CATS Cointegration Analysis:
CATS (Cointegration Analysis of Time Series) is a set of cointegration analysis procedures for use with our RATS software program. CATS was written by Jonathan G. Dennis, Katarina Juselius, Søren Johansen and Henrik Hansen of the University of Copenhagen.
CATS provides a wide variety of tools for analyzing your data and choosing and testing a cointegration model. The program is almost completely menu- and dialog-driven. You simply need to create a "set-up" file which defines your data and sample period and executes the CATS procedure, and then run this file in RATS. From there, you select the desired operations from the CATS menus, and CATS will prompt you for any needed input.
The CATS 2.0 package includes the CATS procedure on CD and a completely revised 200-page manual describing the econometrics of the cointegrated VAR model and how to interpret the output. All features of the program are illustrated by a worked example. The manual also includes a technical appendix describing the mathematics of CATS. Sample data and set-up files for the illustrative examples are also included.
What's new in CATS Version 2.0?
Version 2.0 is a major update to CATS that introduces significant new econometrics capabilities, a re-designed and expanded user interface, and a new, significantly expanded User's Manual.
New Econometrics Features
- Bartlett small-sample correction of the tests for the cointegrating rank and hypotheses on Beta.
- A new "CATSmining" automated model-selection procedure.
- Estimation and hypothesis testing of the I(2) model, including testing hypotheses on the multi-cointegrating relations and the I(1) relations among the system variables.
- Estimation of structural moving average models.
- System reduction tests for lag length determination.
- Missing observations in data allowed.
- Updated recursive estimation routine includes new tests for eigenvalue fluctuation, constancy of the cointegrating space and the log-likelihood function.
- Allows for "backwards" recursion for investigating parameter constancy over the beginning of the sample.
- For most model specifications, CATS now reports the correct critical values and p-values for the rank test. For other models, you can simulate the critical values using a built-in procedure.
- Includes a procedure for estimation and identification of structural moving average models.
New Interface Features
- All-new user interface, with separate menus for various categories of operations, including I(1) analysis, I(2) analysis, graphics, and automated tests.
- All model settings, including the deterministic terms and lag structure, are menu-controlled, so you can now change the underlying VAR model without quitting and re-starting CATS.
- All procedure settings, such as maximum number of iterations and convergence criteria for the switching algorithms, screen output format, and more, can be set via a "Preferences" dialog box.
- The estimated model can now be exported as a RATS "MODEL" making it much easier to compute forecasts and impulse responses.
- The graphs created by CATS can be customized.
- Output can be exported in tex or csv formats.
- Restrictions can be saved and re-loaded, making it easier to replicate analyses or continue your work at a later time.
- CATS offers the option of running in a true batch mode that does not require user interaction to generate basic output. This allows it to be used in a loop.
These features carry over from Version 1.0:
- "Batch" tests for long-run exclusion, weak exogeneity, and stationarity on all model variables (now available from the CATS menu). Also includes a test for unit vectors in Alpha, which corresponds to testing if the cumulated disturbances of any of the variables do not enter the common trends.
- Support for partial systems, models with structural breaks, and various forms of dummy variables.
- Multivariate and univariate tests of the estimated residuals.
- Recursive estimation for assessing constancy of the estimated model parameters, including tests for constancy of the estimated eigenvalues, the cointegrating space, the log-likelihood function, the parameters of an identified system, and the adequacy of one-step-ahead predictions.
- Options for testing hypothesis on the long-run relations in Beta as well as on the adjustment coefficients in Alpha.
- Choice of normalization for each cointegrating vector (CATS 2 simplifies this by suggesting default choices).
- Estimation of the parameters of the moving average model, e.g. the long-run impact matrix C and the loadings to the common trends (with asymptotic t-values).
- A large variety of preset graphics illustrating various key aspects of the estimated model.
New Cointegration Textbook
CATS 2.0 was developed in conjunction with the creation of a new textbook by Katarina Juselius, entitled The Cointegrated VAR Model, from Oxford University Press. Although the book is certainly not required to use CATS, we think anyone interested in cointegration analysis, and particularly anyone using CATS to do cointegration analysis, will find it extremely helpful.