Historical GDP Back-Projection
- Introduced by
- Angus Maddison; with indicator methods from Robert Allen, Paolo Malanima, and Jan Luiten van Zanden
- Year
- 2001
- Type
- descriptive-extrapolation
- Family
- macro-historical-measurement
What does it do?
Historical GDP back-projection estimates long-run income for periods too thinly documented for full national accounting.
Rather than rebuilding sectoral value-added year by year, it anchors to a handful of relatively secure benchmark estimates and fills the gaps between and before them using indirect indicators that move with income, chiefly the share of population living in towns, real wages of building labourers, agricultural productivity, and population density. The logic, associated above all with Angus Maddison and developed further by Allen, Malanima, and van Zanden, is that these indicators bear a stable, theoretically grounded relationship to per-capita output, so their movements can proxy GDP growth where direct measurement is impossible. The method has produced the multi-century per-capita income series that frame debates about pre-modern stagnation, Malthusian dynamics, and the European Little Divergence, while remaining explicitly more uncertain than bottom-up accounts.
The intuition behind the method
Think of GDP across distant centuries as a faded landscape where only a few hilltops, the benchmark years, are clearly lit. Back-projection is the art of inferring the contours between those peaks using shadows cast by things we can still see. If towns grew, the economy almost certainly produced a larger surplus to feed non-farmers, so urbanization shadows income. If a labourer's daily wage bought more bread, living standards likely rose. None of these is GDP, but each is tethered to it by economic logic. The projector calibrates the tether at the lit benchmarks, where both indicator and income are known, then walks backward, letting the indicators steer the path between hilltops. The result is a plausible trajectory rather than a measured one, and its honesty lies in admitting that the further from a benchmark you travel, the more the estimate leans on the assumed relationship rather than the evidence.
History
The approach is inseparable from Angus Maddison, whose decades of estimates produced the first coherent two-thousand-year picture of world income, much of it necessarily back-projected from thin evidence. As the field matured, scholars sought firmer foundations: Robert Allen used real-wage and demand-system reasoning, Paolo Malanima reconstructed Italian income from urbanization and agricultural output, and van Zanden and colleagues formalised the urbanization-income link. The Maddison Project, established after Maddison's death in 2010, now revises his series using these indicator methods and newly available benchmarks, generally lowering pre-modern growth and reshaping the divergence debate around a more gradual, indicator-disciplined narrative.
When to use it
Reach for back-projection when sources are too sparse for full national accounting but a few credible benchmarks and at least one broad indicator series exist. It is the standard approach for extending income series into the medieval and early modern periods, for countries with limited archives, and for constructing the long sweeps needed to test Malthusian or divergence hypotheses. Prefer full accounting where the data support it, because back-projection inherits all the uncertainty of its indicators and the assumed income relationship. Avoid it entirely when no defensible benchmark exists to calibrate the indicators, since the resulting series would then rest on assumption rather than any anchored measurement.
Limitations and considerations
- Rests on the assumed stability of the income-indicator relationship over centuries
- Uncertainty grows with distance from each benchmark anchor
- Indicators may decouple from income during structural change or crises
- Cannot reveal sectoral composition, only an aggregate per-capita trajectory
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Extends income estimates into periods far too sparse for full national accounting
- Economical in data, requiring only benchmarks plus broad indicator series
- Grounded in theoretically motivated links between income and observable proxies
- Readily subjected to sensitivity analysis by swapping indicators or elasticities
Applications
- Constructing two-millennium world and regional per-capita income series
- Testing Malthusian stagnation against gradual pre-modern growth
- Reconstructing medieval and early modern Italian and Dutch income
- Mapping the European Little Divergence between north and south
- Providing income trajectories for countries lacking detailed archives
How it works
How it works
The projection is anchored on benchmark years for which a relatively secure GDP-per-capita estimate exists, typically derived from full national accounting or detailed contemporary surveys. These benchmarks fix the level of the series and constrain its trajectory. The quality of the whole back-projection depends critically on the number, spacing, and reliability of these anchors; widely spaced or weak benchmarks leave long stretches governed entirely by the indicator model.
