Automation: If the machines aren't coming for your jobs, are they coming for your investment returns?

Despite alarmist media reports, the risk to most jobs from automation is low for the foreseeable future — we tend to an optimistic view of mankind’s ability to respond to technological disruption. Nonetheless, certain segments of mid-skill workers may be vulnerable and politicians will need to address the impact on standards of living.

Ed Smith, Asset allocation strategist

Risk of disruption = medium

Technophobia is a recurrent theme in the political history of economic progress. Pericles spent huge sums on major public works to occupy those feared unemployable by new technology; Elizabeth I denied a patent to a knitting machine, convinced it would turn her subjects into beggars; and in 1964 a group of Nobel prize-winning economists organised to alert President Johnson to the dangers of “the automated self-regulating machine”. Time and again, predictions of widespread technological unemployment were wrong. Is the latest wave of technology — robotics and artificial intelligence — any different?

A trip to London’s Science Museum would provide some welcome relief to the technophobic. Its curators have assembled a collection of some of the world’s most advanced robots, but on the second weekend of the exhibition most appeared to have fallen into a state of technological catatonia. Baxter, a robot adept at sorting and packing, is even programmed to show emotions to help him interact with human colleagues. When we first met he was confused. When I checked on him 20 minutes later... still confused. 

Yet in a survey of 1,896 experts from a variety of disciplines, 48% expected technology to displace more jobs than it creates by 2025 (Pew, 2014). Most of the doomsayers are techies, while most of the optimists are economists and historians. One can’t help but see some hubris in the techies’ forecasts, and we fall into the more optimistic camp, at least as far as the next 10 to 20 years are concerned. Mass unemployment is unlikely, but optimists and pessimists agree on one thing: that the change is going to be more rapid and the impact on the economy more rattling than we’ve seen before. 

We start by setting out the debate around the impact on jobs, but we want to move the discussion on. We end by asking how the machines might impact real interest rates, which set the benchmark for all investment returns. The discourse may seem esoteric, but it is pertinent to all of our clients.

Always consider the net effect

The optimists’ argument has it that increasingly cheap robots will displace labour, but with an important corollary: the goods and services produced by the robots will become cheaper, thus stimulating demand. Technological progress also leads to product innovation, thus creating entirely new sectors. The labour displaced by the robots then switches roles to meet this new demand. As ever, we shouldn’t consider whether individual jobs are being replaced by machines without considering the net effect of all changes in employment that result over time. 

The counterargument is usually structured along three broad lines of attack: (i) that we have invented all of the products we are ever likely to consume; (ii) that robots will soon be able to perform the majority of tasks that humans do, so there will be no roles into which labour can switch; or (iii) that robots will cause such a dramatic redistribution of wealth and income in favour of their owners and architects that the total demand for goods and services will fall. 

Use your imagination

The first counterargument is easy to dismiss. Theorists have always missed the capacity for human ingenuity to create entirely new ways to spend money. No less a mind than John Maynard Keynes wrongly predicted that we would be working 15-hour weeks by now because he assumed that the gains to our productivity would accrue more to free time than consumption. In short, discussions of how technology may affect the demand for labour too often focus on existing jobs and neglect the emergence of the yet-to-be-imagined jobs of the future.

The machines are here to help…

The second counterargument — that robots will become so advanced that they will take all of today’s jobs and tomorrow’s (both the cognitive and the routine) — is not so farfetched, but is unlikely to be a concern for most of our lifetimes. And, as we shall see, ignores some fundamental principles of economics. 

This fear has been fuelled by a frequently cited (and miscited) study conducted by two Oxford professors, Frey and Osborne (figure 3), which estimated the susceptibility of employment to automation. It concluded that 47% of US jobs are at risk. However, the citations invariably take this statistic out of context. A job could register as susceptible to automation within the next two decades, even if the probability of that potential being realised was very small. Indeed, Frey qualified his position in a later paper: “Although we cannot exclude the possibility that technology may reduce the overall demand for jobs in the future, this is seemingly not an immediate concern… [and] concerns over automation causing mass unemployment seem exaggerated, at least for now.” (Frey & Rahbari, 2016) 

They also ignore that jobs invariably involve a multitude of tasks. If a machine could perform every task performed by a certain profession, then automation necessarily reduces employment in that profession. But if a profession is only partially automated, employment could well increase (so long as demand for their combined output increases as prices fall). Today in the US there are more cashpoints than human tellers, but there are still twice as many tellers today as there were in the 1970s when cashpoints were first introduced (Bessen, 2015). Cashpoints drove down branch costs and improved the productivity of tellers after they were relieved of the mundane task of counting out money. This allowed banks to open more branches. 

