Time Series Analysis and Forecasting, 2/e (Hardcover)

Time series analysis and forecasting have become increasingly important in many fields of research and application. The goal of this book is to distill and integrate current research results into cohesive and comprehensible methodologies, and to provide a streamlined approach to time series analysis and forecasting.

This book is written in a manner which is suitable for a two-semester course in applied time series analysis and forecasting. It can be used in a one-semester course by excluding more advanced topics in the later chapters. The material is targeted to advanced undergraduate and graduate level students in various fields of study (e.g., business, economics, statistics, engineering, medicine, social sciences, environmental sciences, politics, etc.). With its emphasis on practicality and applicability of methodologies, the book is also useful as a reference source for researchers and practitioners in time series analysis and forecasting. While traditional Box-Jenkins methodology emphasizes the exploration and exploitation of serial correlation in time series, the progressive topics on modeling automation, outliers, heteroscedasticity, and nonlinearity are also of great importance. This book places greater emphasis on these newer issues as well as multivariate approaches in time series analysis and forecasting.

Chapters 2 and 3 cover traditional Box-Jenkins ARIMA methodology for non-seasonal and seasonal time series. These two chapters discuss a vast amount of material in a concise manner. In Chapter 4, an effective approach is introduced to automate the task of ARIMA model identification. This approach can be used in conjunction with transfer function models (Chapter 5) to accomplish automatic transfer function modeling. The idea of automation in model building can be further extended to vector ARMA models and simultaneous transfer function (STF) models discussed in Chapters 14 and 15.

Systematic and non-systematic disturbances cause difficulties and inaccuracies in time series modeling and forecasting. Chapter 6 addresses systematic disturbances caused by calendar variation in monthly time series. Chapter 7 discusses non-systematic disturbances relevant to intervention analysis and outlier detection. As an application of outlier detection and estimation, data mining on time series is discussed in Chapter 9. For completeness, traditional forecasting methods using exponential smoothing are addressed in Chapter 8.

Chapter 10 discusses the use of power transformation for modeling and forecasting of certain heteroscedastic time series. In Chapter 11, conditional heteroscedastic models are introduced to address analysis for time series with heteroscedasticity in a different manner. This class of models includes the well-known ARCH/GARCH models.

Traditional time series models often assume that the relationships within the series and the relationships between series are linear. This assumption may not always be appropriate. In Chapter 12, time-segmented and value-segmented (threshold) time series modeling is presented for handling such nonlinear or non-homogenous relationships. Additional topics on nonlinear time series models such as TAR models are discussed in Chapter 13.

Univariate ARIMA models can be generalized to cover multivariate time series. The theory and application of vector ARMA models for multivariate time series analysis are discussed in Chapter 14. Similar to vector ARMA models, single-equation transfer function models can be generalized to multi-equation transfer function models for multivariate time series analysis and econometric analysis. The theory and application of simultaneous transfer function (STF) models are discussed in Chapter 15. As a special application of vector ARMA models, a causality test using this class of models is discussed in Chapter 1