Andy Lees from MacroStrategy Partnership LLP encouraged me to change my thinking. As he said, to get a more precise picture, you must look at total factor productivity (TFP) – not just labour productivity as is implied in (i) above. That said, economic models measure TFP differently. The standard approach is to define TFP as the portion of real output (GDP) growth that cannot be explained by growth in either labour or capital with labour assigned a weight of 70% and capital 30%.

Having said that, those weights differ, and Andy thinks land should also be expressed as an endogenous variable in (ii) above, and he is not alone in thinking like that – more on that below. In the following, I have opted to keep it relatively simple, though. By ignoring land and other variables, i.e. by focusing on the interaction between labour, capital, TFP and output, everything is quantifiable, which makes the whole exercise easier. Otherwise things would get rather hairy very quickly.

A few assumptions

This is not a cyclical call. I learned many years ago that trying to predict economic cycles is a loser’s game. Even if you get it spot on, financial markets do not always react as you would expect. The current bull market in equities despite economic circumstances being extremely difficult is evidence of that.

There are indeed those who, with some right, can claim that they predicted the Global Financial Crisis in 2007-09 but, if you start digging a little deeper, you often find that they predicted ten of the last five recessions. Perma-bears are, almost by definition, right every now and then, but they conveniently ignore all those predictions that turned out to be incorrect. In the following, I will completely disregard the short-term cyclical outlook.

Rather, my call for GDP growth to disappoint in the years to come is structural in nature. Low economic growth is actually pretty much set in stone, so I am not overly worried about getting it wrong. It is quite simply a continuation of an already existing trend (Exhibit 1) which you can do little about. Over the past several years, I have written extensively about all the reasons why GDP growth is in structural decline, and I do not intend to repeat myself. Therefore, in the following, although I may bring up one or two of those reasons again, I will not argue why that is. I will simply assume you are familiar with my prior work.

That doesn’t imply that economic cycles do not matter anymore. They certainly do, but the high and low points of those cycles will just be lower. If an average OECD economy firing on all cylinders used to deliver annual GDP growth of 3-4%, going forward, the high point may only be 1-2%, depending on which economy we are talking about.

The conundrum

If you think about it, it is actually a bit of a conundrum why productivity growth is as pedestrian as it is, considering we are in the midst of a digital revolution – more on that below. For now, suffice to say that this problem must be addressed. Otherwise economic growth will continue to disappoint for many years to come.

We have at least two, possibly three new technologies in the pipeline that, more than anything else, could address this problem. The first, the mixture of AI and advanced robotics, could become mainstream relatively soon, and that could address the brewing problem of the workforce retiring in large numbers. The other one I feel certain about, at least longer term, is the eventual commercial rollout of fusion energy, which will have a massive impact on productivity growth, as the marginal cost of electricity will be reduced to virtually zero. Unfortunately, we are still (at least) 10-15 years away from that happening, though.

The third innovation that is likely to change the world as we know it, is graphene – a new technology which I have only zoomed in on recently, so it is still very much work in progress as far as I am concerned. Graphene is a bi-product of graphite. It is widely perceived ‘only’ to be a new battery technology which will improve capacity (battery life) and shorten charging times, but graphene can be used for many other things. For example, in the healthcare industry, graphene can be used in connection with various cancer treatment forms, and it can be used to monitor diabetes patients better.

Assuming all other things being equal, more robots will lead to higher labour productivity (as total output per hour worked will rise) and thus to higher GDP growth as per (i) above. The problem is that all other things are not equal, which is why I introduced (ii). Let me explain. The problem in a nutshell is that access to capital – which impacts the supply of robots – is a function of how much capital is available, i.e. it is a function of past and present productivity.

When productivity drops, more capital is consumed than is created. i.e. `Delta`Capital (`Delta`K in (ii) above) turns negative. Governments are prime sinners, but they are admittedly not the only sinners. Every time a new policy is rolled out, if it doesn’t create the return to pay for itself, effectively, capital is destroyed.

In the private sector, as more and more people retire, and as retirees consume down capital (as they do), less and less capital will be available to invest in robots, i.e. the ‘tax’ on productivity will get bigger and bigger as the workforce ages. Likewise, converting to renewable energy forms, which cost up to five times more than the price of conventional energy, is also a ‘tax’ on productivity. Adding to that, capital owners have not re-invested a reasonable proportion of corporate profits in recent years. They have instead chosen to allocate the profits unproductively, thereby consuming down capital.

Whilst common to blame capitalism (Ray Dalio and Niall Ferguson have both written extensively about this topic and have characterised it as a capitalist system in crisis), inequality can only grow in a managed system. In a well-functioning capitalist system, capital and labour share national income in a way that doesn’t change much from year to year, i.e. the gap between rich and poor is quite stable over time. A reverse-to-the mean mechanism will ensure that.

An interesting question raised by Andy is whether deteriorating demographics is due to a change in social values, or if it has been imposed by the stagnation in productivity, making children and traditional family structures unaffordable? You could apply precisely the same question to robots. As Andy said to me:

The advance in science and technology has stalled precisely because it is not being deployed productively; rather than using the latest advancement in computing to, directly or indirectly, push the scientific boundaries further, we are using it for unproductive purposes, thus destroying capital and slowing science. Whilst cheap storage means the advancements and theories are not lost, scientific advancement has slowed heavily in recent decades. Rather than science offsetting the loss from demographics, it is suffering the same loss of capital from unproductive spending.

