herrendorf16may.jpg
VoxEU Column Productivity and Innovation

Structural change and the productivity slowdown

Baumol argued that structural change may lead to a productivity slowdown due to a reallocation of production to service industries with low productivity growth. This column uses a new framework to estimate the effects of Baumol’s disease on future productivity growth in the US. The results suggest that future structural change will not reduce productivity much further thanks to substitutability within the broader service sector.

Many industrialised countries have experienced a slowdown in labour productivity growth since the 1970s. This has led to a large body of research on what the causes of the productivity slowdown are and also to a lively debate about whether future productivity growth is likely to rebound, stay the same, or decline further. For example, Fernald and Jones (2014) and Fernald (2016) investigated what role the two main engines of past productivity growth, human capital accumulation and innovation, have played in the past and may play in the future. Moreover, Gordon (2016) revisited the major innovations of the last 200 years. He concluded that the ‘low-hanging fruit’ have been picked andthatthe impact of future innovations on human welfare will pale compared to past ones. Lastly, Baumol (1967) recognised that structural change is one reason for the productivity slowdown in the US, because it reallocates production to the service industries which tend to have relatively low productivity growth. This is often referred to as Baumol’s (cost) disease, or just the cost disease. Baumol (1967) drew particular attention to the fact that production is even reallocated to stagnant service industries that have hardly any productivity growth, for example education.1

Estimating the future impact of Baumol’s disease

In a recent paper, we investigate the effects of Baumol’s disease on future productivity growth in the context of a new model of structural change (Duernecker et al. 2018). Our work pays particular attention to the question of whether the stagnant service industries will gradually take over the economy in the future. We define productivity as the value added per human capital-adjusted hours (‘value added per efficiency units’). To get sharp results, we focus on how changes in the sectoral composition of the economy affect productivity growth while takingthe processes for total efficiency units and sectoral productivity as exogenously given. Using efficiency units is crucial in this context because it permits thinking about the implications for productivity of reallocating across sectors workers with different levels of human capital.

The first step in the analysis is to establish that Baumol’s disease considerably reduced productivity growth in the post-war US by comparing the average annual growth rates of productivity during the 20-year periods of 1947-1967 and of 1987-2007. Taking 20-year averages serves to smooth out business cycle fluctuations and stopping in 2007 serves to avoid the Great Recession. The data source is WORLD KLEMS, which has human capital-adjusted hours (‘efficiency units’). Our central finding is that in the post-war US, annual productivity growth during 1947-1967 was 0.9 percentage points higher than during 1987-2007. If the 1947 sectoral composition had remained unchanged while the sectoral productivity growth rates had been the observed ones, then the difference between the growth rates in the two periods would have been 0.6 instead of 0.9 percentage points. This means that Baumol’s disease caused a decline of 0.3 percentage points in productivity growth, which is a third of the total decline. In other words, Baumol’s disease has had a sizeable effect on post-war US productivity growth that is too large to ignore.

Looking at the past importance of Baumol’s disease also reveals that in recent years virtually all of the productivity effects of structural change have come from reallocation to and within the service sector, instead of from reallocation within the goods sector. Moreover, the service sector is rather heterogeneous, containing industries that have very high productivity growth (e.g. air transportation) as well as industries that have very low productivity growth (e.g. food and beverage services). These two observations imply that the usual splits in the structural change literature, which treat services as one broad sector and abstract from structural change within the service sector, are not useful in the current context. A natural first step for taking into account structural change within the service sector is to distinguish between service industries with high and low productivity growth, defined as those with productivity growth above and below the average productivity growth in the service sector. It turns out that this way of disaggregating services captures the vast majority of the past productivity effects resulting from structural change, suggesting that it is a promising abstraction for thinking about the future of Baumol’s disease.

The second step in the analysis is to quantify the effects of Baumol’s disease on future productivity growth. Doing this requires a model of structural change in which one can project different trends of sectoral productivity into the future. The model features three sectors that produce goods, services with high productivity growth, and services with low productivity growth. Sectoral value added is produced from labour under constant returns and there is exogenous growth of sectoral productivity and total efficiency units. Since our model does not have capital, the demand side governs how production is reallocated in response to changes in the exogenous variables. It is therefore crucial that the demand side captures the empirically observed effects of changes in the relative prices and of income on the composition of GDP. The specification of Comin et al. (2015) is a natural candidate for achieving this, because it is non-homothetic even in the long run, and so allows for the long-run income effects that are in the data.

We calibrate the model to capture salient features of structural change and sectoral productivity growth in the post-war US economy. Capturing these features requires that there is little substitutability between the value added of the broad goods and services sectors, and that goods are necessities while services are luxuries. As the relative price of services increases together with income over time, these standard features of the demand side imply that resources are reallocated from the broad goods sector to the broad services sector. Matching the targets also requires that the value added from the two service subsectors is substitutable, that the services with high productivity growth are necessities, and that the services with low productivity growth are luxuries. As the relative price of services with low productivity growth increases together with income over time, these features of the demand side imply that resources may be either reallocated to the service sector with low or high productivity growth. Since either possibility may occur for reasonable parameter constellations, it is a quantitative question which of them is empirically relevant.

Future structural change will not reduce productivity much further

Assuming that the recent growth of sectoral labour productivity and total efficiency units of labour continues into the future, we simulate the calibrated model forward. This yields the somewhat surprising result that Baumol’s disease will not be very damaging in the future. Specifically, over the next 60 years, the reduction in productivity growth that results from Baumol’s disease will be less than half of what it has been in the post-war period. The reason for this is that the service sector with low productivity growth does not take over the entire economy. The substitutability between the two service subsectors turns out to be sufficiently strong so that households increasingly substitute away from them when the relative price of services with low productivity growth increases without bound. As a result, the economy reallocates less and less production to the service industries that have little or no productivity growth. This implication of the analysis is robust to changing the details of the forward simulations, and it differs markedly from the canonical model of structural transformation which abstracts from substitutability within the service sector.

In sum, our analysis implies that while past productivity reductions from structural change are here to stay, future structural change will not reduce productivity much further. The crucial reason for this is that there is substitutability within the broad service sector. As a result, people will substitute away from the value added of the stagnant service industries as they get more expensive. The outlook on future productivity growth is therefore considerably more optimistic than Baumol conjectured.

References

Baumol, W J (1967), “Macroeconomics of unbalanced growth: The anatomy of the urban crisis”, American Economic Review 57: 415-426.

Baumol, W J, S A Batey Blackman and E N Wolff (1985), “Unbalanced growth revisited: Asymptotic stagnancy and new evidence”, American Economic Review 75: 806-817.

Comin, D, M Mestieri and D Lashkari (2015), “Structural transformations with long–run income and price effects”, NBER Working Paper 21595.

Duernecker, G, B Herrendorf and A Valentinyi (2017), “Structural change within the service sector and the future of Baumol’s disease”, CEPR Discussion Paper 12467.

Fernald, J G (2016), “Reassessing longer-run US growth: How low?”, Federal Reserve Bank of San Francisco working paper 2016–18.

Fernald, J G and C I Jones (2014), “The future of US growth”, American Economic Review: Papers and Proceedings 104: 44-49.

Gordon, R (2016), The rise and fall of American growth: The US standard of living since the Civil War, Princeton University Press.

Nordhaus, W D (2008), “Baumol’s disease: A macroeconomic perspective”, B.E. Journal of Macroeconomics (Contributions) 8(1).

Endnotes

[1] For subsequent work along the same lines, see Baumol et al. (1985) and Nordhaus (2008).

4,094 Reads