Working Paper, Federal Reserve Bank of Atlanta 2013-5
This paper proposes a moment-matching method for approximating vector autoregressions by finite-state Markov chains. The Markov chain is constructed by targeting the conditional moments of the underlying continuous process. The proposed method is more robust to the number of discrete values and tends to outperform the existing methods for approximating multivariate processes over a wide range of the parameter space, especially for highly persistent vector autoregressions with roots near the unit circle.
Markov chain vector autoregressive processes numerical methods moment matching non-linear stochastic dynamic models state space discretization stochastic growth model fiscal policy