It is not evident for me that we should do anything more than providing support for missing observations in Dynare.
Mixed frequency filtering is a modeling issue of the responsibility of the user. Having missing observations supported makes it possible (easy?) to do it in Dynare as it is.
Specifying how the data of different frequencies intermix is highly model dependent and I don't see what kind of additional standard tools (dedicated functionality) we could provide.
Best
Michel
Johannes Pfeifer writes:
Hi,
I agree with Stéphane. The current Kalman filter implementation treats missing observations as unobserved states. This is standard and the book reference is just one of the standard references that can be cited there.
As Stéphane says, mixed frequency filtering is something different and not (yet) implemented as it requires explicitly setting up observation equations within the model-block to link the different frequencies.
That being said, you might already be able to do mixed frequency filtering and smoothing in Dynare if you can nest it within the standard Kalman filtering framework. But there is no dedicated functionality and syntax.
Best,
Johannes
Johannes Pfeifer Department of Economics University of Mannheim L7, 3-5, Room 242 68131 Mannheim Germany +49 (0)621 181-3430
Am 19.02.2014 11:11, schrieb Stéphane Adjemian:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256
Hi,
I never had a chance to the read the book mentioned by Sébastien... And I won't buy it. So I do not know if we follow the methodology described by Durbin and Koopman (kalman filters with missing observations were implemented in Dynare in 2010 or even before I do not remember, and there is an old literature about this problem). Adding NaNs in the datafile is not the whole story if you want to mix frequencies. You also need to adapt the model. For instance, if your model is quaterly but you also have yearly observed data, you need to add an equation to relate the yearly data to the quaterly variables (except if the yearly variable is a stock which is not the case in general). More generally you need to write the model in the highest frequency and add measurement equations in order link the variables with different frequencies.
Best, Stéphane.
On 19/02/2014 10:43, Sébastien Villemot wrote:
Hi Tobias,
Le mardi 18 février 2014 à 17:48 +0000, Tobias Cwik a écrit :
I hope this email finds you well. I don’t know if you remember me, I was working for Prof. Wieland in Frankfurt. I heard that the Dynare team is working on mixed frequency filtering and smoothing in Dynare. Do you know if there is anything implemented in Dynare already?
Yes, Dynare can handle missing observations in the dataset. Which means that if you have a quarterly series and a annual series, you can mix them, by putting missing observations for the annual series in Q2, Q3 an d Q4. Dynare will be able to run the filter, using the methodology described in Durbin and Koopman (2012): "Time Series Analysis by State Space Methods", second edition, Ch. 4.10.
I put Stéphane in CC, he is the one who programmed this.
Best,
Dynare Team -----BEGIN PGP SIGNATURE----- Version: GnuPG v1 Comment: Using GnuPG with Icedove - http://www.enigmail.net/
iF4EAREIAAYFAlMEgzIACgkQrQOaeEU5AZe7WAD/SYc2SvLAqUCzxJfnkO5TudRv Ld26aIWc4Q5GhK02ISAA/R+bmQsxavC5l34BUFkALFOGbN0MljZNuvzYhPcqnUlo =8Vok -----END PGP SIGNATURE-----
Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev
Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev