National income and product accounts, especially Gross Domestic Product (GDP), are fundamental economic indicators produced by almost every nation.
Simon Kuznets and Richard Stone, both Nobel laureates, pioneered national account creation in the United States and the United Kingdom.
This article aims to clarify what GDP measures, how it's constructed, and why its estimates sometimes differ from other economic indicators.
GDP Measurement Techniques
U.S. GDP and national account estimates heavily rely on detailed economic census data, available every five years.
The challenge involves using this census data alongside monthly, quarterly, and annual economic indicators to produce regular GDP estimates.
Economic indicators like retail sales, housing starts, and manufacturers' shipments of capital goods are often collected for different purposes and may not align perfectly with national accounts' definitions.
Initial monthly estimates of quarterly GDP lack data for approximately 25% of GDP, particularly in the service sector.
These sectors are estimated using past trends and related data like heating/cooling degree days for energy consumption and employment data for medical care and education.
Initial estimates are revised as more complete data becomes available from monthly tabulations, annual surveys with larger sample sizes, and more detailed information.
The Bureau of Economic Analysis (BEA) balances accuracy and timeliness to monitor economic growth, inflation, and key sectors like IT investment and housing.
These estimates inform policymakers, government forecasters, business planners, and investors.
Historical Development of GDP Estimation
Simon Kuznets and colleagues created the first official U.S. economic measure in the 1930s amidst the Great Depression.
Comprehensive economic data was lacking, hindering effective policy responses.
The growing role of government during the Depression highlighted the need for comprehensive national income accounts.
Kuznets' team produced industry-by-industry estimates summing to "national income," primarily using existing IRS tax data and supplemented by other sources like economic censuses and Bureau of Labor Statistics data.
Concerns persisted regarding data gaps, particularly in services, and reliance on business accounting practices and tax data, such as depreciation treatment.
The BEA continues to address these concerns by filling data gaps and aligning tax and financial accounting data with economic accounting concepts.
The Kuznets team then worked on producing more timely estimates by extrapolating from base-year values using employment and payroll data.
In the 1940s, the need for a national production measure became apparent, especially for wartime planning.
GDP estimates aided in assessing overall productive capacity and the impact of shifting from consumer spending to war expenditures.
Measuring production proved more challenging than measuring income due to the issue of double-counting intermediate sales between businesses.
The decision was made to estimate "final sales," excluding intermediate products and equaling incomes earned by production factors.
Indirect estimates were derived from sources like capital goods shipments, construction spending, and government budgets, with consumer spending as a residual.
This measure evolved into gross national product (GNP) as more data became available for consumption, investment, government spending, exports, and imports.
Input-output accounts in the 1950s provided a framework for estimating the economy's size through income, expenditure, and value-added measures.
In 1964, the BEA published its first input-output account directly linked to the national accounts.
The input-output table calculates GDP using three methods: production (value-added), income, and final expenditures.
By the 1960s, the United States had a suite of accounts assessing the economy using these three measurement approaches.
The three measures are conceptually identical but estimated using separate data combinations.
While some sectors, like services, remain difficult to measure, GDP as measured by the expenditure, production, and income methods are typically close, though divergences can occur during business cycle turning points.
Estimates evolve as initial extrapolations are updated with more complete information and reexamined for consistency with the accounts' concepts and definitions.
Evolution of GDP Estimates Over the Estimation Cycle
GDP estimation starts with a "benchmark" estimate, produced roughly every five years, based on economic census data.
Recent benchmark years include 2002, 1997, 1992, 1987, and 1982.
Benchmark estimates rely heavily on the economic census, a mandatory survey covering nearly all U.S. businesses with paid employees.
The BEA sums up consumption (C), investment (I), government spending (G), and net exports ((X–M)) to arrive at nominal GDP.
Nominal GDP estimates are then deflated using price indexes to estimate real, inflation-adjusted GDP.
Between benchmark estimates, GDP is estimated annually and quarterly using data from Census Bureau surveys.
The "advance" estimate, the first GDP estimate for a quarter, is released about one month after the quarter's end.
Around 45% of the advance estimate is based on survey-based monthly data for all three months of the quarter.
Estimates for inventories and exports/imports use two months of data with extrapolations for the third month.
Consumer spending on services relies heavily on extrapolations based on monthly trends.
The "preliminary" quarterly GDP estimate, published two months after the quarter, uses newly available or revised monthly/quarterly survey data for over 75% of the estimate.
The "final" quarterly estimate, released three months after the quarter, reduces trend-based data to 13%.
The Census Bureau’s new Quarterly Services Survey contributes to this decline.
During the summer of each year, the BEA revises estimates for the most recent calendar year and the two preceding years using annual data from various sources.
