Abstract
In likelihood-based estimation of linearized Dynamic Stochastic General Equilibrium (DSGE) models, the evaluation of the Kalman Filter dominates the running time of the entire algorithm. In this paper, we revisit a set of simple recursions known as the “Chandrasekhar Recursions” developed by Morf (Fast Algorithms for Multivariate Systems, Ph.D. thesis, Stanford University, 1974) and Morf et al. (IEEE Trans Autom Control 19:315–323, 1974) for evaluating the likelihood of a Linear Gaussian State Space System. We show that DSGE models are ideally suited for the use of these recursions, which work best when the number of states is much greater than the number of observables. In several examples, we show that there are substantial benefits to using the recursions, with likelihood evaluation up to five times faster. This gain is especially pronounced in light of the trivial implementation costs—no model modification is required. Moreover, the algorithm is complementary with other approaches.
Notes
The algorithm proposed by Herbst and Schorfheide (2013) usually involves tens of millions of likelihood evaluation.
It should be clarified here that the term states is given in a statistical context. It should note be confused with partitioning system variables into “states” and “controls” to facilitate solving the DSGE model i.e., finding Eq. (1) (see, for example, Schmitt-Grohé and Uribe 2004). Here \(s_t\) contains all the variables of the model. If a control engineering approach is taken to solve the model, \(s_t\) would contain both the predetermined (exogenous and endogenous) and control variables.
We also preclude stochastic singularity; i.e., the total number of structural shocks and measurement errors must be greater than or equal to the number of observables.
We combine the standard prediction and updating equations because we are trying to avoid unecessary calculations.
Indeed, looking at reduced rank decompositions of \(P_{t|t-1}\) or differences thereof, has a wide literature known as Square Root Kalman Filtering. I present these related recursions because they are much easier to implement.
The results were broadly the same when only serial computation was considered.
We again note that we do not use the Block KF because it is not a DSGE model.
We set the moving average coefficients on the markup shocks and automatic stabilizer on the government spending shock to zero. This way we can use the two block (AR), sparse Block Kalman Filter.
References
Adjemian, S., Bastani, H., Juillard, M., Mihoubi, F., Ratto, M., & Villemot, S. (2011). Dynare: Reference Manual, Version 4, Dynare Working Papers 1, CEPREMAP.
Aknouche, A., & Hamdi, F. (2007). Periodic Chandrasekhar recursions, arXiv:0711.385v1
An, S., & Schorfheide, F. (2007). Bayesian analysis of DSGE models. Econometric Reviews, 26(2–4), 113–172.
Chib, S., & Ramamurthy, S. (2010). Tailored randomized block MCMC methods with application to DSGE models. Journal of Econometrics, 155(1), 19–38.
Hamilton, J. (1994). Time Series Analysis. Princeton, NJ: Princeton University Press.
Herbst, E., & Schorfheide, F. (2013). Sequential Monte Carlo Sampling for DSGE Models, Federal Reserve Bank of Philadelphia Working Paper.
Jaimovich, N., & Rebelo, S. (2009). Can news about the future drive the business cycle? American Economic Review, 9(4), 1097–1118.
Klein, A., Melard, G., & Zahaf, T. (1998). Computation of the exact information matrix of Gaussian dynamic regression time series models. The Annals of Statistics, 26, 1636–1650.
Morf, M. (1974). Fast Algorithms for Multivariate Systems, Ph.D. thesis, Stanford University.
Morf, M., Sidhu, G., & Kalaith, T. (1974). Some new algorithms for recursive estimation in constant, linear, discrete-time systems. IEEE Transactions on Automatic Control, 19, 315–323.
Schmitt-Grohé, S., & Uribe, M. (2004). Solving dynamic general equilibrium models with a second-order accurate approximate to the policy function. Journal of Economic Dynamics and Conrol, 28, 755–775.
Schmitt-Grohé, S., & Uribe, M. (2012). What’s news in business cycles? Econometrica, 80, 2733–2764.
Sims, C. A. (2002). Solving linear rational expectations models. Computational Economics, 20, 1–20.
Smets, F., & Wouters, R. (2007). Shocks and frictions in US business cycles: A Bayesian DSGE approach. American Economic Review, 97, 586–608.
Strid, I., & Walentin, K. (2009). Block Kalman filtering for large-scale DSGE models. Computational Economics, 33, 277–304.
Acknowledgments
This paper uses material from Chapter 1 of my dissertation at the University of Pennsylvania. I am deeply indebted to my advisor, Frank Schorfheide, for his guidance. I also thank, without implication, participants at the Research Computing Seminar at the Fed Board and John Roberts for comments.
Author information
Authors and Affiliations
Corresponding author
Additional information
Code for this project is available at http://www.edherbst.net.
Appendix
Appendix
1.1 Verification of recursions for \(F_t, K_t, \text{ and } K_{g, t}\).
To show that Chandrasekhar recursions work for the special case discussed above, we show how to write the recursions for \(F_{t}\) and \(K_{t}\) in terms of \(\Delta P_{t|t-1}\). For \(F_t\):
\(K_t\) and \(K_{g, t}\) are similar. For \(K_{g, t}\):
1.2 Verification of the difference equation for \(\Delta P_{t|t-1}\).
Proof of Lemma
From the Kalman Filter, we have that
Subtracting \(P_{t|t-1}\) from both sides, we have that
Using the recursions for \(F_t\) shown in the previous lemma, we have
Grouping like terms and completing the square for \((T-K_{g, t}Z)\), we have
Note that we can write the final product in the equation, using (22), (19), and tediously expanding terms, as,
Rights and permissions
About this article
Cite this article
Herbst, E. Using the “Chandrasekhar Recursions” for Likelihood Evaluation of DSGE Models. Comput Econ 45, 693–705 (2015). https://doi.org/10.1007/s10614-014-9430-2
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10614-014-9430-2