Automation is usually aimed at specific tasks rather than whole occupations. The evidence so far suggests that most of the adjustment to automation to date has occurred through changing task structures within occupations, rather than through workers being forced to switch occupations (Spitz-Oener, 2006). Economists at the Organization for Economic Cooperation and Development (OECD) used detailed analysis of the tasks associated with our jobs to re-estimate the risk of occupational automation projected by Frey and Osborne. On the assumption that a job could disappear entirely if more than 70% of its associated tasks have the potential to be automated, the study found that only 9% of US and 11% of UK jobs are at risk (Arntz et al, 2016). The jobs most at risk are not just those currently requiring skilled hands, but also those related to exchanging information and selling products.

We are still a long way from Artificial General Intelligence (AGI) — machines able to turn their hand to pretty much any task without being programmed specifically to do so. Computers have long since been able to defeat the best human chess players, but they still can’t look at a five-year-old’s picture book and tell us what’s going on. We may be less than 10 years from the day when $1,000 buys us the equivalent processing power of the human brain (Rachel & Smith, 2015), but we have only just been able to emulate the neurological complexity of a flatworm (TechRadar, 2015). Humans are likely to have long-lasting comparative advantages when it comes to reacting to unexpectedly complex situations, spontaneous failures and tasks employing social and emotional intelligence quotients. Indeed, occupations that are relatively social-skill intensive have accounted for nearly all new jobs in the US since 1980. The wages of those employed in jobs requiring limited social skills have fared poorly, even if that job requires high maths skills (Deming, 2015). 


Jobs at risk from automation: Jobs tend to be an amalgamation of tasks, most of which cannot be automated

For sure, machines will cause more and more disruption, but are likely to be more our complement than our substitute. Even if we are on the brink of a technological onslaught, major disruptors still require complementary investment. People need to come up with ways to integrate them. MIT professors Erik Brynjolfsson and Andrew McAfee in their book The Second Machine Age show how this has been the case with most disruptive technologies since steam power. In other research, Brynjolfsson has shown that for each dollar of capital invested in computers, firms tend to make $10 of complementary investments in ‘organisation capital’ (such as business processes, repurposing roles and tasks, and training). That’s a big multiplier.

What’s more, new technologies often have to wait for another technology to come along before they are truly game-changing. The computer did not revolutionise consumer markets until the internet brought new ways to work, shop and play into every home. One new technology begets another; one new industry begets a second. Evidence of these second-round effects was another reason why Frey issued the qualification of his original work with Osborne. 

“Labour is not dead wood to be carved up between tasks. It is a tree whose trunk and branches have lengthened and thickened with time,” says Andy Haldane, with a poetic flourish unusual for a Bank of England Chief Economist (Haldane, 2015).

The substitution of automatable jobs will also depend on the wage level, of course. If intelligent machines do leave a lot of people looking for a job, wages will fall relative to the price of machines (though not necessarily in absolute terms as long as new roles and new products are created), thereby improving employment prospects again. The market clears. Furthermore, economic expansion occurs where inputs are in abundance. If labour becomes cheaper relative to other factors, that influences the range of tasks allocated to it. It also influences the direction of technological change itself by incentivising firms to introduce technologies that allow firms to harness labour and its comparative advantage in certain tasks more intensively (Acemoglu & Restrepo, 2016). 

…but that help won’t come for free

But how great an impact could machines have on wages? Could they skew the distribution of income to such an extent that it would result in a decline in the aggregate demand for goods and services? Remember that the highest earners have the highest propensity to save extra income. Businesses also have a high propensity to save. Therefore if machines compete more with low- and mid-earners, pushing down their wages, and more and more income accrues to the highest earners and firms, less of the overall pot will be spent or invested in the economy. 

We know that the only way for societies to become wealthier — to improve the standard of living and raise real wages — is to keep getting more output relative to the number of inputs (Krugman, 1994). This is called productivity growth and it happens through innovation and technology. But, in theory, technological progress could be so biased towards the newly intelligent machines, their owners and the few high-skilled people they need to work with that average real wages decline, even if comparative advantage keeps people in a job. 

An analogy might help. Imagine everyone is a fisherman, bar one capitalist who supplies nets. Everyone is employed and they fish in a small lake. The wages of fishermen would be determined by how many extra fish an additional fisherman could pull out of the lake for a given number of nets. Now imagine the capitalist supplies better nets (technological change). This would result in more fish per fisherman, and would improve the extra catch one new fisherman could provide, so their wages would rise. 

But now imagine not a better net, but something really game-changing — a fantastical fish magnet, perhaps. This magnet would dramatically increase the catch to the point where the fishermen could catch a surfeit of fish without requiring any new labour. This would change the impact of adding new fishermen. Put another way, the gains from the fish magnet are much larger at lower levels of employment, but the incremental benefit of adding new fishermen is now diminished and real wages wouldn’t increase (and could move lower). The fish magnet is a substitute for much of the fisherman’s role, while the better net is complementary. This technological change is what we call ‘capital-biased’. The share of total income will accrue more to the magnet man than the fishermen. 

