Variables of Interests for SWZ Markov Switching Model

Variable Original Name

File name in Matlab

Line Number

Option Name

Comments

Default Values

nStates

szemldata_a_setup.m

25

?

Number of states

2

indxEqnTv_m

szemldata_a_setup.m

22

?

Rows corresponds to equations

indxScaleStates

szemldata_a_setup.m

26

?

0 (default): no scale adjustment in the prior variance for the number of states so that the prior value is the same as the constant VAR when the parameters in all states are the same.
1: allows a scale adjustment, marking the prior variance bigger by the number of states. Wrong choice because it makes the prior looser as the number of states increases.

0

galp

szemldata_a_setup.m

30

?

Alpha value for squared time-varying structural shock lambda.

1.0

gbeta

szemldata_a_setup.m

31

?

Beta value for squared time-varying structural shock lambda.

1.0

gsig2_lmd

szemldata_a_setup.m

32

?

Case 4 (no state change across variables (i) and across lags (l); only one scale factor for all lagged variables change from state to state). Normal prior variance for structural shock lambda, one scale factor for all lagged D+ in a given equation.

50^2

gsig2_lmdm

szemldata_a_setup.m

34

?

Case 3 (no state change across lags (l) but allows all variables for a given lag to switch states). Normal prior variance for different variables in lagged D+.

50^2

q_diag

szemldata_a_setup.m

36

?

Prior mean: 0.95 for monthly data (about 20 months) and 0.85 for quarterly data (about 6.7 quarters). Diagonal q in the transtion matrix and the duration is 1 over (1-q). To allow an absorbing state, we need to have 1 for all off-diagonal elements and x for the diagonal element.

0.85

indxFlatPriorP

szemldata_a_setup.m

42

?

0: diagonal of transition matrix has some duration
1: flat prior where all alpha's are 1.0 in Dirichlet.

0