I have a model with a lot of lags (e.g. 80) of one variable, and dynare is finding it very difficult to find the steady state. It seems Dynare tells “fsolve” to treat all the auxiliary equations and variables identically to the standard ones, which seems to be resulting in fsolve being pulled towards giving each lag a different value. I can see that treating auxiliary equations identically would simplify your code, but it’s incredibly inefficient when there are large numbers of extra lags. I guess when block is supported on stochastic models, it might just help?
Any advice?
Also, should I adjust the tolerance Dynare uses when solving with a lot of equations? Is it being compared against the vector’s 2-norm (in which case I should) or the vector’s infinity-norm (in which case I probably shouldn’t)?