Formula
\hat{y}_{t_0}\ \text{(benchmark year GDP per capita)}Between benchmarks the analyst assembles indirect indicators known to correlate with per-capita income: the urbanization rate, real wages of unskilled and skilled labour, agricultural yields, population density, and sometimes proxies like literacy or coin output. Each indicator must have wide chronological coverage and a defensible economic link to income. The art lies in selecting indicators that are both well-documented and theoretically connected to the aggregate being reconstructed.
Formula
x_t\ \text{(urbanization, real wage, population density)}A quantitative relationship is established between income and the indicators, calibrated where benchmarks and indicators overlap. This may be a simple elasticity, a regression, or a structural relation such as the demand-system approach linking urbanization to agricultural surplus. The estimated relationship encodes how strongly a movement in the indicator translates into an implied change in income, and its stability across the period is a key, often heroic, assumption.
Formula
y_t = f(x_t) = \alpha + \beta x_t + \varepsilon_t
Using the calibrated relationship, the analyst projects income away from each benchmark by compounding implied growth rates inferred from indicator movements. Where the indicator rises, projected GDP per capita rises proportionally to the estimated elasticity. Interpolation between two benchmarks must reconcile the indicator-implied path with both anchor levels, often by distributing any residual discrepancy smoothly across the interval to avoid artificial jumps.
Formula
y_t = y_{t_0}\prod_{\tau=t_0+1}^{t}(1+g_\tau),\quad g_\tau \propto \Delta x_\tauFinally, segments built from different indicators, sources, or benchmark pairs are spliced into a single continuous series. Splicing requires aligning levels at overlaps and choosing whether to favour one indicator over another in contested intervals. Sensitivity analysis, swapping indicators or relationships, reveals how robust the resulting trajectory is and which historical claims survive across plausible alternative reconstructions of the same period.
Formula
Y_t = \text{splice}(\{y^{(j)}_t\})\ \text{across overlapping segments}Common pitfalls
- Assuming a constant elasticity when the indicator-income link plausibly shifts over time
- Relying on a single indicator without cross-validation against others
- Projecting far beyond the outermost benchmark without flagging the rising uncertainty
- Splicing segments at levels that create spurious discontinuities in growth
Frequently asked questions
Is back-projection just guessing?
No, it is disciplined inference anchored to benchmarks and tethered to indicators with theoretical links to income. The growth path is driven by observable proxies, not free choice, and is calibrated where indicators and income overlap. Its honesty lies in explicit assumptions and in sensitivity analysis that shows which conclusions survive alternative indicators and elasticities, distinguishing it sharply from unconstrained speculation.
Which indicators are most trusted?
Urbanization and real wages are the workhorses, because both have wide source coverage and well-understood economic links to per-capita income. Urbanization proxies the agricultural surplus needed to feed non-farmers, while real wages capture labour's command over staples. Agricultural yields and population density supplement them. Robust reconstructions triangulate several indicators rather than trusting any single one across a long horizon.
How is this related to the Maddison Project?
The Maddison Project is the institutional home of the method, curating and revising the income series Maddison began. Modern updates increasingly replace his impressionistic figures with indicator-disciplined back-projections calibrated to new benchmarks. The result is a continuously revised database in which back-projection is the engine that fills the long pre-statistical stretches between the rarer years of secure measurement.
Related methods
Sources
Maddison, A. (2007). Contours of the World Economy 1-2030 AD: Essays in Macro-Economic History. Oxford University Press.
ISBN: 9780199227204Copied.Allen, R. C. (2001). The Great Divergence in European Wages and Prices from the Middle Ages to the First World War. Explorations in Economic History, 38(4), 411-447.
Open DOI referenceCopied.