Why Doesn't Capital Flow from Rich to Poor Countries?
📄 Summarized from the full manuscript · Human-reviewed for faithfulness before publication
In brief
Standard growth theory says capital should rush from rich countries to poor ones, where returns are higher. Lucas does the arithmetic and finds the predicted gap is enormous -- a return on capital in India roughly 58 times the American one -- while actual investment flows are nowhere near that. He tests four explanations. Adjusting for workers' skills shrinks the predicted gap to about five; adding spillover benefits from a country's own educated workforce can erase it entirely. Fear of expropriation and colonial-style monopoly power round out the candidates. Which one is right matters, because only one implies that transferring capital to poor countries would do any good.
What this paper finds — and why it matters
The simplest neoclassical models of trade and growth make an egalitarian prediction that follows from assumptions about technology alone: if two countries produce the same good with the same constant-returns production function in capital and homogeneous labour, then differences in output per worker must come from differences in capital per worker, diminishing returns put the marginal product of capital higher in the poorer country, and free competitive trade in capital goods sends all new investment to the poorer economy until returns and wages equalise. Lucas puts numbers on it. Taking production per person in the United States as about fifteen times India’s (Summers and Heston, 1988) and a Cobb-Douglas capital share of 0.4 – an average of US and Indian capital shares – the marginal product of capital in India must be about 58 times that in the United States. The point of working the arithmetic is that “there is nothing at all delicate about this standard neoclassical prediction on capital flows”: the observed flows fall so far short that the assumptions must be “drastically wrong,” and the question is which. Four candidates follow. Correcting labour for quality using Anne Krueger’s (1968) estimates – which imply each American or Canadian worker is the productive equivalent of about five Indians or Ghanaians – cuts the US-India income ratio per effective worker from 15 to 3 and the predicted return ratio from 58 to about 5, “a substantial revision” that nonetheless “leaves the original paradox very much alive.” Adding external benefits of human capital, with the spillover exponent estimated at 0.36 from Denison’s US 1909-1959 data, brings the predicted India-US return ratio to 1.04 – eliminating the differential entirely, though Lucas flags as “important and troublesome” that this calculation assumes knowledge spillovers across national borders are exactly zero. The third candidate, political risk, faces a historical objection: contracts in colonial India were enforced as reliably as domestic ones, so why were returns not equalised in the two centuries before 1945? The fourth is a colonial monopoly model in which an imperial power with exclusive control of the colony’s trade and monopsony power over its wages finds it optimal to retard capital flows, implying a colonial return about 2.5 times the European one at a capital share of 0.4 – although Lucas’s own footnote reports contrary evidence from Davis and Huttenback (1989). The conclusion is where the stakes sit: under either human-capital hypothesis, or under the monopoly hypothesis, official capital transfers to poor countries are fully offset by reductions in private investment; only insofar as political risk binds can transfers speed the equalisation of factor prices.
Summary of a classic paper, AI-assisted and human-reviewed. See the linked original for the authoritative claims and full conditions.
Questions & answers
Q1. What is the prediction the paper sets out to test, and how little does it depend on?
That under free competitive trade in capital goods, all new investment should go to the poorer country until capital-labour ratios, wages and returns are equalised – and this follows “from entirely standard assumptions on technology alone.” Lucas states the logic in three steps (p. 92): two countries produce the same good with the same constant-returns production function relating output to homogeneous capital and labour; if production per worker differs, “it must be because they have different levels of capital per worker: I have just ruled everything else out!”; and then “the Law of Diminishing Returns implies that the marginal product of capital is higher in the less productive (i.e., in the poorer) economy.” The conclusion drawn is not that capital flows are somewhat too small but that under the theory “one would expect no investment to occur in the wealthy countries in the face of return differentials of this magnitude.”
Q2. What is the arithmetic, exactly?
