Review of
An
Introduction to the Structural Econometrics of Auction Data
By Harry J. Paarsch and Han Hong
with contributions by M. Ryan Haley
Cambridge, MA: The MIT Press, 2006.
By
J.C. Herbert Emery
Department of Economics
University of Calgary
Harry Paarsch and Han Hong have prepared an impressive introduction to the
structural econometric approach (SEA) to analyzing field data from auctions.
The authors have been pioneers of this subject with over two decades field
research on auctions, and the book provides an ideal showcase for their body
of work. Auctions are of considerable interest to economists interested in
the mechanics of price formation; as an application of game theoretic models,
and as a source of information to guide the design of optimal selling
mechanisms. While considerable research effort has been put towards
developing the theory of auctions and the predictions that that theory can
deliver, the econometric study of auction data is a frontier area for
research. This is an impressive book that is intended as a reference, if
not a text, for graduate students, particularly in the field of industrial
organization, microeconometrics, monetary economics, and given the prevalence
of use of auctions as market mechanisms for the sale of primary commodities,
natural resource and agricultural economics.
The authors have put considerable effort into the pedagogical design of the
book. In addition to the development of "logically-consistent" notation
(described in detail on pages xix-xxii), the book focuses its coverage on the
structural econometric approach to the analysis of auction field data (SEAD)
in the context of the independent private values paradigm (IPVP). In the
former case, the authors state that SEA is internally consistent, unlike the
primary alternative to it, the reduced form approach, since the
economic-theoretic model is mapped directly to an econometric specification.
In addition, the authors believe that SEAD has produced "some of the most
exciting applications of theoretical econometric research in the past decade".
In the latter case, the IPVP provides the simplest case to investigate and
present the empirical techniques employed in SEAD. The benefit of these
choices is a book that presents a wide range of ideas in an intuitive and
sequential manner. The cost of this approach is that, superficially, the book
appears to have a narrow focus and subject. As such, the preface of this
book is extremely important to read since it is there that the authors make
their case for why the pedagogical benefits of their approach dominate the
opportunity cost of the narrower focus. The authors expect that the reader
will be able to see how these applications to the study of auction field data
apply to other contexts without being shown. To support the likelihood of
this outcome, the authors structure the book so that the motivated reader can
work through empirical exercises that will allow them to learn about the
application of the empirical methods, and the relative merits of the
alternative reduced form approach, in a self-directed way. From this
perspective, the inclusion of detailed appendices on statistics, probability
theory, estimation and numerical methods is notable, particularly as they
follow the standardized notation conventions used throughout the book. The
authors have also gone to great lengths to develop the empirical exercises
found at the end of each chapter, with solutions, sample data sets and MATLAB
code (on the accompanying CD). The empirical exercises have been "field
tested" as Paarsch has used them when he teaches his own courses on auctions.
The coverage of topics builds in a logical fashion and the progression of
material leads the reader at a reasonable pace. Chapter 1 discusses the
interesting features of auctions for analysis and the basic intuition for the
estimation problems that must be addressed. Chapter 2 provides an overview of
single object auction theory from the perspective of Harsanyi's theory of
non-cooperative games with incomplete information. In an intuitive and
readable fashion, the chapter presents well-known results from the theory and
outlines the major questions of interest. Chapter 3 introduces the structural
econometrics of auction data in the context of the Vickrey and English auction
settings. The chapter demonstrates the inclusion of covariates and how to
address statistical complications associated with minimum bid prices and
endogenous participation in auctions. Chapter 4 presents many of the same
empirical issues as Chapter 3 but in a more complicated setting. Where the
dominant strategy equilibrium bid functions were only functions of bidder
valuations, in first-price, sealed-bid and Dutch auctions, the equilibrium bid
functions are non-linear functions of bidder valuations, the number of bidders,
and the cumulative distribution function of valuations. Unlike in the Vickrey
and English auction settings examined in Chapter 3, risk aversion is also an
important consideration for equilibrium bidding behavior in first-price,
sealed-bid and Dutch auction settings. Where the authors assess that the
non-linear relationships in the bid function introduce a number of
"interesting" computational and econometric issues, for the non-specialist,
they represent a significant increase in complexity with respect to learning
and understanding the application of SEA to auction field data. Chapter 5
extends the treatment of material in Chapter 3 to the setting of a multi-unit
auction. In this setting the scope for strategic behavior on the part of
bidders makes the determination of equilibrium bidding behavior much more
complex but also more interesting than in the single unit auction setting.
