I am trying to test how the presidential elections can affect the trends in the US stock market. I am thinking the following empirical model: Rt=U(St)+β[Rt-1-U(St-1)]+δ(St)Vt, P(St|St-1)=exp(C+γ1(pre-election_8)+ γ2(post-election_8) /1+ exp(C+γ1(pre-election_8)+ γ2(post-election_8). Pre-election and post-election are the dummy variables if the month is in the specified period before or after an election. Would you please kindly let me know if the model is reasonable to you? Following is my code. Many Thanks
Code: Select all
Set return= 100.0*log(price/price{1})
Compute nlags=1
Compute gstart=1927:1, gend=2019:03
@MSVARSetuo(lags=nlags, switch=mh)
# return
Nolin(parment=common) mu phi sigma
frml loglmsvar = log(%MSVarProb(t))
dec equation p1eq p2eq
dec vector v1 v2
nonlin(parmset=tv) v1 v2
function %MSPTV time
type rect %MSPTV
type integer time
local rect p(1,2)
compute p(1,1)=%logistic(%eqnvalue(p1eq,time,v1),1.0)
compute p(1,2)=%logistic(%eqnvalue(p2eq,time,v2),1.0)
compute %MSPTV=p
end
@MSPChangeFnc %MSPTV
function TVPInit
type vector TVPInit
compute p=%MSPTV(gstart)
compute TVPInit=%MSVARInit()
end
equation p1eq *
# constant pre_election_8 post_election_8
equation p2eq *
# constant pre_election_8 post_election_8
compute v1=log(.70/(1-.70))*%unitv(%eqnsize(p1eq),1)
compute v2=log(.05/(1-.05))*%unitv(%eqnsize(p2eq),1)
maximize(parmset=tv+common,start=(pstar=TVPInit()),$
pmethod=simplex,piters=10,method=bfgs) loglmsvar gstart gend