Bayesian learning of graphical vector autoregressions with unequal lag-lengths
- 435 Downloads
Graphical modelling strategies have been recently discovered as a versatile tool for analyzing multivariate stochastic processes. Vector autoregressive processes can be structurally represented by mixed graphs having both directed and undirected edges between the variables representing process components. To allow for more expressive vector autoregressive structures, we consider models with separate time dynamics for each directed edge and non-decomposable graph topologies for the undirected part of the mixed graph.
Contrary to static graphical models, the number of possible mixed graphs is extremely large even for small systems, and consequently, standard Bayesian computation based on Markov chain Monte Carlo is not in practice a feasible alternative for model learning. To obtain a numerically efficient approach we utilize a recent Bayesian information theoretic criterion for model learning, which has attractive properties when the potential model complexity is large relative to the size of the observed data set. The performance of our method is illustrated by analyzing both simulated and real data sets. Our simulation experiments demonstrate the gains in predictive accuracy which can obtained by considering structural learning of vector autoregressive processes instead of unstructured models. The analysis of the real data also shows that the understanding of the dynamics of a multivariate process can be improved significantly by considering more flexible model classes.
KeywordsBayesian analysis Granger-causality Graphical models Statistical learning Vector autoregression Markov chain Monte Carlo Greedy optimization
- Abramovitz, M., & Stegun, I. A. (Eds.) (1965). Handbook of mathematical functions with formulas, graphs, and mathematical tables. New York: Dover. Google Scholar
- Bernardo, J. M. (1999). Nested hypothesis testing: the Bayesian reference criterion. In J. M. Bernardo, J. O. Berger, A. P. Dawid, & A. F. M. Smith (Eds.), Bayesian statistics (Vol. 6, pp. 101–130). London: Oxford University Press. With discussion. Google Scholar
- Brüggemann, R., Krolzig, H.-M., & Lütkepohl, H. (2002). Comparison of model reduction methods for VAR processes. EUI Working Paper, ECO, 2002/19. http://hdl.handle.net/1814/791.
- Dahlhaus, R., & Eichler, M. (2003). Causality and graphical models in time series analysis. In P. J. Green, N. L. Hjort, & S. Richardson (Eds.), Highly structured stochastic systems (pp. 115–137). London: Oxford University Press. Google Scholar
- Dash, D. (2005). Restructuring dynamic causal systems in equilibrium. In: R. Cowell & Z. Ghahramani (Eds.), Proceedings of the tenth international workshop on artificial intelligence and statistics (AIStats). Society for artificial intelligence and statistics. Available electronically at http://www.gatsby.ucl.ac.uk/aistats/.
- Eichler, M. (2001). Graphical modelling of multivariate time series. Technical report, Universität Heidelberg. arXiv:math.ST/0610654.
- Eichler, M. (2006a). Fitting graphical interaction models to multivariate time series. In Proceedings of the 22nd conference of uncertainty in artificial intelligence. Arlington: AUAI Press. Google Scholar
- Eichler, M. (2008). Causal inference from multivariate time series: What can be learned from Granger causality. In C., Glymour, W. Wang & D. Westerstahl (Eds.), Proceedings from the 13th international congress of logic, methodology and philosophy of science. King’s College Publications, London. Google Scholar
- Friedman, N., Murphy, K., & Russell, S. (1998). Learning the structure of dynamic probabilistic networks. In G. F. Cooper & S. Moral (Eds.), Proceedings of the 14th annual conference on uncertainty in artificial intelligence (UAI-98). San Mateo: Morgan Kaufmann. Google Scholar
- Granger, C. W. J. (1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 37, 24–36. Google Scholar
- Granger, C. W. J. (2001). Essays in econometrics: collected papers of Clive W.J. Granger. Cambridge: Cambridge University Press. Ghysels, E., Swanson, N.R. & Watson, M.W. (Eds.). Google Scholar
- Lauritzen, S. L. (1996). Graphical models. London: Oxford University Press. Google Scholar
- Lynggaard, H., & Walther, K. H. (1993). Dynamic modelling with mixed graphical association models. Master’s thesis, Aalborg University. Google Scholar
- Moneta, A., & Spirtes, P. (2005). Graph-based search procedure for vector autoregressive models. LEM Working Paper 2005/14, Sant’Anna School of Advanced Studies, Pisa. Google Scholar
- Oxley, L., Reale, M., & Tunnicliffe, W. (2004). Finding directed acyclic graphs for vector autoregressions. In J. Antoch (Ed.), Proceedings in computational statistics 2004 (pp. 1621–1628). Heidelberg: Physica. Google Scholar
- Robert, C. P., & Casella, G. (2005). Monte Carlo statistical methods (2nd ed.). New York: Springer. Google Scholar
- Seinfeld, J. H. (1986). Atmospheric chemistry and physics of air pollution. New York: Wiley. Google Scholar
- Spirtes, P., Glymour, C., & Scheines, R. (2000). Causation, prediction, and search. Cambridge: MIT Press. Google Scholar
- Stanghellini, E., & Whittaker, J. (1999). Analysis of multivariate time series via a hidden graphical model. In D. Heckerman & J. Whittaker (Eds.), Proceedings of the seventh international workshop on artificial intelligence and statistics. San Mateo: Morgan Kaufmann. Google Scholar
- Valdés-Sosa, P. A., Sánchez-Bornot, J. M., Lage-Castellanos, A., Vega-Hernández, M., Bosch-Bayard, J., Melie-García, L., & Canalez-Rodríguez, E. (2005). Estimating brain functional connectivity with sparse multivariate autoregression. Philosophical Transactions of the Royal Society B-Biological Sciences, 360, 969–981. CrossRefGoogle Scholar
- Winker, P., & Maringer, D. (2004). Optimal lag structure selection in VEC-models. Computing in Economics and Finance 2004 155, Society for Computational Economics. Google Scholar