These data are more accurate due to larger sample sizes from mandatory annual surveys and comprehensive administrative data.
Benchmark revisions incorporate more accurate data and update concepts/definitions to reflect economic changes, such as recognizing investment in computer software in 1999.
Final Demand or Expenditures Approach
Consumer Spending
Benchmark years use the "commodity-flow" method based on business records to develop estimates of final sales to consumers by product category.
This method starts with total sales by producers, then adds transportation costs, trade margins, sales taxes, and imports, while deducting inventory changes, exports, sales to businesses, and sales to government.
For example, in 1997, domestic apparel manufactures' shipments were $66 billion, with $56 billion added for imports, $99 billion for transport/trade markups/taxes, and $12 billion subtracted for exports/inventory changes. Sales to government/businesses further subtracted 12 billion, resulting in $197 billion for consumer spending on apparel.
This estimate is reconciled with retail trade data from the economic census.
Annual estimates for post-benchmark years use "best-change" indicators from various sources; the Census Bureau's retail trade surveys are most important.
"Best change" involves multiplying the percent change from the retail trade survey category that corresponds to the detailed benchmark estimate by the best level from the benchmark or annual revision.
The "retail-control method" extrapolates about one-third of consumer spending monthly, quarterly, and annually.
A problem with the retail-control method occurs when the structure of the economy changes, such as increased sales to businesses and government through "big-box" stores, leading to an overstatement of consumer spending.
For goods/services with accurate price/quantity data or without consistent receipts data, the BEA uses the "price-times-quantity method."
An important example is new motor vehicle spending, using unit sales data and allocating them to consumers, businesses, and government using vehicle registration data, multiplied by prices by vehicle type.
The remaining price-times-quantity extrapolators are primarily for services like brokerage, cellular telephone, and cable television. These extrapolators are often replaced by data from the Quarterly Services Survey for final quarterly estimates.
The Quarterly Services Survey, initiated in 2004, samples approximately 6,000 service providers with paid employees and covers about one-quarter of the 55percent of economic activity accounted for by services (excluding retail/wholesale trade).
For some categories lacking monthly/quarterly expenditure data or price-times-quantity data, simple trends are used, such as population growth and the consumer price index for "personal care" services.
Some components of consumer spending involve implicit transactions that lack explicit market transactions, requiring imputation.
Examples include the rental value of owner-occupied housing and the value of financial services provided without explicit fees.
These imputed estimates account for about one-eighth of consumer spending, based on data from the BEA for 2005.
For owner-occupied housing, best-level rental values are based on estimates from the Census Bureau’s Residential Finance Survey, extrapolated by annual and quarterly estimates of home ownership costs derived from the consumer price index.
Financial services imputations estimate the shares of net interest receipts associated with checking, bookkeeping, loan processing, and investment services.
These unpriced services are estimated as differences between interest rates paid to depositors and received from borrowers compared to a reference rate, representing the cost of services banks provide without explicit charges.
The United States does not primarily use household surveys for estimating consumption expenditures due to their tendency to underestimate small, infrequent expenditures, spending by non-primary respondents, and "sin" goods/services.
Business surveys provide more reliable estimates of consumer spending than household surveys.
Investment
Benchmark estimates for equipment and software investment are derived using a commodity-flow method based on economic census data, excluding sales to other businesses.
Investment in structures is measured directly using Census Bureau Value of Construction Put in Place report data with adjustments for commissions and taxes.
Inventories reported in the economic census are adjusted to replacement cost, accounting for valuation methods, turnover rates, and prices at acquisition and withdrawal times.
Annual/quarterly estimates of equipment investment are extrapolated using the Census Bureau’s Manufacturers’ Shipments, Inventories, and Orders Survey.
A judgment-based value for the third month’s change in business inventories is required for the quarterly advance GDP estimate, which is often a significant source of revisions.
Annual/quarterly estimates of investment in most structures are extrapolated using the Census Bureau’s Value of Construction Put in Place report, combining survey, regulatory, and trade source data.
Brokers’ commissions, capitalized and included in residential housing investment, are extrapolated using Census Bureau data and trade-source data.
Exports and Imports
Mandatory U.S. Customs reports collected by the Census Bureau provide a monthly census of goods transactions, allowing best-level estimates for benchmark, annual, and quarterly data.
Adjustments are made to exclude noncommercial gold shipments and include petroleum in pipelines, but these adjustments are less extensive than for other GDP components.
Judgment-based estimates for the third month of goods exports/imports are required for the advance GDP estimate.
Revisions to inventories and foreign trade tend to be offsetting, with higher imports potentially leading to upward revisions to inventories.
