Estimating the Distribution of Consumption-based Taxes with the Consumption based Taxes with the Consumer Expenditure Survey

Ed HarrisEd HarrisKevin PereseCongressional Budget OfficeTax Analysis Division

DisclaimerDisclaimer

Analysis and conclusions presented here are my own and should not be interpretedare my own and should not be interpreted as those of the Congressional Budget OfficeOffice.

Uses of the CEUses of the CE

• Primary Use: Distribution of consumption-Primary Use: Distribution of consumptionbased taxes – Excise taxes– Cap and Trade– Value-Added Tax?

• Estimated distributional effect of these taxes depends critically on relationship ybetween consumption and income observed in the CE

Average Excise Tax Rate By Income Quintile

3.0

Percent of Income

2 0

2.5

Lowest

1.5

2.0

0.5

1.0Middle

Highest

0.0

197

9

98

0

981

98

2

98

3

98

4

98

5

98

6

987

98

8

98

9

99

0

991

99

2

99

3

99

4

99

5

99

6

1997

99

8

99

9

00

0

001

00

2

00

3

00

4

00

5

00

6

1 19 1 19 19 19 19 19 1 19 19 19 1 19 19 1 19 19 1 19 19 20 2 20

20

20

20

20

Average Gain or Loss in Households' Purchasing Power from the Greenhouse-Gas Cap-and-Trade Program in H.R. 2454: 2020 Policy Measured at 2010 Levels of IncomeIncome

Loss From Compliance Costs

Gain From Allowance Allocations and Other

Transfers

Net Gain or Loss in Household Purchasing

Power

Average Dollar Gain or Loss per Household

Lowest Quintile           ‐430 555 125Second Quintile           ‐560 410 ‐150Middle Quintile           ‐685 375 ‐310Fourth Quintile           ‐825 455 ‐375Highest Quintile          ‐1,400 1,235 ‐165Unallocated ‐120 130 10All Households 900 740 160All Households ‐900 740 ‐160

Gain or Loss as a Percentage of After‐Tax Income

Lowest Quintile           -2.5 3.2 0.7Second Quintile           -1.5 1.1 -0.4Middle Quintile           -1.3 0.7 -0.6Fourth Quintile           -1.1 0.6 -0.5Highest Quintile          -0.7 0.6 -0.1Unallocated 0 2 0 2 0 0Unallocated -0.2 0.2 0.0All Households -1.2 1.0 -0.2

Source: Congressional Budget Office, "The Economic Effects of Legislation to Reduce Greenhouse Gas Emissions", September 2009, Table2

Approach To Cap and Trade Estimates

• Estimate price effect of the cap and tradeEstimate price effect of the cap and trade program on final goods

• Impute expenditures by category from the• Impute expenditures by category from the CE to CBO’s base distributional database (which is based on income tax records(which is based on income tax records supplemented with data from the CPS)A l i ff t t di t ti t• Apply price effect to spending to estimate the effect across income groups

Input-Output Model: Price Change Results

Food 0.5%Clothing 0 2%Clothing 0.2%Nondurables 0.4%Electricity 8 8%Electricity 8.8%Natural Gas 11.4%Gasoline 4 2%Gasoline 4.2%All Expenditures 0.7%

Assumes Total allowance revenues of about 0.7% of GDP

CE and NIPA aggregatesCE and NIPA aggregates

• CE aggregates generally below NIPACE aggregates generally below NIPA aggregates

• Applying price increases from NIPA based• Applying price increases from NIPA based I/O model to spending in the CE does not yield the same revenueyield the same revenue

• Differential across expenditure categories• Adjusting for these has distributional

implications

Imputing Consumption: Preparing the CE

• We convert quarterly cross-sections from the q yInterview Survey to annual panel files– Reweight complete and incomplete interviews

Adj t t f di di• Adjustments for diary spending• Adjustments for renters with no reported

utility spendingutility spending• Pool multiple panels• Two Methods to impute from adjusted CE: p j

– Hot deck imputation for most of sample– Regression imputation for high income

householdshouseholds

Imputing Consumption:Statistical Match SOI/CPS & CE

• Hot deck routine with both rigid and flexible matching i icriteria

– Fixed: Region– Flexible: Age (+/- 1 year increments)

Income (+/- 2% increments)Family Type (+/- 1 child only)

• For each record in base data file, match to a CE ,record within the same cell

• Carry over ratio of consumption to income, expenditure shares of different itemse pe d tu e s a es o d e e t te s

• Applied to:• Single households <$150,000 income• Married households <$300,000 incomea ed ouse o ds $300,000 co e

Consumption to Income ratios, 2004

BLS Published Income and Consumption by Income Class, 2004BLS Published Income and Consumption by Income Class, 2004

Population Average Income

Average Consumption

Consumption/ Income

< $5,000 4.553 $2,626 $17,029 6.49< $10,000 7.218 $7,856 $14,596 1.86< $15,000 8.950 $12,554 $19,444 1.55< $20,000 8.177 $17,427 $23,023 1.32, , ,< $30,000 14.172 $24,892 $27,741 1.11< $40,000 13.125 $35,107 $33,273 0.95< $50,000 11.374 $45,052 $38,204 0.85< $70,000 18.069 $59,920 $47,750 0.80> $70,000 30.644 $118,332 $76,954 0.65

Total 116.282 $54,680 $43,395 0.79

NOTE:BLS consumption concept does NOT equal CBO consumption concept

Consumption and income data are constructed based on b th d di d t both survey and diary data.

Source: BLS, Consumer Expenditure Survey, 2004 Table 2. Income before taxes: Average annual expenditures and characteristics.

