XFE vs. Trimmed Mean, etc.

Karl Smith’s objection to my whimsical call for a 175 basis point cut in the federal funds rate seems to depend on what price index is used. My preference was for the most recent 12-month change in the deflator for market-based personal consumption expenditures excluding food and energy:

The best price index we have is the market-based core personal consumption deflator, which gives an inflation rate of 1.7%…over the most recent 12-month period.

I really do think that’s probably the best simple indicator of the inflation rate (though without the constraint of simplicity I would make a lot of changes, such as using a weighted average rather than a 12-month figure, putting some nonzero weight on food and energy, taking other kinds of price indices into account, and so on). Rumor had it a few years ago that Alan Greenspan liked it too. Anyhow, Karl uses a slightly different PCE-based deflator:

Using the 12 month trimmed mean PCE deflator of 2.2, we get 4.3 rounded down to 4.25 for a 100 bps cut.

The difference in those inflation figures accounts for the difference in the interest rates prescribed by our respective Taylor rules. There are a couple of differences in the indices we use: mine excludes non-market based prices and all food and energy, whereas his excludes only the most extreme price changes, wherever those should happen to occur.

Though in real life I think Karl’s 100 basis point cut would have been a better idea than my 175 basis point cut (for reasons of interest rate smoothing and risks of market instability rather than basic Taylor rule considerations) and indeed, I think the Fed’s 50 basis point cut was probably just the right thing to do under the circumstances (given that there will be another chance at the end of October), I stand behind my preference of price index. Regarding the market-based feature, I’ve always preferred accuracy to comprehensiveness when it comes to price indexes, and I really don’t trust prices that have to be computed by statisticians rather than observed in a market context. And I have a couple of reasons for preferring the “ex food and energy” core to the “trimmed mean” core.

First, I see a core index not so much as a way of filtering out volatility (which, if you’re going to do it, shouldn’t you apply a time series filter as well as a cross-sectional one?) but as a way of filtering out price changes that specifically aren’t likely to be repeated. One thing much of the food and energy component has in common is that the prices are determined in speculative markets for storable commodities. Oil is the most obvious example: any change in the observed price of oil roughly represents a revision of the market’s best guess as to what the price of oil will be in the future, less storage and financing costs. If there’s a jump in the price of oil, and we want to know whether that jump will be repeated, we can infer that the people who know the most about it, on average, don’t think so. If they expected the jump to be repeated, futures traders would have bid up the prices of distant contracts, and spread traders would have bid up the spot price further in anticipation of purchases by arbitrageurs with storage capacity (who could sell the distant futures, buy the physical commodity at a lower price, and lock in a profit), and the original jump would already have been large enough to eat up most of the hypothetical repeat jump.

Granted “food and energy” is not the ideal proxy for “storable commodities with speculative markets plus goods and services whose prices depend primarily on such commodities,” but it’s a pretty good first stab. Excluding such items amounts to outsourcing part of the task of forecasting inflation to the private sector. When it comes to general macroeconomic conditions, the Fed may have better information than the market, but when it comes to pricing of specific commodities, it’s not really plausible that the Fed’s information is even as good as that of the market, where people with specialized knowledge (and often private information) stand to make and lose large amounts of money daily on price changes.

By contrast, if there’s a jump in the price of some non-speculative item – say, for example, hairstyling services – there is no a priori reason to think that the jump won’t be repeated. There are empirical reasons – price jumps tend not to be repeated – but without knowing something about the fundamentals of the hairstyling market, it’s hard to know whether a given experience is expected to represent the rule or the exception. So, at least as compared to oil prices, I would not be greatly inclined to exclude hairstyling services temporarily from my price index just because it had a big jump in one month or one year. To do so amounts to making a naïve inflation forecast, and if we’re going to make inflation forecasts, why not go to someone who knows how to do a better job of it, rather than just using naïve rules of thumb?

The other reason I like excluding food and energy is that I don’t think it’s optimal for the Fed to try controlling such prices. The point of controlling prices overall (the point, for example, of having an inflation target), as I see it, is to provide a nominal anchor for monetary policy, so as to avoid a situation where nearly all prices start rising at a faster and faster rate (roughly as they did in the 1965-1980 period). As I argued last year, the choice of nominal anchor is a matter of convenience. You could choose gold, but that turns out (as we learned in the early 1930s) to be a very inconvenient choice. You could also choose a comprehensive basket of goods and services, but that’s probably not the most convenient choice either.

If the price of important commodities such as oil were to continue rising rapidly year after year, it might necessitate a downward path for real wages. Given that nominal wages can be sticky downward, rather than forcing a difficult decline in nominal wages (and most likely one or several recessions) by trying to keep the overall price level stable, it would make more sense to let the energy component of prices rise while letting nominal wages remain stable. Taken to its logical conclusion, my argument might imply that the Fed should target wages, or some combination of wages and stickier prices. The details have yet to be sorted out, but it is clear that typically non-sticky prices, such as those that largely determine the cost of the food and energy component of personal consumption, are not a convenient part of the nominal anchor.