VoxEU Column Frontiers of economic research Global economy Global governance

Double-counting of investment

GDP counts investment twice – when it occurs and when rental income results. This column proposes an amendment to the national accounting system that only includes investment once. This would ensure that national income accounts do not overstate the resources available for consumption. It also has major implications for the estimation of the capital share in income.

Much of macroeconomic analysis and policy relies on calculated aggregates, such as GDP and national income. Proper ways to construct the national income and product accounts to measure these aggregates were once at the forefront of economic research – notable is the work in the 1930s and 1940s by Simon Kuznets. These advances appear today in revised form in US data from the Bureau of Economic Analysis and in UN data in the System of National Accounts. But economic analysis underlying the national income accounts has largely disappeared as a field of research at leading academic institutions. This situation is unfortunate because there remain major conceptual economic issues in the national accounts, one of which concerns the treatment of investment.

GDP counts investment twice – when it occurs and when rental income results

GDP is a measure of production, which equals consumption plus gross investment.1 In setting up a system of national accounts, Kuznets (1934, 1941) stressed that the only true final goods are consumption goods at various dates. Therefore, a reasonable test for a national accounting system for production and income is how well it measures the potential for consumption over time. As Kuznets (1941: 46) put it: “Is it the value of goods produced that leads to the most valid appraisal of the positive contents of economic activity? Since the final aim is to satisfy the wants of ultimate consumers, we might perhaps more properly center attention on ultimate consumption”.

If, as Kuznets argued, the only final goods are consumption goods in various periods, a reasonable requirement for a measure of national product or income is that it accurately reflect, subject to data constraints, the resources available for consumption. More specifically, a necessary condition from an intertemporal perspective is that, at least conceptually, the present value of production and income should equal the present value of consumption. However, the usual measures of national or domestic product and income fail this test because they double-count investment. Gross or net product includes gross or net investment when it occurs, and includes the corresponding present value a second time when additional rental income results from the enhanced stock of capital. Thus, from the standpoint of the intertemporal budget constraint for consumption, aggregates such as GDP and national income overstate the resources available for consumption.

I quantify the double-counting problem within a standard model used by economists, the steady state of the neoclassical growth model (Barro 2019). With reasonable parameters, GDP overstates the potential for consumption by 28%, while national income exaggerates this potential by 9%. Thus, for example, in international comparisons, countries that invest and save larger fractions of their incomes artificially appear to be too rich when gauged by per capita GDP.

The proposed ‘permanent income’ measure only counts investment once

I propose an amendment to the national accounting system to correct for this double-counting problem. The core idea is that gross investment should be fully expensed at the time it occurs. In the steady state, this adjustment means that net product and national income equal consumption (consistent with Kuznets’s argument). But, outside of the steady state, the adjusted measure – which I call permanent income – deviates from current consumption because the 100% expensing applies not to the current flow of gross investment but rather to the long-run flow. This measure corresponds to the concept of permanent income introduced by Milton Friedman (1957: Chapter 2). As an example, if a nation chooses to consume less and invest more out of a given current output, today’s permanent income does not change, even though today’s consumption falls.

The proposed permanent income variable is implementable with existing data. Starting from GDP, net domestic product (equal to national income for a closed economy) is computed in a standard way by subtracting estimated depreciation flows. In a simple model, these flows equal the product of the depreciation rate and the stock of capital. The calculation of permanent income requires only an extension from this usual depreciation rate to an effective rate that adds in the economy’s expected long-run rate of economic growth. In practice, this growth rate seems easier to gauge than the depreciation rate, which actually differs widely across different types of capital.

I show that current national accounting practice also exaggerates capital’s income share. Using typical parameter values, the capital income share based on GDP is around 40%. With the conventional adjustment to allow for depreciation, the computed capital income share for national income is reduced to 24%. With the additional adjustment to calculate permanent income, the share falls further, to 16%. Hence, the proper accounting treatment of investment makes a major difference in calculating the division of aggregate income between capital and labour.

In the existing national accounts, the over-counting problem is straightforward for businesses’ equipment and structures. However, it takes on different forms for other types of capital, which include intellectual property, household durables, housing, inventories, government capital, and human capital. As an example, the recent revisions of the US and UN national accounts to capitalise intellectual property (software, R&D, and ‘artistic originals’) resulted in substantial increases in reported levels of product and in capital income shares. These adjustments likely worsened accuracy from the standpoint of gauging consumption potential. Thus, although reasonable for some purposes, the capitalisation of investment flows for ideas or other forms of capital can deliver misleading results.

Another issue is that the treatment of investment as a final good affects the Bureau of Economic Analysis’s input-output tables. Goods treated as perishable (materials and intermediates) show up as flows from one industry to another. Goods viewed as durable investments appear instead, along with consumption, as final-good uses of products. The Bureau of Economic Analysis does generate supplementary information in its Capital Flow Tables, which were last produced for 1997 and which provide an input-output analysis for newly produced structures, equipment, and software. However, these tables apply to investment outlays, not to the flows of rental services on capital that would enter as inputs into production. Conceivably, the information that underlies the Capital Use Tables could be employed to construct an input-output analysis for rental services.

Fundamentally, the dynamic problems that I consider arise in the standard national accounts because the implicit conceptual framework is static. The setting is not well grounded in intertemporal budget constraints and, therefore, does not handle appropriately the economic role of investment and capital stocks. I try to show the source and extent of these problems and also suggest revised procedures that improve accuracy while remaining practical.


Barro, R J (2019), “Double-counting of investment”, NBER Working Paper 25826.

Friedman, M (1957), A theory of the consumption function, Princeton, NJ: Princeton University Press.

Kuznets, S (1934), National income, 1929-32, Washington DC: US Government Printing Office.

Kuznets, S (1941), National income and its composition, 1919-1938, New York: National Bureau of Economic Research.


[1] Consumption can include public as well as private consumption, and investment can include net foreign investment.

8,714 Reads