Measuring Pure Inflation

Here are Ricardo Reis and Mark Watson who will tell you what they have done and why it matters. This is interesting work:

Relative Goods' Prices and Pure Inflation, by Ricardo Reis and Mark W. Watson, NBER WP 13615, November 2007 [open link]: ABSTRACT This paper uses a dynamic factor model for the quarterly changes in consumption goods' prices to separate them into three components: idiosyncratic relative-price changes, aggregate relative-price changes, and changes in the unit of account. The model identifies a measure of "pure" inflation: the common component in goods' inflation rates that has an equiproportional effect on all prices and is uncorrelated with relative price changes at all dates. The estimates of pure inflation and of the aggregate relative-price components allow us to re-examine three classic macro-correlations. First, we find that pure inflation accounts for 15-20% of the variability in overall inflation, so that most changes in inflation are associated with changes in goods' relative prices. Second, we find that the Phillips correlation between inflation and measures of real activity essentially disappears once we control for goods' relative-price changes. Third, we find that, at business-cycle frequencies, the correlation between inflation and money is close to zero, while the correlation with nominal interest rates is around 0.5, confirming previous findings on the link between monetary policy and inflation.

...

6. What have we done and why does it matter? In this paper, we ... used different estimation techniques and specifications to robustly estimate pure inflation, and proposed a simple method to compute macroeconomic correlations while controlling for goods’ relative price changes.

Our first finding was that pure inflation can differ markedly from other conventional measures of inflation, like the PCE deflator or its core version. It is smoother, less volatile, and in particular in the 1990s, its ups-and-downs are quite different from those in other measures of inflation. This should be useful to economic historians since it provides an alternative account of the movements in inflation in the last half-century. Relative to existing measure of inflation, pure inflation has the virtue of separating absolute from relative-price changes, which is a crucial distinction in economic theory. Moreover, pure inflation matches more closely the concept that many economists seem to have in mind when discussing aggregate movements in prices and monetary policy (typically based on intuition that comes from a one-good world).

Our second main finding was that pure inflation was quantitatively significant (it accounts for about 5% of individual price changes), but only accounts for 15-20% of the variability in inflation measured by conventional price indices, like the PCE deflator, the GDP deflator, or the CPI. This has at least two implications for the work of economic theorists building models to explain inflation. First, it shows that comparing the predictions of one-good models with common measures of inflation is flawed. The difference between these measures and pure inflation is large enough that it can easily lead to mistakenly accepting or rejecting models. Second, our estimates provide a new test statistic with which to test the pricing assumptions of models with many goods. An important ingredient (and topic of debate) in recent models of nominal rigidities is a model of pricing that implies slow adjustment of prices to monetary policy shocks together with frequent price changes. The fraction of the variability of cost-of-living inflation accounted for by pure inflation can be an important statistic to diagnose the success of these models at fitting the data.

Third, we found that, once we controlled for relative goods’ prices, the Phillips correlation became quantitatively insignificant. Therefore, the correlation between real quantity variables and nominal inflation variables that we observe in the data can be accounted for by changes in relative prices. This implies that models that break the classical dichotomy via nominal rigidities in good’s price adjustment are likely more promising than models that rely on money illusion on the part of agents.

Fourth, we found that pure inflation is partly related to monetary policy variables. The link to the growth rate in monetary aggregates is weak, but the correlation with nominal interest rates at business cycle frequencies is strong (approximately 0.5).

To conclude, economic theories have strong predictions on whether and when there should be pure inflation and what its effects would be, and discussions of monetary policy often revolve around its relation with pure inflation. However, observing pure inflation is naturally difficult, since the concept itself is more a fruit of thought experiments than something easily observed. As a result, there have been few systematic attempts to measure it in the data. The goal of this paper was to make some progress on measuring pure inflation and understanding its effects. Our estimates are certainly not perfect. We hope, however, that they are sufficiently accurate that future research can look deeper into the time-series and the moments that we provide, and that by stating the challenges and putting forward a benchmark, we can motivate future research to come up with better estimators. Likewise, we are sure that our findings will not settle the debates around the key macroeconomic correlations. Our more modest hope is that they offer a new perspective on how to bring data to bear on these long-standing questions.

Comments

[Gil]

Does that mean we should take a leaf out of Libertarian's handbook and go back to the gold standard then?

[johnchx]

This looks very interesting. I've been skeptical about whether there was any way to operationalize the intuition of "the" price level, and I'm looking forward to reading how Reis and Watson approach the problem.

