEX. • Robustness approach (when assessing PSID data) Define reference group at the MSA/race level Use PSID confidential geo-code data to get MSA info for each household. For the state/race moments of the income distribution, we use CPS data from 1990-2002 (total income of men aged 18-49). For the MSA/race moments of the income distribution, we use census data from 2000 (total income of men aged 18-49). We explored many different income measures as a robustness exercise. An Important Caveat • Throughout our analysis, we are taking the choice of reference group as being “exogenous”. • We believe that there are many interesting potential implications that may arise if we endogenize residential choice patterns (i.e., allow people to choose their reference group). • We are thinking about these implications in future work. Reference Group Income Distribution and Visible Spending • How do moments of the reference group income distribution interact with visible spending? ln(visibleisr ) 0 sr ( s gr ) ln(TotalExpenditurei ) X i i where Γs and Γr are vectors of state and race fixed effects, respectively. • Regression estimated via IV (as described above) where current income, education and occupation controls are used as instruments for total expenditure. • Figure 3 plots the estimated δsr against the mean state income for the particular race/state cell (from the CPS as described above). Key results: Systematic negative relationship between mean income of state and the propensity to consume visible goods (all else equal). 1 Figure 3 AL .5 KY 0 AR MANVNV NJ AR KS SC IL VA TX TX OH OK WA WI KS DC OR AL MD CO MD NY MA MIDC CA AK PA CO CA TN WIIA FL AZ AZINNJHI AK SC NY IL KY MN NC NC LA MN CT OH CT MI OKMOIN GA IA VA FL WA AL OR HI PA AR PA KY IN LA SC MO IA OH KSWI MI MA TN GA LA TN MONC TX IL ORAZ GA MN WA CT NV NY AK FL NJ CA OK CO VA -.5 HI 9.6 MD 10 10.4 10.8 Log of Mean Income of Race-State Cell White Black Hispanic DC 11.2 Examining Within Race Regressions ln(visibleis ) 0 1 ( ky ) 2 ( Dky ) ln(TotalExpenditurei ) X i i where: μ is the log of the mean income for persons race/state cell (from CPS) D is the dispersion of income in a race/state measured by the coefficient of variation (from CPS). Note: We also control directly for “housing” costs (which are location specific). Table 6: Within White Results Dependent Variable (1) (2) (3) (4) Log All Less Visible and Housing (5) -0.60 (0.14) -0.70 (0.14) -0.58 (0.13) 0.23 (0.06) -0.01 (0.05) -0.72 (0.30) -0.63 (0.28) 0.59 (0.10) -0.06 (0.03) -0.13 (0.06) 0.01 (0.03) -0.15 (0.02) Log Visible Expenditure Log of Mean Income of Own Race in State Coefficient of Variation of Income for Own Race in State Log of Individual Housing Expenditures * Log Food * We also instrument individual housing expenses with state housing prices (from 1990 and 2000 census) Table 7: Within Black and Hispanic Results Dependent Variable (1) Log of Mean Income of Own Race in State Coefficient of Variation of Income for Own Race in State Log of Individual Housing Expenditures * Log Mean Income of All in State -0.44 (0.13) Log Visible Expenditure (2) (3) (4) Log All Less Visible and Log Food Housing (5) (6) -0.51 (0.12) -0.45 (0.13) -0.64 (0.15) 0.12 (0.08) -0.02 (0.03) 0.25 (0.17) 0.26 (0.18) 0.26 (0.17) -0.14 (0.07) -0.02 (0.04) -0.09 (0.08) -0.16 (0.09) 0.16 (0.04) -0.14 (0.03) 0.60 (0.31) Explaining the differences across races How much of the race gap can be explained by differences in reference group income? Specifically, compare: ln(Visible Expenditure) = βo + β1 Black + β2 Hispanic + φ ln(Total Expenditure) + θ X + η with ln(Visible Expenditure) = βo + β1 Black + β2 Hispanic + β3 Mean Incomeik + β4 Coefficient of Variationik + γ ln(Total Expenditure) + δ X + η Table 8 (The Payoff) Variable 1 2 3 Black Coefficient 0.26 (0.02) 0.28 (0.02) -0.03 (0.07) -0.005 (0.07) -0.04 (0.07) Hispanic Coefficient 0.23 (0.03) 0.26 (0.03) -0.01 (0.08) -0.01 (0.06) -0.04 (0.07) Log of Mean Own Group State Income -0.53 (0.12) 4 -0.51 (0.11) Coefficient of Variation State Fixed Effects 5 -0.52 (0.11) 0.17 (0.12) No Yes No Yes Yes Table 8 (The Payoff) Variable 1 2 3 Black Coefficient 0.26 (0.02) 0.28 (0.02) -0.03 (0.07) -0.005 (0.07) -0.04 (0.07) Hispanic Coefficient 0.23 (0.03) 0.26 (0.03) -0.01 (0.08) -0.01 (0.06) -0.04 (0.07) Log of Mean Own Group State Income -0.53 (0.12) 4 -0.51 (0.11) Coefficient of Variation State Fixed Effects 5 -0.52 (0.11) 0.17 (0.12) No Yes Yes Yes Yes Table 8 (The Payoff) Variable 1 2 3 Black Coefficient 0.26 (0.02) 0.28 (0.02) -0.03 (0.07) -0.005 (0.07) -0.04 (0.07) Hispanic Coefficient 0.23 (0.03) 0.26 (0.03) -0.01 (0.08) -0.01 (0.06) -0.04 (0.07) Log of Mean Own Group State Income -0.53 (0.12) 4 -0.51 (0.11) Coefficient of Variation State Fixed Effects 5 -0.52 (0.11) 0.17 (0.12) No Yes Yes Yes Yes Table 8 (The Payoff) Variable 1 2 3 Black Coefficient 0.26 (0.02) 0.28 (0.02) -0.03 (0.07) -0.005 (0.07) -0.04 (0.07) Hispanic Coefficient 0.23 (0.03) 0.26 (0.03) -0.01 (0.08) -0.01 (0.06) -0.04 (0.07) Log of Mean Own Group State Income -0.53 (0.12) 4 -0.51 (0.11) Coefficient of Variation State Fixed Effects 5 -0.52 (0.11) 0.17 (0.12) No Yes Yes Yes Yes Summary • Document a set of facts that both Blacks and Hispanics spend a considerable more on visible consumption items than similar Whites. • This behavior is persistent within all sub groups and exists in the data since 1984. There is some evidence that this behavior dissipates with age. • A model of conspicuous consumption and signaling fits the data very well. • Controlling for the mean income of the group from which the individual is drawn explains the majority of the racial gap in visible consumption. • Moreover, the model is race blind. The model is supported when looking at behavior within races (either Whites or Blacks). Part 4. Potential Implications • How does the propensity to spend on visible goods effect the spending on other categories? o If we wish to promote Black spending on items such as education or health care, we need to understand the incentives to purchase status by investing in visible consumption. o May effect they way we design social programs. o Question: To what extent is visible spending differences correlated with spending differences in spending on other categories, like health care and education? o Question: Can conspicuous consumption be a potential explanation for observed saving/wealth differences across races? Unresolved Questions • How does conspicuous consumption affect saving in a dynamic model? Need to take a stance on why people value the “status” • Can any of the observed “saving” gaps between blacks and whites be explained by differences in spending on conspicuous spending? • How do people signal status in different settings? Do these finer models of signaling and status matter for anything “bigger”. • How are residential sorting patterns affected by conspicuous consumption motives? The reference groups – along some dimension – is endogenous!