Re: st: Qustion about a state space model

[Jorge Eduardo Pérez Pérez <perez.jorge@ur.edu.co>]

Your model is a time-varying parameters regression model. Stata is
unable to estimate this model. -sspace- only estimates "linear
state-space models with time-invariant coefficient matrices"
_______________________
Jorge Eduardo Pérez Pérez


On Wed, Mar 2, 2011 at 6:13 PM, Masafumi Yabara <m.yabara@gmail.com> wrote:
>
> Dear Stata members,
>
> I'm trying to run a state space model as below to see the trend of
> stock market integration, following the literature on the subject
> (e.g., I.-W. Yu et al./Journal of Banking & Finance 34 (2010) 2874-2885).
>
> Signal equation: ln Rt - ln St = At + Bt (ln Rt - ln Gt) + error1(t)
> State equation 1: At = A(t-1) + error2(t)
> State equation 2: Bt = B(t-1) + error3(t)
>
> where Rt: equity market index level of a dominant regional market (in
> my case, Kenya) at time t
> St: equity market index level of a country of interest (in my case,
> Tanzania) at time t
> Gt: equity market index level of a dominant external market (in my
> case, US) at time t.
>
> My question is, how I can construct the model in STATA, because my
> model did not run.
>
> Here are what I did.
> I typed the following commands in STATA.
> The variables "nsetse" and "nseiundu" correspond to "ln Rt - ln St" and
> "ln Rt - ln Gt" in the signail equation above, respectively.
>
> . constraint 1 [u1]L.u1 = 1
> . constraint 2 [u1]e.u1 = 1
> . constraint 3 [u2]L.u2 = 1
> . constraint 4 [u2]e.u2 = 1
> . constraint 5 [nsetse]u1 = 1
> . constraint 6 [nsetse]u2 = nseindu
> . constraint 7 [nsetse]e.nsetse = 1
> .
> . sspace (u1 L.u1 e.u1, state noconstant) ///
> >           (u2 L.u2 e.u2, state noconstant) ///
> >           (nsetse u1 u2 e.nsetse, noconstant), constraints(1/7)
>
> Then I got the following error messages, and could not solve it
> although I modified the scales of the variables and tried different
> techniques (options) and a number of different settings.
>
> (note: constraint number 6 caused error r(111))
> searching for initial values .
> (setting technique to bhhh)
> Iteration 0:   log likelihood = -284.60372
> ......
> (switching technique to nr)
> Iteration 5:   log likelihood = -282.73863  (backed up)
> optimization terminated because of numerical instability:
> Hessian is not negative semidefinite r(430);
>
> The tricky part of the model is, to me, that an unobserved state (Bt) is
> multiplied by an independent variable (ln Rt - ln Gt) in the signal equation.
> As a matter of fact, the result says the constraint No.6 is causing errors.
> I guess my way of constructing the model is wrong, but I can't figure out
> how I can resolve the problem.
> Any advise and suggestion are very much appreciated.
>
> Thank you so much in advance for your help.
>
> Best regards,
>
> Masafumi Yabara
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/