Supplementary material to accompany the paper: Time-varying asymmetry and tail thickness in long series of daily
financial returns by Baej Mazur and Mateusz Pipie, Cracow University of Economics, Cracow, Poland.

All the computations are carried out using Ox language: OxMetrics 6.30 (64 bit) www.doornik.com.
Please note that depending on the OS version the Ox import/export might result in very specific (sometimes incompatibile) layout of matrices exported as MS Excel files. In some cases there are empty columns addded between 'AA' and 'BA' - proper load/save operations might require dealing with the issue. 

There are separate files for the purpose of in-sample estimation and prediction (using Bayesian inference methods). In-sample estimation is carried out using Metropolis-Hastings algorithm with random-walk proposal (MH-RW). Please note that the estimation procedure relies upon specification/calibration of candidate densities (location/covariance matrices as well as tuning parameters) - care should be taken for this step (the computer code does not include automated calibration). Note that certain parts of the code should be commented out (or in) for the purpose of calibration.
Out-of-sample estimation is conducted using either sequential MH-RW runs (which is computationally demanding) or (for simpler models) using a parallelized sequence of MH steps with independent proposal (numerically effective but not necessarily feasible for more complicated models). The 'parallel' version runs a sequence of forecasts within one process whereas the 'sequential' version runs only one forecast. The parallelization for the MH-IS version is at the level of computer code, so it requires a single process/thread to run. However, for the sequential (MH-RW) version effective application usually requires a number of separate processes that operate independently (as separate parallel processes). For the purpose the code includes features for splitting the recursive experiment into a number of components. 

The archive contains the following sub-directiories:
source_data: the S&P500 daily returns (.xlsx file with the dataset used)
in_sample:
M_1 - AR(1)-GJR-GARCH with t conditional distribution
M_2 - AR(1)-GJR-GARCH with AST-ST conditional distribution denoted by M(0,0,0,0,0) in the paper
M_3 - AR(1)-GJR-GARCH with AST-ST conditional distribution denoted by M*(0,2,0,2,2) in the paper
M_4 - AR(1)-GJR-GARCH with AST-ST conditional distribution denoted by M**(0,3,0,3,3) in the paper
FIGARCH_type_MODELS: for results presented in Appendix B

out_of_sample: contains two essential variants of the code used for the purpose of Bayesian prediction (and risk assessment). Note that full predictive results involve very large files in certain cases and might require a lot of memory. We therefore include basic files (not including full results) for used for the predictive experiments, although more detailed results are available from the authors by request (blazej.mazur@uek.krakow.pl).

