Photo by Agto Nugroho on Unsplash https://bit.ly/2Uea8Gm

The use of both industrial and service robots has been expanding rapidly during the past decade. There is little doubt that this is leading to changes in income inequality. Automation replaces tasks previously performed by low-skill labour, though it also creates new tasks, usually more highly skilled. Automation explains, in part, the poor labour market performance of low-skilled workers in advanced economies since 2000.

While history presents many technical innovations that have had this effect temporarily, artificial intelligence (AI) offers machine learning and adaptability sufficient to represent a truly extraordinary technical leap. This suggests the possibility that, at least under the extreme “singularity” scenario, the majority of current human tasks could be replaced. In the near term, we can be confident that the use of capital, which includes intelligent machines and software, will continue replacing low-skilled labour in production. Indeed, in the advanced economies, low-skilled labour shares of total income have fallen in recent decades, while the shares of skilled labour and capital have risen and measures of inequality, such as Gini coefficients, have continued to rise.

National policy questions

Automation is a natural response by firms facing competitive pressure. It offers an exciting sense of modernity to their workers and to customers and it increases flexibility tailored to customer demand. Yet, by raising incomes to capital and reducing them to the low-skilled, it worsens inequality and raises the cost of transfers to displaced or “underpaid” workers.

Some governments have invested heavily in research and development (R&D) and the production of robots and artificial intelligence, while others have left it to their firms to lease or acquire the fruits of this effort from abroad. A first policy question, then, is as to whether there is any economic first-mover advantage from leadership in automation. Given the distributional consequences, this will depend on whether the welfare criterion chosen is (1) Rawlsian, emphasising the performance of low-skilled households, (2) Benthamite, here interpreted as the aggregation of pecuniary advantage, or whether it simply maximises (3) capital or (4) gross domestic product (GDP).

A second policy question is as to how governments should offset negative inequality effects. Potential policies to mitigate declining low-skilled working conditions include higher wages through collective bargaining or mandated minimum wages, a wider spread of capital ownership and the redistribution of income via fiscal policies. Wage bargaining and minimum wages have in the past been linked to unemployment and could therefore exacerbate the effects of automation on the overall performance of low-skilled households. Moreover, the rise of the “gig economy” – that is, the software assisted separation of tasks – has made labour organisation more difficult. And the experience of Thatcherite Britain has discredited efforts to spread ownership of capital. This leaves fiscal intervention (taxes and transfers).

Importantly, automation and its policy responses are taking place in a global economy that is integrated via financial flows and trade. Both the adoption of new technology and the fiscal measures that redistribute its effects generate spill-overs to other nations. These stem from the successful adopter’s increased capital income, increased saving, lower real interest rates and increased investment spending. When the adopting country is large, this reduces interest rates internationally, appreciating its currency during the investment surge and depreciating it later as labour and capital costs fall.

Spill-overs also occur due to fiscal responses. These can either raise sovereign debt or raise tax rates applying to consumption expenditure or income to labour or capital. This choice determines the scale of sovereign debt, rates of interest, low-skill wage costs and investment. Particularly for large economies, these changes also influence global financial tightness, comparative productivity and exchange rates.

A global macroeconomic model

We employ a global macroeconomic model that identifies six regions, the United States, the European Union, Japan, China, Australia and the Rest of the World. In each region, there are three households: low-skilled, skilled and capital-owning. Regional production depends on low-skilled and skilled labour, capital and tradable intermediate goods. In the simulations discussed here, we take a long run perspective. This implies that there are clearing markets, retention of fiscal balance at 2016 levels and capacity adjustments that keep constant expected net rates of return on capital.

The way we characterise the implementation of automation is to extrapolate to the year 2036 the falling low-skilled labour share based on the experience of the United States in the past two decades. Skilled workers will gain some tasks but lose others and so their share is held constant. The residual income share goes to the owners of machines, which increasingly embody artificial intelligence.

To represent automation drives, the demands of the available technology in the model are altered in these ways, and we observe the consequences in the absence of new fiscal responses. To identify strategic interactions, we do this in the United States, the European Union and China, collectively and one region at a time.

For regions that implement automation, the low-skill wage falls, capital returns rise and inequality rises as expected. Negative spill-overs to other economies are small, in part because exchange rate adjustments improve the purchasing power of incomes abroad.

Increased, if concentrated, wealth spills over to other regions, mainly because of the declines in long bond yields that occur in response to higher saving in the implementing region. With lower real long bond yields in integrated global financial markets, investment rises everywhere, though to a dominant extent in the implementing region.

When all regions implement the technology, the inequality increases are pervasive. The widespread introduction of the technology raises global capital income and saving rates by substantially more, pushing real bond yields further down and asset prices further up. Notwithstanding the rise in inequality, this has the positive consequence of greater capital growth globally.

Under a Rawlsian criterion that values only the welfare of the poorest, the dominant strategy is therefore to restrict automation. The low-skilled are losers in all implementing regions. If the criterion were Benthamite, the dominant strategy is to implement automation. The same is true if the criterion is to maximise real GDP.

Welfare effects of fiscal responses

Our analysis focuses on governments avoiding greater inequality through the “earned income tax credit”, which leaves low-skill wages unconstrained but compensates low-skilled households through the tax system. Earlier research has shown that this intervention is superior to the alternative of a “universal basic income” since it sustains employment. Implementing regions offer transfers that are sufficient to hold their inequality levels (Gini coefficients) constant. These transfers are financed by increasing either the consumption tax rate or the rate of capital income tax.

  1. Consumption tax financing

In this case, the burden of new revenue is shared across households according to their consumption expenditures. The results suggest that only under a Rawlsian criterion would the United States consider this fiscal intervention. The preferred option of the United States would be to press China to implement the policy, because global investment would then be redirected away from China and toward the United States.

The perspective of the European Union is more positive. Though its capital owners would prefer not to implement, its low-skilled and skilled workers would be net beneficiaries and its GDP would expand slightly. The European Union would also gain from implementation by China, for the same reasons that the United States would. By contrast, while a Rawlsian strategy in China would see it implemented, it would reduce the real purchasing power of skilled and capital owning households and slow GDP growth, and so would be a political non-starter.

  1. Capital income tax financing

Financing transfers by increases in the tax rates on capital income is the equivalent of Bill Gates’ taxation of robots. Under a Rawlsian criterion all regions would implement the policy as before, though by small margins the low-skilled in each region would prefer unilateral rather than collective implementation. By the other criteria the United States would not implement it, the European Union might on Benthamite grounds and the Chinese would not. Stabilising inequality by this means would require capital income tax rates to rise by about 15 percentage points for the three regions. This would be politically difficult, though more affordable in the case of China than the consumption tax option.

If, by some means, the three large economies were forced to implement transfers to stabilise their inequality levels, a Benthamite criterion would have them preferring to finance it by capital income taxation, while a real GDP criterion would see governments preferring the consumption tax.

Overall, the dominant strategy in most cases is for governments to refrain from such fiscal interventions. These results therefore suggest pessimism about the feasibility of fiscal interventions to stabilise inequality in the face of accelerated automation. Only a lurch to policy criteria of the Rawlsian type would see this happen.

 

This blog piece is based on our working paper “Automation, taxes and transfers with international rivalry”, CAMA Working Paper 44/2018.

Like or share this piece:

Leave a comment

Your email address will not be published. Required fields are marked *

*