View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

For release on delivery
7:30 p.m. EDT
September 29, 2005

Inflation Modeling: A Policymaker's Perspective

Remarks by

Donald L. Kohn

Member

Board of Governors ofthe Federal Reserve System

at the

Quantitative Evidence on Price Determination Conference
Martin Building
Washington, D.C.

September 29,2005

An occasion like this one is a natural opportunity to reflect on how policymakers'
understanding of the inflation process has progressed over time. l Clearly we have come a long
way since the early 1970s. Most important, we have absorbed the central lesson of Milton
Friedman's 1968 address to the American Economic Association--that any tradeoff between
inflation and unemployment is only temporary because of the dynamic nature of expectations.
We have also taken on board the practical application of this lesson that monetary policy must be
vigilant about anchoring inflation expectations. Operationally, maintaining price stability
requires abiding by the Taylor principle of raising nominal interest rates more than one for one in
response to movements in inflation, especially those movements perceived as persistent. It also
requires that policy tighten or ease systematically to bring aggregate demand in line with the
economy's productive potential, not only because output stabilization is a policy objective in its
own right but also because such actions help to head off undesirable changes in inflation down
the road.
These basic precepts, embraced by central bankers everywhere, have almost certainly
contributed to the improved performance of inflation over the past decade or two, and this better
price performance probably has helped to damp business cycles. Of course, economists and
policymakers still have a great deal to learn about the interactions of monetary policy, the real
economy, and inflation. But as the conference papers illustrate, we are making progress.
Still, when it comes to inflation modeling and policymaking, as my grade school report
cards used to say: There is room for improvement. Specifically, I wonder whether current
inflation research adequately addresses the questions that bedevil me as a policymaker. Tonight
I cannot resist the opportunity presented by having access to a captive group of researchers to

I The views I express here are my own and not necessarily those of other members of the Federal Open Market
Committee. Flint Brayton and David Reifschneider, of the Board's staff, contributed to these remarks.

-2share these concerns with you, with the intention of providing some constructive suggestions for
future work.
What properties am I looking for in a model of inflation and the economy overall? First,
the model should provide a coherent analytical framework that the policymaker can use to
interpret incoming data and to choose a proper policy response. Second, the model should
provide an accurate empirical description of the economy as it relates to both forecasting and the
influence of policy on the outlook. My experience on the Federal Open Market Committee
(FOMC) as a consumer of model-based forecasts and analysis suggests that work remains to be
done on both fronts--and, in particular, on meeting both objectives with the same model.
As regards the first property, the conceptual--albeit informal--framework that many
policymakers like to use is largely "bottom up" and features costs and expectations. It starts with
wages and the prices of other inputs into production, and after taking into account productivity, it
sees prices set as a markup over unit costs. Wage determination plays a key role in this
framework and is influenced by such factors as inflation expectations, productivity, and labor
market slack; the markup, too, is important because it varies over time depending on changes in
the competitive environment, expected future costs, and other factors.
At first glance, the empirical structural models currently favored in academic research
and discussed at this conference--such as the New Keynesian Phillips curve--appear to conform
to the informal policymaker model: Expectations formation is dealt with formally, and price
inflation, at least in some empirical representations, is directly tied to measures of unit labor
costs. On closer inspection, however, one sees that the specifications of these models typically
ignore important factors bearing on the inflation outlook.

-3One category of neglected factors is price shocks--changes in the levels of key inputs,
such as energy or imports. Policymakers spend a great deal of time discussing the circumstances
under which such shocks can lead to persistent changes in the rate of inflation. Yet, despite their
historical importance for aggregate inflation, energy prices, for example, are controlled for in
only one of the structural models discussed at this conference. And this importance is not
necessarily a concern of the past: Prices for oil and natural gas have soared since 2003, directly
boosting the energy component of the consumer price index as well as raising the production
costs, and ultimately to at least some degree the prices, of non-energy goods and services. As a
policymaker, I can assure you that any model of inflation that did not take account ofthese
effects, and how they might or might not affect ongoing rates of inflation, would have been of
little practical use to the FOMC over the past few years.
Reduced-form regressions suggest that the response of core inflation to energy prices has
diminished over the past twenty years. Does this smaller response reflect a change in the
expectations formation process that has come about because the public perceives that inflation
will remain low, perhaps because the monetary authority is now seen to be more vigilant in
reacting to price pressures? Or does it reflect a reduction, from the late 1970s until a couple of
years ago, in the persistence of energy price movements that has prompted firms to be less
worried about passing temporary cost increases onto customers? Determining which of these
explanations is most important is a critical issue for monetary policy right now, when futures
markets indicate that people expect the current elevated price of energy to persist.
But economists are not well positioned to provide much evidence on this issue, given the
relative paucity of empirical work on expectations formation. Certainly the standard approach
used to estimate structural models is not that helpful because it simply assumes an answer--

