Economists can't agree on where rates should settle

At this year's Jackson Hole symposium, economists showed once again that they cannot agree on where the neutral rate of interest (r*) lies. Policymakers are debating whether rates are restrictive enough. The answer depends on r*, the number that tells central banks whether to keep tightening or start easing, and the profession's estimates are split by far more than a rounding error.

Why the number matters now

The Federal Reserve has raised its policy rate to a range of 5.25 to 5.5 per cent, and core inflation on the Fed's preferred measure was still just above four per cent in June. Subtract one from the other and the real policy rate is positive for the first time in years. Whether that is tight enough depends entirely on where neutral sits. If r* is near zero, real rates above one per cent are strongly restrictive and the Fed can stop. If r* has risen towards two per cent, current policy is barely leaning on the economy at all.

The dispute has sharpened because the economy has not slowed as expected. Unemployment is close to its lowest level in half a century, consumer spending has held up, and long-dated Treasury yields have risen to their highest since 2007. Each of these facts can be read as evidence that policy is not as tight as the models say.

Models against markets

Semi-structural models such as the Holston-Laubach-Williams framework still put r* close to historical lows, around one per cent, which would make current policy very tight. Market pricing and the economy's continued momentum suggest the real neutral rate has moved well higher. Policymakers are flying a fast jet through thick fog with a badly calibrated altimeter.

How the filters work

The Holston-Laubach-Williams approach treats r* as an unobserved variable and infers it from the data. The logic runs as follows: if output is above its estimated potential and inflation is rising, the real interest rate must be below neutral; if output is weak and inflation falling, it must be above. A statistical filter updates the estimate each quarter as new figures on growth, inflation and interest rates arrive. The method is elegant, but its output depends on estimates of potential output that are themselves uncertain and often revised.

The filters also move slowly by design. They are built to distinguish lasting shifts from temporary noise, so they take years to recognise a genuine change in the neutral rate. The confidence bands around the central estimate are wide enough that neither a neutral rate of zero nor one of two per cent can be ruled out with conviction. The pandemic made matters worse: the collapse and rebound of output in 2020 and 2021 was so large that the models had to be modified to stop it distorting the results.

What markets are saying

The alternative is to read r* from market prices. Forward real rates, derived from inflation-protected Treasuries, show what investors expect short-term real rates to be several years ahead, once the current cycle has passed. Those rates have risen sharply over the past two years and now sit well above the model estimates. On this reading, investors have concluded that the world of near-zero real rates is over.

Market measures have their own biases. Forward rates include a term premium, compensation for the risk of holding long-dated bonds, which has risen as governments issue more debt and central banks shrink their balance sheets. Part of the increase in forward real rates may be pay for that risk rather than a change in expected neutral rates. Separating the two requires another model, which reintroduces the problem the market approach was meant to avoid.

A history of wrong guesses

Central banks have misjudged unobservable variables before. Research by Athanasios Orphanides showed that the Fed's real-time estimates of the output gap in the 1970s were badly wrong, which led policymakers to believe they were leaning against inflation when they were feeding it. The error was not visible until years later, when revised data became available.

A more recent example is 2018. Jerome Powell said in October that the policy rate was still a long way from neutral; weeks later, after a slump in share prices, he suggested it was just below the range of neutral estimates. The Fed raised rates in December, paused in January and cut three times in 2019. That episode, in which the Fed discovered it had tightened enough only when markets told it so, is a fair preview of what may happen again.

The case for not knowing

The strongest counterargument is that central banks do not need a precise r* at all. Powell has argued for years that policy should be guided by outcomes, not by unobservable benchmarks: if inflation keeps falling and the labour market cools, policy is tight enough, whatever the models say. The Fed's current approach of deciding meeting by meeting on incoming data is designed for exactly this kind of uncertainty.

That approach works in principle, but it collides with the lags of monetary policy. Rate rises take a year or more to reach their full effect on spending and hiring. By the time the data show the economy has slowed, the tightening that caused it is long in the past, and more may already be in the pipeline. An outcomes-based strategy without a reliable r* is effectively a strategy of driving by the rear-view mirror.

What to watch

The Fed's September projections will show whether officials have raised their estimate of the long-run policy rate, which has stood at 2.5 per cent in recent rounds. Bank lending surveys will show whether tighter credit is spreading. Forward real rates and the term premium will reveal whether markets are pricing a higher neutral rate or simply demanding more compensation for risk.

Without a reliable estimate of the neutral rate, central banks are left to learn by trial and error, and they will probably discover they have gone too far only when something breaks. The most likely break points, over the next year, are the ones that already cracked in the spring: regional banks and commercial property, where borrowers refinance at rates that will look restrictive whatever the models conclude.