Andy made another good point. He suggested I take a look at the inflation-adjusted yield curve which, even in the US, is now very negative. According to the US Treasury (see here), as at the 9th September, the 10-year real yield in the US was -1.00%, implying that US GDP will fall by about 10% in real terms over the next 10 years. Effectively, the US bond market is telling us that advanced robotics and other, new technologies will not offset the decline in GDP growth caused by adverse demographics.

Some further thoughts on TFP

One last point from Andy:

If we consider the economy as consisting of [a mixture of] fixed (labour, plant and equipment, and technology, etc.) and operation capital, then TFP and GDP should be one in the same. Unfortunately, this is not so because our accounting for depreciation has differed from the actual depreciation of such equipment, resulting in a record age of capital stock, including the workforce. This difference between the paper accounts and the hard record is the capital we have consumed, initially through underinvestment, and in more recent years through its actual decline. As with trade where, at a system level, imports and exports must equate, so all input and output are an identity, so every output either creates or destroys capital.

Let’s spend a minute on the model behind (ii) above where only capital and labour is assessed separately. Assuming the input from labour is constant from one year to the next, TFP will only improve if GDP grows faster than the capital stock. In other words, throwing more capital into the work process, for example by replacing humans with robots, won’t necessarily have a positive impact on GDP growth unless the net gain (from more capital but less labour) is positive.

Let’s stay with that example. As robots replace humans, labour productivity will almost certainly rise, but TFP could potentially fall for the reason I just mentioned. At the very least, the two measures of productivity could lead to very different results. In an increasingly digitised economy where capital matters more and labour less, it is therefore important that we increasingly zoom in on TFP when we assess productivity in society. A simple way to measure TFP is as follows:

(iii) TFP = Total Output / Weighted Average of Inputs

In other words, TFP is a measure of total output divided by total input with the (quantifiable) input factors being labour and capital. However, labour and capital are not the only two factors affecting TFP (Exhibit 2).

I have already mentioned land, but factors such as the quality of the infrastructure, environmental and other regulations and the quality of the educational system all matter a great deal, and the list is long. In practice, though, the two input factors used in almost all TFP models are capital and labour. In economic theory, the Cobb-Douglas production function typically used to represent total output is:

(iv) TFP =Y / `K^(α)` * `L^(ß)`

where Y is total output (GDP), K is capital, α is the output elasticity of capital, L is labour and ß is the output elasticity of labour. Now, TFP represents the increase in total output that is not the result of an increase (or a decline) in those two input factors. It is actually more relevant to look at the change over time than to look at absolute numbers. By using simple arithmetic, (iv) can be expressed as follows:

(v) ∆Y = α * `Delta`K + ß * `Delta`L + `Delta`TFP

… which measures the relationship between growth in total output, growth in capital and labour, and growth in TFP. It will be an uphill battle to argue that (v) is less correct than (i), so Andy’s point is well taken.

How the two measures compare

As I have already pointed out, the two measures of productivity do not always go hand in hand and I suspect that, in an increasingly digitised world, that will happen more frequently. Labour productivity first: In the UK, labour productivity, when measured as output per hours worked, has been virtually unchanged since the Global Financial Crisis (Exhibit 3) and is now approx. 15% below the G7 average with only Italy underperforming the UK.

Much lower labour productivity in the UK when compared to most other G7 countries could quite conceivably be the result of very low penetration of advanced robotics in the UK. As you can see from Exhibit 4 below, the UK is far, far behind most other industrialised countries in terms of the number of industrial robots installed per 10,000 employees.

Switching to TFP, neither has that measure of productivity performed particularly well in the UK since the Global Financial Crisis (Exhibit 5), and the reason could be at least partially the same, but another dynamic is worth pointing out. More and more capital is misallocated in the UK, and that has particularly been the case in the years since the Global Financial Crisis. Vast amounts of capital that could, and should, be deployed productively have instead been deployed unproductively.

When the supply of risk-willing capital is finite – and it almost always is – then TFP will be negatively affected by more and more capital being misallocated. The list is long, but allow me to mention just a few sources of misallocated capital in recent years:

1. Massive amounts of property speculation (as far as the UK is concerned, mostly buy-to-let in and around London).
2. A rising percentage of corporate profits being pocketed by capital owners since the Global Financial Crisis rather than being reinvested.
3. The rising cost of servicing the growing populace of elderly. It is both expensive and non-productive, at least in economic terms (or, as Andy pointed out, in a free economy, in the past, the elderly would have been employed by the family or community to look after children etc. but, in today’s welfare state, their funding has been divorced from any work obligations).
4. The rising cost of servicing the growing mountain of debt. Essentially, more and more capital is used to service existing debt rather than being used to enhance productivity.

The UK has been affected by all these four factors. Nobody should therefore be surprised that much capital has been misallocated in this country which could, at least partially, explain the poor performance of TFP in the UK.

Summing it all up

All this becomes relevant when trying to identify those companies and countries looking to benefit from the digital revolution. Going forward, you would want to invest in countries (and in companies) where TFP is rising and where the penetration of advanced robotics is growing rapidly. Regrettably, neither is the case in the UK.

That said, you cannot assume that the implementation of advanced robotics will automatically fix all the problems left behind by a retiring workforce. As I said earlier, only if the net gain from more capital but less labour is positive will `Delta`GDP improve. I often come across even seasoned industry insiders who argue that there is no reason to worry about the poor demographic outlook. The digital revolution will sort that out, they say, but I hope that, with this paper, I have demonstrated that it is not so simple.

Niels C. Jensen with assistance provided by Andy Lees, MacroStrategy Partnership LLP

10 September 2020