Macroeconomists generally agree that while the rational expectations assumption is very strong, it is a more useful approximation to actual expectations than the now ancient approaches of assuming adaptive expectations or the even older approach of regressing inflation on lagged inflation and using the fitted values as expected inflation. The old reduced form approaches are extremely vulnerable to the Lucas critique – the relationship between future and past inflation depends on policy and it is hard to believe that actual people would do so poor a job of forecasting as to assume they are economic constants. The argument for using rational expectations due to Lucas, Sargent and Wallace was extremely influential. In particular, the earliest version suggested that a single minded determined commitment to disinflation could lead to lower inflation without causing reduced output. I wasn’t an economist at the time, but I definitely have the impression that this mathematical result was considered to be very relevant to the real world so there was a hope that a sufficiently firm public commitment to disinflation would allow disinflation with only very temporary disruption to output. In the event, the disinflation of the early 80s came with huge increases of unemployment. This pattern can be reconciled even with the assumptions of price flexibility and fully rational expectations if the true priorities of the monetary authority are assumed not to be public. In other words, the authority’s claim that it is determined to cut inflation at any cost might not be credited. In such a model of asymetric information about the monetary authority’s objective function, credibility must be obtained through the costly signal of accepting high unemployment. Again, this can be due to fully rational Bayesian updating by market participants.
This still implies that an authority which is, in fact, dominated by inflation hawks will, on average, cause lower expected inflation conditional on lagged inflation. The difference due to the true preferences might even start out at zero, but it will grow over time. If we have learned that Paul Volcker was a more determined enemy of inflation than Arthur Burns and G. William Miller. We must expect that economic agents learned this before his Fed chairmanship ended. It would seem that by 1982 at the latest, his ruthless determination to defeat inflation at almost any cost was very obvious.
It seems to me that the claim that rational expectations are an improvement over the fitted value of inflation on lagged inflation must imply the belief that expected inflation was surprisingly low on average while Volcker was Fed chairman compared to what one would guess based on lagged inflation. At the very least this should be true if the sample is restricted to the period when Burns, Miller or Volcker was Fed chairman.
In fact the opposite is true. The median forecast of inflation from the Livingston survey of accepted experts was surprisingly high when Volcker was chairman. This is true even for the December 1970 through June 1979 subsample, that is for inflation outcomes which occurred when Burns, Miller or Volcker was chairman.
Footnote: Finf6 is the median of Livingston survey forecasts of the annualized rate of consumer price inflation over the next 6 months. Linf6 is the lagged annualized CPI inflation rate over the previous six months, volcker is an indicator which is 1 if Paul Volcker was chairman of the Federal Reserve Board on the date of the forecast inflation. Forecasts were made each June and December. The sample of forecasts starts with forecats made in June 1946 about prices in December 1946 and ends with forecasts made December 2003 (when the survey was redesigned).
Compared to the sample as a whole, expected inflation was 4% higher than one would expect while Volcker was chairman. Clearly there is something wrong with this regression. In fact in the first years of the survey the median Livingston forecaster grossly under-predicted inflation. This may reflect early problems with the survey or choosing the panel. It may also reflect the extreme difficulty of forecasting inflation just as World War II price controls were being eliminated and soon thereafter. The second regression restricts the sample to outcomes in June 1951 and later – this also corresponds to the first period when William McChesney Martin was chairman, so there are two fewer Fed chairmen in the reduced sample.
The coefficient drops to the still dramatically high 2.5% of extra expected inflation when Volcker was chairman.
Now Martin, and Alan Greenspan also have solid reputations as enemies of inflation. The real test compares Volcker to the notorious inflators Arthur Burns and G. William Miller.
Annualized 6 month CPI inflation was unexpectedly 1.44% higher when Volcker was chairman compared to Burns or Miller. The reported standard error is the Huber-White heteroskedastic consistent standard error (the ordinary standard error is almost identical giving a t-statistic of 2.94).
A simple regression of inflation on lagged inflation is a very crude way to forecast inflation. It is therefore a very crude way to model forecasts. It would seem that it is easy to improve on this model of forecasts using standard concepts such as regime shifts and credibility. In fact, the presumed improvements move economists’ guesses about Livingston survey participants’ inflation expectations away from the truth.
The inclusion of lagged 6 month CPI inflation and that only means that the survey forecasts are, in effect, being compared to very crude estimates. It is very possible that participants coincidentally had unrelated good reason to forecast high inflation when Volcker was chairman. This would imply that the coefficient on the Volcker dummy should be positive also when actual not forecast inflation is the dependent variable. However, that coefficient is negative.
Now it is possible to reconcile the results reported here with rational expectations – this is always possible. For example, if the sole aim of Livingston survey participants was to make life unpleasant for Robert Lucas, then their behavior makes sense. One could also argue that the average economic agent is more expert than the selected experts. However it is striking that the coefficients have the opposite sign of those implied by influential arguments which appeal to rational expectations.
The most remarkable fact is that the general view in the profession is that Volcker was and is an extremely credible enemy of inflation, the Livingston Survey is very well known. Analysis can’t be much simpler than my analysis here. Yet as far as I know, my calculations are arithmetically correct and my results haven’t been reported in the literature.
The boring details after the jump.
The data on forecasts are taken from the Livingston web site
http://www.phil.frb.org/research-and-data/real-time-center/survey-of-professional-forecasters/data-files/ . Starting in 1946 each June and December, participants forecast economic variables including the consumer price index 6 months and 12 months after the date of the survey.
Here I use the median forecast of the consumer price index 6 months after the date of the survey. The forecast inflation rate (finf6) is the square of the ratio of that forecast and the price level in the base period (provided by Livingston) minus one. That is it is the forecast of the annualized rate of inflation over the next six months. Calculating corresponding achieved inflation is not completely trivial. The Livingston survey asked for the CPI index but changing the base year from time to time and not exactly when the BLS changed the base year. A standard time series of the CPI will have one unchanging base year and deviate from the value survey participants were attempting to forecast. Fpr the inflation outcome (inf6) I use the annualized rate of increase of the base period CPI from one wave of the survey to the next unless the base year was changed (giving rates of ,for example, around -50%). For those intervals I used the annualized rate of increase from June to December (or from the previous December to June) of the CPI index from FRED http://research.stlouisfed.org/fred2/ . This is not ideal because the base year for the FRED data is 2009 but it only affects the measurement of inflation for three intervals: from June 1953 to December 1953, from December 1961 to June 1962, from December 1970 to June 1971, and from December 1987 to June 1988. In the 1940s inflation was very hard to forecast (the study started almost exactly when WWII price controls were eliminated). It is also possible that the Livingston team re-evaluated who they considered to be an expert after the terrible performance of their panel in the first few years. In any case, I generally use data only from the 1951 on. The survey was redesigned in 2003, so I don’t use data from more recent years.
If the idea of inflation fighting credibility has any credibility, one would also expect that Volcker’s anti-inflation credibility increased over time reaching the current very high level. The learning process would imply that early in Volcker’s chairmanship, forecasts of inflation were higher than optimal
forecasts made with the benefit of hindsight of his general record. This would imply a negative trend in forecast errors (forecast minus outcome) during Volcker’s chairmanship.