-4-

rational expectations, typically accompanied by full central bank credibility. But how well does
this assumption match reality? True, financial market participants do seem to respond to
incoming economic data in a generally forward-looking and logical manner, but we also observe
asset price levels, volatility, and implied risk premiums that are sometimes difficult to
understand. And I doubt that any central bank has achieved perfect credibility in the markets.
Moreover, it is not obvious that investors' expectations always line up with those of households
and firms, the ones that should matter for wage and price setting. We really know very little
about the precise manner in which these agents form their beliefs about the future, in part
because of a lack of comprehensive data on expectations. Nonetheless, economists are certainly
grappling with this issue, as evidenced by the growing interest of late in models that incorporate
rational inattention, sticky information, learning, and imperfect credibility. Behavioral
economics, with its focus on how people perceive and act on information in making decisions,
may also provide some insights into the modeling of expectations.
If plausible departures from rational expectations and full credibility are empirically
verified, then our structural models need to take that into account. A new set of questions would
then be on the table in policy analysis. Among other questions that could be addressed would be
how policy actions (as opposed to inflation outcomes) influence expectations and how sensitive
Federal Reserve credibility is to short-run departures from low inflation. Such knowledge would
be extremely useful in current circumstances in gauging the scope for monetary policy to offset
the short-run output effects of higher energy prices without triggering adverse longer-run
inflation consequences.
Another area needing further attention is wage setting. My impression, reinforced by the
papers at this conference, is that many of the newer empirical structural models of price inflation

.

- 5posit a central role for real marginal cost but seem to have little to say about its determination or
that of its main component, labor compensation. This neglect may reflect the difficulty of
specifying the aggregate dynamics of an atomistic labor market characterized by pronounced
heterogeneity, significant informational imperfections, and important adjustment costs. By
comparison, these complications seem less severe in product markets, which for the most part are
more transparent and easier to analyze. Moreover, as I will discuss in a minute, empirical work
on wage determination may be especially hindered by issues of data quality.
Let me give an example of why I think it is important to have a firm grasp of how wages
are set. Many economists, both inside and outside the Federal Reserve, think that the
acceleration in labor productivity in the mid-1990s subsequently helped to restrain the rate of
price inflation. This conclusion requires that the adjustment of nominal wages to the higher
trend in productivity was slow, an outcome that put downward pressure on unit labor costs and,
hence, on prices. Its consistency with the data notwithstanding, I would be more comfortable
with this hypothesis if it were supported by a structural model of wage determination that was
firmly grounded in theory and microevidence.
As noted earlier, the design of sound structural models is only a start. As a policymaker,
I need those models to provide accurate forecasts and empirically well-grounded policy analysis.
The analysis contained in the papers by Rudd and Whelen, Kiley, and Laforte are valuable first
steps in comparing the empirical properties of alternative structural and reduced-form models.
Nonstructural specifications, despite their shortcomings with regard to the desirable
model properties I noted earlier, do have an important role to play in policymaking--in particular,
forecasting. Structural models do not as yet fully encompass the information used by reducedform specifications, and the historical relationships summarized in these latter models have often

