Technology and work: The productivity puzzle

Technology and work: The productivity puzzle

Phuah Eng Chye (12 May 2018)

When we talk with the optimists, we are convinced that the recent breakthroughs in AI and machine learning are real and significant. We also would argue that they form the core of a new, economically important potential General Purpose Technologies (GPT). When we speak with the pessimists, we are convinced that productivity growth has slowed down recently and what gains there have been are unevenly distributed, leaving many people with stagnating incomes, declining metrics of health and well-being, and good cause for concern. People are uncertain about the future, and many of the industrial titans that once dominated the employment and market value leaderboard have fallen on harder times. These two stories are not contradictory. In fact, any many ways, they are consistent and symptomatic of an economy in transition…The breakthroughs of AI technologies already demonstrated are not yet affecting much of the economy, but they portend bigger effects as they diffuse. More importantly, they enable complementary innovations that could multiply their impact. Both the AI investments and the complementary changes are costly, hard to measure, and take time to implement, and this can, at least initially, depress productivity as it is currently measured.” Erik Brynjolfsson, Daniel Rock, Chad Syverson (2017) “Artificial intelligence and the modern productivity paradox: A clash of expectations and statistics.”

Even if the effect of technology on employment was unpredictable, the logical assumption was that the employment slack would be offset by the positive impact of technology on productivity. This leads us to the next puzzle, namely why productivity growth has slowed despite rapid technological innovation.

Maurice Obstfeld and Romain Duval note “between the 2000–2007 and 2011–2016 periods, total factor productivity (TFP) growth dropped from 1 to 0.3% in advanced economies and from 2.8 to 1.3% in emerging and developing economies. These numbers leave out the 2008–10 crisis period, during which productivity naturally plummeted. For advanced and low-income countries, the sharp deceleration in TFP occurred on the back of a slowdown that had already started prior to the crisis.” Many explanations have been put forward to explain the productivity puzzle.

