Federal Reserve Bank of New York President John Williams said Thursday that detecting structural changes in the economy while they are unfolding remains among the most consequential challenges confronting policymakers. Delivering remarks at the Reykjavík Economic Conference in Iceland, Williams focused on the particular difficulty of recognizing changes in productivity growth in real time and the implications that delayed recognition has for monetary policy.
Williams said the topic has taken on extra prominence given current attention on artificial intelligence and discussion of its potential economic effects. He framed the issue around how economies respond when the underlying rate of productivity growth shifts and emphasized that firms, households, and policymakers typically adjust their expectations only gradually.
Historical episodes and the recognition problem
To illustrate his point, Williams reviewed several historical episodes. He noted a productivity slowdown in the 1970s after a roughly 25-year span of postwar gains, an acceleration beginning in the mid-1990s, and another slowdown that emerged in the mid-2000s. The 1970s slowdown was associated with stagflation, while the productivity boom in the late 1990s and early 2000s coincided with a period of strong growth and low inflation, he said.
Williams emphasized what he called the recognition problem - the tendency for shifts in trend productivity growth to appear clear only in hindsight. Using the data he cited, year-over-year productivity growth can vary widely - swinging from negative 2% to positive 7% - around a long-run average of just over 2%. That volatility, he said, makes it difficult to separate genuine shifts in trend from ordinary data noise in real time.
As an example, Williams pointed to estimates from the Council of Economic Advisers during the 1970s. Those estimates fell slowly through much of the decade and then dropped sharply in 1979, long after the slowdown had begun, he said. He said a similar pattern of delayed recognition also appeared in the mid-1990s and again in the mid-2000s.
Modeling gradual recognition and macroeconomic implications
Williams described simulations from the New York Fed's dynamic stochastic general equilibrium model to show how gradual recognition changes predictions compared with standard theory. Under classic models, an increase in trend productivity raises real interest rates and can lead to a contraction in hours worked, investment, and output. These model predictions, he said, contrast with the historical association in which episodes of elevated productivity growth often align with robust economic performance and heightened investment.
In the New York Fed simulations, when trend productivity accelerates but agents recognize that acceleration only gradually, they initially treat the shift as largely a one-time change rather than an ongoing trend. That interpretation affects behavior: higher productivity lowers firms' costs, which puts downward pressure on inflation until prices and wages have fully adjusted, Williams said. The magnitude and persistence of the inflationary effect depend on how quickly businesses and households update their expectations about future productivity.
Real-time identification remains challenging
Williams reiterated that identifying structural change in real time is extraordinarily difficult. He noted that shifts in productivity growth are relatively rare and carry substantial uncertainty, meaning estimates of trend productivity are accompanied by wide confidence bands. Because expectations about future growth tend to evolve slowly, the gradual recognition of a true change in productivity can materially alter the macroeconomic path implied by models and therefore complicate monetary policy decisions.
In sum, Williams argued that both the practical difficulty of distinguishing trend shifts from regular volatility and the slow adjustment of expectations combine to produce significant challenges for policymakers seeking to respond appropriately to changes in underlying productivity growth.