[marcello]

didn't Friedman and others always state that inflation is equal to the growth of the money supply ? Mish and other bloggers repeats this over and over. But if I read the article correctly, these authors are saying that "pure inflation" is actually not correlated _at all_ with money supply. ??

what gives ?

(warning- not an economist)

[Winslow R.]

Turns quite a few assumptions upside down, including imperfect information model by Mankiw. Hope this paper holds up under closer examination....

"Therefore, measuring pure inflation and controlling for relative prices does not,
by itself, explain why the link between measures of monetary policy is weak in the shortrun
and strong in the long-run. The results in table 5 show that the two usual measures of
the stance of monetary policy only account for a modest amount of the variability in pure
inflation. This leaves open the question of what drives changes in the unit of account in
17
the United States.10"

I'd suggest deficit spending.

[reason]

Gil,
could explain your comment please. I don't understand the connection.

[reason]

Being Lazy - Winslow did you read the whole paper? If so can you tell me if they found any long run relationships between pure inflation, the CPI and PPE and money. And if so which of the three did it most closely track over the long run?

[reason]

I know I should be jumping up and down in excitement about this, because I have often been pushing the importance of relative price changes. But if you understand that I think relative price changes are important because of welfare considerations (and especially relative welfare considerations), and productivity interpretation, then you must realise that this sort of data mining doesn't really answer the questions I'm interested in.

[Gil]

nope.

[robertdfeinman]

I'd like to make a process comment.

Before the rise of the blogosphere a technical paper such as this would be seen by very few. Even if electronic dissemination paths existed, such as email lists, pre-prints or listservs the audience was small and homogeneous.

With the increasing availability of online publications and the attention being given to them by blog authors the discussion has been broadened. Those from other disciplines can now chime in and provide viewpoints that were previously lacking.

This may eventually be more important that formal refereeing which seems to be subject to bias, especially in the social sciences.

I hope this will lead to a higher quality of material being published. Fear of widespread criticism (if not ridicule) can be a powerful motivator to more careful writing.

[Winslow R.]

"Often, the three measures move together. For example, the two spikes in inflation
in the mid and late 1970s are common to all three measures of inflation. Likewise, the
disinflation of the early 1980s shows up both in core inflation as well as pure inflation,
and it is the largest contraction in pure inflation in the sample. In the 1990s, movements
in pure inflation do not match those in core PCE inflation. The disinflation of 1991-92 is
particularly pronounced in pure inflation, but not in core inflation and, in the late 1990s
and early 2000s, core inflation was low but pure inflation was particularly high."

"Core inflation is more closely tied to pure inflation
than its headline counterpart, but the squared coherence is still only around 20%-25%.
For median CPI inflation, the squared coherence is less than 20%.8"

[skeptonomist]

"The link to the growth rate in monetary aggregates is weak, but the correlation with nominal interest rates at business cycle frequencies is strong"

It seems to me that this is true of PPI inflation as well, that is long-term correlation is better. What is the correlation coefficient for dPPI vs. dM over the long term? Moreover

dPPI + dGDPreal = dM2

over 10 years is excellent as I think is well known (or should be). You can't expect a really good correlation of inflation with dM without including dGDP.

[Callahan]

How's this for a measure??, ... I was MUCH better off financially ten years ago, at nearly half the wages I earn now.

To me this is definitely INFLATION. Seems those cheap chinese products don't help me much. It's the high cost of health care, gasoline, natural gas, food, and nearly every other g.d. necessitiy.

[skeptonomist]

I wrote "PPI" above when I meant "CPI-U", although I suspect the results would be similar for PPI.

[skeptonomist]

"...at business-cycle frequencies, the correlation with nominal interest rates is around 0.5, confirming previous findings on the link between monetary policy and inflation."

No big surprise here - the Fed closely tracks inflation (if it is not trying to counter a recession). But this correlation is excellent on the scale of a month or two (depending on maturity), let alone "business-cycle frequencies" (if that means long-term).

[reason]

Thanks Winslow. It would be nice to see a plot of the cumulated trends. It sounds to me like something has broken around 2000.

[Winslow R.]

This is from an earlier paper by the same authors...

"It is possible that when the
relative price of apples and oranges changes, makers of fruit salad change the amount of
each fruit in the salad in a way that is missed by the statistical agency’s price collectors.
If they mistakenly record the change in price as if the salad was the same, then the task of
separating changes in the numeraire from changes in relative prices is impossible. To
make any progress, we must assume (or hope) that this is not the case."

http://www.princeton.edu/~mwatson/papers/npi_may_2007.pdf

I'd also have a concern that the quality of the oranges may change with the price collectors seeing it as the same salad.