Today and tomorrow finds me in the fine state of Texas, at a conference on "Price Measurement for Monetary Policy," co-sponsored by the Federal Reserve Banks of Cleveland and Dallas. First up was a paper by Diana Weymark and Mototsugau Shintani from Vanderbilt University, titled "Measuring Inflation Pressure and Monetary Policy Response: A General Approach Applied to US Data 1966-2001." The idea is take a particular model of the economy and monetary policy, and ask the following sort of questions: How responsible was the central bank in determining inflationary outcomes, where the central banks influence covers both active policy (or interest rate) choices, as well as the evolution of inflationary expectations.
The experiment is a tricky one, and several people commented on whether the precise nature of the measures used to answer this question exactly captured what the authors had hoped. But the results are certainly provocative:
One of the most striking aspects of [our results] is the similarity of the Fed’s policy response over the 35 year period under study. The [policy stabilization] index measures the proportion of inflationary (or deflationary) pressure removed by monetary policy. The average [index] values show that under all five chairmen, the Fed countered positive inflationary pressures and magnified deflationary pressures...
Volcker and Greenspan are now generally credited with having taken a tough stand against inflation. However, our... indices show that monetary policy under Martin and Burns was also effective in counteracting inflation pressure... monetary policy under Miller was much less effective in combating inflation than under the previous two Chairmen. However, according to [our results], inflation pressure during Miller’s tenure as Fed Chairman was 3 times higher than it had been during the previous eight years... that Miller did try to use periods of deflationary pressure to bring about significant reductions in inflation. However, these efforts were not very successful because... the deflationary episodes were considerably weaker than the inflationary pressures that developed during this period...
... the Fed’s lack of success under Miller may in part be attributable to an inability to convince economic agents of the Fed’s commitment to price stability.
I'm not sure that result would be robust to other approaches to measuring Fed effectiveness -- and certainly the management of inflation expectations can be mainly placed upon the doorstep of the central bank -- but, heck, might as well let the rehabilitation of G. William Miller's reputation get its due.
UPDATE, Paper 2: Wherein Lafyette College's Julie Smith asks the question "Better Measures of Core Inflation?" "Better" here means a more accurate forecast of CPI inflation one or two years out than you can get with the median or trimmed-mean CPI. The answer the author gives is "yes", and the key (for you stat geeks) is to (a) individually estimate models for the individual component prices of the CPI market basket, but (b) estimate your statistical models for the components jointly. If you are interested in such forecasting issues, the paper is worth a look.
Paper 2, Update 2: Asked from the floor: If what we are interested in is a good forecast of future inflation, why not use a full forecast model, based on all kinds of data, rather than a core inflation measure? And if forecasting is not what we are nterested in, what's the point of core inflation in the first place? Good question that.
Update, Paper 3: One answer to that last question is provided in the paper "Policy-Sensible Benchmark Core-Inflation Measures for the Euro Area and the U.S.", by Stefano Siviero and Giovanni Veronese from the Bank of Italy. The essence of the argument is that we are not primarily interested in doing forecasting well, but in doing policy well. A central bank should focus on core inflation if doing so helps them better achieve its ojectives (typically taken to be some combination of low headline inflation, maintaining GDP growth near its potential, and relatively smooth interest rates). Although the results are tentative, Siviero and Veronese shout out a warning to fans of the usual core inflation suspects:
Our findings suggest one cannot recommend that the most popular core inflation measures be used to support monetary policy-making. Specifically, we find that it is arguably inappropriate to remove all erratic components from headline inflation: by reacting to core inflation measures that do so, monetary policy effectiveness may be seriously impaired, even if one's reaction is designed in such a way so as to be optimal on the basis of standard welfare criterion.
The message here is one that ought to be communicated more clearly by monetary policymakers: It can be quite appropriate for a central bank to focus on some measure of core inflation, not because it is something the central bank thinks you should care about but because it helps to control the thing you do care about -- in this case, overall or "headline" inflation.
UPDATE, Paper 4: Next up -- Core Inflation as Idiosyncratic Persistence: A Wavelet Approach to Measuring Core Inflation , by Richard Anderson,* Federal Reserve Bank of St. Louis; Fredrik Andersson, Lund University; Jane Binner, Aston University; Thomas Elger, Lund University. Here we find another shot at motivating core inflation as something a central bank uses to better control overall inflation, and control it in a way that generates the best outcomes for its customers -- i.e., you and me. The actual game in this paper is, nonetheless, straighforwardly statistical. The idea is essentially the following familiar idea: Every individual price in the economy is a combination of an underlying trend and temporary ups and downs, and the statistical problem is to separate the two.
