Time Series Models for Business and Economic Forecasting

Article excerpt

Philip Hans FRANSES. Cambridge, U.K.: Cambridge University Press, 1998. ISBN 0-521-58404-3. x + 280 pp. $69.95.

The main purposes of analyzing time series are to test economic hypotheses or generate forecasts. Time Series Models for Business and Economic Forecasting concentrates on the second topic on an intermediate level. It differs considerably from other econometric textbooks. Franses reviews the more recent developments in modeling time series to focus on generating ex ante forecasts: seasonal unit roots, period models, aberrant observations, and common features. For each method, intuitive motivation and practical considerations are discussed in detail, making the book very readable. Real-life examples taken from scientific applications are used to discuss important issues in applied analyzes of business, financial, and macroeconomic data.

Among the standard topics in time series analysis, the book deals with univariate and multivariate methods. The issues for the univariate approach are ordered depending on the variation importance of a time series: modeling trends, analyzing seasonality, looking for aberrant observations, specifying conditional heteroscedasticity and accounting for nonlinearity. Each issue is reviewed in a separate chapter containing an introduction and conclusion, The most important issues are considered in the multivariate framework as common features where common trends and cointegration, common ARCH, or common nonlinearity is briefly discussed. This approach reflects the author's preference and research interest. Franses believes sufficiently in the importance of these issues to specify models used for forecasting, and he stresses the significance of forecasts to assess models and to satisfy the main purpose of time series modeling.

Two chapters come before these chapters. One describes the analyzed data and their specific features; the other explains useful concepts to model univariate time series that are nonseasonal, linear, and nontrending and have constant variance. Referring to the work of Box and Jenkins (1970) these include the identification of models, model estimation, diagnostic measures, model selection, and forecasting. These concepts allow one to model the cyclical component of a time series.

Because of the book's aim to have the reader construct models to be used for out-of-sample forecasting, it does not cover forecasting with regression models that use exogenous variables. Nevertheless, determining forecasts often relies on autoregressive models. Therefore, the book requires knowledge of basic econometric methods, like regression analysis, estimation, testing, matrix calculus, and the ability to use standard econometric software programs like Eviews or RATS. Moreover, the book does not serve as the primary source of the general time series analysis theory. It does not include the rigorous formulation of the methods. But it does provide relevant references where the more technical elaboration is conducted and the most recent developed methods are applied. The author uses many plots and dozens of tables to demonstrate and evaluate the results of the forecasts. He stresses that the reader needs time to generate the forecasts, and that the modeling of aberrant data needs a priori knowledge when th ey occur. …