Benchmark and annual estimates of international trade in services are based on data from the BEA's international transactions accounts, incorporating data from benchmark surveys of international trade in services and quarterly surveys.
Recent reports suggest that international transactions accounts' coverage of international services is better than domestic services coverage, with ongoing improvements being implemented.
Government
Benchmark estimates of federal government final expenditures are based on budget data.
For state/local governments, benchmark estimates are based on Census of Governments data.
Transfer payments are excluded from final expenditure measurements as they are not payments for production but rather income redistribution.
Only about one-third of total federal spending represents government expenditures for final goods/services.
Government capital expenditures are treated as investment, and depreciation is part of the cost-based measure of government spending.
The BEA adjusts for timing changes associated with government accounting rules.
The Office of Management and Budget publishes a chapter in the federal budget presenting the budget on a national income and product account basis.
Annual/quarterly estimates of federal spending are best-level estimates derived from the federal budget and Monthly Treasury Statements.
Annual estimates of state/local spending are based on Census Bureau annual surveys of state/local government finances.
Quarterly estimates of state/local spending involve trend extrapolations based on monthly employment/earnings data and construction data.
Adjusting for Inflation
The BEA deflates nominal components of the final expenditures measure of GDP at the most detailed level using price indexes from the Bureau of Labor Statistics (CPIs, PPIs, IPIs) to estimate real GDP.
Real output is estimated using quantity indexes to extrapolate real output from the base year for components using price-times-quantity methods.
Deflated detailed components are aggregated using a Fisher index, or chain-type index, incorporating current-period weights.
Chained current-period weights provide more accurate estimates of real GDP growth/inflation than fixed-expenditure weights.
Gross Domestic Income
Gross domestic income, the income-based measure of the economy, should conceptually equal GDP.
Data on incomes are available in tax/financial accounting records.
Challenges include adjusting tax/financial data to match economic concepts, time lags in data availability, and accounting for income misreporting.
Income from legally prohibited goods and services is excluded from the accounts.
Compensation
Benchmark, annual, and most quarters' estimates of wages/salaries are best-level estimates from the Quarterly Census of Employment and Wages, covering 98% of U.S. jobs.
Wage/salary estimates for the most recent quarter are based on the Bureau of Labor Statistics' Current Employment Statistics program.
Benchmark estimates of supplements to wages/salaries are comprehensive measures, including employer contributions for social insurance, pensions, and private insurance.
Estimates of government social insurance are based on Social Security Administration data.
Estimates of private health insurance contributions are based on the Medical Expenditure Panel Survey (Centers for Medicare and Medicaid Services).
Estimates of federal health insurance contributions are based on Office of Personnel Management data.
Estimates of pension contributions are based on Department of Labor data.
Estimates from the BEA are sometimes compared with estimates from the Bureau of Labor Statistics, causing confusion. From 1965–2005, real weekly wages for production/nonsupervisory workers in the Bureau of Labor Statistics series declined at an annual rate of 0.3 percent, whereas real compensation per worker for all workers from the BEA grew at an annual rate of 1.4 percent over the same period. The difference in the two series reflects the rapid growth in benefits and irregular pay, as well as the incomes of supervisory and nonproduction workers, all of which are not covered in the Bureau of Labor Statistics hourly wage series.
Corporate Profits
Benchmark estimates for profits come from annual federal tax data prepared by the IRS Statistics of Income program.
Tax-based data requires adjustments for depreciation/inventory expenses, income misreporting, and certain expenses allowed for tax purposes but not for national accounts.
Adjustments are made to ensure the time series captures economic income from current production with costs valued at full market value and is invariant to changes in tax laws and tax reporting incentives.
Annual profits data for recent years/quarters are extrapolated based on the Census Bureau’s Quarterly Financial Report and publicly available corporate financial data.
Financial data on profits are adjusted for nonrecurring gains/losses and stock-option/pension expenses.
Estimates of profits in the national income and product accounts can differ significantly from those of Standard and Poor’s or other financial reports due to the conceptual differences and adjustments made.
Other Capital-Type Incomes
Benchmark estimates for rental income of persons, net interest, and proprietors’ income also come from tax data and are subject to income misreporting adjustments.
The largest adjustment is to proprietors’ income, raising reported income by more than 50%.
Rental income consists of imputed net rental income of owner-occupied housing and income of individuals renting investment properties/second homes.
Owner-occupied rental income is calculated as described earlier for consumer spending imputation.
Net rental income is computed by subtracting mortgage interest, taxes, maintenance, and other expenses.
Net interest is extrapolated using the percent change in an index of interest paid constructed by multiplying moving averages of interest rates times the values of corresponding types of outstanding assets and liabilities from the Federal Reserve Board’s flow-of-funds accounts.
