Estimating Market Power and Strategies

Book Reviews wide range of users and to promote the use of MATLAB among statisticians and other data analysts. After their major success with its first edition, the authors now present a second edition with many additions. The volume now comprises 15 chapters covering almost all key statistical tools that are vital for sound inferential data analysis. After a succinct introduction, the book teaches working with probability concepts, sampling concepts and the generation of random variables. It then provides tools for exploratory data analysis. Following these basic tools, it introduces sophisticated tools such as Monte Carlo methods for inferential statistics, data partitioning, probability density estimation, supervised and unsupervised learning, parametric and non-parametric models, Markov chain Monte Carlo methods and spatial statistics. The coverage of each topic is comprehensive and comprises wellillustrated examples and formulae. Besides, it also includes many useful appendices. These appendices provide information on basic concepts related to MATLAB, and a partial list of functions in the ‘Statistics toolbox’, version 6.0, that is available for purchase from MathWorks. It also provides the list of the ‘Computational statistics toolbox’, a downloadable toolkit from Web sites http:// lib.stat.cmu.edu, http://www.crcpress. com/e-products/downloads/ or http:// www.pi-sigma.info/. These appendices also provide toolboxes for exploratory data analysis along with a brief description of various data sets that are used in the book. Moreover, a detailed bibliography and exercises at the end of each chapter add to the usefulness of the book in understanding ‘How to work with MATLAB’ effectively. I had little experience in working with MATLAB and I wished to learn how to use this tool in deriving meaningful inferences from data. This book has helped me to learn how to implement Monte Carlo methods in inferential analysis and parametric and non-parametric models. A key point of the book is the availability of MATLAB code, example files and data sets from the book’s Web site (http://lib.stat.cmu.edu or http://www.crcpress.com/e-products /downloads/). Such accessibility of relevant information helps one to get in tune with this powerful software. Moreover, the authors tried their best to keep theoretical aspects to the minimum possible—a feature that should attract a wide readership. Nonetheless, to take full advantage from this book the reader should devote quite some time on practical sessions. A major feature of the book lies in enabling readers to learn easily about the practical implementation of various methods. In my opinion, it is one of the best books aimed at facilitating learning of MATLAB. Besides statisticians, 943 this book has equal potential for teaching nonstatisticians about the workings of MATLAB. Therefore, it should be useful to researchers across all disciplines. I recommend this book particularly to those who are interested in learning how to use MATLAB in computational statistics and data analysis. Varinder Jain Centre for Development Studies Thiruvananthapuram E-mail: [email protected] Estimating Market Power and Strategies J. M. Perloff, L. S. Karp and A. Golan, 2007 Cambridge, Cambridge University Press 340 pp., £45 (hardbound), £19.99 (softbound) ISBN 978-0-521-80440-0 (hardbound), 978-0-521-01114-3 (paperbound) This book is aimed at graduate students and researchers in the fields of econometrics and industrial organization. It focuses on developing techniques to carry out the empirical work that is needed to support the economic theory analysis of competition, particularly focused on the analysis of mergers and for competition cases. The book sets out various techniques for estimating market power (the ability to sustain prices profitably above cost) and strategies that are followed by competing firms (the game theoretic plans used by firms to compete with rivals). Although the bulk of the text concentrates on the theory that is considered and developed in the book, approximately 10% of the book is given over to questions and answers at the end of each chapter, which aid understanding. The inclusion of case-studies also helps to contextualize the theory—two cases are considered: airlines and cola drinks. Both data sets have been used in previously published studies, which allows the authors to build on existing work to demonstrate new theory (which is clearly the intention, rather than to provide a current assessment of market conditions). The case-studies also note a practical problem which is common to much econometric analysis—the availability of data— whereas marginal revenue is relatively easy to obtain (being closely related to the price of a good), marginal cost is more difficult to uncover and often must be estimated. Supplementary to the main text, there is a selfcontained statistical appendix that concentrates on the ‘information theoretic’ methods that are used in the book—an approach that is suitable for situations where data are sparse or aggregated and where 944 Book Reviews the data generation process is unknown. These entropy methods aim to make the best possible use of available data and use an ‘information criterion’ to choose one of the infinite distributions that are consistent with the observed data in situations where the data are not sufficient to allow estimation of all parameters. The text assumes a high level of econometric knowledge, and, although the authors do provide a short summary of relevant economic theory, statistical readers without prior training in economics would probably find it necessary to undertake additional study to benefit from the technical material. However, the book is extremely well referenced throughout (with a comprehensive 13page bibliography), so any reader needing further guidance would have no difficulty in sourcing relevant material. This book would be of interest to those readers wishing to advance their knowledge and understanding of how econometrics can be used to investigate questions of market power. Judith Corbyn London Clinical Prediction Models—a Practical Approach to Development, Validation and Updating E. W. Steyerberg, 2009 New York, Springer xviii + 500 pp., £53.99 ISBN 978-0-387-77243-1 The development and application of prediction models is often suboptimal in medical research. This is the motivation which led Dr Steyerberg to write this well-structured and highly didactic book, which is aimed mainly at epidemiologists and applied statisticians. Clinical Prediction Models consists of four parts. Part 1 is a simple and effective introduction to study designs and statistical techniques for the development of continuous, categorical, ordinal and survival prediction models. Part 2 argues for the use of a seven-step procedure to model development: step 1, data inspection; step 2, coding of predictors; step 3, model specification; step 4, model estimation; step 5, model performance; step 6, model validation; step 7, model presentation. Part 2 contains very good sections on missing data, the bootstrap, shrinkage and penalization which purposely avoid the mathematical details that too often discourage the use of these important methods in medical research. Part 2 offers also an excellent discussion of calibration and discrimination, two aspects that are often confounded or wrongly employed in medical research. Part 3 gives a detailed account of why most clinical prediction models fail to work outside the context in which they were developed and offers a reasoned and practical approach to testing external validity. Part 3 comprises also two sections on the updating of prediction models for new settings; these are particularly relevant, not only because this topic is rarely mentioned in other textbooks but also because, practically, it is often more important to test or refine previous models than to develop new models. Part 4 analyses two well-known clinical studies as case-studies for model development. Such casestudies will help the reader to understand how to implement in practice the seven-step procedure for model development that is proposed by the author. In keeping with the applied nature of the book, data sets and R code for most of the examples and for all the case-studies can be downloaded from the associated Web site (http://www.clinical predictionmodels.org). In my opinion, Clinical Prediction Models provides a very good intermediate level treatment of model development, validation and updating applied to medicine; it also bridges the gap between basic regression textbooks, which do not discuss prediction models in detail, and more advanced books on model development, which are often avoided by the average applied statistician or epidemiologist because of their mathematical detail. Giorgio Bedogni Liver Research Centre Trieste E-mail: [email protected] Introduction to Nonparametric Estimation A. B. Tsybakov Berlin, Springer 214 pp., £47.99 ISBN 978-0-387-79051-0 My reaction on seeing the title of this book was ‘Nonparametric estimation of what?’. The answer is densities, regression functions and some closely related concepts such as Gaussian white noise models. As is often the case ‘Introduction to’ in the title may be interpreted here as ‘a fairly advanced treatment of’. The book is an updated translation of a French version that was published in 2003 and is based on lecture notes for a postgraduate course. Owing to the nature of their training and work demands applied statisticians often do not have the mathematical background or time to cope easily with the advanced mathematics that are needed by, or at any rate used by, many theoretical statisticians.