AI and govt bond yields – onwards and upwards?

There is some evidence of upward pressure on real interest rates from higher productivity growth in Figure 1, which shows US nominal yields were around 5-6% during the faster productivity growth of the Goldilocks era, from the 1990s to 2007, but although there are parallels with Goldilocks in the post-Covid era, there are also distinct differences (see below).

Longer term studies of productivity generally show the main drivers to be (1) capital per worker, (2) technology gains, and (3) resource re-allocation. These factors drove rapid productivity gains in the period before the Global Financial Crisis (GFC), apart from in Japan, though productivity growth started to slow before the GFC in the G7, particularly in manufacturing, as the benefits from Moore’s Law^[3] and re-shoring faded. But in 2008, the GFC crushed domestic demand, financial intermediation and investment growth, ushering in secular stagnation and weak productivity growth for over a decade to 2020. Real and nominal yields adjusted quickly to zero rates, central bank QE and the weaker demand for capital once the scale of the GFC, and credit crunch became clear, and then remained at, or near historic lows until after Covid, as Figure 1 shows. 

The long period of very weak productivity growth, from 2007 – 2020, and low yields, has been explained by zero rates “zombifying” economies, pushing the cost of capital well below the marginal product of capital^[4], and distorting the allocation of capital, allowing unprofitable companies to survive. Proponents of this view appeal to Austrian and Swedish capital theory and suggest recessions and periodic bouts of "creative destruction” are required in capitalist economies to rid them of unprofitable companies, and boost productivity^[5], even if this risks deep recessions. A clear example of this phenomenon is the Japanese economy with its low productivity and GDP growth alongside zero rates since the early 1990s,^[6] and there is evidence of long productivity “tails” elsewhere in the G7 since the GFC. Figure 2 shows how US manufacturing productivity broadly flat-lined in the period since the Fed adopted zero rates in 2008, even if this does not establish causation. There are, however, very recent signs of recovery. 

Turing to the impact of AI on productivity growth and yields, the overall effect on the labour market depends on whether the displacement effect on jobs, in reducing employment, exceeds the re-instatement effect of new jobs being created, and the timing. An example from the Goldilocks era was that many displaced employees found work in new occupations like website-design. For workers in jobs partially exposed to AI, productivity may be augmented by AI. This would also increase the scalability of labour, and ease labour shortages driven by demographics and ageing labour forces.

But inter-temporal, or timing effects are important here, and there are cross-currents on interest rates. If the displacement effect of lost jobs from AI proves substantial, and unemployment initially increases, it could squeeze domestic demand and inflation lower, and force a recession, particularly if reinforcing the impact of higher short rates. That might allow inflation to fall initially, easing pressure on both nominal and real yields, even if longer term growth potential increases, putting upward pressure on rates and yields. This narrative would suggest steeper yield curves would develop in the medium term. However, flat, or inverted, yield curves continue to dominate fixed income markets, as the US Treasury curve gradient in Figure 3 shows. 

Inverted, or flat, curves may be cyclical, after G7 tightening cycles, but could also reflect the view AI will drive a replay of Greenspan’s flat yield curve conundrum^[7] and low inflation rates from the Goldilocks era. Evidence of stable inflation breakevens, and the disinflation in 2023 support this view. But although there are some parallels between Goldilocks and the post-Covid eras, there are clear differences as well, notably ageing labour force demographics, much higher government debt/GDP ratios, geo-economic fragmentation and higher energy prices.

Indeed, the recent surge in demand for electricity and data centres is a consequence of the AI boom. These differences between Goldilocks and the post-Covid era are reflected in stubborn service sector inflation, which is still running at 5.2% in the US, and 5.9% in the UK, and would justify a higher term premium in longer maturities.

In conclusion, there are a number of structural crosscurrents in the global impact of AI on bond yields and curves, most notably improved productivity, and some initial displacement of labour, on the one hand versus higher equilibrium rates, or the Fed’s R*, on the other. Caution is also required in drawing inferences from the Goldilocks era, particularly about the shape of the yield curve, when the case for steeper curves is strengthened by higher debt/GDP ratios, and higher energy prices.

1. See “ Measuring the natural rate of interest” “, New York Federal Reserve, https://www.newyorkfed.org/research/policy/rstar/overview.

2. See “ The shape of things to come” “, LSEG, FTSE Russell, August 2023.

3. Moore’s Law states that the number of transistors in a micro-chip doubles within 2 years, and broadly implies computers become more powerful and efficient, driving down costs and increasing output.

4. See “   Selected papers on economic theory by Knut Wicksell”, “ pp.67-92, Erik Lindahl, 1958.

5. See , “ Capitalism, Socialism and Democracy”, Joseph Schumpeter, 1942.

6. See “ Productivity Trends in Japan — Reviewing Recent Facts and the Prospects for the Post-COVID-19 Era” “ — T Yagi, K Furukawa, and J Nakajima, Bank of Japan, July 2022.

7. Alan Greenspan, Federal Reserve, Semi-annual US Congressional testimony, February 2005