This paper builds an unobserved components model that combines a multivariate filter approach with a Cobb-Douglas production function. This combination allows potential output estimates to incorporate more economic structure than the traditional production function approach, while retaining the ability to conduct growth accounting exercises. The model is a backward-looking state space model estimated with Bayesian methods employing the Kalman filter to jointly decompose six key observable variables (real GDP, unemployment rate, labour force participation rate, hours worked per person, a measure of core in ation and wage in ation) into trend and cyclical components. To do so, it relies on several reduced form relationships across the cyclical components, such as a wage and a price Phillips curve and an Okun's law type relationship, while it also assumes common trends for a few variables and allows for hysteresis effects. The model is estimated on aggregate euro area data with Bayesian methods. The paper finds that the resulting output gap estimates have good revision properties and reasonable forecasting performance in particular in terms of GDP and core in ation vis-a-vis a set of benchmarks.
state-space model production function Bayesian estimation