[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
st: Factor models in panel data (was: Factor models)
st: Factor models in panel data (was: Factor models)
Fri, 31 Jul 2009 14:57:22 -0400
First of all, thanks to Bob for all the references on dynamic factor models.
I think that some of these are useful for my purposes but not entirely, mainly due to the panel structure of my data.
As I mentioned earlier, I have data on domestic macroeconomic variables for 40+ countries over time (e.g., GDP growth, dollar reserves, inflation, etc.) AND data on global financial conditions that are common to all countries in my data set at any given point in time (e.g., the level of the U.S. interest rate, volatility in the EMBI index, volatility of the VIX index, etc.).
The problems I see with the implementation of a standard factor model (as implemented by -factor- in Stata) are the following:
1. I think that, as it is, this command doesn't take into account the panel structure of the database, does it? For instance, if my data has 80 quarters per country and 40 countries and it's sorted by country and time period, then the change between observation 79 and 80 for any given domestic macroeconomic variable is going to be very different than the change between observation 80 and 81, because the former two correspond to periods 79 and 80 of country 1, while the latter two correspond to period 80 of country 1 and period 1 of country 2.
2. While the domestic macroeconomic variables change by country and by time period (e.g., each country has different inflation rates in 1995 Q2), global financial conditions are the same for all countries at a given time period (e.g., the U.S. interest rate is the same for ALL countries in my data base in 1995 Q2). So, can I still put these common variables together in the same database where I have my domestic variables before I do the factor analysis?
3. Some of my variables as I use them in my regression model are in logs (e.g., log of the U.S. interest rate), or "scaled" appropriately to compare them across countries (e.g., reserves/imports, external debt/GDP, etc.)? Should I leave them like this in my factor analysis, or should I just use the actual variables I'm interested in (e.g., U.S. interest rates in levels, reserves and external debt without scaling by imports or GDP, respectively, etc.)?
Thank you very much once again.
> From: email@example.com
> To: firstname.lastname@example.org
> Date: Sat, 25 Jul 2009 02:49:33 -0400
> Subject: Re: st: Factor models
> There is a substantial literature on dynamic factor models. Mario Forni and Lucrezia Richlin have written on this in 1996 and 1998 . In September 2003, Mario Forni, Marc Hallan, Marco Lippi, and
> Lucrezia Richline published "The Generalized Dynamic Factor Model: One-sided
> Estimation and Forecasting," as an LEM working paper (2003/13). Marco Lippi and Daniel Thornton
> have written a working paper for the St. Louis Federal Reserve bank (WP2004-013a) in
> 2004, entitled "A dynamic factor analysis of the response of U.S. interest to News."
> Stock and Watson in 1989, 1991, and later have worked on this subject. In 2005 they
> published, "Implications of Dynamic Factor models for VAR analysis," in which they have dealt with dynamic factors and model identification with long and short run restrictions in VAR form. Tao Chen, Elaine Martin, and Gary Montague have written an article in Computational Statistics and Data Analysis (2009) v. 59 entitled, "Robust probabalistic PCA with missing data and contribution for outlier detection." Rangan Gupta and Alain Kabundi in 2008 (Journal of Economic Literature) have written "A Dynamic Factor Analysis for Forecasting Macroeconomic Variables in South Africa." Some
> time ago, Allessandro Federici and Andrea Mazzitelli in 2005 at an Italian Stata User's Group presented "Dynamic Factor Analysis with Stata," based on the Coppi and Zannella(1978) and later work.
> In 2004 Ben Bernanke has written on factor analysis with BVAR models. Siem Jan Koopman has also incorporated PCA in his state space models. James Hamilton, in his 1994 classic, Time Series Analysis, has shown how state space models lend themselves to dynamic factor analysis.
> And these are small sample of the articles that have emerged on this subject. Andrew Harvey in 1989 in "Forecasting, Structural time series models, and the Kalman Filter" has also shown on these factors are the unobserved signal components extracted from the data generating process.
> These are some of the articles on the subject.
> Bob Yaffee
> Robert A. Yaffee, Ph.D.
> Research Professor
> Silver School of Social Work
> New York University
> Biosketch: http://homepages.nyu.edu/~ray1/Biosketch2009.pdf
> CV: http://homepages.nyu.edu/~ray1/vita.pdf
> ----- Original Message -----
> From: kokootchke
> Date: Friday, July 24, 2009 7:49 pm
> Subject: st: Factor models
> To: statalist
>> Hi everyone.
>> I would like to know if anyone could provide some references to
>> factor models (latent factor models, dynamic factor models, etc.).
>> I am studying the effects of domestic macroeconomic variables vs.
>> global economic conditions on bond spreads and I believe that I could
>> use factor models to improve my analyses.
>> I have never used these models before and have found some papers by
>> Stock and Watson, Lippi et al., etc... which are useful... but I would
>> like to read something a bit easier to get my feet wet, and then start
>> building up from there.
>> Thank you very much!
>> Best regards,
>> NEW mobile Hotmail. Optimized for YOUR phone. Click here.
>> * For searches and help try:
>> * http://www.stata.com/help.cgi?search
>> * http://www.stata.com/support/statalist/faq
>> * http://www.ats.ucla.edu/stat/stata/
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
Windows Live™ SkyDrive™: Store, access, and share your photos. See how.
* For searches and help try: