A popular strategy for estimating output gaps is to anchor them to structural economic relationships. The resulting output gaps, however, are often highly sensitive to numerous auxiliary assumptions inherent in the approach. This complicates their use in policymaking. We illustrate the point using the Phillips curve, arguably the most popular structural relationship in this context. Depending on the specification, we show that conditioning on this relationship either introduces a trend in the output gap - which is conceptually unappealing - or has little effect on it - which defeats the purpose of the exercise. Moreover, the estimated gaps perform poorly in real time, with large ex-post revisions. The opaqueness of the approach, which increases greatly with the dimension of the estimated system, can mask these problems. In order to address these limitations, we propose a more parsimonious and transparent approach to embedding economic information that is less vulnerable to misspecification. As an illustration, we apply the corresponding parsimonious multivariate filter to US data. We find that proxies for the financial cycle, notably credit growth, but also unemployment contain significant information and help generate robust real-time output gap estimates.
Stored only when both papers are in the corpus — 50 citations exist upstream.