Applied time series modelling and forecasting
Brian Sloboda2004, International Journal of Forecasting
Sign up for access to the world's latest research
Abstract
Applied Time Series Modelling and Forecasting provides a non-technical approach to applied econometric time series models, which involve non-stationary data. This monograph emphasizes the why and how of econometric time series modeling and places less emphasis on the analytical details. Additionally, it extends the topical coverage of Harris (1995) by discussing the econometric analysis of panel tests for unit tests and co-integration as well as for financial time series data. Additionally, the authors present the latest techniques in structural breaks and season unit root testing, testing co-integration with a structural break, seasonal co-integration in multivariate models, and approaches to structural macroeconomic modeling. Chapter 1 provides an overview (or review for some readers) of the basic analytics of time series analysis, which provides the foundation for the remainder of the monograph. Chapter 2 includes details of short-and long-run relationships of the time series, with the first part of the chapter focusing on the examination of long-run relationships between the economic time series. Moreover, the discussion delves into distinguishing between stationary and non-stationary variables. Neglecting to examine such distinctions could result in a spurious regression, which might imply a statistically significant long-run relationship when no causal relationship exists. In fact, the establishment of a long-run relationship naturally leads to the concept of co-integration, which allows analysts to examine if there is a causal relationship between the economic time series. The final part studies the short-run relationships of economic time series. The discussion stresses that the estimation of short-run models can be problematic and argues that the method of differencing the data is not a good solution since this would remove information about the long-run behavior of the time series. Thus, a remedy would be the error correction model (ECM) since the ECM would contain information about the long-and short-run aspects of the economic time series. Chapter 3 presents the details of the presence of a unit root in the time series, and the discussion begins with the Dickey -Fuller (DF) test for a unit root showing that a t-test of the null hypothesis of nonstationarity is not based on the standard t-distribution. Much of the discussion deals with what elements should be included in the testing procedure. That is, whether the inclusion of the trend and the intercept (i.e., deterministic components) or just one of them would result in different results in the testing procedure. The chapter continues by going into the issues of which of the deterministic components should enter the testing process by examining a sequential testing procedure espoused by Perron. The Dickey -Fuller test is modified to examine a more complicated time series known as the augmented Dickey-Fuller, one that entails the addition of lagged dependent variables to the test equation. The discussion continues with the issues of how many lagged terms should be incorporated into the test and related issues such as the power and size properties of the augmented Dickey -Fuller test. The final part of the chapter discusses the empirical issues of seasonal unit roots, including the integration of structural breaks. Chapters 4 and 5 provide a discussion of the estimation of co-integration of single equation models (Chapter 4) and multiple co-integration models (Chapter 5). The most commonly applied method in testing for co-integration, up to the early 1990s, was the twostep estimation procedure of Engle and Granger (1987). This single-equation method for estimation of co-integration is based on the restrictive assumption of a single co-integration relationship, which is estimated via the OLS procedure. However, with the case of more 0169-2070/$ -see front matter D
Key takeaways
- Additionally, it extends the topical coverage of Harris (1995) by discussing the econometric analysis of panel tests for unit tests and co-integration as well as for financial time series data.
- In fact, the establishment of a long-run relationship naturally leads to the concept of co-integration, which allows analysts to examine if there is a causal relationship between the economic time series.
- In the last section, the monograph extends the discussion of a single co-integration model to seasonal analysis.
- Chapter 5 discusses the Johansen procedure, which models multiple co-integration methods, a maximum likelihood estimation procedure proposed by Johansen estimate long-run equilibrium relationships.
- This monograph provides a comprehensive, nontechnical introduction to econometric time series, which can serve graduate students and practitioners of time series econometrics.
