Intractable Index Number Problems
No formula keeps V = P × Q meaningful everywhere: what each system gives up, and what FAND publishes instead.
Every value in an economic account is a price times a quantity. The identity cannot fail, because whichever of the three terms is not observed is computed from the other two. So the question is never whether V = P × Q holds. It is which term was observed, which was computed, and whether the computed one means anything to the people who act on it.
What no formula can do
A user wants four things from a system of price and quantity indexes:
- Additivity. The quantity of wheat plus the quantity of rice equals the quantity of food, at every level of the hierarchy.
- Transitivity. Comparing 1950 with 2025 directly gives the same answer as going through 1990; comparing Bolivia with Egypt directly gives the same answer as going through the United States.
- Characteristicity. Each comparison is made at prices that belong to the years or places compared, not at someone else’s.
- An observed price. The P in the identity is a price somebody paid.
No index has all four. Ragnar Frisch showed in 1930 that Irving Fisher’s tests for an index number cannot all be met at once, and the axiomatic literature since has turned that into impossibility theorems.1 Every statistical system therefore decides what to give up.
| System | Keeps | Gives up | Where its price comes from |
|---|---|---|---|
| National accounts over time (2008 SNA; the BEA since 1996) | Characteristicity, by chaining year to year; observed prices | Additivity: chained components do not sum to the total | Price surveys; the implicit deflator is value over chained quantity |
| ICP across countries, 1970–1985 (Geary–Khamis) | Additivity and transitivity | Characteristicity: one set of “international prices” for all | Survey prices averaged into world prices |
| ICP since 2005 (GEKS) | Transitivity and characteristicity | Additivity | Survey prices; the detailed price data are confidential |
| World Bank wealth accounts (CWoN 2024, chained Törnqvist) | Chaining | Additivity, and an observed price | Value over a physical count: nothing observed |
| FAND | Values at current prices; observed P and Q for production | Any published composite “real volume” | Market prices for production; one deflator for the benchmark’s dollar |
The national accounts chose chaining. The 2008 SNA recommends annually chained Fisher indexes and accepts chained Laspeyres volumes with their implicit Paasche prices; the BEA has published chained Fisher measures since 1996.2 Both publish implicit deflators. What they cannot publish is a set of chained components that adds up. The BEA prints a “residual” line and warns that where relative prices move fast, contributions computed from chained dollars can mislead even a few years from the reference year.3 The price is there, but as an index relative to a reference year, not as a level anyone could compare across goods.
The International Comparison Program (ICP) chose the other way, twice. Its rounds through 1985 used the additive Geary–Khamis method. After the debates of 1988–89 it moved to the non-additive EKS method, because additive “international prices” lean toward the prices of the richest economies and make the poorest look larger than they are: the Gerschenkron effect.4 The ICP now says that for uses such as poverty measurement it prefers non-additive methods that avoid the effect.5 Gerschenkron’s own point was that price levels are not commensurate between economies on different production possibility frontiers. World Bank staff argued in the 1980s that it applies to expenditure-side PPPs as well as to production. Every PPP carries it, in proportion to the weight a method gives to economies at each level of development, and re-benchmarking does not end it.
The wealth accounts are the extreme case
In the national accounts, V is observed and P comes from price surveys; only Q is computed. In the World Bank’s Changing Wealth of Nations (CWoN), V is itself a model: rents capitalized over an assumed asset life, lifetime earnings discounted to the present. Since the 2024 edition, Q is a chained Törnqvist index of physical quantities, such as the size of fish stocks or the number of workers, weighted by those modeled values.6 The price that completes the identity, V ÷ Q, is a value per worker or a value per tonne in the ground. Nobody pays it, and nobody can check it against a transaction.
OconEco’s comparisons show what such a price does:
- Gold against oil. At market prices the ratio ran from 9 to 43 barrels per ounce over 1995–2020. The ratio implied by CWoN’s gold and oil wealth series ran from 31 to 45. In the wealth accounts the 2008 oil peak and the 2020 collapse look much alike.7
- Human capital. The implicit deflator of Japan’s human capital changes direction 14 times in 25 observations. The ratio of the produced-capital deflator to the human-capital deflator ranges from 0.15 to about 7.5 across ten economies.8
These are not errors by the Bank’s team. They are what any system produces when P is the residual of two models. The Bank’s own advisers judged its earlier practice, the GDP deflator for all wealth, unsuitable, and proposed consumption prices for one purpose and volume indexes for another.9 The 2024 edition took the volume route, and in doing so moved its price further from anything a market quotes.
These comparisons were computed by OconEco from the Bank’s published files, outside FAND’s production pipeline.
Prices are what the theory says agents act on
Growth accounting does not ignore prices; it uses them up. Factor prices enter as the cost shares that weight inputs, and the dual measure of productivity is built from them.10 One relative price, investment goods against consumption, has a literature of its own: Hsieh and Klenow found that investment is dear relative to output in poor countries because consumption is cheap there, not because machines cost more.11 That reading rests on ICP price levels, and so carries the Gerschenkron effect with it.
What the frameworks do not carry is the price level of what agents trade, item by item and place by place. The national accounts publish price relatives. The ICP keeps its detailed price data confidential (OG-004). Quality adjustment and the treatment of goods that enter or leave the basket are, in FAND’s terms, weight of discussion: relative values assigned where no transaction set them, published inside a figure that reads like a ledger entry.
