The fourth industrial revolution is bringing about numerous challenges and opportunities. This column summarises selected takeaways from a recent ECB conference which brought together leading minds from academia, institutions and the private sector on the expected effects of digitalisation on the economy, including labour markets, productivity, investment and inflation, and possible implications for monetary policy.
On 4-5 July, the ECB hosted the conference, “Challenges in the digital era”, which featured contributions from several leading economists on the effects of digitalisation on the real economy. In this column, we summarise three themes that were keenly debated: labour market implications of digitalisation; effects on inflation, market power and monetary policy; and the productivity promises of digital technologies.
Work, wages and technology: Past, present and future
David Autor presented work on the geography of labour markets and the work of the past and the future, using US data. He showed that real non-college wages have been falling during the last 40 years, as many mid-skilled, mostly urban, non-college jobs have disappeared. In the period after WWII, high-skilled occupations in urban factories and offices evolved in close association with mid-skilled jobs, where employees were both college and non-college workers.
Instead, in recent decades, trade and automation have caused an occupational realignment by depressing the demand for administrative and clerical jobs; while low-skilled jobs have experienced an inflow of non-college and some college workers, only the more educated workers have moved towards highly paid jobs. Consequently, employment polarisation has increased over time – disproportionately so in cities, where the real wage gap between college and non-college workers has widened and the urban wage premium for non-college workers has fallen.
Yet, opportunity is still concentrated in cities, especially ‘superstar’ cities, although not for less-educated workers. Middle-skilled jobs are scarcer in cities than in rural/suburban areas, representing a critical change compared to a generation ago. Non-metropolitan areas are changing more slowly than the major cities and are getting older, albeit with a relatively stable structure of jobs, skills, and wages.
Similarly, cities account for a large share of employment in ‘new jobs’ as both workers in technology-complemented ‘frontier jobs’ and ‘wealth’ occupations (in-person services to wealthy urban workers, such as yoga instructors and baristas) are disproportionately present in urban areas. However, while workers at the frontier are more educated and earn higher wages than the average worker, ‘wealth jobs’ were paid the average wage in 2015.
On the other hand, ‘last mile’ occupations (jobs with nearly automated tasks and a residual human component) are as represented in cities as in less populated areas and pay low wages. As 38% of last-mile workers are non-college workers, against 26% of the overall workforce, a major future challenge is to rebuild career paths, not just jobs, for non-college workers.
The implications of digital technologies for employment were an important theme. Anna Salomons (Bessen et al. 2019) provided the first estimates of the effects of automation on workers at firms that automate (using Dutch data). While there is a displacement effect, it is far lower relative to mass layoffs and plant closings.
Displacement works primarily through separations for long-time and older workers, as well as an increase in non-employment following separation, partially due to early retirement; this translates into a decline in earnings, despite no evidence of wage scarring, of about 10% less of one year’s earnings over five years. The effect is pervasive across industries and workers’ skills.
James Bessen (Bessen and Righi 2019) evaluated the outcomes of firms with a high share of software developers in their workforce, who are likely to foster IT-driven innovation in a firm, relative to other firms. IT shocks are associated with a positive effect on productivity, sales and, with the exception of employment in manufacturing, also employment. Additional results point to a declining labour share of revenue as a result of an IT event and to a slowing response after 2002.
Instead, Gino Gancia discussed recent research (Blanas et al. 2019) which showed that ICT capital investment correlates to sectoral employment gains, but software capital to employment losses. At the firm level, he provided new evidence of adjustment (Bonfiglioli et al., 2019): robot adoption leads to productivity gains for the firm and wage gains for non-production employees, but lower firm employment and no post-adoption increase in sales, suggesting that firms may simply be raising markups. Overall, a common thread is that, at the firm level, automation has overall more negative effects than ICT in general, consistent with recent theories.
Bessen also took a broader view of the effects of technology on employment, stressing the role of market satiation. For instance, in the US textile industry, productivity growth was associated with higher employment until 1950, as people owned few clothes and demanded more; further price declines failed to boost employment as consumers had already ‘filled their closets’. He showed that computerisation has positive employment effects in non-manufacturing industries but the opposite for manufacturing, where demand is less elastic.
