# DSGE models (Dynare) are only based on simulations and approximations? (without data)

I'm beginner in DSGE models. Are they only based on simulations and approximations? Don't we have any real data-sets (for example interest rate, labor force, etc) to estimate the equations? I opened some .Mod files and I can only see equations, variable names and other functions. If yes, why do we have this structure in DSGE models and why can't we use real data-sets in these models?

• In his Congressional Testimony, Robert Solow said "...the DSGE approach is not a good way to understand macroeconomic behavior...". www2.econ.iastate.edu/classes/econ502/tesfatsion/… The DSGE straitjacket is known to fail in forecasting---the ultimate test for usefulness---contest against the old reliable Cointegration variety models. – user8647 Jun 25 '16 at 18:50

It doesn't seem like you've tried very hard to find estimation using Dynare (google results of "Dynare estimation"). Dynare is in fact capable of doing estimation and typically people use some sort of data to estimate the parameters that govern their model.

## What is a DSGE model?

A DSGE model is a series of equations (and variables and parameters) that describe a "model" economy. Once you write everything down and take into consideration the optimizing decisions made by the agents, it is simply a system of stochastic difference (differential) equation that you can imagine writing in some form along the lines of:

$$E_t F(y_{t+1}, y_t, y_{t-1}, u_t) = 0$$

## What does Dynare do?

The majority of introductory examples only use Dynare to solve the model and run some simple simulations. In addition to being able to solve the model, Dynare is also capable of doing Bayesian estimation of the parameters that make up the model by feeding in data.

### Solves and simulates the Model

Dynare takes this system of equations (and the given parameter values) and solves them using perturbation methods. One of the outputs of the solution is a law of motion for all of the variables in the economy -- i.e. one can write something along the lines of

$$y_{t+1} = F(y_t, u_{t+1})$$

This law of motion allows you to generate model analogues of the data that you are interested in -- i.e. capital, labor, consumption etc...

### Estimates the model

Once Dynare has a law of motion and has obtained simulated moments, it can compare the output of the model to data that you have provided. Dynare uses Bayesian estimation (though it is possible it does other types, but I'm not familiar with it) to estimate parameters which means that you need to provide it with a prior for each variable that you would like to estimate. It then uses a simple Metropolis Hastings algorithm to obtain estimates for the distribution over parameter values.

## References

I would recommend you read both of the following references.

• Tommaso Mancini Griffoli. Dynare User Guide: An introduction to the solution & estimation of DSGE models. 2007-2008. found here
• Stefanie Flotho. DSGE Models - solution strategies. December 2009. found here
• Thank you for answer. So after solving the entire system we import data or data should we used in solving process? – user2991243 Jul 16 '15 at 16:00
• Read the user guide for more information. Simplified version of what it does: 1) Begins with a set of parameter values. 2) Solves the model given those parameter values. 3) Checks how likely the outside data is given those parameter values. 4) Repeats until it finds the parameter values that are "most" likely (this is step that is simplified because it estimates them in a Bayesian way and I'm describing maximum likelihood). – cc7768 Jul 16 '15 at 16:03