Fix calls to dynamic routines in identification

The attached file contains a moving average process. It crashes identification in calls like

    [residual, g1 ] = feval([M_.fname,'_dynamic'],yy0, ...
                            repmat(oo_.exo_steady_state',[M_.maximum_exo_lag+M_.maximum_exo_lead+1]), M_.params, ...
                            oo_.dr.ys, 1);

Here, the period number (last argument) is always set to 1 instead of presumably M_.maximum_exo_lag+1 .

dsge_mada_debt1.mod

@JohannesPfeifer The issue here is preprocessor and not the calls to dynamic model.

By using the attached modified mod file, where I introduce a fake call to stoch_simul in line 580, dynare properly eliminates all lead/lags in exo variables and lead/lags >1 in endo vars, implying that the internal representation has always M_.maximum_exo_lag = 1.

In turn, if I fix the calls to dynamic model etc., I get then an error in subsequent calls to th_autocovariances.m due to the same issue of lead/lag representation (it seems th_autocovariances.m expects a model without exo lags, or it is buggy ... ).

So, the way I see, it should be the preprocessor that treats a call to identification as if there was one to stoch_simul, thus representing the model without lags in exo vars and with max endo lead/lag = 1.

dsge_mada_debt1.mod

Dear all, to state it more explicitly, the routines are designed and expect to receive a model representation without lagged exo and with max lag = 1. this seems the standard representation of the stochastic form of the model, right? or do you see a major problem with this restriction/assumption?

@rattoma ok, I'll set that up. To be clear, I will do this both for identification and for sensitivity.

added preprocessor label

assigned to @houtanb and unassigned @rattoma

mentioned in commit preprocessor@4c71d143

@houtanb great many thanks!

closed via commit 618e2c54

This seems like a bugfix we need to cherry pick for 4.5.

mentioned in commit 0f8c5ef2