Sunday, November 25, 2012

Econometric Modelling With Time Series

That sounds like a snappy title, but it's been taken already!

There's a new econometrics book that's about to be released that looks really interesting. It's titled, Econometric Modelling With Time Series: Specification, Estimation and Testing. To be published by Cambridge University Press next month, this volume caught my eye, not only because of its title, but also because one its co-authors is a former Monash U. colleague of mine, Vance Martin (now at the University of Melbourne). Vance is joined by co-authors Stan Hurn and David Harris.

The publisher's "blurb" reads as follows:

"This book provides a general framework for specifying, estimating, and testing time series econometric models. Special emphasis is given to estimation by maximum likelihood, but other methods are also discussed, including quasi-maximum likelihood estimation, generalized method of moments estimation, nonparametric estimation, and estimation by simulation. An important advantage of adopting the principle of maximum likelihood as the unifying framework for the book is that many of the estimators and test statistics proposed in econometrics can be derived within a likelihood framework, thereby providing a coherent vehicle for understanding their properties and interrelationships. In contrast to many existing econometric textbooks, which deal mainly with the theoretical properties of estimators and test statistics through a theorem-proof presentation, this book squarely addresses implementation to provide direct conduits between the theory and applied work."
These authors really know what they're doing, so I'm looking forward to getting hold of a copy of their book.


© 2012, David E. Giles


  1. This does look very useful. There's a draft online at

  2. Dear Prof Giles,

    I read a paper with the following justification for using non-stationary data in a time series regression,

    "The typical solution to avoid spurious regression ... is differencing the series until we get a non-stationary series, I(0). However, if either the dependent or the independent variable or both is stationary, then the regression is misspecified by differencing the series. By first differencing, we impose the constraint that the parameter on the lagged variable is unity, which may not be true if the series is stationary. In such circumstances, including lagged values of the dependent and independent variables as regressors helps avoid the problem of spurious regression. In this case, a set of parameters exists for which the error term is stationary and the t-statistics for the individual coefficient estimates will have the usual asymptotic normal distribution."

    Sorry for the lengthy para.
    Would you comment on this please?

    Thank you!

    1. Montero - thanks for the comment. The key part of your quote seems to be that one or both of the variables is already stationary. In that case, differencing the stationary variable(s) will result in over-differencing. An over-differenced series is still stationary, but it isn't I(0).

      If both series are non-stationary (say, I(1)), then the quote does not apply. In that case, differencing to make them I(0) is appropriate unless they are also cointegrated. In the cointegrated case, we have 2 options: fit the model in the levels of the data (this will be the long-run cointegrating relationship); or fit an error-correction model - this uses the differened data, the lagged differences, and an error correction term.

      I'd be interested to know where the quote was from. Commenting on something out of context is always a challenge!


  3. Dear David,

    Thanks for the response.Yes they have the dep var stationary and some indep vars non-stationary-which might make the quote right. I know that quoting without writing the source is not the best thing to do. However, it was a manuscript that I was invited to revise.