The RATS Software Forum

TomDoan wrote:Could you cite a specific example?

Hi, Tom. Thank you for your quick response.For example, 4.4 Application 1 from pp 78-81 of Kim and Nelson, "State-space Models
* with Regime Switching", the reported p=0.9008 and q=0.7606 in the text book. But, by the example code, p= 5.880e-013 and q=0.02458.

What does the delta method have to do with that? Those are fundamentally different point estimates. I have no idea where they got their results---the likelihood maximizer with that data set is, in fact, with a zero transition probability out of the "recession" regime which is what RATS is estimating. If you do the same estimation with Hamilton's original data set, the model reproduces Hamilton's results whether you use the logistic or direct estimation. It's possible that the data set provided by K&N isn't the actual one used in the book, since the log likelihood doesn't match when you feed in their estimates.

Note that the empirical work done for the book is very sloppy, so you might want to read the text but ignore the results.

TomDoan wrote:What does the delta method have to do with that? Those are fundamentally different point estimates. I have no idea where they got their results---the likelihood maximizer with that data set is, in fact, with a zero transition probability out of the "recession" regime which is what RATS is estimating. If you do the same estimation with Hamilton's original data set, the model reproduces Hamilton's results whether you use the logistic or direct estimation. It's possible that the data set provided by K&N isn't the actual one used in the book, since the log likelihood doesn't match when you feed in their estimates.

Note that the empirical work done for the book is very sloppy, so you might want to read the text but ignore the results.

Thank you for your explanation .