Rachel Ngai pointed out a similar hump-shaped pattern for female hours in home production from 1870-1960, which rose as higher living standards created a demand for cleaner homes and better food, and then fell once demand was satiated. The increasing share of employment in services and female reallocation from home to market production is also a result of the interaction of structural transformation and demand.
At this stage of the digitalisation process, as Ngai emphasised, retraining has a central role in minimising employment losses and facilitating transitions within sectors. Similarly, Manuel Trajtenberg stressed the relevance of non-formal education, which should enhance non-cognitive skills, while Romain Duval suggested that labour market institutions and social insurance may have to be rethought. Institutions should shift support to workers, not jobs, and include policies to increase the portability of social rights to facilitate mobility even across borders. Ultimately, whether the development of the current technological process will benefit/exacerbate social challenges is highly dependent on the policies that will be adopted.
Market power in the digital age
The implications of technology on market power, and market structure more generally, were at the centre of attention. One such implication was on pricing. Alberto Cavallo studied how online competition affects retail pricing behaviour, using daily prices from the Billion Prices Project. Over 2008-2017, median price duration fell from 7 to 3.5 months, with online retailers at the lower end of that range. This effect was not driven by sale prices or retailer composition, or by the sign of the price change. He also found that direct competition with Amazon reduces implied durations by about 20%, suggesting a meaningful role for online competition in the increase in price flexibility.
In addition, he found that online retailers tend to have an especially strong preference for single pricing across locations. This may be due not only to online transparency but also a desire not to risk customer anger at geographical price differentials. Taken together, uniform pricing and more flexibility make prices more sensitive to aggregate national (rather than local) shocks, implying higher pass-through of costs to prices. These findings have important implications for issues such as inflation measurement, price stickiness, price dispersion and welfare.
Chiara Criscuolo provided new evidence on industry concentration, suggesting that of the countries sampled (Belgium, Spain, France, Finland, Italy, the UK, Sweden, the US, and Japan) concentration increased by 5 percentage points on average over 2002-2014. The main drivers of concentration were investment, openness to trade and FDI, digital intensity, and regulation. Of these, intangible investment was a particularly strong predictor of concentration changes, especially strong in globalised, already concentrated and digitally intensive country-industries, consistent with the scalability of intangible capital. Taken at face value, this would imply ‘good’ concentration (driven by firm performance), but more work is needed to detect possible frictions such as rent-seeking behaviour or slow technology diffusion.
Anton Korinek argued that the rise of superstars (firms with high market shares, rising mark-ups, intangible-capital intensive) is the natural outcome of digital innovation (advances in collection, processing, and provision of information). Digital innovation involves a virtuous circle of cost reductions, innovations, increasing market share and subsequent costs savings passed on to customers. In this process, demand for labour, though initially falling, recovers as output expands. This constellation, however, posed dilemmas for policymakers in terms of anti-trust and macro stabilisation objectives.
Christos Genakos cautioned that the question, “what is the effect of concentration on prices or markups?”, is not well-posed (because of measurement, endogeneity, technological differences across sectors, and so on). He called for more detailed industry studies precisely because the mechanisms at work are not the same across industries. This could help answer questions on sources of market power – whether they arise from network effects (reflecting digital platforms), monopsony power, rent seeking, globalisation, or from particular antitrust regimes.
Chad Syverson considered what market power, relative to perfect competition, implies for monetary policy pass-through. By lowering the cost of capital, monetary accommodation shifts out firms’ marginal cost (MC) curve. As profit maximising firms produce until marginal revenue (MR) equals MC, the slope of the MR curve determines the new equilibrium. Under perfect competition (i.e. a flat MR curve), a monetary expansion would lead to a larger output expansion than under a monopoly (i.e. a steep MR curve). As such, companies with high market power respond in general less to changes in costs, and hence monetary policy, than perfectly competitive firms.
