Below we go into more details regarding each specific routine included
in the folder "Comparison":

(1): Main_No_GARCH: This file generates data sets with 2 breaks and
estimates the model using the SSP-KF method in Grassi et al. (2015),
its modified version discussed in the reply to the referee, and the
method used in Koop and Korobilis (2012, 2013) based on forgetting
factors. The output consists of the "average parameter distance" APD
of the corresponding methods, which allows the user to compare the
performance of different models.

(2): Main_GARCH: This file generates data sets with 2 breaks while
also considering GARCH type errors and estimates the model using the
SSP-KF method in Grassi et al. (2015), its modified version discussed
in the in the reply to the referee and the method used in Koop and
Korobilis (2012, 2013) based on forgetting factors. The output
consists of the "average parameter distance" APD of the corresponding
methods, which allows the user to compare the performance of different
models.


The remaining .m files are those related to the functions performing
each specific estimation routine: 

(1a): funcKK: This file contains the estimation routine based on the
forgetting factor used in Koop and Korobilis (2012, 2013). The input
parameters are: data, the forgetting factor, lambda, and the kappa
parameter associated with the EWMA estimator. The corresponding output
are the filtered estimates of the latent states/parameters.

(2a): funcSPx: This file contains the Self-Perturbed estimation
routine based proposed in Grassi et al. (2015).  The input parameters
are: data, the perturbation parameter, beta, and the kappa parameter
associated with the EWMA estimator. The corresponding output are the
filtered estimates of the latent states/parameters.

(3a): funcSPx_Ft: This file contains Self-Perturbed estimation routine
of Grassi et al. (2015) modified to incorporate the changes in EWMA
discussed in the end of section 2.  The input parameters are: data,
the perturbation parameter, beta, and the kappa parameter associated
with the EWMA estimator. The corresponding output are the filtered
estimates of the latent states.

Finally, we have to stress that the performances of the three
estimators can be improved by selecting dynamically the design
parameters.