In practice, the evidence does suggest rising real wages, albeit with more inequality. Or, more crudely, the poor gain, but not as much as the rich. How much inequality rises from here depends on how substitutable machines become for labour. As we have discussed, it seems we are decades away from perfect substitutability in most professions. If machines are not perfect substitutes, humans’ unique talents become increasingly valuable and productive as they combine with the accumulating, newly intelligent machines. This increase in labour productivity outweighs the fact that the machines are replacing some humans, and wages rise with output. But as we have seen, harnessing the true productive potential can take time.

Using a theoretical model, economists at the IMF suggest a baseline of 20 years for the productivity effect to outweigh the substitution effect and drive up wages. This lag corresponds with the evidence from the last 200 years of disruptive innovation, as figure 4 shows (Haldane, 2015).

Historic and contemporary evidence suggests that mid-skill workers are most affected by the initial disruption. Today’s mid-skill jobs are the ones that haven’t come back after each of the last three recessions (Siu & Jaimovich, 2012). While over time, new mid-skill jobs will emerge, they may come too late for the displaced individuals — state-provided retraining is very unsuccessful. These workers tend to add to the supply of low-skill workers, placing downward pressure on wages as income inequality rises. 

Complex economic simulations, depending on their parameters, suggest a wide variety of outcomes in the race between man and machine, but they almost universally conclude that the average worker will likely end up taking home a decreasing share of total income (cf. Benzell et al, 2014) as humans become more reliant on machines for the continued improvement of their productivity. 

Figure 4: Real wages and employment
Technology has not proved a problem over the past 200 years in the UK.

Automation versus ageing

It is imperative, however, to join the automation debate with another great conundrum of our era: ageing. Ageing workforces are close to transitioning to shrinking workforces. Machines may take some jobs, but there will be fewer workers to apply for them anyway. Furthermore, as we age, we tend to consume more services than goods (think healthcare). Services tend to be more labour intensive, and it is in the provision of services — particularly those requiring human interaction — where machines are likely to be least substitutable for humans. This should further alleviate concerns that machines will decimate workers’ pay.

Will the robots eat my interest rate?

Ageing is frequently blamed for lower rates of growth and lower rates of returns. Certainly it is partly culpable, but there are a number of other factors and we now ask how intelligent machines could alter the outlook.

Contrary to popular belief, real interest rates are not driven by central bankers, but by the ‘neutral real rate’. The neutral real rate is often described as the ‘Goldilocks’ rate of interest; the rate consistent with stable economic growth that does not cause inflation to overheat. It is linked to two factors. 

First, it is linked to the potential growth rate of the economy (potential, meaning an economy operating without any slack caused by a downturn in the business cycle). If you think of growth as a proxy for the return on investment, lower returns lead to less demand for capital and that pushes down on real rates. Years of ultra-low central bank policy rates have not caused the economy to overheat because trend growth and the neutral rate have fallen, and so policy is not nearly as expansionary as many pundits make it out to be. Central bankers are slaves to the neutral rate. 

It is also linked to the desire to save or invest. At a global level, savings and investment must be equal — all investment must be funded. If you’re wondering how debt works here, a loan is accounted for simultaneously as an investment the lender makes in the borrower’s debt and as money (savings) deposited in the borrower’s bank account. The neutral rate of interest is what balances the two. Desired saving will tend to rise as real rates increase, because higher rates generate higher returns on saving; desired investment will tend to fall as real rates increase because the real rate is a key component of the cost of capital, so as real rates rise it becomes more costly to invest. 

The natural rate of interest has declined by approximately 4.5 percentage points since the 1980s. Studies attribute about a third of the decline in the neutral rate over the past few decades to lower potential growth, and most of the remainder to changes in the preference for savings and investments (Rachel & Smith, 2015; Laubach & Williams, 2003). Below we set out the key drivers behind these moves and ask how intelligent machines may influence them. We summarise the impact on real interest rates in figure 5.

Figure 5: Real wages and employment
Factors driving down neutral rates over the past 30 years.

Potential growth

Potential growth is a function of more workers, more invested capital and higher productivity. Machines can’t make more workers, and in the short term could lead to fewer eligible workers participating in the job market if older workers find it more difficult to acquire the skills needed to work with new machines (Fujita and Fujiwara, 2014). We discussed how new technology often begets more investment, but offsetting that are the observed trends of corporate capital hoarding, constrained banking and the falling cost of machines. If the new technology is truly disruptive it will likely require the scrapping of old capital and a new wave of investment. Productivity growth has stalled, but remember that there is often a lag after the initial technological change. 