Consumption-to-Income Ratios by Pre-tax-Income Quintiles, CEX 1994 - 2004

250%

300%

Quintile 1

200%

250%

150%Quintile 2

100%

Quintile 3

Quintile 4

50%Quintile 5

0%1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Source: BLS, Consumer Expenditure Survey, Table 1. Quintiles of income before taxes: Average annual expenditures and characteristics, multiple years

Potential Adjustments That Reduce C-I ratios at the Bottom

Adjustments Made:D l i d• Drop very low-income records

• Use income averaged between 1st and last interview• Estimate income taxes based on reported income, use ratio of

consumption to after-tax incomeconsumption to after tax income• Adjustments to consumption definition

Explored But Not Done:p• Limit to prime-age individuals• Cap consumption-income ratios unless observed dis-saving can

explain

• Even with these adjustments, C-I ratios are quite high for bottom of the distribution

• Any adjustments to hit PCE totals exacerbate this problemAny adjustments to hit PCE totals exacerbate this problem

High Income RegressionsHigh Income Regressions

• Both income and expenditure amounts are ptop coded in CE

• Impute expenditure amounts based on regression models for high incomeregression models for high-income households

• Need to extend analysis significantly beyondNeed to extend analysis significantly beyond the income range covered in the CE

• Separate models for electricity, gasoline, fuel il t l d t t l ditoil, natural gas, and total expenditures

• Use regression results up to 1M in income, after that hold C-I ratio constantafter that hold C I ratio constant

Comparison of High Income UnitsComparison of High Income Units

CE SOICE SOI

Units above 100,000Number of Units (M) 18.9 16.1( )Average Income $164,000  $254,000 Share of Income 43.2 51.2

bUnits above 150,000Number of Units (M) 7.3 7.1Average Income $236,000  $425,000 Share of Income 23 8 37 7Share of Income 23.8 37.7

Source: BLS Table 2301. Higher income before taxes: Average annual expenditures and characteristics, Consumer Expenditure Survey, 2006 and IRS Statistics of Income, Individual Income Tax Returns 2006 Table 1.2

Top Quintile Income and Consumption SharesTop Quintile Income and Consumption Shares

0.6

0 5

0.55

0.45

0.5

CE IncomeCE Spending

0.35

0.4 CPS Income

CBO Income

0.3

High Income RegressionsHigh Income Regressions

ln(Consumption) by ln(Pre-tax-income), CEX 2004

$13

$14

$15

$10

$11

$12

umpt

ion)

$7

$8

$9

$10

ln(C

onsu

$5

$6

$7

10 10.5 11 11.5 12 12.5 13

ln(pre-tax-income)$60,000 $160,000

High Income Regressions Projections

Ln(Consumption) = Ln(Pre-tax-Income)

$700,000

$800,000

70%

80%

$400 000

$500,000

$600,000

onsu

mpt

ion

40%

50%

60%

tion-

to-In

com

e R

atio

Predicted Consumption(Left Axis)

$200,000

$300,000

$400,000

Pred

icte

d C

20%

30%

40%

Pred

icte

d C

onsu

mpt

Predicted Consumption-to-Income Ratio

$0

$100,000

$500,000 $1,000,000 $1,500,000 $2,000,000 $2,500,0000%

10%

Predicted Consumption to Income Ratio(Right Axis)

Pre-tax-income

High Income RegressionsEffect of Top-coding

ln(Expenditures) = ln(Pre‐tax Income)S C O i

50%

60%

50%

60%

BLS vs. CBO estimates

40%40%

ome Ratio

20%

30%

20%

30%

Expe

nditures‐to‐Inco

BLS

CBO

10%10%

CBO

0%0%

$500,000 $1,000,000 $1,500,000 $2,000,000 $2,500,000

Pre‐tax Income

High Income RegressionsEffect of Top-coding

ln(Gasoline Expenditures) = ln(Pre‐tax Income)BLS CBO i

1.80%

2.00%

1.80%

2.00%

BLS vs. CBO estimates

1.20%

1.40%

1.60%

1.20%

1.40%

1.60%

come Ratio

0.60%

0.80%

1.00%

0.60%

0.80%

1.00%

Expe

nditures‐to‐Inc

0.20%

0.40%

0.20%

0.40%

BLS

CBO

0.00%0.00%

$500,000 $1,000,000 $1,500,000 $2,000,000 $2,500,000

Pre‐tax Income

Final Consumption-Income RatioFinal Consumption Income Ratio

0.21.8

1 2

1.4

1.6

0.10.8

1

1.2Total Consumption (left scale)

0.2

0.4

0.6Energy Consumption (right scale)

00

Lowest Quintile

Second Quintile

Middle Quintile

Fourth Quintile

Highest Quintile

(right scale)

Quintile Quintile Quintile Quintile Quintile

EvaluationEvaluation

• Data from CE is very valuable. Especially access y p yto the micro data. Only source for: – Detailed consumption

V i ti f ti b hi i– Variation of consumption by age, geographic region,

• But• But…– Observed consumption-income pattern is difficult to

explain– Differential reporting error across income groups– Raises questions about the expenditure shares

derived from the CEderived from the CE

Suggested ImprovementsSuggested Improvements

Major• Top-down reconciliation of income and consumption as part of

the interview process– Perhaps something like to the diary, where focus is total

spending/savingspending/saving• High-Income oversampleMinor

– Pool all interviews for a CU create panel weights– Pool all interviews for a CU, create panel weights– Impute from diary to interview, so one complete file– Continue research into reconciling differences with PCE

• Provide cross-walk (or adjustment factors) between NIPA PCE and ( j )UCC codes

– Study income misreporting with a one-time match to administrative records