8/n SECOND: Why am I concerned about accuracy?
Background:
In HA models, linearization-based methods start with Reiter (JEDC, 2009). It is nonlinear in individual states but imposes linearity in aggr states. It solves Khan & Thomas (ECMA, 2008) model about 100x faster than Krusell-Smith method.
Background:
In HA models, linearization-based methods start with Reiter (JEDC, 2009). It is nonlinear in individual states but imposes linearity in aggr states. It solves Khan & Thomas (ECMA, 2008) model about 100x faster than Krusell-Smith method.
Comments
These methods promise to be much faster than even Reiter’s method.
So, I am posting my concerns here, hoping that it will lead to a discussion of the pros and cons of these approaches and that hopefully, the authors can respond and persuade us.
My read of the SSJ paper: accuracy concerns are at least not exacerbated. Their achieve ~identical IRFs to Reiter (and Smets-Wouters when applied to RA case). Section 6 describes how to get "nonlinear perfect foresight" IRFs
Linearization methods work well bc the full Krusell-Smith (nonlinear) solution turns out to be pretty (log) flat in aggregate states. However, while this is true for small shocks — 1 to 2 sigma devs from trend — for larger shocks nonlinearities do matter. -->
Plus, idiosyncratic shocks are countercyclical, with their variance rising or skewness becoming more negative during recessions.
We are already well-equipped to deal with average recessions!
So, new methods are only useful to the extent that they can solve complex models accurately for BIG shocks.