In short, if we are on the cusp of a new era of intelligent machines, we would be likely to see an improvement in growth as a result of investment and productivity — and therefore slightly higher real rates — but there are still other structural forces blowing in the opposite direction.

Ageing and saving

An eight percentage point increase in the number of people of working age relative to the total population has increased desired savings to such an extent that it may have lowered real rates by almost
1 percentage point (Rachel & Smith, 2015). People in the second half of their working lives save the most, while older people tend to draw on their savings. On current trends, this will reverse from about 2020. If automation causes older workers to drop out of the workforce rather than retrain, it could lead to a faster drop in savings. On the other hand, a technology-driven healthcare revolution (as described in the article on personalised medicine on pages 7—10) could increase life expectancy and this could cause retirees to consume less each year from their savings and people in work to save even more.

Inequality

Firms and the wealthy have a higher propensity to save. So if they take a larger slice of the pie, desired savings increase. One of the few things that everyone seems to agree on in the man versus machine debate is that income and wealth inequality will increase, especially in the initial stages of disruption when skills and retraining opportunities are deficient.

Emerging market savings glut

Rising commodity prices and decisions to increase reserves after the Asian crisis saw the desired savings of emerging markets (EM) rise more than investment. If machines are less complements and more substitutes for EM workers, then machines could cause EM household savings to decline. EM governments may also increase investment in education and infrastructure to boost their comparative advantages in order to offset some of the impact of machines. Moreover, if EMs are equally as good at coding and managing the machines, then firms may choose to increase investment there too. That said, if automation in manufacturing leads to Western firms withdrawing their production, capital could flow out of EM economies. The effect of automation on EM savings is uncertain.

Falling capital goods prices

The prices of investment goods (such as computers) have fallen. When this happens, a given project requires less investment expenditure, but more projects can be undertaken. The net effect across the economy depends on substitutability again — whether investment displaces consumers so requiring fewer projects to meet future demand. As we have discussed the next generation of intelligent machines are not likely to be perfect substitutes for most occupations, but the gains are likely to be somewhat biased towards capital owners. New opportunities should alleviate some of the downward pressure falling investment prices place on rates. 

Falling public investment

The global government investment to GDP ratio has declined by about one percentage point, for reasons more to do with politics than economics. If technology causes higher inequality, it may necessitate greater public investment in order to compensate the working classes for the increase in inequality. On the other hand, capital is taxed at a lower rate than labour and if automation leads to a rising capital share, tax receipts will be lower making infrastructure spending difficult, particularly at a time of high debt. 

Higher net returns

Investment decisions are based on the difference between return on capital and cost of capital. The IMF estimates that the global return on capital has risen relative to rates, meaning that fewer investments need to be made in order to deliver required returns to investors, thereby pushing down on real interest rates. In so far as machines are productivity enhancing, the return on capital should rise. Whether or not it rises faster than the cost of capital depends upon all of the factors above. 

So on balance, machines are likely to exert some upward pressure on real interest rates. This is good news for holders of cash. Higher real interest rates would hurt bond holders as yields normalise in accordance with a higher neutral real rate. But as yields rise it increases the ability for bonds to satisfy the required returns of new investors, thereby decreasing the incentive to move into credit or equities.

Conclusion

The relationship between jobs and technology is far more nuanced than many commentators appreciate. Hopefully, this will provide some reassurance to clients who might have feared that their children or grandchildren face a life of robot-induced penury. Given my interaction with Baxter at the Science Museum, any such dystopia is decades away; but we still take a more optimistic view of mankind’s ingenuity and believe that technology will unleash as many new employment opportunities as threats.

Given many workers may not have the right skills to work alongside new technologies — and that retraining programmes are poor (Card et al, 2010) — machines are likely to disrupt many working lives. History suggests this trend is likely to affect particular groups of mid-skill workers, possibly concentrated in particular regions. As a result, greater income inequality and a decline in the average worker’s share of total income could exacerbate social and political tensions. 

If the initial analysis is right about last year’s EU independence referendum decision and the election of President Trump being the product of working- and middle-class anger about declining standards of living, technology could create further political populism. Like globalisation, it is a perfect faceless enemy for demagogues, appearing to line the pockets of the global elite. The challenge for politicians is to find better ways to ameliorate the impact of change on specific groups and communities and to articulate better the wider social and economic benefits of technological progress. Investors should take note: our research last year uncovered a new relationship between policy uncertainty and investment valuations, unprecedented for at least 25 years.

On the other hand investors concerned about secular stagnation or the impotence of monetary policy to deal with the next recession when interest rates are near zero should welcome the age of automation, for it appears that it is likely to exert a net upward pressure on the structural rate of interest appropriate for the economy. As ever, we must consider the net effects.

View the full disruptive technology report here.

Visit our disruptive technology hub.

Rate this page:
Average: 5 (1 vote)