Personal Taxes

The Feasibility of a Quarterly Distribution of Private Earnings

BEA Working Paper Series, WP2021-8

The Feasibility of a Quarterly Distribution of Personal Income

Authors Dennis Fixler, Marina Gindelsky, and Robert Kornfeld0F1

U.S. Bureau of Economic Analysis

Contact Marina.gindelsky@bea.gov

Date December 2021

Abstract The U.S. Bureau of Economic Analysis (BEA) conducted a feasibility study to evaluate whether it is possible to produce a quarterly distribution of personal income and construct inequality metrics that are valid, informative, and transparent. The primary obstacles to producing such estimates are the lack of available quarterly microdata and inability to follow households over time (panel data). Therefore, we cannot account for household behavioral responses to shocks, such as applying for transfers after a wage loss, or participating in the gig economy, and we miss interdependency of these income sources. In this paper, BEA presents estimates of an interpolated quarterly distribution for 2007-2018. The estimates are driven by changes in aggregate income composition, such that the average of the quarterly estimates for each year is equal to the annual estimate. Many sources of data were considered to improve the quarterly estimates. An in-sample forecast exercise using a simplified methodology shows reasonable results during stable growth years but significantly underestimates inequality during periods of economic volatility (Great Recession and recovery). Forecast results can incorrectly show rising (falling) inequality during quarters when it is falling (rising).

Keywords Distribution data estimation, well-being, national income accounting

JEL codes C81, C82, D31, E01, I3

1 The authors would like to acknowledge the key contributions of Dave Wasshausen, Dylan Rassier, Tina Highfill, Ben Mandel, and Erich Strassner.

The views expressed in this paper are those of the authors and

do not necessarily represent the U.S. Bureau of Economic Analysis, or the U.S. Department of Commerce.

2

1. Introduction

This report is a feasibility study of quarterly estimates of the distribution of personal income (PI) and disposable personal income (DPI). Though most research on the distribution of income has focused on annual data (BEA, 2020; Auten and Splinter, 2019; Piketty et al. 2018; Congressional Budget Office, 2018), there is some interest in higher frequency quarterly estimates. Table 1 below summarizes some frequently cited inequality estimates, their publication frequency and the years available. Thus far, BEA annual estimates have been as current as those of other authors. The only quarterly estimates available are those extrapolated by the Federal Reserve.

Table 1. Available Inequality Estimates

Estimates

Measure

Frequency

Availability

BEA

Personal income and

Annual

2007-2018, soon 2000-2019

disposable personal income

Auten and Splinter (2019)

National income

Annual

1960-2015

(pre-tax and post-tax)

Piketty, Saez, and Zucman

National income

Annual

1913-2019

DINA tables

(pre-tax and post-tax)

CBO

Household income

Annual

1979-2018

pre-and-post-tax

Census Bureau

Household money income

Annual

1967-2019

World Inequality Database

National income and

Annual

1913-2019

wealth quantiles

Federal Reserve

Household wealth

1989Q3-2021Q1 (2019SCF is

Distributional Financial

Quarterly

latest available microdata, so

quantiles

Accounts

results extrapolated post 2018)

There are several general criteria for distributional national accounts. First, distributional statistics must be valid. Accordingly, we seek to construct metrics that are objectively measurable using the smallest possible set of assumptions to reduce measurement error. A high degree of measurement error is counterproductive in that it may lead to erroneous conclusions about the economy and a potentially inappropriate policy response. The validity of the estimates relies heavily on high quality microdata. Examples of such microdata are survey and administrative datasets produced and carefully compiled by the Census Bureau, the Internal Revenue Service (IRS), the Bureau of Labor Statistics (BLS), and other federal agencies. Although these datasets have shortcomings, such as representativeness at the top (or bottom), limited income variables, and inconsistencies, they are the best available microdata sources for annual income.

Next, distributional statistics should be informative. Given the wide suite of inequality measures currently available, produced by government statisticians, think tanks, and academics, measures produced by BEA should provide additional valuable information linking macro growth (aggregate

3

income consistent with the U.S. national income and product accounts (NIPAs)) with micro households. Personal income is the measure that is the major component of aggregate income (approximately 87 percent of gross domestic income).