-6-

proven to be more useful guides to future inflation developments. Such models often include
energy- and import-price tenns along with measures of slack and expectations. They infonn the
staff s judgmental inflation forecast as well as my own thinking about the outlook.
Data measurement issues add to the challenge of developing better empirical models for
policy work. One important example concerns hourly labor compensation. The measure
reported in the national accounts is often revised significantly, displays substantial volatility
from quarter to quarter, and has components that may not coincide with the labor costs relevant
to business pricing decisions. These deficiencies, among others, also affect published estimates
oflabor's share in the nonfarm business sector--the us~al measure of real marginal cost used in
empirical work. Other measures of labor compensation, such as the employment cost index,
have their own particular deficiencies and can yield conflicting signals of trends in labor costs.
An unpleasant implication of these various data problems is that they may make difficult the
development of structural models incorporating wage behavior that are reliable enough for policy
analysis, despite their attractiveness to policymakers like me.
Of course, measurement issues do not afflict only models that exploit data on labor
compensation. Many inflation models in use, both structural and reduced-fonn, include some
measure of resource utilization as a detenninant of price inflation, such as the gap between the
unemployment rate and the non-accelerating inflation rate of unemployment (NAIRU). And as
we know, the latter cannot be directly observed. Although economists have made some progress
in estimating potential output, policymakers should be cautious about responding aggressively to
estimated movements in economic slack.
Paradoxically, the ability to confinn our estimates of output relative to potential in a
timely way may have been diminished by the success of the Federal Reserve and other central

..
-7-

.
•

banks in achieving low and stable inflation. Although controversial, some evidence from
reduced-form price equations suggests that inflation has become less sensitive to economic slack
in recent years, possibly as a consequence of more firmly anchored inflation expectations.
Consistent with this evidence, the Board staffs rule-of-thumb estimate of the sacrifice ratio rose
from around 2 or 3 in the mid-1980s to around 4 currently.2 Imbalances between demand and
potential supply would thus now be slow to show through convincingly to inflation, but when
they do, they may be costly to correct.
The various specification and empirical issues that I have raised tonight are major
challenges, and based on our experience using the FRBjUS model here at the Board, they may
prove quite difficult to overcome. FRBIUS has many of the desirable attributes mentioned
earlier. In particular, its wage and price equations are variants of the New Keynesian Phillips
curve, with core specifications that are derived from cost minimization and equations that are
estimated assuming rational expectations. However, in incorporating the effects of such factors
as energy costs, import prices, and productivity into the model's wage-price block, the staff
found it necessary to sacrifice some theoretical niceties in order to get close to the predictive
accuracy provided by reduced-form models. Perhaps future researchers will be able to resolve
this tension in some clever way, but some tradeoff between purity of design and forecasting
accuracy probably will always be unavoidable. In addition, the staffs experience with FRBIUS
illustrates some of the problems inherent in using real-time wage data. Although the policy
analyses generated using the model have been quite useful from my perspective, and the model
has compiled a reasonable forecasting record overall, the accuracy of its price inflation forecasts
has suffered from their sensitivity to current readings on labor's share.

The sacrifice ratio is defined as the cumulative amount of unemployment greater than the NAIRU (in percentage
points) required to reduce the inflation rate by 1 percentage point, all else being equal.

2

..
-8These remarks have presented a long list of "helpful" suggestions for future research.
You might ask: Does successful monetary policymaking really require all this additional
knowledge? Do not central banks already have what they need to do a good job? If
policymakers continue to recognize the critical role oflong-term price stability, keep a close eye
on inflation expectations, and adhere to the Taylor principle, would not things be OK, as they
have been for much of the past twenty years?
This view strikes me as too complacent. Even if it is true that things tend to tum out OK
on average under the present state of knowledge, macroeconomic performance could be better
yet if policymakers were able to take advantage of a fuller understanding of the dynamics of the
economy. And we would do well to be cautious about attributing the good macroeconomic
performance entirely to good monetary policy. Decomposing the sources of the Great
Moderation is a difficult business, and a number of researchers interpret the evidence as
suggesting that monetary policy was not the most important factor. Luck as well as structural
changes in the economy may have had a lot to do with the current low level and apparent
stability of U.S. inflation. Ifso, and if our luck turns and we experience a series of adverse
shocks, our ability to formulate policies that deliver sound performance may depend upon a
much better understanding of the inflation process and of expectations formation.

.•