The application of the idea, however, is not so simple, or familiar. Unless you are conversant in things like wavelets, neo-Edgeworthian index numbers, and stuff like that -- or are interested in finding out about them -- you are likely to find the paper a bit of a slog. However, the authors do offer up what is becoming a theme at this conference: Whatever a core inflation measure ought to be, something like the CPI ex food and energy is not likely to be it.
Update, Paper 5: Looking for horse race? Rob Rich and Charlie Steindel are your boys. Here's what they do, in "A Comparison of Measures of Core Inflation"...
This paper examines several proposed measures of U.S. core inflation: an ex food and energy series, an ex energy series, a weighted median series, and an exponentially smoothed series. We evaluate the performance of the candidate series using criteria such as ease of design, accuracy in tracking trend inflation, as well as explanatory content for within-sample and out-of-sample movements in aggregate inflation. The empirical analysis principally focuses on the methodologically consistent Consumer Price Index (CPI) which is only available starting in 1978. As a check for robustness, we also provide a summary for Personal Consumption Expenditure (PCE) inflation starting in 1959.
... and here is what they find:
... we find no compelling evidence to focus on a particular measure of core inflation, including the series that excludes food and energy prices. We view the results as consistent with the diversity of findings reported in previous studies, and suggest they are a consequence of the design of the individual core inflation measures and their inability to account for the variability in the nature and sources of transitory price movements.
In other words, if what core is all about is disentangling longer-run trends in price movements from temporary ups and downs -- the idea of the Anderson et al paper in the previous update -- none of the stanard simple measures of core are uniquely (or consistently) qualified for the task.
UPDATE, papers 6 and 7:
The Rich and Steindel paper in the previous update did not include trimmed-mean measures of core as part of the analysis. If your core-inflation jones is aching as a consequence, Andrea Brischetto and Tony Richards (Reserve Bank of Australia) and Mike Bryan (Federal Reserve Bank of Cleveland) will ease your pain in "The Performance of Trimmed Mean Measures of Underlying Inflation" and "Monitoring Inflation in a Low Inflation Environment," respectively. (In case you have forgotten, an x-% trim just means lopping of the x-% most extreme prices within a distribution of prices, whatever they may be.) Brischetto and Richards look at various trims (corresponding to different percentages of the price-distribution which are excluded) for the Australia, the Euro area, Japan, and the U.S. They find:
Based on data for four economies, we find that trimmed means tend to outperform headline and exclusion-based core measures on a range of different criteria, which indicate that trimmed mean measures can be thought of as having a higher signal-to-noise ratio than either of the other measures. This makes trimmed means more useful for extracting information about the current trend in underlying inflation from the relatively noisy monthly or quarterly CPI data.
This does not seem entirely consistent with the Rich and Steindel analysis, but Mike Bryan stressed that, first, trimmed-mean estimators represent a technique for reomoving high frequency -- that is, very short-term -- noise in the inflation process. Looking for improved forecast performance over a longer horizon is not likely to be productive.
Second, Mike stressed that it is precisely in low inflation environments that trimmed-mean core measures really shine. Here's what I found really interesting: If you apply statistical tests to find when the inflation trend changed in the United States, you will find there was a change September 1981 -- when the average inflation rate fell from 9.4 percent per year to 4.1% per year -- and January 1991 -- when the trend fell from 4.1 percent to 2.7 percent. Now consider ask the question"if you were watching a measure in core inflation in real time, when would you have picked up theses changes in trend?" It turns out, the choice of core or headline, this core or that core, wouldn't have mattered much when the change was big, as in 1981:
When the change was small, however -- as in 1991 -- you could pick the shift up fairly quickly with core measures -- trimmed measures in particular -- and not for nearly six years with the headline number:
Steve Cechetti (Brandeis University) reacted to all of this, and closed things with this observation: A core inflation measure should be smooth, track the trend in inflation, and give you information about breaks in trend relatively quickly. Core inflation measures should not be the target for monetary policy, or (necessarily) be the best possible forecast for inflation.
An interesting end to an interesting day.