All eyes this week will be on May’s inflation rate, which will be reported Thursday.
While there’s little doubt that the Consumer Price Index will be markedly higher than the pandemic-depressed level of a year earlier, there’s widespread disagreement about whether this higher rate will be more than temporary. The report is widely anticipated because April’s reading broke a 12-year record, prompting a broad decline in stocks.
I’m reminded of George Bernard Shaw’s classic quip that if you laid all the economists end to end they still wouldn’t reach a conclusion.
I’m taking a different approach in this column: Determining which of several forecasting models of expected inflation have stood the test of time. Of the three that I measured, the one with the best record is forecasting that inflation over the next five years will average just 1.48% annualized.
This model was created a number of years ago by the Cleveland Federal Reserve Bank. The model has a number of inputs, including surveys of professional forecasters, the breakeven inflation rate (the difference between the yields on Treasuries and TIPS of similar maturities), and inflation swaps (derivatives in which one party to the transaction agrees to swap fixed payments in return for payments tied to the inflation rate).
To appreciate the strength of its track record, I tested it against several other well-known inflation forecasts: The University of Michigan’s consumer survey of expectations for inflation over the subsequent five years, and the five-year breakeven inflation rate by itself.
For each I calculated a statistic known as the r-squared, which represents the degree to which changes in the model correctly forecasts changes in the CPI over the subsequent five years. The r-squared ranges from a theoretical maximum of 100% (which would mean that the model perfectly forecasts changes in the CPI over the subsequent five years) to a minimum of 0% (which would mean the model has no predictive power).
I ran the tests on monthly data extending back to 2003, which is the earliest point at which readings from all three models was available. The Cleveland Fed model’s r-squared was highest at 23%, followed by the University of Michigan consumer survey at 20%, and the five-year breakeven inflation rate at 0%. These results are good news for those of us worrying about inflation, because the Cleveland Fed model’s five-year inflation forecast is the lowest of the three — as you can see from the accompanying table.
You may find it frustrating that these r-squareds are lower than 100%. But there are so many extraneous factors that affect inflation, including sheer randomness, that higher r-squareds are rarely seen in the real world. You should know that most of the models that get Wall Street’s attention have r-squareds that are insignificantly different than zero.
You might take issue with the good news story of the Cleveland Fed model because the University of Michigan survey, whose r-squared was almost as high as for the Cleveland Fed’s model, is projecting a five-year inflation rate that is more than double — 3.4% annualized versus 1.5%.
This objection fades when you realize that the high r-squared for the University of Michigan survey emerges on the assumption that you bet against it. In other words, higher values are associated with lower subsequent inflation, and vice versa. It therefore might even be seen as good news, in this ironic and contrary way, that the consumers interviewed by the University of Michigan are expecting high inflation over the coming five years.
The investment implication? Wait to see if the stock market plunges in reaction to the high headline inflation that is reported Thursday. If it does, a gutsy contrarian trade would be to buy, on the grounds that the market is overreacting to what is likely to be little more than a temporary blip.
Mark Hulbert is a regular contributor to MarketWatch. His Hulbert Ratings tracks investment newsletters that pay a flat fee to be audited. He can be reached at firstname.lastname@example.org.
Read More:Is inflation going higher or lower? We checked with the model that has the best record