Profit-seekers act on relative prices. The public accounts give those who govern volumes. Businesses and households are both among the governed, but only those close to a transaction see its prices. A government that steers by index growth can look past the windfall gains that relative prices hand some for-profits, and the costs those gains shift to households. That is an information asymmetry between the governed and those who govern. The standards hold it in place: once established, a measurement standard resists revision, because the institutions that produce and use it are organized around it.12 A ledger of consent records the specifics of exchanges so that the governed can see them. A public ledger that publishes volumes and withholds prices records less than consent requires.
What FAND publishes instead
- Value at current prices is the primary statement, converted by the World Bank’s Atlas method rather than at PPP, with the United States, the benchmark, on its own BEA accounts (OG-004 and OG-005).
- Where P and Q are both observed, a year’s production of copper and its price, FAND carries both, and the value is their product (OG-005).
- For “real” figures, one deflator, the US GDP price index, applied to every type of capital and every economy. If quality change is absorbed into Q, as the orthodox decomposition claims, the law of one price must hold, and one price means one deflator (OG-012). The deflator removes the change in the benchmark’s own price level and nothing else. It makes no claim about consumption or production volume, so relative prices between types of capital and between places stay in the numbers as data. FAND holds the choice of one deflator over many as a testable proposition, not a settled one.
- PPPs serve as a screen for take-off, not as a conversion in the balance sheet.
What FAND gives up is a published composite “real volume” of wealth.
Directions for further work
- Publish the four-part decomposition already built for Fossil Fuels and Minerals: value change as volume at the year’s prices, the benchmark’s inflation, the commodity’s relative price, and entry and exit, four dollar figures that add exactly.
- Carry P and Q beside V on every leaf where both are observed, each with its pedigree, so a user can rebuild any composite with the formula of her choice.
- Rerun the CWoN implicit-price comparisons on each new edition, asset by asset on the 2024 Törnqvist volumes.
- Set out in one note why operational thresholds use exchange rates rather than PPPs, the case the World Bank’s 1989 Board paper on per capita income made and OG-004 summarizes.
Related reading
- Operational Guidelines: OG-004 (PPP conversion), OG-005 (current-price primacy), OG-012 (the quality–price entanglement).
- NUSAP-lite: how each number is graded.
- The ledger of consent: what a ledger records, and what it cannot.
- Take-off: where PPPs do serve.
- Four Types of Capital or Three: FAND against CWoN.
Footnotes
-
Ragnar Frisch, “Necessary and Sufficient Conditions Regarding the Form of an Index Number Which Shall Meet Certain of Fisher’s Tests,” Journal of the American Statistical Association 25, no. 172 (1930): 397–406; Wolfgang Eichhorn and Joachim Voeller, Theory of the Price Index: Fisher’s Test Approach and Generalizations, Lecture Notes in Economics and Mathematical Systems 140 (Berlin: Springer, 1976). Characteristicity is László Drechsler’s term: “Weighting of Index Numbers in Multilateral International Comparisons,” Review of Income and Wealth 19, no. 1 (1973): 17–34. ↩
-
European Commission, IMF, OECD, United Nations and World Bank, System of National Accounts 2008 (New York, 2009), chapter 15, as summarized in IMF, Quarterly National Accounts Manual (2017), chapter 8, Box 8.1; BEA, NIPA Handbook: Concepts and Methods of the U.S. National Income and Product Accounts, chapter 4. ↩
-
BEA, “Appendix A: Additional Information About the NIPA Estimates,” Survey of Current Business (July 2014), Statistical Conventions. ↩
-
Yuri Dikhanov, “Calculating and Publishing Additive ICP Results,” ICP Technical Advisory Group (World Bank, November 2021); Yuri Dikhanov, “Sensitivity of PPP-Based Income Estimates to Choice of Aggregation Procedures” (World Bank, 1997); Alexander Gerschenkron, A Dollar Index of Soviet Machinery Output, 1927–28 to 1937 (Santa Monica: RAND, 1951). ↩
-
World Bank, “ICP 2021: Methodology — PPP Calculation and Estimation” (2025). ↩
-
World Bank, The Changing Wealth of Nations 2024: Revisiting the Measurement of Comprehensive Wealth (Washington, DC, 2024). ↩
-
John O’Connor, “The Gold/Oil Price Ratio: Market Prices, FAND Valuation, and CWoN Implicit Deflators”, OconEco working paper (April 2026): OconEco computation from the CWoN 2024 reproducibility package (World Bank, FR_WLD_2024_195: gold and oil wealth in current and constant dollars) and the World Bank Commodity Markets Outlook annual prices (March 2026). ↩
-
OconEco computation from the CWoN 2024 data (World Bank, 2024): implicit deflators as current-dollar over constant-dollar wealth, ten economies, 1995–2020. ↩
-
Robert Inklaar, Wulong Gu and Erwin Diewert, “Deflation in the Changing Wealth of Nations Database” (paper prepared for the World Bank, November 2022). ↩
-
Dale W. Jorgenson and Zvi Griliches, “The Explanation of Productivity Change,” Review of Economic Studies 34, no. 3 (1967): 249–283. ↩
-
Chang-Tai Hsieh and Peter J. Klenow, “Relative Prices and Relative Prosperity,” American Economic Review 97, no. 3 (2007): 562–585. ↩
-
Theodore M. Porter, Trust in Numbers: The Pursuit of Objectivity in Science and Public Life (Princeton: Princeton University Press, 1995). ↩