Proprietors’ income is extrapolated by a variety of indicators of activity, ranging from shipments and sales in industries populated by small firms to employment and earnings.
Adjusting Gross Domestic Income for Inflation
Unlike the final demand measure captured in GDP, there is no clear basis for adjusting income components for inflation.
Gross domestic income is adjusted using the overall GDP deflator.
Statistical Discrepancy
In concept, GDP as measured by the final expenditures approach should equal gross domestic income, but in practice, they differ, leading to a statistical discrepancy.
Over time, the two estimates are similar in level, growth rate, and cyclical pattern of growth.
Countries differ in their treatment of the statistical discrepancy.
The United States places a high value on the consistency between component estimates in the national accounts and corresponding economic indicators.
The BEA publishes the statistical discrepancy to help users access the accuracy of GDP.
Most adjustments are made to national income components due to long lags in data availability.
Initial gross domestic income estimates could be used to improve the accuracy of the GDP estimates at turning points in the business cycle.
It is useful to look at growth in both GDP and gross domestic income in assessing the current state of the economy.
Value-Added (or Production) Approach
The BEA has historically estimated industry value added in two ways: as a residual in input-output tables and directly using value-added incomes in GDP-by-industry estimates.
The BEA has integrated the input-output accounts and the GDP-by-industry estimates using the best available information from both.
The input-output approach imposes consistency by reconciling data on commodity outputs, inputs, and final demand.
In constructing the benchmark input-output table, the BEA sorts data into intermediate, final, and value-added categories and makes adjustments to reclassify sales of secondary activities or products to provide more homogenous groupings of economic activities within the input-output framework.
Gaps in the data are filled using a mixture of public- and private-sector information.
Value added can be misallocated due to the difficulty in allocating intermediate inputs across industries.
The resulting input-output table is balanced by adjusting data from the weakest sources, such as intermediate inputs or the residual value-added item, profits.
Annual Value-Added Estimate
The new methodology used by the BEA for calculating value added by industry combines directly measured value-added (income) data by industry with value-added estimates from the benchmark input-output accounts.
For years in which a benchmark input-output table is available, the method subjectively rates the accuracy of each type of value-added estimate for each industry using measures of variance.
The new method solves for value added as a weighted average of the directly measured value-added estimates and the benchmark input-output value-added estimates with the weights based on the measures of variance.
Annual estimates are extrapolated using annually balanced input-output and GDP-by-industry estimates, which are in turn reconciled to the expenditure-side estimates of GDP.
The major extrapolators are sales and receipts data from the Census Bureau and income data from the national income and product accounts.
Real Value Added, Gross Output, and Intermediate Inputs
Real value added is estimated by "double deflation": Detailed gross output by industry estimates are deflated by industry-specific gross output price indexes, and intermediate inputs are deflated by industry-specific intermediate input price indexes.
The price indexes used for this deflation are mainly producer price indexes from the Bureau of Labor Statistics, which has dramatically expanded its coverage of service industries in recent years.
Accuracy: Getting the General Picture Right
Because GDP estimates are based on administrative records and other nonsample data, confidence intervals and standard errors cannot be used to measure accuracy.
For the last five benchmark revisions of GDP, which correspond to the census years 1982, 1987, 1992, 1997, and 2002, the nominal level of GDP was revised an average of 1.1 percent, and the growth rate between benchmark years was revised an average of 0.26 percentage point.
The initial estimates of real GDP successfully indicated the direction of change in GDP an average of 98 percent of the time.
The mean revision between the advance estimate and the latest estimate was only 0.4 percentage point over the 1983–2006 period, which is not statistically significant or an indication of bias.
The U.S. national accounts meet or exceed internationally accepted levels of accuracy and comparability.
Challenges and Conclusion
Several significant challenges remain.
One problem that concerned Kuznets and his team—the lack of adequate data measuring the services sector—is still important.
A second set of challenges relates to the development of better estimation methods for components that are, by their nature, difficult to value.
The Bureau of Economic Analysis is also expanding its coverage of intangibles, such as research and development, which are not generally bought and sold in the marketplace.
Lack of integration and problems of inconsistency across these programs have hampered analysis of such issues as the downtrend in personal saving and the underlying causes of improved growth and productivity over the last decade.
Given the already heavy reliance on extrapolation in producing the advance GDP estimates, it is unlikely that the estimates can be made available much faster.
The detailed data needed from the economic census for the input–output tables are not available until four years after the reference year.
The Bureau of Economic Analysis will continue to investigate new sources of data for the national accounts and to improve the estimating methods applied to these data in order to release timely, accurate, and relevant data in a comprehensive and consistent framework.