# [R-SIG-Finance] Hi this is not a R-problem per se but an econometric problem of course

Brian G. Peterson brian at braverock.com
Mon Jun 15 14:33:36 CEST 2009

``` From your description of the system you are trying to estimate, it
seems that you should be looking at unit roots and cointegration instead
of the pure *LS methods.

Regards,

- Brian

KAUSHIK BHATTACHARJEE wrote:
> Dear All,
> I have 4 dependent  time series variables --y1t,y2t,y3t,y4t..
> For any fixed 't', they occer in sequence..first y4t, then y3t,...last y1t.
>   So I have a model like this.....
> y4t= y3tlag1+y2tlag1+y1tlag1+y4tlag1+ error4t
> y3t=  y4t +     y2tlag1+y1tlag1+ error3t
> y2t= y3t + y4t + y2tlag1+ y1tlag1+ error2t
> y1t= y2t + y3t + y4t + y1tag1 + error1t
>
> considering it a triangular (or recusrive--as mentioned in Maddala) system--using OLS I am getting some results. However, when I am considering them occuring for a given time period 't'--so trying tro estimate them jointly by  using 2 Stage Least Squares  (or 3SLS)...I am getting entirely different results in the sense that all the coefficients that are significant in case of OLS are insignificant in 2SLS.
>
> My uestion is : Just because I am changing the method of estimation , why the parameter estimates are  changing so much?
> Any comments/ reference so that where to look for?
> Even it it is model misspecification / non-inclusion of imp variables...it is true for both the case.. So why this difference?
>
>  Kaushik Bhattacharjee
>
--
Brian G. Peterson
http://braverock.com/brian/
Ph: 773-459-4973
IM: bgpbraverock

```