In order for the statistics to be informative, they must have a high signal to noise ratio. That is, they should be attuned to business cycle fluctuations as they pertain to real changes in the income distribution, without introducing measurement error that obscures the trends. We should not expect to see highly volatile distributional metrics during times of stability and growth; conversely, we should expect to see significant shifts in the income distribution with the implementation of new government transfer programs, or with significant job losses, as these reflect business cycle changes in the macro- NIPA series.

Finally, distributional statistics should be transparent. It is important for data users and policymakers to understand how the estimates are constructed and what assumptions are made. We have strived to be as transparent as possible to enable users to better understand the estimates and to be able to compare our results and methods to those of previous studies.

This report proceeds as follows. Section 2 presents a conceptual framework for estimating quarterly distributions of income, beginning with defining BEA’s income measures with which we are striving to produce distributional measures. Section 3 presents our initial, prototype estimates of the quarterly distribution of PI and DPI. Section 4 presents a forecasting exercise for the Great Recession. Section 5 provides an overall summary and conclusions.

2. Conceptual Framework, Data, and General Issues

We begin this section by defining personal and disposable income in the NIPAs. With definitions in hand, we next present results from the annual estimates to provide a sense of how quarterly distributions may vary, followed by a discussion on interpreting quarterly results not based on a panel of households. We discuss how we might expect the distribution of PI and DPI to vary over quarters, based on evidence from BEA’s estimates of the annual distribution of income and other considerations.

2.1 Defining PI and DPI

In many ways, PI and DPI are not simple measures of cash income received in a given period. PI includes wages and salaries, supplements (employer contributions for employee pension and insurance funds), the income of farm and nonfarm sole proprietors and partnerships, rental income of persons (which consists mostly of imputed owners’ equivalent rent, interest income (which includes monetary and substantial imputed interest), dividend income paid to shareholders (but not capital gains from rising stock prices), many types of government social benefits (including health insurance such as Medicaid and Medicare and refundable tax credits), and other transfers, less contributions for government social insurance. DPI equals PI less personal current taxes.

Each of roughly 65 disaggregated components of PI and DPI is subject to seasonal adjustment to remove the average effect of variations that normally occur at about the same time and in about the same magnitude each year-for example, the effect of work schedules, weather, or holidays. If households regularly receive lower pay in certain months, then seasonal adjustment procedures will smooth income over the year such that estimates of the quarterly distribution of top-line PI will not capture these

4

routinely lower income levels in a specific quarter. The quarterly PI and DPI measures, will, on the other hand, capture unusual changes in income, such as onetime bonuses or income losses, or income changes that accompany periods of recession or recovery.

BEA’s estimates of PI and DPI reflect a mix of cash-basis treatment (when pay is received) and an accrual-basis treatment (when the income is legally earned). The international guidelines for national accounts, as explained in the 2008 System of National Accounts, recommend an accruals treatment. In practice, the timing of receipt of most components of PI is very similar under a cash- or accrual-basis treatment: most income from wages, social benefits, and other types of income is accrued and paid in the same quarter. For example, large stimulus measures such as the tax credits approved in the second quarter of 2008 as part of the Economic Stimulus Act of 2008, and the economic impact payments approved in the second quarter of 2020, appear mainly in the quarter in which they were enacted. The accrual and cash treatment of some components of PI and DPI, however, differ markedly. Employer contributions to defined benefit (DB) plans, for example, are estimated not as cash contributions but as the accrued change in the employers’ liability for entitlements, whether or not the plan is fully funded; DB plan benefits are not counted. Personal current tax payments, refunds, and tax settlements are recorded over the period in which they are earned (typically smoothed over a year), not when actually paid.

Some components of PI are not cash payments that are readily available for spending. PI includes an imputation for owners’ equivalent rent, as well as imputed interest and dividends, which homeowners do not actually receive as cash. It includes Medicaid and Medicare, government social benefits that are payments for services and not available for discretionary spending. Because of the inclusion of these noncash and imputed income components, PI is not a simple measure of any abrupt swings in quarterly discretionary income that households may experience tied to sudden economic and policy changes.

