Can robots replace humans in monetary policy?

Federal Reserve Chairman Jay Powell and President-elect Donald Trump agree that running the U.S. central bank is the biggest job in government.

Their reasons are different. Shortly before winning the election, Mr. Trump was characteristically dismissive, saying the main benefit of being Fed chairman was the admiration that comes with the role. “That’s the biggest job in government. You come into the office once a month and say, ‘Let’s flip a coin,’ and everyone talks about you like you’re God,” he said. spoke.

Last week, Mr. Powell responded by declining to mention the coin toss but otherwise offering some agreement (7 minutes, 20 seconds into this video). “I really love this job,” he said. “And it’s a special place to be surrounded by dedicated people and know that the work you do really matters to people. It’s a great honor to do that work.”

Forget the coin toss, President Trump’s comments raise important questions. To what extent can monetary policy setting and analysis be automated?

Robot setting policy

Setting monetary policy rules has a long history. This is because monetary policy should be simple in principle. We have a consistent theory and accurate data, and applying one to the other yields the optimal policy path for interest rates (or money supply, in the case of money supply). (based on the monetarist tradition).

But these rules never worked. The most famous monetary policy algorithm is the Taylor rule, which links interest rates to the deviation of inflation from the target and the degree of spare capacity in the economy. Therefore, the Taylor rule states that if inflation is high and all resources are fully utilized, interest rates should be high. Low interest rates are needed to stimulate the economy when people and businesses are out of work or when inflation is well below target. The underlying theory is a new Keynesian approach that assumes that the output gap and inflation deviation can be measured accurately.

A new study by the Bank for International Settlements shows how inadequate the Taylor rule is at predicting interest rates across most developed countries.

The intention of the authors of this article is to augment the Taylor rule with better economic theory, current central bank monetary policy principles, and better data, all with the aim of steering monetary policy away from supply shocks. The aim is to be more sensitive to demand shocks.

In this world, higher interest rates are an appropriate response to demand-driven inflation because they suppress economic activity and address the underlying problem. But if inflation is caused by a supply shock (e.g., a rise in oil prices), then the interest rate question is moot. Central banks should “scrutinize” the initial price impact if it is temporary. Raising interest rates would have too big an effect too late on the shock, hitting demand after inflation subsides and ultimately creating deflationary forces.

When it comes to large-scale supply shocks, such as the post-COVID-19 inflation period, there are nuances that have raised concerns that second-order effects could trigger a wage-price spiral. In this case, higher interest rates would be needed to anchor inflation expectations and prevent workers and businesses from taking advantage of the initial supply shock to raise profits and real wages. But according to the BIS study, the empirical results are clear. A “more conservative policy response to supply rather than demand-driven inflation” is needed.

Data work here is very important. The BIS draws on academic research (mainly by Adam Shapiro of the San Francisco Fed) that attempts to break down inflation into demand and supply components, and uses a “targeted Taylor rule” that is strongly responsive to demand to We considered whether we could better explain the bank’s policies. As shown in the graph below, there is inflation-driven inflation and moderate supply-driven inflation. The BIS investigation was clear. The asymmetric Taylor rule approach may explain policy well.

Given this result, a natural question that BIS has not considered is whether robots can apply asymmetric policy rules on behalf of policymakers. According to BIS research director Hyun Sung Shin, the answer is no.

“The actual monetary policy strategy is a little more complicated than the asymmetric Taylor rule,” he told me. why?

First, the data is far from perfect. How to divide inflation into demand-driven and supply-driven components is still far from agreed. The graph above looks reasonable as the most recent inflation has been driven primarily by supply, but not exclusively. In July, I highlighted other studies that came to exactly the opposite result, particularly regarding Europe. These distinctions are themselves outputs of the model and are subject to errors and uncertainties, especially when measured in real time.

Data problems persist regarding the inflation component. Ideally, we want accurate forecasts of inflation, rather than recently measured rates, to prevent the rules from going backwards. Approaches that rely on estimates of the output gap use hypothetical data that cannot be known with precision.

This theory also does not necessarily hold true for parameters such as the degree of implicit relationship between inflation and surplus productive capacity. In reality, many events occur outside of the strict model parameters.

Even if we can better explain how central banks operate, human judgment and disagreement will still be needed for some time.

Can robots interpret policy?

If robots cannot easily replace central bankers, can robots interpret them as well or better than humans?

The BIS Quarterly Review also has an interesting article on how to best use large-scale language models in economics.

Rather than rehash the findings, I would like to highlight some of the ongoing analysis that my colleagues and I are working on at the FT, led by Joel Sass. We have used a large-scale language model to interpret central bankers’ speeches on a hawkish and dovish scale.

The Fed’s results are in the graph below. Click on the chart to see that each dot represents the Fed Governor’s speech and includes key passages extracted by artificial intelligence. The question now is whether this will put central bank watchers out of work.

After considerable polishing, there is no doubt that this model has produced excellent results, and the Fed’s speech is judged to be hawkish when (or just before) interest rates are rising. The Fed has become seen as dovish as it prepares to cut interest rates.

But let’s be brutal. There is a bit of a “no shit” element to this result, as speeches are seen as more hawkish when interest rates are rising and more dovish when interest rates are falling, so this model Questions remain as to how much value it brings. This model can also gather information from across the Internet and, for all we know, may use the federal funds rate as an input to its valuation.

But let’s not get upset about this. This model is highly effective at parsing vast amounts of text with incredible accuracy, allowing speech to be “read” and extracted valuable information very quickly.

Central bankers can be as dense and long-winded as they like. Tools have emerged to extract signals from long pieces of prose.

Is Powell programmable?

My computer programming skills are quite old, having dabbled in BASIC as a child and Modula-2 as a junior researcher. But it occurs to me that there is a simple algorithm that can explain Jay Powell’s recent policies.

Recall that in September, when the Fed cut interest rates by 0.5 percentage points, Chairman Powell said the U.S. economy was in “good shape” to warrant deep rate cuts and that he wanted them to continue.

In an interview last week, Powell said the Fed “can afford to be a little more cautious” given the health of the economy, calling it the envy of the world.

So Powell appears to be programmable. The following algorithm applies (with apologies to all you proper programmers):

10 Select the policy according to your preference.

20 They argue that this is appropriate because the U.S. economy is in good shape.

Go to 30 10

What I’ve read or seen

Barring any surprises, the Bank of England plans to cut interest rates four times by the end of next year, Andrew Bailey said. There may be some surprises, but

South Korea’s central bank governor Lee Chang-yong says he is more worried about President Donald Trump’s trade policy than domestic political turmoil.

India replaced hawkish central bank governor Shaktikanta Das with Sanjay Malhotra, even though inflation remains a problem.

China reaches out for economic stimulus again

important charts

The Fed is proud of its reliance on data. Not only is this backward-looking, but the data that most affects civil servants, the growth in monthly salaries, is abysmal.

The U.S. monthly employment report released last week revealed that payrolls increased by 227,000 in November. However, the average absolute number of revisions for this series through the third month of publication is more than a quarter of that, at 57,000.

So what can we say? The U.S. labor market is somewhere between a pretty depressed market and a gangbusters going. In other words, we don’t know too much about monthly U.S. payrolls, and they carry too much weight.