With US production per person about fifteen times India’s and a Cobb-Douglas capital share of 0.4, the implied Indian marginal product of capital is about 58 times the American one. The sources and steps are stated plainly (p. 92): production per person in the United States “is about fifteen times what it is in India,” from Summers and Heston (1988, Table 3, pp. 18-21); output per worker follows y = Ax raised to the capital share, with a common intercept in both countries; and in terms of production per worker the marginal product of capital scales as output per worker raised to a power of one minus the capital share over the capital share. Setting the capital share to 0.4, “an average of U.S. and Indian capital shares,” again for both countries, yields a factor of fifteen raised to 1.5 – “about 58 times the marginal product of capital in the United States.” Lucas is explicit about why he bothers: “I worked out the arithmetic for this example to make it clear that there is nothing at all delicate about this standard neoclassical prediction on capital flows.”
Q3. How much of the gap does adjusting labour for human capital remove?
It cuts the predicted return ratio from about 58 to about 5, by cutting the income ratio per effective worker from 15 to 3. The first candidate explanation is that the introductory calculation treats effective labour input per person as equal across countries, “ignoring differences in labor quality or human capital per worker” (§I, p. 92). Lucas uses Anne Krueger’s (1968) study, which he calls “the best attempt to correct measured labor inputs for differences in human capital” while noting its estimates rest on 1950s data – defended on the grounds that “the percentage income differentials between very rich and very poor countries have not changed all that much in the last 25 years and, in any case, a rough estimate is better than none at all.” Krueger’s method combines each country’s mix of workers by education, age and sector with US estimates of how those factors affect productivity as measured by relative earnings, and her Table III (p. 653) reports, for 28 countries, the per capita income each could attain with US physical capital per worker: from about .38 for India, Indonesia and Ghana to unity for Canada and .84 for Israel (§I, p. 93). Because these figures have the dimension of relative human capital stocks raised to labour’s share, taking labour’s share at .6 implies relative human-capital endowments from about .2 to unity – “each American or Canadian worker was estimated to be the productive equivalent of about five Indians or Ghanians.” Lucas adds a sanity check in wage terms: compensation per employed civilian in the United States in 1987 was about $24,000, “so this estimate implies that a typical worker from India or Ghana could earn about $4800 in the United States.” Reinterpreting income as income per effective worker makes the US-India ratio 3 rather than 15 and the predicted return ratio 3 raised to 1.5, “= 5 rather than 58.”
Q4. Why does Lucas insist this revision does not resolve the puzzle?
Because a factor of five is still far larger than observed flows can bear, and because a complete resolution along these lines would create a worse problem with labour migration. On the first point: “This is a substantial revision, but even so, it leaves the original paradox very much alive: a factor of 5 difference in rates of return is still large enough to lead one to expect capital flows much larger than anything we observe” (§I, p. 93). The second point is the more interesting one and is worth stating in full, because it is the constraint that disciplines the whole exercise: under constant returns, equal capital returns imply equal wages for equally skilled labour, so if human-capital adjustment alone eliminated the return differential there would be no economic motive for labour flows either – “Yet we see immigration at maximal allowable rates and beyond from poor countries to wealthy ones. We do not want to resolve the puzzle of capital flows with a theory that predicts, contrary to the evidence provided by millions of Mexicans, that Mexican workers can earn equal wages in the United States and in Mexico.”
Q5. What is the external-benefits-of-human-capital hypothesis, and how is its key parameter estimated?
That an economy’s technology level is the average human capital of its workers raised to a power, with that exponent estimated at 0.36 from Denison’s US 1909-1959 data. Lucas notes one could always resolve the puzzle by assuming returns are equalised and backing out the implied technology-level differences, and says that is “almost what I will do in this section, but I will do so in a way that has more content,” by making the technology term an external effect of average human capital, multiplying the productivity of a worker at any skill level exactly as a technology intercept would (§II, p. 93). The estimation uses Denison’s (1962) comparison of US productivity in 1909 and 1958: over 1909-1959 US output per man-hour grew about one percentage point a year faster than capital per man-hour, and Denison estimates human-capital growth of .009 attributed entirely to schooling; combined with a capital share of .25 these imply a spillover exponent of .36 – “That is to say, a 10 percent increase in the average quality of those with whom I work increases my productivity by 3.6 percent” (§II, p. 94). Lucas attaches the caveat himself: “This estimate is based on the assumption that the total stock of human capital grows at the same rate, .009, as that part of the stock that is accumulated through formal schooling. I do not have any idea how accurate an assumption this is.”