2.2 Annual Estimates: Review of Methodology and Results

A review of BEA’s estimates of the annual distribution of PI and DPI (published December 2020) helps inform our sense of how quarterly distributions may vary. In order to calculate the distribution of PI, we first identify a NIPA total to be distributed (the 65 disaggregated components described above). Because of measurement error and various timing and coverage issues, the sums of each type of household income in microdata data tend to be less than the corresponding BEA aggregates. However, we must allocate the NIPA total to each household. Therefore, we identify variables from the Current Population Survey, Annual Social and Economic Supplement (CPS ASEC) and from external data sources that can be used to allocate this total such that when summed (with household weights), the micro data total will equal the NIPA estimate. Because the ASEC data provide individual-level detailed disaggregated data on many income types, which sum to household, they are critical for the annual estimates, and thus frequently used by inequality researchers.

In order to construct a numerical example of this process, we can suppose there is only one source of income and four sample households, representing a population of 40 people, and our goal is to distribute a NIPA total of $10,000. However, for this example, the CPS total is assumed to be $100*20+$0*10+$500*5+$400*5 = $6,500. For each household, the share of the CPS total is calculated. Then, this share is multiplied by the NIPA total, as in the example below.

5

Example 1. Scaling CPS Values to Annual NIPA Values

Household

Original CPS

Population

Share of total = original

Imputed value

value

weight

CPS value/CPS total

= share * NIPA total

1

100

20

100/6,500 = 0.0154

0.0154*10,000 = 154

2

0

10

0/6,500 = 0

0*10,000 = 0

3

500

5

500/6,500 = 0.0769

0.0769*10,000 = 769

4

400

5

400/6,500 = 0.0615

0.0615*10,000 = 615

When imputed values are multiplied by population weight, they sum to the NIPA total, the same way as the original CPS values sum to the CPS total, when multiplied by the weight: $154*20+$0*10+$769*5+$615*5 =$10,000. For income items that use additional data sources, we first adjust the original CPS value, then proceed the same way.

After all components have been scaled to NIPA totals and added together to compute household and (subsequently) personal income, equivalized personal income is calculated by dividing personal income by the square root of the number of household members. This is done in order to compare households of different size. For example, if household income is $10,000 and there are four members of the household, equivalized household income is $5,000 (half of $10,000). Equivalized rankings of income are used for all income inequality metrics. A more detailed description of the annual methodology (Gindelsky 2020) can be found here.

From 2007 to 2018, the annual share of PI (and DPI) received by all deciles, except for the top 10 percent, varied by less than a percentage point, even over the course of the financial crisis (table 2). The shares of PI received by the bottom decile varied from 1.8 to 2.0 percent, while the share received by the next decile varied from 3.3 to 3.6 percent. At the top of the distribution the share ranged from 34.8 to 38.3 percent for the top decile and from 11.7 to 14.8 percent for the top 1 percent. The results for DPI are similar. In many instances, last year’s share of income for a percentile group is a reasonable predictor of the next year’s share of income. While the increase in inequality over the last several decades is widely acknowledged among economists, BEA’s results imply that this trend has occurred gradually.

The annual estimates also show that different deciles rely mainly on different types of income (table 3). For the lowest deciles, wages and government social benefits are the most important sources of income. For the highest deciles, wages, asset income (interest and dividends) and business income (partnerships and sole proprietorships) are the most important sources of income. As a result, the effect of changes in specific types of income will vary across low- and high-income households; changes in government benefits will have larger effects on the bottom and middle deciles, while changes in asset income will have larger effects on top deciles.

This is an excerpt of the original content. To continue reading it, access the original document here.

Disclaimer

BEA – Bureau of Economic Analysis published this content on 03 December 2021 and is solely responsible for the information contained therein. Distributed by Public, unedited and unaltered, on 03 December 2021 23:31:09 UTC.


Publicnow 2021

Related Articles