About the connection(s) of conditional forecast and det_simulation

I try to find the shocks within a given horizon through conditional_forecast by combining the domestic data and a well-defined model. This method has been adopted and the related paper has been published in EER.

Similarily, I set preference shocks, labor supply shocks, fiscal shocks, and monetary policy shocks and obtain the values of these shocks via conditional_forecast. The impulse responses of the relevant variables in the oo_ file are consistent with the data.

However, I re-simulated these shocks using deterministic simulations and found that the impulse response functions of the variables were completely inconsistent with the data I had collected.

The last two lines of the m file correspond to the mod files for conditional prediction and cross-validation, respectively. I want to know why this difference occurs? Is it because one of them is based on VAR forecasts, while the other is based on simulations of a general equilibrium model?

fluc_consum.zip (93.6 KB)

Your deterministic simulation is a perfect foresight one. Agents are aware of the full path of shocks.

Thanks a lot. I don’t intend to use stochastic simulation.
Can I understand you as saying that the mechanism behind conditional forecast is not equivalent to the mechanism behind perfect foresight solver?

Yes. Conditional forecast paths ask, which surprise shocks are required to happen in each period to achieve the desired values of the variables. Perfect foresight asks, which sequence of fully known shocks at the beginning of the simulation are required to achieve a particular outcome for the endogenous variables.

In this case, how should I perform cross-validation? That is, after obtaining the four shocks in the model through conditional forecasting, how can I simulate obtaining C_data and pi_data from real-world data?

Use simult_. See tests/conditional_forecasts/3/fs2000_conditional_forecast_initval.mod · master · Dynare / dynare · GitLab

1 Like