DSGE Models: Problem of Trends

There are trends (deterministic and stochastic) in the most macroeconomic time series. Dynamic Stochastic General Equilibrium (DSGE) models have to take into account these data features. Data detrending is one of the popular approaches that imply exogenous (to the model) decomposition of time series into cycle and trend components, and dropping of the last one. The aim of the paper is to analyze the consequences of such approach. This paper shows that the methods described above distort the model, save some specific conditions. If one of the following conditions remains, then detrending disturbs the model unsystematically. Trend is eliminated from each time series separately. One variable has different nonlinear transformation than the other (example: one variable is in-logs while the other in-levels). Correlation of trend divergence (i.e. difference between trends of one and another variable) with exogenous shocks is incorrect (correct correlation can be nonzero). If trends are dropped from the model, then detrending distorts the model systematically. The author presents numerical results of detrending analysis and creates DSGE model. Then the model was estimated on multiple arrays of simulated data with different detrending schemes including the absence of detrending. Data detrending leads to 1.5–3 times higher errors of parameters estimation. More flexible detrending scheme leads to worse results (HP filter produces the worst result). However, if trend is eliminated from the data and DSGE model without trend is used then estimation errors increases additionally by 4–10 times.