Inf lation forecasting is a truly dismal science
FOR CLOSE to 40 years, the US and much of the world were blessed with low inflation. No longer. Depending on whom you believe, the dramatic price and wage spike of the last few months is either a flash in the pan or the beginning of something more ominous. Both sides in this debate speak with confidence about what will happen next. But the history of inflation forecasting suggests that humility is in order.
Every oracle relies on a different set of tools to predict the future. In antiquity, animal entrails — ideally, the liver of a sacrificed sheep — would be scrutinized for clues of things to come. In the postwar US, professional economists seeking to predict inflation opted for something a little less sanguinary: the Phillips Curve, named for the economist William Phillips.
The new approach had many advantages, and sparing sheep was the least of them. The Phillips Curve posited a predictable inverse relationship between unemployment and inflation: If unemployment went up, inflation went down, and vice versa. This relationship between the two variables permitted economists to weigh the implications of policy decisions as well as predict future inflation rates.
In time, economists developed different versions of the Phillips Curve model. Most revolved around an ideal baseline rate of unemployment. Go above this sweet spot, and inflation would fall; go below it, and inflation would rise. This rate was dubbed the Non-Accelerating Inflation Rate of Unemployment, shortened to the unlovely acronym Nairu.
In the 1970s, Milton Friedman and other economists attacked the theoretical assumptions that animated the model, arguing that it would fail to hold up over longer time horizons. But the curve remained the crystal ball of choice for forecasters hoping to figure out where inflation was headed in the short term.
The new consensus was neatly captured by the economist Alan Blinder. In 1997, he wrote that “the empirical Phillips curve has worked amazingly well for decades,” and counseled its continued use by policymakers.
Some economists began questioning Blinder’s claim. In 2001, two economists at the University of California at Los Angeles — Andrew Atkeson and Lee Ohanian — published a paper based on an experiment that compared the predictive prowess of the Phillips Curve to a model that was simple to the point of parody: forecasting next year’s inflation by averaging the previous four quarters’ rates. In other words, next year’s inflation will be the same as the previous year’s. That’s it.
“We establish this naive forecast as our benchmark,” they explained, “not because we think that it is the best forecast of inflation available, but rather because we think that any inflation forecasting model based on some hypothesized economic relationship cannot be considered a useful guide for policy if its forecasts are no more accurate than such a simple atheoretical forecast.”
Atkeson and Ohanian pitted their model against two different variations of the Phillips Curve as well as the Federal Reserve’s internal forecasting metrics. The result? The naive model held its own against all contenders, equaling, and in some cases, besting the sophisticated, multivariable forecasting models beloved by economists.
Subsequent studies largely corroborated these findings, but added some important caveats.
Researchers such as James Stock and Mark Watson found in 2008 that the accuracy of forecasts rooted in Phillips Curve models improved when the unemployment rate deviated significantly from the Nairu, and faltered when the rate approached the ideal.
But it’s worth recalling that these temporary improvements in forecasting power were relative to a “model” that a child could have invented. (What will inflation be next year? Same as last year!)
This point was made even more pungently by a subsequent study by the economists Marie Diron and Benoit Mojon, who came up with their own, equally interesting thought experiment. Their comparative “model” was even simpler: take a central bank’s inflation target and then use that number as a consistent prediction for inflation every single year.
In the US, the unofficial target number was 2%. Diron and Mojon’s construction therefore predicted 2% inflation, year in and year out. They did the same thing with other countries. Guess what? They managed to beat the complex multivariable models for one long stretch running from 1995 to 2007. Not bad.
For those who continue to cling to the hope that inflation is predictable, there is another method a bit more reminiscent of those oracular sheep: simply ask what the rest of the herd thinks will happen. In other words, question ordinary people (or professional forecasters) about their expectations for inflation and average the answers.
This approach, a landmark study from 2007 showed, delivers better results than any of the standard forecasting models. In this particular inquiry, the alsorans included the Phillips Curve as well as predictive methods that take signals from the bond markets — namely, data about the structure of debt.
That inflation expectations are reliable is understandable. If you believe prices will rise, you will act in ways that could insure this comes true. Put differently, inflation expectations are as much blueprints for action as they are predictions. They’re self-fulfilling.
A more recent study corroborated these results, showing that inflation expectations of both ordinary consumers and professional forecasters generally topped other methods in a decade-by-decade matchup going back to the 1960s. Still, in two of those decades, modified versions of Atkeson and Ohanian’s “model” managed to come out on top.
If asking other people what they think will happen is the best approach, surely the professional forecasters have the edge, yes? Even here, the evidence is contradictory. While the professionals have a better track record over the very long term, there is an embarrassingly long stretch of time, 1984 to 2006, where average Americans narrowly edged out the professionals.
Let that sink in for a moment. Ordinary Americans — significant numbers of whom believe in haunted houses (42% or 58%, depending on the poll); devil possession (41%); the lost city of Atlantis (57%); and astrology (30%) — consistently outperform Federal Reserve economists, bond market professionals and, in many years, professional economic forecasters.
It’s something to keep in mind next time a trained economist tells you about a model showing that inflation will rise, fall, or stay the same. In truth, the collective wisdom of people who have never heard of the Phillips Curve is likely to provide a better guide to what lies ahead.