  • Weak demand. The weak demand argument, usually associated with falling investments and secular stagnation, blames weak economic conditions for causing low productivity. Jacques Bughin, Hans‐Helmut Kotz and Jan Mischke found “about half of the recent drop in productivity growth is from weak demand, particularly in Europe”. They explained weak demand operates through three channels to dampen productivity growth; namely by discouraging investments (low demand, ample capacity and public sector disinvestment), through economic of scale effects and the shift to lower value products.
  • Waning productivity gains from technology. Jacques Bughin, Hans‐Helmut Kotz and Jan Mischke attributed “the other half (of the recent drop in productivity growth), and a more significant factor in the US, was the waning of the productivity boom that began in the mid-1990s”. It has also been argued that technological gains have been waning because the pace of innovation has slowed at the technological frontier[1].
  • Sector transition. Karen Harris, Austin Kimson and Andrew Schwedel note “the shift to a service-led economy moves a large percentage of workers out of high-productivity growth industries into low-productivity growth industries…in the US – Between 1993 and 2014, auto manufacturing output per worker increased by 128%. Over that same period, output per worker in hospitals increased by just 16%. In contrast, employment increased 28% in hospitals, while it declined by about the same amount in auto manufacturing. Capital investment in auto manufacturing supported high labor productivity growth. While the overall quality of medical care in hospitals may have increased over time, it remains a field in which human touch is highly intensive, with fewer options for capital investment to boost labor productivity”. In this regard, the earlier transition from agriculture to manufacturing shifted work from low to high-productivity jobs which expanded employment and income. The concern is that the shift from manufacturing to services will run headlong into the “productivity lag”[2] which limit the prospects for productivity gains.
  • Aging. Gustavo Adler, Romain Duval, Davide Furceri, Sinem Kiliç Çelik, Ksenia Koloskova and Marcos Poplawski-Ribeiro estimate aging can “affect TFP growth and, all else equal, may have played a role in slowing TFP gains – perhaps by as much as 0.2–0.5 percentage point per year on average across advanced economies, and about 0.1 percentage point on average across emerging and developing economies, from the 1990s through the 2000s”. Mårten Blix points out “what makes the challenge ahead so great is the magnitude of retirements ahead…there were more retirees than entrants to the labour market…When the young are to enter – and the elderly to leave – the labour market, there is a clear risk that productivity growth will, at least temporarily, be dampened because many are learning new jobs. Notably, in many occupations craftsmanship is important, and the benefits that the young have in terms of speed and ability to learn can be contrasted to the benefits of experience and expertise that the older workers have. To the extent that skills are not transferred in full, there is a tangible risk that productivity growth may be held back for several years.” Daron Acemoglu and Pascual Restrepo documents that labor markets “undergoing more major demographic change have invested significantly more in new robotic and other automation technologies…because ongoing demographic changes are increasing the scarcity of middle-aged workers and industrial automation is most substitutable with middle-aged workers…in the presence of demographic change, industries with the greatest opportunities for automation are experiencing more rapid growth of productivity and greater declines in labor share relative to other industries.” MGI adds “slowing population growth and aging in advanced economies are also reducing demand momentum. Not only is the consumer pool expanding more slowly, but as the population ages, demand tends to shift to industries such as health care and the public sector that have seen slower (measured) productivity growth.”
  • Financial crisis-related effects. Gustavo Adler, Romain Duval, Davide Furceri, Sinem Kiliç Çelik, Ksenia Koloskova and Marcos Poplawski-Ribeiro note advanced economies in the aftermath of the global financial crisis “has displayed TFP hysteresis – persistent TFP loss from a large and seemingly temporary shock…weak corporate balance sheets, combined with tight credit conditions, have undermined TFP growth, partly by constraining investment in intangible assets in distressed firms…the boom-bust financial cycle and its corollary of weak corporates and banks has also increased misallocation of capital within and across sectors…an adverse feedback loop of weak aggregate demand, investment, and capital-embodied technological change…elevated economic and policy uncertainty may have further weakened TFP growth, partly by tilting investment away from higher-risk, higher-return projects”.
  • Top firms and concentration. Erik Brynjolfsson, Daniel Rock and Chad Syverson assert “the benefits of the new technologies are being enjoyed by a relatively small fraction of the economy, but the technologies’ narrowly scoped and rivalrous nature creates wasteful gold rush-type activities. Both those seeking to be one of the few beneficiaries, as well as those who have attained some gains and seek to block access to others, engage in these dissipative efforts, destroying many of the benefits of the new technologies.” As a result, the differences in terms of productivity, market shares, profit margins and wages have widened between top performing and laggard firms. “There are concerns that industry concentration is leading to substantial aggregate welfare losses due to the distortions of market power. Furthermore, growing inequality can lead to stagnating median incomes and associated socio-economic costs, even when total income continues to grow”.
  • Diffusion lag. Some attribute the productivity slowdown to the lag in the realization of productivity benefits. Erik Brynjolfsson, Daniel Rock and Chad Syverson point out “there are two main sources of the delay between recognition of a new technology’s potential and its measureable effects. One is that it takes time to build the stock of the new technology to a size sufficient enough to have an aggregate effect. The other is that complementary investments are necessary to obtain the full benefit of the new technology, and it takes time to discover and develop these complements are and to implement them…In the case where the intangible artificial intelligence (AI) capital stock is growing faster than output, then TFP growth will be underestimated, while TFP will be overestimated if capital stock is growing more slowly than output…Furthermore, suppose the relevant costs in terms of labor and other resources needed to create intangible assets are measured, but the resulting increases in intangible assets are not measured as contributions to output. In this case, not only will total GDP be undercounted but so will productivity, which uses GDP as its numerator. Thus periods of rapid intangible capital accumulation may be associated with lower measured productivity growth, even if true productivity is increasing”.
  • Mismeasurement. The productivity slowdown has been blamed on mismeasurement or the difficulty in estimating the benefits of rapid innovation. An IMF Staff Discussion Note article suggest “the presence of effects causing underestimation of GDP growth is not in doubt, but a stable measurement error in the GDP growth rate would not cause productivity growth to slow. The question, therefore, is whether measurement error got larger around the time the estimated rate of productivity growth slowed”. In its review of other studies, it concludes the measurement error has not increased significantly and is unlikely to be the main factor behind the productivity slowdown.

The behaviour of weak productivity growth is all the more puzzling given the recovery in employment. The McKinsey Global Institute (MGI) note since 2010, “steady hiring has continued to expand employment and hours have continued to grow for longer than in most previous recoveries, creating what we term a job-rich recovery.” But it was accompanied by a “productivity-weak recovery, with low value added but high hours worked growth, and a broad-based decline with a distinct lack of productivity-accelerating sectors.”

MGI suggests the collision of three waves explain these patterns. Wave 1 was due to “the waning of a productivity boom that began in the 1990s dragged down productivity growth by about one percentage point. Around 2005, a decade-long productivity boom from a PC, software, and database system ICT revolution and the restructuring of domestic operations and global supply chains was ending. By then, retail supply chain management tools were broadly implemented and manufacturing offshoring momentum slowed.”

Wave 2 was due to the “financial crisis aftereffects, including weak demand and uncertainty, caused another percentage point drag. After the crisis hit, sectors such as financial services went from boom to bust, and companies reacted to weak demand and uncertainty by holding back investment, driving capital intensity growth down to the lowest rates since World War II. Weak demand further depressed productivity growth through negative economy of scale effects and downshifts in product and service mix.” MGI observed “the first two waves have dragged down productivity growth by 1.9 percentage points on average across countries since the mid-2000s, from 2.4 percent to 0.5 percent.”