Q6. What does that parameter do to the predicted return gap?
It closes it: the predicted India-US return ratio becomes 1.04. Taking the Krueger estimate that five Indians equal one American, “the predicted rate of return ratio between India and the United States becomes… 1.04. That is, taking the external effects of human capital into account in the way I have done entirely eliminates the predicted return differential” (§II, p. 94). Lucas is careful to note this is not built in: “the value of [the spillover exponent] estimated from the 1909-58 U.S. comparison exactly eliminates the return differential in a 1959 India-U.S. comparison,” and he says he is “surprised how well it works in a cross-country comparison.”
Q7. What is the assumption that Lucas himself flags as troublesome?
That knowledge spillovers across national boundaries are exactly zero. “But it is important and troublesome, I think, to note that the cross-country comparison is based on the assumption that the external benefits of a country’s stock of human capital accrue entirely to producers within that country. Knowledge spillovers across national boundaries are assumed to be zero” (§II, p. 94). He observes that ordinary experience suggests some external benefits of individual knowledge are local – confined to single cities or even neighbourhoods – while “others are worldwide in scope,” and then states the binding limit on the exercise: “without some real evidence on the scope of these external effects, I do not see how to advance this quantitative discussion any further.” The summary he draws from Sections I and II together is deliberately a range rather than a point: correcting for human-capital differentials “reduces the predicted return ratios between very rich and very poor countries from about 58 at least to about 5, and possibly, if knowledge spillovers are local enough, to unity.”
Q8. How does the paper set up the capital-market-imperfection explanations?
By noting that the discussion so far has been static, and that a return differential at a point in time implies flows of goods over time – which in a one-good setting are borrowing contracts that someone must be able to enforce. Lucas describes the pattern: goods flow from advanced country A to backward country B for a phase, followed by a phase “(which lasts forever) in which goods flow from B to A in the form of interest payments or repatriated profits” (§III, p. 94). For that to be a competitive equilibrium, “it is evident that there must be an effective mechanism for enforcing international borrowing agreements. Otherwise, country B will gain by terminating its relationship with A at the point where the repayment period begins, and, foreseeing this, country A will never lend in the first place.” This is the imperfection usually labelled political risk.
Q9. What is the objection to political risk as the answer?
The historical record before 1945, when contract enforcement in much of the Third World was as effective as at home, and capital-to-effective-labour ratios were still not equalised. “A serious difficulty with political risk as an explanation for the inadequacy of capital flows lies in the novelty of the current political arrangements between rich and poor nations” (§III, pp. 94-95). Until around 1945 much of the Third World had been under European-imposed legal and economic arrangements for decades or centuries, and “a European lending to a borrower in India or the Dutch East Indies could expect his contract to be enforced with exactly the same effectiveness and by exactly the same means as a contract with a domestic borrower.” Hence the question Lucas poses and does not answer: “Even if political risk has been a force limiting capital flows since 1945, why were not ratios of capital to effective labor equalized by capital flows in the two centuries before 1945?” His response is candid – “I do not know the answer to this question” – followed by the suggestion that the colonial powers should not be assumed to have simply enforced laissez-faire.
Q10. What is the colonial monopoly model, and what does it predict?
