Will AI-driven productivity growth allow the Fed to lower rates?

Greetings from the newsroom in London. I’m standing in for Chris today as he takes a well-deserved breather.

On Friday, the US Supreme Court struck down the bulk of Donald Trump’s tariffs. Then the US president used a separate legal avenue (so-called Section 122, which may also be liable to challenge in the courts) to declare a blanket 10 per cent tariff on all countries.

In a sign that “plan B” was well prepared, less than 24 hours later he said he was topping it up “immediately” by an extra 5 percentage points only for it to take effect on Tuesday at 10 per cent. Markets largely shrugged off the news, given the low expected impact of the new tariff rate on the US’s inflation and growth outlook. Send your predictions for what comes next: joel.suss@ft.com.

Productivity growth = lower interest rates?

Kevin Warsh — nominee to be the next Federal Reserve chair — argues that productivity growth stemming from AI will be “structurally disinflationary”, just as the advent of the internet was in the 1990s. This gives the Fed latitude to lower interest rates in the face of higher economic growth, he argues.

Not so fast, according to textbook economic theory. In classic macroeconomic models, higher productivity growth leads to higher real rates.

This happens primarily via savings and investment channels. First, companies anticipate higher profits and thus seek to invest more. This boosts demand for loans, pushing interest rates higher. 

Second, households anticipate higher wages and thus save less. The reduction in the savings rate lowers the supply of loanable funds in the economy, also pushing up interest rates.

In a speech last week on AI and productivity, Fed governor Michael Barr argued that these forces imply a higher setting for the policy rate when the economy is at equilibrium. “Last year I raised my long-term estimate of r* modestly because of higher productivity,” he said. “I expect that the AI boom is unlikely to be a reason for lowering policy rates,” he added.

Other Fed rate-setters have echoed this textbook view in recent weeks, including Philip Jefferson, who expects increased productivity growth to “result in an increase in the neutral rate, at least temporarily”.

In short, Warsh will struggle to convince his colleagues — even if they share his conviction that an AI-induced productivity boom is under way. 

Data complications 

But while textbook models are valuable for helping us think through the implications of higher productivity, often the data is disobliging. The long-run relationship between productivity growth and interest rates has largely not followed the theory. 

Below is a chart showing US real rates and productivity growth since 1890. Instead of the positive association textbooks would predict — that is, productivity correlating with rates — the relationship has been negative. 

This is not just a US phenomenon. Over the past hundred-plus years, UK productivity growth has coincided with lower real rates.

Of course, there was more going on over this time period. As pointed out to me by Jonathan Haskel, a former member of the Bank of England’s Monetary Policy Committee, increasing lifespans is one confounding factor. Higher life expectancy exerts both downward pressure on rates through increased savings and upward pressure on productivity growth.

The simple point is that we shouldn’t expect the data to behave as the stylised models suggest.

Causality might also run in the opposite direction. When rates rise, unproductive firms that are propped up by low borrowing costs fold, improving aggregate productivity via a “cleansing effect”. Similarly, if rates are lowered, investment and productivity should rise. Other researchers have questioned whether real interest rates are driven by investment and savings decisions in the first place.

Further complications

The textbook theory faces other issues when it meets reality.

Take households. The permanent income hypothesis, in which households save less in anticipation of higher future wages, is hard to square with the fear that AI will bring permanent unemployment to many.

The technologists driving the AI revolution are certainly not downplaying this possibility. (See, for example, Microsoft AI chief executive Mustafa Suleyman telling the FT that most white-collar work will be automated within 12-18 months.)

And even if new jobs are created, as they always have been in the wake of new technology, the pace of change and the labour market disruption could have negative aggregate demand implications. Who will buy all the AI-produced goods and services if unemployment shoots up and the labour share of total income heads to zero?

While the extreme downside scenario is unlikely, the fear of mass job losses is counter to the theory that suggests households expect real income gains in the face of productivity growth.

That being said, debt-financed corporate investment supporting the over $600bn in AI capex this year should, other things being equal, push up interest rates. The household saving rate in the US is also now historically low.

Moreover, estimates of r* have edged higher in recent years and may well continue to do so alongside accelerating productivity growth. Tighter immigration, high fiscal debt and a reduction in overseas demand for US Treasuries could all also exert upward pressure on interest rates. 

Don’t bet on it

In sum: there is a range of narratives that can be spun about how real interest rates will change in a world of higher productivity growth.

This is not a reassuring conclusion if you want to lower interest rates now. But it does bolster the case for continued data-dependent policymaking.

Warsh, in his invocation of the 1990s productivity boom under former Fed chair Alan Greenspan, has argued for ditching data dependence. “If you are looking at the [economic] data, my view is you are backward-looking; you are going to be late. You are not going to realise the country is able to have non-inflationary growth faster. So you are going to have to make a bet,” he said.

But Greenspan was, in reality, incredibly data dependent. Federal Open Market Committee transcripts reveal that he relied on detailed sectoral numbers when he famously identified that productivity growth was being measured incorrectly in 1996.

Given the uncertainty about AI and its productivity effects, the right response from policymakers is to, like Greenspan, study the data. A bet on lower rates comes with significant inflation risks.

What I’ve been reading and watching

One last chart

Over at the FT’s Monetary Policy Radar, we’ve pitted prediction markets against financial markets. When it comes to anticipating Fed decisions since 2023, Polymarket has an edge, but only slightly. The chart below shows the average absolute error in percentage points for the respective markets in the days before FOMC meetings.

Central Banks is edited by Harvey Nriapia