Hello!
I am estimating a standard NK model on an observable x, via Bayesian estimation, using Metropolis Hastings algorithm. The observable x is a series separately computed by me.
I would like to test how “good/realistic” x is. Is there a test (eg. likelihood ratio) to compare the model estimated on x to the same model estimated on some other data or to another solution of the same model, to test statistically my observable x?
Thank you very much,
Luisa
Dear Luisa,
it is not conceptually clear what you are trying to achieve. Usually, you take the data as given and compare models on that data. That would be Bayesian model comparison.
Dear Professor,
I estimated a data series applying a technique from the IO literature and my objective is to analyze the macroeconomic implications of this series (ie. does it imply a realistic pattern of inflation, output gap etc). The goal is to test whether the IO technique is valid. For this purpose, I estimate the NK model on the data series.
Since the results I get are a product of both the NK model and the data series, I am interested in a comparison with the same model, not affected by the data series which I am trying to test.
Thank you.
Sorry, but I still don’t understand the goal. Usually, you would use a theoretical model, simulate artificial data, and then text whether the series x you compute using your technique recovers the truth.