MGI attributes wave 3 to digitization. They note that “digitization, often involving a transformation of operating and business models, promises significant productivity-boosting opportunities but the benefits have not yet materialized at scale. This is due to adoption barriers and lag effects as well as transition costs; the net effect on productivity in the short term is unclear…analysis show that transition costs can include an initial duplication of structures and investment, cannibalization of incumbent business, and the diversion of management attention”.

MGI highlights “the broad-based pattern of job-rich but productivity-weak recovery across most countries raises the question of why companies are increasing employment without corresponding increases in productivity growth. It also highlights the importance of examining demand side drivers for slow value-added growth and low productivity growth…looking across more than two dozen sectors, we find few jumping sectors today, and the ones that are accelerating are too small to have an impact on aggregate productivity growth. For example, only 4 percent of sectors in the United States were classified as jumping in 2014, compared with an average of 18 percent over the last two decades, and they contributed only 4 percent to value added. The distinct lack of jumping sectors we have found across countries is consistent with an environment in which digitization and its benefits to productivity are happening unevenly.”

In addition, “since the Great Recession, capital intensity, or capital per hour worked, in many developed countries has grown at the slowest rate in postwar history. Capital intensity indicates access to machinery, tools, and equipment and is measured as capital services per hour. An important way productivity grows is when workers have better tools such as machines for production, computers and mobile phones for analysis and communication, and new software to better design, produce, and ship products, but this has not been occurring at past rates. A decomposition of labor productivity shows that slowing growth of capital per hour worked contributes about half or more of the productivity growth decline in many countries.”

MGI notes “with exceptionally few jumping sectors, and accelerating sectors that are too small to have a major impact on aggregate productivity growth. Not only do we see a distinct lack of productivity-igniting sectors, but we also see a broad-based slowdown of productivity growth across sectors after 2010. More than 65 percent of sectors experienced a productivity-growth slowdown” in most countries. “Three sectors – manufacturing, retail and wholesale trade, and information and communication services – together contributed between 54 and 84 percent of the productivity-growth decline.” It adds that “weak capital intensity growth accounts for at least half of the measured productivity growth decline across countries.”

Robert D. Atkinson and John Wu provide the interesting insight that “many believe that if innovation only accelerates even more, then new jobs in new industries and occupations will make up for any technology-created losses. But the truth is that growth in already existing occupations is what more than makes up the difference. In no decade has technology directly created more jobs than it has eliminated. Yet, throughout most of the period from 1850 to present, the U.S. economy as a whole has created jobs at a robust rate, and unemployment has been low. This is because most job creation that is not explained by population growth has stemmed from productivity-driven increases in purchasing power for consumers and businesses. Such innovation allows workers and firms to produce more, so wages go up and prices go down, which increases spending, which in turn creates more jobs in new occupations, though more so in existing occupations (from cashiers to nurses and doctors).”

In this regard, “in the 1960s, a decade where productivity and median wages increased more than 30 percent, the fastest-growing occupations were janitors and building cleaners (as commercial real estate expanded significantly); laborers and freight, stock and material movers (as we needed more workers to transport the rapidly growing number of goods American consumers were buying); maids and house cleaners (as more Americans could now afford household help and because of the increase in women entering the labor force); and secretaries and administrative assistants (as the office economy grew). None are occupations that are intrinsically created by new technology. But technological innovation is central to the acceleration of the economy and the creation of these jobs. Together, jobs in these four occupations accounted for 56 percent of all net job creation in the 1960s.” Robert D. Atkinson and John Wu therefore conclude “the single biggest economic challenge facing advanced economies today is not too much labor market churn, but too little, and thus too little productivity growth”.

There is growing concerns over the consequences of the productivity puzzle. Maurice Obstfeld and Romain Duval warn the persistence of low productivity growth “would have profound, adverse implications for progress in global living standards, the sustainability of private and public debts, and the space for macroeconomic policies to respond to future shocks. In conditions of high income inequality, low growth also undermines social cohesion, with adverse political repercussions”.

But there are expectations of a recovery. MGI suggests there are grounds to “expect productivity growth to recover and see the potential for at least 2 percent growth a year over the next ten years, with 60 percent coming from digital opportunities.” MGI explains “digitization has not yet reached scale, with a majority of the economy still not digitized. MGI has calculated that Europe overall operates at only 12 percent of digital potential, and the United States at 18 percent, with large sectors lagging in both. While the ICT, media, financial services, and professional services sectors are rapidly digitizing, other sectors such as education, health care, and construction are not.”

