The Fed Board has published a paper by Michael T. Kiley titled “Anchored or Not: How Much Information Does 21st Century Data Contain on Inflation Dynamics”. Please, find below our review and comment, including possible implications for the Fed staff.
To keep in mind
Research shows that, unless one deliberately ignores the history of US inflation pre-2000, the persistence of current inflation is estimated very high and somehow comparable (although still lower) to the “accelerationist” (“unanchored”) period. A direct implication of this result is that high inflation of 2021 may continue in 2022 and beyond, although deviations of inflation from its average are ultimately expected to be reabsorbed.
What the paper does
Kiley (2022) is a technical paper. Here we provide the intuition. The intuition is that what one thinks of today’s inflationary process can depend on the assumptions of the model used. In particular, it can depend on whether one puts a lot of weight on the prior that today’s data generating process is similar to the (“accelerationist”) pre-2000 world or whether one thinks that today’s process has nothing to do with the “unanchored” period. Kiley (2022) approach is to “let the data speak”: do the post-2000 period add any substantial information to the pre-2000 sample? If so, how?
(For the record: in a Phillips curve model, the degree of persistence of the process is determined by the autoregressive coefficient. If the autoregressive coefficient is one, the process becomes “unanchored” as any shock ends up having a permanent effect on inflation. On the other hand, in the “anchored” world, the autoregressive coefficient sums up to one with the coefficient of long-term inflation expectations; to the extent that expectations are stable, the inflation process remains “anchored” to the long-run mean determined by expectations)
In the limiting case of zero conviction in the prior (that is, zero conviction that the pre-2000 period can be a guide for today’s inflation), Kiley’s Bayesian approach is equivalent to the approach in which a break is assumed in the Phillips Curve and only post-2000 data is used to estimate the inflation process (which is what most reduced form anchored Phillips curve models do). What Kiley (2022) does is to compare the posteriors of the autoregressive coefficient according to the different priors. The idea is intuitive: if the posteriors heavily depend on the degree of conviction (prior), then today’s process would be probably different in nature. On the other hand, if the posteriors do not depend on the degree of conviction, then one cannot reject that today’s inflationary episode is similar in nature to the “unanchored” period.
Results
The figure below shows Kiley’s main result. In particular, the figure shows the coefficient of lagged inflation according to different prior weights on pre-2000 data.
The degree of persistence in the posterior estimates is high in all cases except those with no weight on the 20th century prior (the purple dashed line). The coefficient on lagged inflation exceeds 0.9 with equal weights on prior and data (the blue line), exceeds 0.85 when the weight on the prior is 0.20, and is near 0.7 when the weight on the prior is 0.05; in the limiting case of essentially no weight on the prior, the coefficient on the lags is small at about 0.2 (in line with simple least squares for the 2000-2019 period).
Kiley’s results show that a Bayesian decisionmaker aware of the inflation process from 1960-1999 would view the current inflation process as similar to that from 1960-1999, unless they place very little weight on the prior information (that is, unless they decide to deliberately ignore that sample). A direct implication of this result is that the inflation process may signal substantial persistence, suggesting the high inflation of 2021 may continue in 2022 (and beyond). Nonetheless, the coefficient on the lags of inflation is less than one in all specifications—so the process is not of the “accelerationist” type, implying that deviations of inflation from its average may be long-lived but ultimately “transitory”.
Comment and implications for the Fed staff
Kiley (2022) paper is extremely interesting. Kiley (2022) shows that unless one deliberately ignores the history of US inflation pre-2000, the data suggest that today’s degree of persistence is similar to the pre-2000 period, although the data rejects the “accelerationist” hypothesis.
Kiley (2022) results have important implications for the Fed staff (and the FOMC). In our view, the Fed staff relies on anchored Phillips curve models estimated over the 1998 to present sample (similar to what other contributions in the literature do). However, by doing so, the Fed staff is assuming that the inflationary process is stable (an assumption based on the stability of observed long-term inflation expectations). In other words, in our view what the Fed staff does is equivalent to the limiting case (zero prior on pre-2000 data) described in Kiley (2022). But the problem of the Fed staff approach is what we have already stressed in previous notes: if a model assumes the stability of the inflationary process, it is currently prone to large out-of-sample errors. In fact, relying only on an anchored model is a perfect recipe to miss structural breaks.
Having said that, Kiley (2022) also shows that it is still possible to reject the null of an “accelerationist” process. In other words, the current shock is probably persistent but not permanent, yet.
The good news for the Fed staff is that long-term inflation expectations continue to be stable and the data reject the “accelerationist” hypothesis. The (very) bad news, however, is that the data do not reject the hypothesis that high inflation might persist for years, possibly much more than what suggested by the anchored models the staff seems to rely on.
Clearly, the FOMC cannot ignore such a big risk.