Robert Gordon’s new NBER paper, ‘Is US Economic Growth Over?’ has been making waves in the past few weeks. Here are a few thoughts on it.
A short version of Gordon’s argument is here. His provocative topline is:
The paper … suggests not just that economic growth was a one-time thing centred on 1750-2050, but also that because there was no growth before 1750, there might conceivably be no growth after 2050 or 2100. The process of innovation may be battering its head against the wall of diminishing returns. Indeed, this is already evident in much of the innovation sector …
Much of this is familiar from Tyler Cowen’s book ‘The Great Stagnation’, which kicked up a similar dustcloud in 2011. Gordon goes on to identify six ‘headwinds’ that may push back innovation in the States. These are: an ageing population, low skills and poor state education systems, inequality, globalisation and outsourcing, environmental constraints and government/consumer debt overhangs.
*
Gordon is very clear that the analysis applies only to the US, but is also keen for researchers in other countries to pitch in. So, in no particular order:
1/ Many of the six ‘headwinds’ are much stronger in the US than elsewhere (a weak public education system in particular). As Gordon says, ‘my guess is that a Canadian or Swedish economist looking at the past and future of his or her country would not be nearly so alarmed’. (p23)
2/ Equally, some of these headwinds might reverse direction. For example, climate change may create new economic opportunities, not just constraints.
3/ This is very long wave analysis, and frankly it’s too early to tell if Internet-based innovation – Gordon’s ‘third industrial revolution’ – has really come to an end. That data only runs from 1960-2007; but previous revolutions took at least 100 years to fully diffuse. It feels premature to say that ‘the productivity impact of IR3 evaporated after only 8 years.’ (p13)
4/ There’s also something slightly odd about the historical treatment – which is done in terms of the country at the technology frontier. Gordon looks at the UK from 1300-1906, then the US til the present day. But it’s not obvious the underlying country drivers of growth are the same. And as Gordon acknowledges (p6), the future frontier country might not be the US.
5/ The analysis overlooks the dynamic nature of some innovations. Gordon suggests that urbanisation ‘only happens once’ – true, but its effects are persistent. Agglomeration economies can trigger virtuous cycles of urban development, raising growth over very long periods.
6/ Similarly, innovations like ‘travel speed’ haven’t changed much (p2) but their diffusion has been massive – e.g. far more people have access to plane travel than in 1958. This must have some impact on economic welfare through market size, if not on productivity.
7/ Gordon is perhaps a bit unfair on internet innovations. As he rightly points out, for most people running water and central heating are more ‘important’ than broadband. But he doesn’t really consider the internet as a general purpose technology with multiple affordances – many of which we’ve only just started to grasp. Equally, he doesn’t really consider big data, mobile, cloud or social technologies.
8/ The paper doesn’t really explore reasons why ‘IR3’ technologies haven’t fed through into labour productivity stats. There’s now a massive organisational literature which provides some answers. Work by Erik Brynjolfsson tells us that to make the most of ICTs, firms need to do substantial complementary investment in management and organisational structures. Research by John Van Reenen, Nick Bloom and others also highlights the importance of good management in triggering productivity gains for firms.
Some sectors have understood this better than others – such as ICT, retail and financial services, which have seen substantial technology-related productivity jumps.
In other words, just ‘putting a computer on the desk’ isn’t going to have much effect. And looking at average productivity changes hides some big sectoral differences. But this is how Gordon’s paper is thinking about it. A closer, finer-grained analysis might turn up some different answers.