What i want to do is to estimate a TVAR-like model but instead of using a single threshold for the whole system i would like to specify a threshold for every endogenous variable. So far so good, coming to eviews i created the following system:
Code: Select all
s=(c(1)+c(2)*s(-1)+c(3)*m(-1)+c(4)*int(-1)+c(5)*y(-1)+c(6)*pi(-1))*i1+(1-i1)*(c(7)+c(8)*s(-1)+c(9)*m(-1)+c(10)*int(-1)+c(11)*y(-1)+c(12)*pi(-1))
m=(c(13)+c(14)*m(-1)+c(15)*s(-1)+c(16)*y(-1)+c(17)*int(-1)+c(18)*pi(-1))*i5+(1-i5)*(c(19)+c(20)*m(-1)+c(21)*s(-1)+c(22)*y(-1)+c(23)*int(-1)+c(24)*pi(-1))
int=(c(25)+c(26)*int(-1)+c(27)*m(-1)+c(28)*s(-1)+c(29)*y(-1)+c(30)*Pi(-1))*i2+(1-i2)*(c(31)+c(32)*int(-1)+c(33)*m(-1)+c(34)*s(-1)+c(35)*y(-1)+c(36)*Pi(-1))
y = (c(37)+c(38)*y(-1)+c(39)*int(-1)+c(40)*m(-1)+c(41)*s(-1)+c(42)*Pi(-1))*i3+(1-i3)*(c(43)+c(44)*y(-1)+c(45)*int(-1)+c(46)*m(-1)+c(47)*s(-1)+c(48)*Pi(-1))
Pi=(c(49)+c(50)*Pi(-1)+c(51)*s(-1)+c(52)*int(-1)+c(53)*y(-1)+c(54)*m(-1))*i4+(1-i4)*(c(55)+c(56)*Pi(-1)+c(57)*s(-1)+c(58)*int(-1)+c(59)*y(-1)+c(60)*m(-1))and estimated it by using Full Information Maximum Likelihood. The i1-i5 variables are the usual indicator functions which follow a simple rule (like i1 = (s(-1)<t1(-1), where t1 =@movav(s, 3)).After that i transformed the whole thing into a model object and performed simulations/forecasts. What do you think? Is this valid?
thanks in advance,
best regards