An imperial power with exclusive control over the colony’s trade but facing a free colonial labour market chooses capital per worker to maximise repatriated profit, and the resulting first-order condition sets the colony’s marginal product of capital above the world return – about 2.5 times it, at a capital share of 0.4. The setup, “very much in the spirit of Adam Smith’s (1776/1976) analysis of an earlier phase of colonialism,” assumes at one extreme that the colony has no capital of its own and no ability to accumulate any, so the imperialist chooses capital per worker and repatriates the entire income (§III, p. 95). The objective is total production less competitively determined wage payments less the opportunity cost of capital at the world return; the first-order condition equates the colony’s marginal product of capital to the world return plus the derivative of the colonial real wage with respect to capital per worker. Lucas identifies the active channel precisely: “It is the imperialist’s monopsony power over wages in the colony that is crucial. His optimal policy is to retard capital flows so as to maintain real wages at artificially low levels.” With the Cobb-Douglas technology and a capital share of .4, “the return on capital in the colony should be about 2.5 times the European return. These are quantitatively interesting rents.” He takes institutional support from “the carving up of the Third World by the European powers, and the frequent granting of exclusive trading rights to monopoly companies.”
Q11. How honest is the paper about evidence against the monopoly model?
It reports the contrary evidence in its own footnote and raises an internal objection in the text. The footnote is direct: “According to Lance Davis and Robert Huttenback (1989), investment in the late British empire was open to firms from any country on competitive terms, which would obviously be inconsistent with this model. Moreover, they do not find rates of return in the British colonies that exceeded European returns for similar investments” (§III, fn. 1, p. 95). The text objection is about scale: in a country like India or Indonesia, where most of the workforce was and is in traditional agriculture, “it is hard to imagine that the ability to control capital inflows from abroad gave the imperialists much monopsony power over the general level of wages,” because European capital must have been a small fraction of total capital, most of which was land. Lucas concludes that for the mechanism to matter “it must have been because only a small part of the colonial labor force was skilled enough to work with imported capital in, say, goods manufacturing” – which in turn “would obviously need a more refined view of the nature of human capital than one in which five day-laborers equal one engineer.”
Q12. Does the monopoly story end with colonialism?
Lucas argues it need not have, and points to post-colonial restrictions on inflows as possible private taxation of capital. “Insofar as monopoly control over trade in capital goods was an important factor in the determination of capital-labor ratios prior to 1945, I do not see any reason to believe it ceased to be a factor after the political end of the colonial age. Monopoly returns are not of interest to Europeans only” (§III, pp. 95-96). He cites “much unsystematic evidence of heavy private taxation of capital inflows in Indonesia, the Phillipines, in the Iran of the Shah, and other poor economies that are otherwise attractive to foreign investors,” and treats the standard rationales for restricting inflows – mistrust of foreigners, reluctance to let development proceed “too fast” – as warranting “a Smithian skepticism.” The evidential status of this passage is explicitly informal (“much unsystematic evidence”), and the summary should not upgrade it.
Q13. Why does the paper say it matters which explanation is right?
Because three of the four hypotheses imply that transferring capital to poor countries accomplishes nothing, and only one implies it helps. “The central idea of virtually all postwar development policies is to stimulate transfers of capital goods from rich to poor countries” (§IV, p. 96). Under either of the human-capital-based hypotheses, “such transfers will be fully offset by reductions in private foreign investment in the poor country, by increases in that country’s investments abroad, or both” – because returns are already equalised, so there is no unmet arbitrage for aid to fill. Under the monopoly hypothesis, transfers are “also be fully offset by reductions in private investments,” and the reason is put memorably: “Giving goods to a monopolist does not reduce his interest in exploiting potential rents.” Only political risk leaves room: “Only insofar as political risk is an important factor in limiting capital flows can we expect transfers of capital to speed the international equalization of factor prices.”
Q14. What does Lucas recommend instead, and how strongly?
Human-capital accumulation, and tying aid to openness to competitive foreign investment – both framed as conditional judgements rather than established results. “In a world of largely immobile labor, policies focused on affecting the accumulation of human capital surely have a much larger potential. So too, I think, do policies in which aid of any form is tied to the recipient’s openness to foreign investment on competitive terms” (§IV, p. 96). Both clauses carry hedges the paper does not remove – “surely have a much larger potential”, “So too, I think” – and neither is supported by evidence presented in this paper; they follow from whichever combination of the four hypotheses turns out to hold, a question the paper explicitly leaves open.
