The modern approach to modelling business cycles in Macroeconomics consists of proposing a set of impulses or shocks that move the economy out of its long run equilibrium and a propagation mechanism that transforms these random shocks into business cycles. Dynamic Stochastic General Equilibrium (DSGE) models, based on this methodology, have been the workhorse of Macroeconomics for the past two or three decades. Agents in these models behave optimally but subject to constraints coming from their budgetary restrictions and the market structure or institutional environment. This optimal behaviour and restrictions form the key propagation mechanism of shocks generating business cycles.
The parameters of these models are commonly estimated using Bayesian methods and used for policy analysis and forecasting. With Bayesian methods, the researcher has a prior about the probability distribution for parameter values which is updated by data, i.e. observations from the real world. However, a common practice in these models is to assume the range of shocks that affect the economy: technology, preference, government spending, monetary policy, markup shocks, etc. These are the sources of macroeconomic uncertainty. However, shocks are not observable, and we have to infer them from the data we use to estimate DSGE models. When estimating the role and importane of these shocks with Bayesian methods, researchers impose
them rather than select them. This is because the prior used for the standard deviation of these shocks is almost universally considered to exclude zero. Hence, shocks are considered fundamental by assumption.
What we ask in this paper are the following questions: first, suppose that a shock is not “fundamental” in the sense that it is not an important source of uncertainty. Does it matter if we impose this shock when estimating the model? The answer to this question is yes. Imposing inexisting shocks affects dramatically the values of the parameters that drive propagation in the model, leading to biased estimates. Second, we propose a method to select rather than impose the shocks entering a DSGE model, i.e. the drivers of macroeconomic uncertainty. The method is simple and implementable and it allows the prior for the standard deviation of shocks to include zero. We can thus select shocks and avoid the estimation bias arising from imposing inexisting shocks.
Finally, using our method, we estimate an industry-standard medium scale DSGE model due to Smets and Wouters (2007) widely used in the profession. Our results are quite revealing. We find that price and wage markup shocks and government spending shocks do not appear to be fundamental drivers of macroeconomic uncertainty. I.e., these shocks are not fundamental or structural and are simply likely to be capturing measurement error.
Read the complete paper here.