A narrative approach to a fiscal DSGE model

Abstract

Structural DSGE models are used for analyzing both policy and the sources of business cycles. Conclusions based on full structural models are, however, potentially affected by misspecification. A competing method is to use partially identified SVARs based on narrative shocks. This paper asks whether both approaches agree. Specifically, I use narrative data in a DSGE-SVAR that partially identify policy shocks in the VAR and assess the fit of the DSGE model relative to this narrative benchmark. In developing this narrative DSGE-SVAR, I develop a tractable Bayesian approach to proxy VARs and show that such an approach is valid for models with a certain class of Taylor rules. Estimating a DSGE-SVAR based on a standard DSGE model with fiscal rules and narrative data, I find that the DSGE model identification is at odds with the narrative information as measured by the marginal likelihood. I trace this discrepancy to differences in impulse responses, identified historical shocks and policy rules. The results indicate monetary accommodation of fiscal shocks.Narrative approach to a fiscal DSGE model 803 signal-to-noise ratio of the measurement equation. With a proper prior, the models both here and in Caldara and Herbst (2019) require a Metropolis-within-Gibbs sampler. Unlike their setting, I allow for possibly missing instruments, an important issue in applied work.My result that the proxy SVAR identification is valid for a class of DSGE models is the second building block for my DSGE-SVAR: I provide conditions under which the instrument-identified VAR in Mertens and Ravn (2013) correctly identifies shocks and policy rule coefficients in models with standard Taylor-type policy rules, such as Leeper, Plante, and Traum (2010) and Fernandez-Villaverde, Guerron-Quintana, Kuester, and Rubio-Ramirez (2015). This property of the narrative VAR contrasts with traditional VARs that identify shocks through contemporaneous zero restrictions. DSGE models that match the VAR then need to assume that economic agents only react to policy shocks with a delay. The narrative VAR approach is valid if the data are generated from a widely used class of DSGE models, without restricting the timing. The key condition for my result is that the information set in the VAR captures the variables policy-makers pay attention to. This theoretical result mirrors the empirical result in Caldara and Herbst (2019) that credit spreads may be an important policy variable.My application contributes to the literature on fiscal and monetary policy DSGE models, which is important from a substantive point of view: With monetary policy constrained by the zero lower bound (ZLB), "stimulating" fiscal policy has gained a lot of attention and influential papers such as Christiano, Eichenbaum, and Rebelo (2011) have used quantitative DSGE models for the analysis of fiscal policies. Since the fiscal building blocks of DSGE models are less well studied than, say, the Taylor rule for monetary policy (e.g., Clarida, Galí, and Gertler (2000)), assessing the fiscal policy im...

Monetary Policy, Real Activity, and Credit Spreads: Evidence from Bayesian Proxy SVARs

“…To the best of our knowledge, bootstrap algorithms developed to construct confidence intervals in proxy SVARs only apply to strong instruments. Moreover,Lunsford and Jentsch (2016) show that the choice of bootstrap algorithms can yield very different confidence intervals for impulse responses.8 This feature is one major differentiation of our analysis from other Bayesian approaches-for example,Bahaj (2014) andDrautzburg (2016).…”

The role of startups for local labor markets

“…Mertens and Ravn (2013) consider two particular decompositions: lower and upper triangular factorizations of S 1 S ′ 1 . While Drautzburg (2020) showed that certain policy rules justified these factorizations in the case of fiscal or monetary policy shocks, we are not concerned with policy shocks. We thus pursue a different type of identification.…”

“…See Drautzburg ((2020), appendix A.4) for closed‐form expressions of η , κ , and in terms of the reduced‐form VAR covariance matrix .…”

Proxy VAR Models in a Data-Rich Environment

“…Secondly, this study relates to the Bayesian Proxy VAR literature, most directly to the small-scale Bayesian Proxy VAR model by Caldara and Herbst (2019) whose framework I extend to allow for latent factors. Other studies employing external instruments in the Bayesian paradigm are Drautzburg (2016) who estimates a narrative DSGE-VAR model, Bahaj (forthcoming) who applies high-frequency identification in a multi-country framework, and Arias et al (2018) who propose a Proxy VAR framework amenable to an importance sampler. Lastly, Kerssenfischer (2019) investigates the relation between informational insufficiency and identification in a frequentist setting.…”