Finally, in Chapter 6, Paarsch and Hong briefly present directions for future
SEAD research.
While this book is titled an "Introduction" to SEAD, the intended audience for
this book are individuals with a strong background in econometrics (the reader
has had courses in probability, statistics, and econometrics at the level of
Russell Davidson and James G. MacKinnon (2004) Econometric Theory and
Methods) and some ability to program using C or FORTRAN, at least with the
aid of a programming environment like GAUSS or MATLAB. Compared to the reduced
form approach, SEAD requires advanced training and abilities in econometrics.
To put it another way, anyone prepared to work through this book is already
sufficiently tooled up to perform, and publish, reduced form analysis of
data. Working through the book represents a significant investment to be able
to apply structural econometric analysis. From this perspective, "entry costs"
for reduced form analysis are much lower than for SEA and many applied
economists will likely be content to focus their efforts in that direction.
To convert more applied economists to the SEA of data, a stronger statement of
the gains from SEA over the reduced form approach for the analysis of auction
field data would be useful. This is no small matter. The persistence of the
popularity of reduced form econometrics approaches reflects, at least in part,
the belief (or rationalization) that the results obtained for a relatively low
cost are "not too bad".
Unfortunately, the authors do not take on this perception of the utility of
reduced form econometrics which is unfortunate since it leaves the magnitude
of the contribution of SEAD ambiguous. Paarsch and Hong state that they focus
on SEA exclusively since reduced form analysis is not "internally consistent"
with the economic theory that motivates the econometric specification (page
xvi). The authors also state that, despite their strong opinions on the
relative merits of SEA versus reduced form approaches, "we have explicitly
chosen not to judge the relative merits of the various approaches. Instead we
have provided the reader with a variety of different empirical exercises a
diligent learner will be able to form his or her own opinion concerning the
merits as well as the difficulties of the two approach." (page xvii) In my
opinion, it would have been nice to see the sermon in writing so that more
researchers could learn the reason why the SEA can give us so much more than
the reduced form approach. This would have been more consistent with how the
authors address debates within the community of structural econometricians
such as the use of parametric versus non-parametric approaches. In that
particular debate, the authors do a great job explaining the trade-offs
between the approaches. In this context, on page 105 the authors state that
"parametric models are an approximation to the real process. As in all
empirical work, the researcher must decide where to make the approximation.
We assume that when researchers constrain themselves to parametric models,
they understand the approximations present in their research."
In the context of the more general debate over the SEA versus reduced form
approaches to the analysis of auction field data, is the advantage of SEA
largely qualitative (it is less of an approximation than the reduced form
approach which is intrinsically desirable), or are the advantages
quantitatively important (estimates of interest for policy making are
substantially different)? A particular place where the authors could have made
the point in strong way would have been in Chapter 3 where the discussion of
Paarsch's work on British Columbia timber auctions is presented. Paarsch's
calculation of the optimal reserve price is an interesting policy analysis in
its own right, but for the purposes of a book intended to teach the
application of SEA, it would have been interesting to see the optimal reserve
price that would have been implied using a reduced form approach. Comparing
the policies implied by the two approaches using the same data would give the
reader a sense of what is gained from the SEA in return for the added effort
and cost. In my opinion, addressing this debate directly would create
additional demand for this book as it would then be written to convert more
applied researchers to use the SEA as opposed to marketing to the converted.