This does not mean that lower (but non-zero) market power will result in higher pass-through; this will depend on how firm incentives change with power. Higher power implies a steeper demand curve, but whether it implies a steeper MR curve also depends on whether the demand curve flattens or steepens as output changes, and on the size of the change in the optimal quantity as competition changes.
Growth and productivity: Can digital technologies deliver?
A major theme of the conference was the modern productivity paradox of great innovations, predominantly driven by artificial intelligence (AI), and lacklustre productivity growth. Chad Syverson (Brynjolfsson et al. 2017, 2018) argued that the resolution of this paradox, as with the Solow paradox of the 1980s, lies in the particular nature of the technology at hand (computers then, artificial intelligence now). Such general-purpose technologies (GPTs) are characterised by the ability to improve over time and spawn complementary innovations and hence suffer from implementation lags.
Substantial investments in physical, human, managerial, and intellectual capital need to be made for the productivity effects of these technologies to materialise. Intriguingly, the productivity effects of two disparate GPTs – electricity and IT- have a remarkably similar behaviour, featuring a very long lag (around three decades) before rapidly accelerating and then slowing down.
He also suggested that the slowdown is not the result of mismeasurement. That said, there certainly is mismeasurement, in particular due to the intangible nature of digital technologies (data, algorithms, firm-specific human capital, new processes), which are poorly measured. As intangibles are both outputs (investment) and inputs (capital), TFP will be underestimated initially, when the growth rate of this new investment and its return is high, and overestimated later when enough capital has been accumulated, giving rise to a mismeasurement J-curve.
Pat Bajari (Bajari et al. 2019) demonstrated the value of using AI techniques (text analysis and computer vision) and online transactions (rather than posted) prices to augment hedonic models, in order to address the classic problem of estimating quality-adjusted inflation indices and hence address another source of mismeasurement.
John Fernald and co-authors (Esfahani et al. 2019) documented that the productivity slowdown in advanced economies has been offset, at a global level, by stronger productivity growth in emerging economies. Since labour and product markets in emerging economies are less efficient, they put a drag on world output growth. As such, the authors augmented a standard growth accounting framework to take into account distortions in labour, product and capital markets, including markups. Their main finding is that even though the growth of world labour productivity (ALP) is highly volatile, the country-industry contribution to world growth is much smoother. The bulk of this volatility reflects shifts in the misallocation of capital and, especially, labour in the world economy.
Diego Comin (Anzoategui et al. 2019) showed evidence of procyclicality in technology-related firm decisions (R&D expenditure, technology adoption) and the protracted effects of the Great Recession on several relevant metrics. He then introduced a mechanism that endogenises growth at business cycles frequencies in an otherwise standard model, with a central role for R&D and technology adoption to drive cyclical fluctuations. The estimated model showed that the productivity slowdown in the Great Recession was primarily the result of a decline in the speed of technology adoption, which was itself caused by the recession.
Pat Bajari discussed the value of data in prediction (Bajari and Chernozukhov, 2019); there is a misconception that big data is always better, but in fact data richness has diminishing returns, and how data is used to improve models can be important. He further emphasised the role of digital technologies in revolutionising decisions of firms; cheaper CPU power made it possible to deal with big data and, through continuous incremental enhancements in data handling, data science has allowed firms to move from heuristics to scientific decision-making in many different areas (inventory, truck-load and itinerary, human resource management).
As for specific technologies, Peter Gal (Gal et al. 2019) showed evidence of large productivity gains from broadband internet, cloud computing and back-office integration systems. These effects are in general stronger for more productive firms, with the exception of cloud computing, which allows firms to avoid large fixed costs and is less demanding in terms of complementary investments.
He also presented results on service platforms (Bailin Rivares et al. 2019), showing large within-firm effects from platforms that are ‘aggregators’ of incumbents (e.g. Booking, TripAdvisor) on productivity, profits and employment. Competing ‘disruptors’ (AirBnb, Uber) instead had no effects on incumbent productivity, but the competition shock did reduce markups, wages and employment for affected firms. Productivity in services is typically lower than manufacturing (due to lower uniformity, lower competition, limited scale), and raising productivity in these sectors is crucial given their increasing share of total activity in modern economies.
