When developing DSGE models, which will be used to analyze the fiscal or monetary poli­cy, it is important to take into account all the features observed in these data. If the model does not match the data, the estimates become inaccurate, and the conclusions drawn become unreliable. An important characteristic of data that is often given little attention is the presence of trends. Typically, trends are either removed using various filters or modeled. In the first case, a large amount of information is removed from the data, which leads to a significant decrease in the accuracy of parameters’ estimation. In the second case, only the trend in the labor productivity is most often used. Modeling only one trend means that, for example, the components of GDP in real terms should grow at a single growth rate on average, which is actually far from the case. This problem is especially relevant for developing countries, including the Russian Federation.

In this work, Russian data is analyzed in which significantly different average growth rates of GDP components are found. An approach is proposed to simulate sector-specific non-statio­nary productivity. The obtained estimates allow to conclude that the inclusion of additional trends helps to achieve better performance in sense trend-cycle decomposition of the data observed. Also, the simulation analysis shows that using the model without additional trends leads to a significant decrease in the parameter estimation accuracy. Thus, it is crucial to take into account the presence of trends in the data while creating practice-oriented DSGE models.