What You Match Does Matter: The Effects of Data on DSGE Estimation

This paper explores the effects of using alternative data sets for the estimation of DSGE models. I find that the estimated structural parameters and the model's outcomes are sensitive to the variables used for estimation. Depending on the set of variables the point estimate for habit formation ranges from 0.70 to 0.97. Similarly, the interest-smoothing coefficient in the Taylor rule fluctuates between 0.06 and 0.76. In terms of the model's predictions, if interest rates are excluded during estimation, the estimated structural coefficients are such that the model forecasts a strong deflation following an expansionary monetary expansion. More importanlty, three ways to assess different observable sets are proposed. Based on these measures, I find that that including the price of investment in the data set delivers the best results.