Solow's own 1957 paper found that seven eighths of the growth in US output per hour over the previous four decades was unexplained by capital accumulation. Moses Abramovitz, writing in the same period, called the residual a measure of our ignorance, and the phrase has stuck because it is accurate.
What does the Solow residual contain?
What the residual actually contains is everything the two measured inputs missed. It contains genuine technological progress, which is what people mean when they cite it. It also contains improvements in the quality of labour that a headcount does not see, because a graduate and a school leaver enter L identically, and improvements in the quality of capital, because a 2024 machine tool and a 1994 machine tool enter K at deflated cost.
It contains reallocation gains, which are large in a country moving workers out of agriculture and into services and have nothing to do with any new idea. And it contains every measurement error in output, in the capital stock and in the labour series, with the sign wherever the errors happen to fall.
Advertisement
How should you read a TFP number?
Two consequences follow that professionals treat as reflexes.
The first is that TFP growth is not a policy variable. You cannot instruct an economy to raise its residual, because the residual is defined by what your inputs failed to capture. What you can do is name a mechanism, human capital or competition or contract enforcement, and then measure whether it moved.
The second is that a decomposition is only comparable with another built the same way. India's 1.480 and America's 1.182 in the worked examples share a depreciation rate, an investment-based capital construction and a decade, which is what lets them sit on one page. They do not share a labour concept. The India build counts workers and the United States build uses an hours index, and a headcount series overstates labour input growth whenever the average working week shortens, so the residual takes the difference. That single gap is the caveat the side-by-side carries, and it is named again where the two builds are set out.
Lift a TFP figure from one source and set it beside a figure from another and you are usually comparing conventions rather than economies.
Capital deepening versus technological progress
Two economies grow at 5.0% a year. The first does it with capital growing at 9.0% and labour at 1.0%; the second with capital growing at 3.0% and the same labour. With α = 0.35, the first economy's capital contributes 3.150 points and its residual is 1.200 points, or 24.00% of growth. The second economy's capital contributes 1.050 points and its residual is 3.300 points, or 66.00%.
Same headline. Completely different futures.
The first economy is growing by capital deepening: raising capital per worker. Diminishing returns guarantee that this process decelerates, because each additional unit of capital per worker produces less output than the last, and it stops entirely at the steady state where new investment only covers depreciation and the growth of the workforce. The second economy is growing mostly on the residual, which the model treats as unbounded.
What is the forecasting rule for capital deepening?
The forecasting rule this yields is the most practical thing in growth accounting. Capital deepening decelerates by construction; the residual does not. So extrapolate the residual with a flat or declining hand, and force the capital term to slow as the capital-output ratio rises.
India's decade to FY2024 took 3.354 of its 5.90 points from capital, more than half, on a stock growing at 8.39% against output at 5.90%. That cannot continue, because the capital-output ratio would rise without limit and the marginal product of capital would fall toward zero. India's ratio rose from 2.20 to 2.78 across the decade, and worked example 7 is where that arithmetic gets applied honestly.