Q15. What, then, does the paper actually settle?
Not which explanation is correct, but how large the anomaly is under standard assumptions and how much of it each candidate can absorb. The paper never claims to have identified the right answer – it asks “which combination, if any, of the four hypotheses I have advanced is adequate to account for the absence of income equalizing international capital flows” (§IV, p. 96) and leaves the combination undetermined, having recorded one unanswered historical question (“I do not know the answer”), one unmeasurable parameter (the geographic scope of knowledge spillovers), and one piece of contrary evidence on the colonial model. What it does establish is quantitative: an unadjusted return gap of about 58, an adjusted gap of about 5, a candidate mechanism that could close even that, and a set of policy conclusions that depend entirely on which of these is doing the work.
Key terms in this paper
Definitions below follow the paper's own usage.
- The neoclassical capital-flow prediction
- the paper's opening arithmetic, derived from "entirely standard assumptions on technology alone." Two countries producing the same good with the same constant-returns production function in homogeneous capital and labour must, if output per worker differs, differ in capital per worker; the Law of Diminishing Returns then puts the marginal product of capital higher in the poorer country; and if trade in capital goods is free and competitive, new investment occurs *only* in the poorer economy until capital-labour ratios, wages and returns are equalised. Lucas stresses the point of doing the arithmetic explicitly is to show "there is nothing at all delicate about this standard neoclassical prediction on capital flows."
- Effective labour (human-capital-adjusted)
- labour input measured in quality-adjusted units, which is how the paper's first candidate explanation reinterprets its own equations. Anne Krueger's (1968) estimates combine each country's mix of workers by education, age and sector with US estimates of how those factors affect productivity as measured by relative earnings, yielding the per capita income each country could attain with US physical capital per worker. Reading income per *effective* worker in place of income per worker cuts the US-India ratio from 15 to 3 and the predicted return ratio from 58 to about 5.
- External benefits of human capital
- Lucas's second candidate: the assumption, following his own 1988 paper and Romer (1986), that an economy's technology level is the average human capital of its workers raised to a power, so that average skill multiplies every worker's productivity exactly as a technology intercept would. The exponent is estimated at .36 from Denison's 1909-1959 US comparison -- "a 10 percent increase in the average quality of those with whom I work increases my productivity by 3.6 percent" -- and applied to Krueger's cross-country human-capital estimates it brings the predicted India-US return ratio to 1.04. Lucas flags two limits: the estimate assumes the whole human-capital stock grows at the schooling-driven rate of .009 ("I do not have any idea how accurate an assumption this is"), and the cross-country application assumes knowledge spillovers across national boundaries are exactly zero.
- Political risk
- Lucas's term for the class of capital-market imperfection in which no effective mechanism enforces international borrowing agreements, so the borrowing country would gain by walking away when repayment begins and, foreseeing this, the lender never lends. Its difficulty as an explanation is historical rather than logical: before about 1945 much of the Third World was under European-imposed legal arrangements in which a European's contract with an Indian borrower "could expect his contract to be enforced with exactly the same effectiveness" as a domestic one, which leaves unexplained why capital-to-effective-labour ratios were not equalised in the two centuries before 1945.
- Colonial monopoly (monopsony) model
- the fourth candidate, a model in the spirit of Adam Smith's analysis of colonialism: an imperial power with exclusive control over trade to and from a colony, but facing a free labour market there, chooses the colony's capital per worker to maximise output less competitively determined wages less the opportunity cost of capital. The first-order condition sets the colony's marginal product of capital above the world return by the derivative of the colonial wage with respect to capital per worker -- "It is the imperialist's monopsony power over wages in the colony that is crucial. His optimal policy is to retard capital flows so as to maintain real wages at artificially low levels." With a capital share of .4 this implies a colonial return about 2.5 times the European one.