# Generating Fan Charts for Perfect Foresight Simulations with expectation errors

**URL:** <https://forum.dynare.org/t/generating-fan-charts-for-perfect-foresight-simulations-with-expectation-errors/28093>\
**Category:** ML/Bayesian estimation\
**Created:** [14 November 2025 10:53 UTC](https://forum.dynare.org/t/generating-fan-charts-for-perfect-foresight-simulations-with-expectation-errors/28093 "2025-11-14T10:53:54Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Sarv1926](https://forum.dynare.org/letter_avatar_proxy/v4/letter/s/e19adc/32.png) [@Sarv1926](https://forum.dynare.org/u/Sarv1926)\
**Post date:** [14 November 2025 10:53 UTC](https://forum.dynare.org/t/generating-fan-charts-for-perfect-foresight-simulations-with-expectation-errors/28093/1 "2025-11-14T10:53:54Z")

</div>

**Dear all,**

I’m working with a DSGE model analyzing a policy reform path using perfect foresight simulations. I’ve already estimated shock standard deviations using Simulated Method of Moments (SMM). I’d like to generate fan charts showing uncertainty around the policy path by running multiple simulations with different shock realizations introduced at specific periods.

I’m wondering about the best approach for drawing these shocks: Should I simply draw from independent normal distributions N(0, σ²\_SMM) for each shock, or would it be preferable to perform Bayesian estimation and use posterior draws from the MCMC sampler?

If using the Bayesian approach, would draws from `oo_.posterior_simulations` be appropriate for repeated perfect foresight simulations with expectation errors? I’m particularly interested in whether this would better capture parameter uncertainty and shock correlations compared to the simpler approach.

Any guidance on methodology or implementation would be greatly appreciated.

---

<div class="post-metadata">

**Author:** ![jpfeifer](https://forum.dynare.org/user_avatar/forum.dynare.org/jpfeifer/32/5044_2.png) [@jpfeifer](https://forum.dynare.org/u/jpfeifer)\
**Post date:** [19 November 2025 08:46 UTC](https://forum.dynare.org/t/generating-fan-charts-for-perfect-foresight-simulations-with-expectation-errors/28093/2 "2025-11-19T08:46:34Z")

</div>

It very much depends on what you are trying to do. Both method of moments and Bayesian estimation deliver measures of parameter uncertainty. That allows drawing parameters from the estimated distribution. Generally, that is a bit easier for a Bayesian approach because you naturally get random draws from the posterior. But you can also construct parameter draws based on the method of moments.