GLS Estimation of Dynamic Factor Models

Breitung, Jörg; Tenhofen, Jörn

doi:10.1198/jasa.2011.tm09693

Cited by 55 publications

(50 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Third, in models with heteroskedastic or autocorrelated errors, the sequential LS technique can be used to compute the implied ML estimator that is equivalent to minimizing the weighted sum of squared residuals (cf. Breitung and Tenhofen (2011)), which is equivalent to the (pseudo) ML estimator. It is unclear how this could be achieved with the sequential PC approach.…”

Section: S(fmentioning

confidence: 99%

See 1 more Smart Citation

Analyzing Business and Financial Cycles Using Multi-Level Factor Models

Breitung

Eickmeier

2014

SSRN Journal

View full text Add to dashboard Cite

Research QuestionIn recent years (dynamic) factor models have become increasingly popular for macroeconomic analysis and forecasting in a data-rich environment. A serious limitation of the standard approximate factor model is that it assumes the common factors to affect all variables of the system. However, the efficiency of the Principal Component (PC) estimator for the common factors may deteriorate substantially if there are factors that load on subsets of variables only. Those factors may also be of independent interest. In an international context, for example, factors that load on variables associated with specific regions ("regional factors") could be (and have been) analyzed besides the global factors which link all variables in the model. Alternatively (or in addition), a block structure may represent economic, cultural or other characteristics. This paper addresses the question how such models can be estimated in a fast and easy way. ContributionThis paper makes several contributions. First, we provide a comprehensive comparison of existing estimation approaches for multi-level factor models and propose two very simple alternative estimation techniques based on sequential least squares and canonical correlations (which avoids any iterations). Second, we extend the sequential least squares estimation approach to a three-level factor model (with, for example, global, regional and variable-specific factors) with overlapping blocks of factors. Such factors structures are challenging as they cannot be estimated one level after another. A final contribution are three applications in which we study international comovements of business and financial cycles as well as asymmetries over the business cycle in the US. ResultsOur suggested estimation techniques provide (point) estimates in a tiny fraction of a second (in typical (macro) settings) compared to a Bayesian estimator that requires several hours. We also shows that based on Monte Carlo simulations that, in some circumstances, our proposed estimators tend to outperform alternative (two-step principal component and quasi maximum likelihood) estimation methods.In the first application we apply several estimation methodologies for twolevel factor models to an annual real activity dataset of more than 100 countries between 1960 and 2010. We estimate global and regional factors which turn out to be very similar across methods. We confirm Hirata et al.'s main finding that regional (business cycle) factors have become more important and global factors less important over time.In the second application, we use a large quarterly macro-financial dataset for 24 countries between 1995 and 2011. We estimate a global factor, regional factors, as well as variable-specific (macro and financial) factors. We find that financial variables strongly comove internationally, to a similar extent as macroeconomic variables. Macroeconomic and financial dynamics share common factors, but financial factors independent from macro factors also matter for financial variables. Finally, ...

show abstract

Section: S(fmentioning

confidence: 99%

“…Although the resulting LS estimator is consistent in the case of heteroskedastic errors (since the LS estimators are robust against heteroskedastic errors), the asymptotic efficiency may be improved by using a generalized leastsquares (GLS) approach (cf. Breitung and Tenhofen (2011)). …”

mentioning

confidence: 99%

Analyzing Business and Financial Cycles Using Multi-Level Factor Models

Breitung

Eickmeier

2014

SSRN Journal

View full text Add to dashboard Cite

show abstract

“…Bai (2003), Breitung andTenhofen (2011), Choi (2007), Connor and Korajczyk (1988), Doz et al (2011b), Fan et al (2011), Goyal et al (2009), Inoue and Han (2011, Stock and Watson (2002ab), Wang (2010), among others. The present paper considers the likelihood-based estimation of the model.…”

Section: Introductionmentioning

confidence: 99%

Maximum Likelihood Estimation and Inference for Approximate Factor Models of High Dimension

Bai

2016

Review of Economics and Statistics

View full text Add to dashboard Cite

show abstract

“…Key papers include Bai (2003), Bai and Ng (2002, 2010, Reichlin (2000, 2005), Hallin and Liska (2007), Onatski (2009Onatski ( , 2010 and Stock and Watson (2002b). Additionally, the properties of estimators based on generalized least squares are also discussed in Breitung and Tenhofen (2011), Doz and Reichlin (2011a) and Jungbacker and Koopman (2008). Doz and Reichlin (2011b) suggest a two step estimator by combining principal component and maximum likelihood methods.…”

Section: Unemploymentmentioning

confidence: 99%

Diffusion Index Model Specification and Estimation Using Mixed Frequency Datasets

Kim

Swanson

2013

Recent Advances in Estimating Nonlinear Models

View full text Add to dashboard Cite

Standard-Nutzungsbedingungen:Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden.Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen.Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in AbstractIn this chapter, we discuss the use of mixed frequency models and diffusion index approximation methods in the context of prediction. In particular, select recent specification and estimation methods are outlined, and an empirical illustration is provided wherein U.S. unemployment forecasts are constructed using both classical principal components based diffusion indexes as well as using a combination of diffusion indexes and factors formed using small mixed frequency datasets. Preliminary evidence that mixed frequency based forecasting models yield improvements over standard fixed frequency models is presented.JEL classification: C22, C51.Keywords: forecasting, diffusion index, mixed frequency, recursive estimation, Kalman filter.

show abstract

GLS Estimation of Dynamic Factor Models

Cited by 55 publications

References 19 publications

Analyzing Business and Financial Cycles Using Multi-Level Factor Models

Analyzing Business and Financial Cycles Using Multi-Level Factor Models

Maximum Likelihood Estimation and Inference for Approximate Factor Models of High Dimension

Diffusion Index Model Specification and Estimation Using Mixed Frequency Datasets

Contact Info

Product

Resources

About