However, “several factors dampen the pace of digital adoption, including consumer preferences for digital vs. traditional products and services, the risks associated with major business model changes including fear of cannibalization, capability gaps, and concerns about technological obsolescence. These factors make some firms cautious about digitizing too fast, delaying adoption and the realization of the benefits of digitization. Naturally, the risk of business model changes and cannibalization is much larger for companies with large existing legacy businesses than it is for new digital-only players, creating very different incentives for incumbents than new entrants in any industry.”

MGI expressed caution in that “digitization may change industry structure and economics in ways that could constrain innovation and productivity growth.” In particular, “the rise of platforms with large network economies of scale raises concerns about rising market power, but the implications are uncertain…These digital platforms are often not limited by geographic boundaries but have the potential for global reach, creating potential for market power on a scale never seen before. While the economic cost associated with monopolies has been well established, digital platforms may exhibit unique characteristics that make the implications of that market power differ from many past monopolistic industries”. In this regard, “rising corporate concentration has not yet reduced incentives to innovate but could do so in the future… and translate into weaker incentives to innovate and invest in raising productivity”.

Jacques Bughin, Hans‐Helmut Kotz and Jan Mischke add that “leakages may challenge the realisation of this digitised demand potential”. They point to “concern that some demand drags may be more structural – or secular – than purely crisis-related. Broad-based income growth has diverged from productivity growth for a long while now. A declining labour share of income and a rising trend in income inequality have been eroding median wage growth. Moreover, the rapidly rising costs of housing exert a dampening effect on consumer purchasing power. It appears increasingly difficult to make up for weak consumer spending (of largely liquidity-constrained households) via higher investment…rising returns on investment discourage capital expenditures relative to dividends. Demographic trends may further diminish investment needs through an ageing population having less need for residential and infrastructure investment. These demand drags are occurring while interest rates – endogenously reflecting expected mediocre growth perspectives – are hovering near the zero lower bound. All of this holds back the pace at which capital per worker increases, impacts company incentives to innovate, and thus puts a structural damper on productivity growth. In a low-pressure economy, the virtuous circle does not get under way.”

Overall, solving the productivity puzzle is seen as the key to addressing deflationary pressures from demographic aging and secular stagnation. In this regard, it is also commonly held out that productivity gains are deemed as a pre-condition to overcoming wage stagnation. But productivity may turn out to be the policy red herring. The concept of productivity loses relevance when manufacturing costs dwindle to a small percentage of GDP and the economy becomes largely intangible and service-driven. This is illustrated by the tendency of the profit share of GDP to rise in tandem with service sector expansion and financialisation[3]. In tandem with this, the bigger concern is to address the income distribution bottlenecks that are causing wages to lag productivity growth and the labour share of income to decline.


Daron Acemoglu, Pascual Restrepo (March 2018) “Demographics and automation”. National Bureau of Economic Research. Working Paper 24421.

Erik Brynjolfsson, Daniel Rock, Chad Syverson (November 2017) “Artificial intelligence and the modern productivity paradox: A clash of expectations and statistics.” NBER Working Paper No. 24001.

Gustavo Adler, Romain Duval, Davide Furceri, Sinem Kiliç Çelik, Ksenia Koloskova, Marcos Poplawski-Ribeiro (2017) “Gone with the headwinds: Global productivity”. IMF Staff Discussion Note. file:///C:/Users/user/Downloads/sdn1704%20(4).pdf

Jaana Remes, James Manyika, Jacques Bughin, Jonathan Woetzel, Jan Mischke, Mekala Krishnan (February 2018) “Solving the productivity puzzle: The role of demand and the promise of digitization”. McKinsey Global Institute.

Jacques Bughin, Hans‐Helmut Kotz, Jan Mischke (22 March 2018) “Strong aggregate demand: Critical for reaping benefits of digitisation”.

Karen Harris, Austin Kimson, Andrew Schwedel (7 February 2018) “Labor 2030: The collision of demographics, automation and inequality”. Bain & Company.

Maurice Obstfeld, Romain Duval (10 January 2018) “Tight monetary policy is not the answer to weak productivity growth”.

Mårten Blix (February 2013) “Future welfare and the ageing population”. Interim report from the Commission on the Future of Sweden.

Phuah Eng Chye (26 August 2017) “The services economy: Revisiting Baumol’s cost disease.”

Phuah Eng Chye (2 September 2017) “The services economy: Comparing the manufacturing and service paradigms”.

Robert D. Atkinson, John Wu (8 May 2017) “False alarmism: Technological disruption and the U.S. labor market, 1850–2015.” The Information Technology and Innovation Foundation (ITIF).

[1] Gustavo Adler, Romain Duval, Davide Furceri, Sinem Kiliç Çelik, Ksenia Koloskova and Marcos Poplawski-Ribeiro.

[2] Phuah Eng Chye “The services economy: Revisiting Baumol’s cost disease.”

[3] Phuah Eng Chye “The services economy: Comparing the manufacturing and service paradigms”.

Leave a Reply