[R-SIG-Finance] Cointegration, more than one structural break

[mat]

karla hernandez villafuerte a écrit :
> Hi!
>  
> I have taken your recomendation and I am trying to apply the 
> methodology suggested by Kejriwal and Perron in the paper "Testing for 
> Multiple Structural Changes in Cointegrated Regression Models". I 
> think that if I use the "strucchange" package and the function 
> "breakpoints", I will get the break points and dates with the same 
> methodology described in the paper.
true. So the good news for this paper is that all the estimation part is 
already available thanks to the strucchange pkg
>  
> In order to prove if the number of breaks found by the "breakpoint" 
> function is significant I want to apply the sup-Wald test. I am not 
> sure if a normal supWald test is correct of if I need to implement 
> some kind of modifications. In my case all the coefficients are 
> allowed to change. I send you the program that I am using.
I feel you will need to look more precisely into the paper:
-KP don't use fluctuation if I remember well, no?
-the test statistic (here wald test) is right, but they use also 
modified forms  in case of autocorellation or endogeneity
-the critical values have to be taken from their paper

Hope this helps

Matthieu
>  
> > ##BREAKS POINTS CALCULATION
> >
> > lrun
>
> YX[, 1] ~ YX[, 2] + dattowork[, 1] + 0
> >
> > #Empirical Fluctuation process
> > flproc <- efp(lrun, type = "RE")
> > print(sctest(flproc))
>  
>         RE test (recursive estimates test)
> data:  flproc
> RE = 3.6801, p-value = 6.893e-12
>  
> > #Bai and Perron
> > breaks<-breakpoints(lrun, h = 0.15, breaks = 3, hpc = c("none"))
> > print(breaks)
>          Optimal 4-segment partition:
> Call:
> breakpoints.formula(formula = lrun, h = 0.15, breaks = 3, hpc = c("none"))
> Breakpoints at observation number:
> 54 127 155
> Corresponding to breakdates:
> 1994(7) 2000(8) 2002(12)
>
> > dummy<-ts(rep(1,length(YXSR[,1])),start=startlag,frequency=12)
> > dummy1<-dummy; 
> dummy1[breaks$breakpoints[1]:(breaks$breakpoints[2]-1)]<-0
> > dummy2<-dummy; 
> dummy2[breaks$breakpoints[2]:(breaks$breakpoints[3]-1)]<-0
> > lrunbreak<-update.formula(lrun, . ~ . + dummy1 + dummy2)
> > longrbreak<-dyn$lm(lrunbreak)
>
> > Waldt<-wald.test(vcov(longrbreak), longrbreak$coefficients, Terms=3:4)
> > print(Waldt)
> Wald test:
> ----------
> Chi-squared test:
> X2 = 28.0, df = 2, P(> X2) = 8.2e-07
>
>  
> In advance thank you for your help!,
>  
>  
> Karla Hernández
>  
>  
>  
>  
>  
>  
>  
>
>  
> ------------------------------------------------------------------------
> Date: Wed, 27 Jan 2010 15:08:05 +0100
> Subject: Re: [R-SIG-Finance] Cointegration, more than one structural break
> From: matthieu.stigler at gmail.com
> To: karlavhv142 at hotmail.com
> CC: r-sig-finance at stat.math.ethz.ch
>
> Hi
>
> Testing for multiple breaks in the cointegration framewk is 
> unfortunately for now not possible in R.
>
> For one break, you have a test in the urca/vars package, as well as 
> the Gregory Hansen Test (I had implemented it so I can send once raw 
> code if needed).
>
> If you want to implement a test for multiple breaks in cointegration, 
> have a look at the work of Kejriwal and Perron.
>
> Hope this helps
>
> Mat
>
>
>
> 2010/1/25 karla hernandez villafuerte <karlavhv142 at hotmail.com 
> <mailto:karlavhv142 at hotmail.com>>
>
>
>
>
>     Dear Members:
>
>
>
>     I want to analyse the cointegration rank using the Engle and
>     Granger methodology and taking into account the presence of more
>     than one structural break in the long run relation.
>
>     I can not find how to do that using R. If some one have a
>     suggestion I would be very glad.
>
>
>
>     Thank you!
>
>
>
>     Karla H.
>
>     _________________________________________________________________
>
>
>            [[alternative HTML version deleted]]
>
>     _______________________________________________
>     R-SIG-Finance at stat.math.ethz.ch
>     <mailto:R-SIG-Finance at stat.math.ethz.ch> mailing list
>     https://stat.ethz.ch/mailman/listinfo/r-sig-finance
>     -- Subscriber-posting only. If you want to post, subscribe first.
>     -- Also note that this is not the r-help list where general R
>     questions should go.
>
>
>
> ------------------------------------------------------------------------
> Con Hotmail hay una forma muy fácil de encontrar tus emails: 
> organízalos por carpetas. Conoce cómo 
> <http://www.descubrewindowslive.com/hotmail/organizacion.asp>