Working Paper: CEPR ID: DP5638
Authors: Richard Dennis; Kai Leitemo; Ulf Söderström
Abstract: Robust control allows policymakers to formulate policies that guard against model misspecification. The principal tools used to solve robust control problems are state-space methods (see Hansen and Sargent, 2006, and Giordani and Söderlind, 2004). In this paper we show that the structural-form methods developed by Dennis (2006) to solve control problems with rational expectations can also be applied to robust control problems, with the advantage that they bypass the task, often onerous, of having to express the reference model in state-space form. Interestingly, because state-space forms and structural forms are not unique the two approaches do not necessarily return the same equilibria for robust control problems. We apply both state-space and structural solution methods to an empirical New Keynesian business cycle model and find that the differences between the methods are both qualitatively and quantitatively important. In particular, with the structural-form solution methods the specification errors generally involve changes to the conditional variances in addition to the conditional means of the shock processes.
Keywords: misspecification; optimal policy; robust control
JEL Codes: C61; E52; E58
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| robust control methods (C53) | mitigate the effects of model misspecification (C51) |
| structural-form methods (C51) | change conditional means and variances of shock processes (C22) |
| state-space methods (C32) | alter conditional means (Y60) |
| choice of modeling approach (C52) | affect derived economic equilibria (D59) |
| differences in equilibria from methods (C62) | important behavioral implications for monetary policy (E52) |
| structural-form methods (C51) | more activist monetary policy response (E63) |
| evil agent (Y60) | conditional means and variances of shocks (C32) |
| choice of method (C52) | qualitatively and quantitatively different outcomes in economic models (C54) |