A fundamental relationship of mainstream economic theory at the heart of the Federal Reserve’s strategy for setting interest rates has been a poor guide for policy makers for at least three decades, according to a study by the Philadelphia Fed’s top-ranking economist.

The paper, co-authored by Philadelphia Fed Director of Research Michael Dotsey, shows that forecasting models based on the so-called Phillips curve, which asserts a link between unemployment and inflation, don’t actually help predict inflation.

“Our results indicate that monetary policymakers should at best be very cautious in their reliance on the Phillips curve when gauging inflationary pressures,” Dotsey and Philadelphia Fed economists Shigeru Fujita and Tom Stark wrote.

Their study is timely. Fed officials have been surprised by a deceleration in U.S. inflation over the past several months despite a continued decline in unemployment, the opposite of what the Phillips curve relationship would predict.

Minutes of the last meeting of the central bank’s rate-setting Federal Open Market Committee in July revealed that “a few participants cited evidence suggesting that this framework was not particularly useful in forecasting inflation,” while “most participants thought that the framework remained valid.”

If the majority view on the FOMC is that the Phillips curve framework is still valid, it implies that central bankers should continue raising interest rates with unemployment at a 16-year low, because they expect inflation will rise in the medium term even though prices pressures have been disappointingly soft.

Kansas City Fed President Esther George, who has been more forceful than many of her colleagues in recent years about the need to raise rates, lent support to that view on the sidelines of this week’s annual gathering of central bankers from around the world in Jackson Hole, Wyoming.

“There may in fact be something wrong with the models, I don’t know, I think that continues to be a question that many economists are asking,” George said during a TV interview with Bloomberg’s Michael McKee that aired Thursday. Even so, she favors another rate increase this year.

The Philadelphia Fed economists found that rising unemployment was sometimes able to help predict lower inflation, but falling unemployment didn’t help predict higher inflation. They noted that was particularly the case during the 1970s and early 1980s when the Fed responded to runaway inflation by raising rates so high that the U.S. economy fell into recession.

“Our evidence may indicate that using the Phillips curve may add value to the monetary policy process during downturns, but the evidence is far from conclusive,” they wrote. “We find no evidence for relying on the Phillips curve during normal times, such as those currently facing the U.S. economy.”