The rising role of intangible assets was another major theme as regards to both productivity and market structure. Janice Eberly showed that intangibles are associated with greater concentration in the US, which could either be the result of changes in technology in a competitive environment or of market power. The heterogeneous nature of this relationship suggests caution when designing policies that aim to promote intangible capital and limit concentration.
Jonathan Haskel suggested that the presence of intangibles substantially changes both our understanding of modern economies and also our measurement thereof; he gave the example of AI as a combination of fast hardware (a tangible), using new software (a measured tangible) which searches databases (unmeasured intangibles).
He further noted, as did other speakers, the particular nature of financing for intangibles; they are often sunk and highly firm-specific, unlike tangible capital which may be resold. As such, they are unsuitable as collateral for bank financing. Although intangible investment did not take as big a hit in the crisis as did tangible, its rate of growth was hampered, and in fact the slowdown of TFP growth was strongest in countries with slower intangible capital deepening; indeed, he showed evidence that bank lending is skewed towards real estate and away from intangibles.
Romain Duval (Ahn et al. 2019) showed the strong effects of expansionary monetary policy in mitigating the effects of financial frictions on intangible investments. Reinhilde Veugelers singled out the relative dearth of specialised non-bank finance as a major barrier for digital investment in Europe, and showed that businesses in the EU are falling behind in the digital R&D frontier, relative to both their US and Chinese counterparts. The differences with the US at the level of the firm were not major, so the growing digital divide between the EU and US must be driven by composition effects, as old/small firms are significantly less likely to be digitally active.
To read original report, click here
References
Ahn, J, R Duval and C Sever (2019), “Macroeconomic Policy, Product Market Competition, and Growth: The Intangible Investment Channel”, Working Paper.
Anzoategui, D, D Comin, M Gertler and J Martinez (2019), “Endogenous Technology Adoption and R&D as Sources of Business Cycle Persistence”, American Economic Journal: Macroeconomics, forthcoming.
Autor, D (2019), “Work of the Past, Work of the Future”, American Economic Association: Papers and Proceeding 109(5): 1–32.
Bailin Rivares, A, P Gal, V Millot and S Sorbe (2019), “Like it or not? The impact of online platforms on the productivity of incumbent service providers”, OECD Economics Department Working Papers, No. 1548.
Bajari, P, V Chernozhukov, A Hortaçsu and J Suzuki (2019), “The Impact of Big Data on Firm Performance: An Empirical Investigation”, AEA Papers and Proceedings 109: 33-37.
Bajgar, M, C Criscuolo and J Timmis (2019), “Supersize Me: Intangibles and Industry Concentration”, Mimeo.
Bessen, J (2019), “Automation and Jobs: When Technology Boosts Employment”, Economic Policy,forthcoming.
Bessen, J, M Goos, A Salomons and W van den Berge (2019), “Automatic Reaction: What Happens to Workers at Firms that Automate?”, Working Paper.
Bessen, J and C Righi (2019),”Shocking Technology: What happens when firms make large IT investments?”, Working Paper.
Blanas, S, G Gancia and S Y Lee (2019), “Who is Afraid of Machines?”, Working Paper.
Brynjolfsson, E, D Rock and C Syverson (2017), “Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics”, Working Paper.
Brynjolfsson, E, D. Rock and C Syverson (2018), “The Productivity J-Curve: How Intangibles Complement General Purpose Technologies”, Working Paper.
Cavallo, A (2018), “More Amazon Effects: Online Competition and Pricing Behaviors”, Working Paper.
Esfahani, M, J Fernald and B Hobijn (2019), “World Productivity: 1995-2014”, mimeo.
Gal, P, G Nicoletti, T Renault, S Sorbe and C Timiliotis (2019), “Digitalisation and Productivity: In Search of the Holy Grail – Firm-level Empirical Evidence from EU Countries”, OECD Economics Department Working Papers, No. 1533,
Korinek, A and D X Ng (2019), “Digitization and the Macro-Economics of Superstars”, Mimeo.