Hello
My apologies for the inconvenience and thanks a lot for the answers provided!
For 1.2 and 1.3, one common theme was whether the season parameter depended on
periodogram of the series' involved or whether it is the frequency at which the
data is collected. The details are below:
Sample Data: (Total 62 data points)
Date logdiff logexch
2003-Jan -1.07173 4.77811
2003-Feb -1.07496 4.78164
2003-Mar -1.09263 4.78582
2003-Apr -1.08933 4.79206
2003-May -1.09413 4.76149
2003-Jun -1.09396 4.71402
2003-Jul -1.09613 4.69993
2003-Aug -1.09630 4.69711
2003-Sep -1.10494 4.67488
2003-Oct -1.08094 4.66070
2003-Nov -1.09220 4.69236
Commands used for 1.1:
m1co1 = ca.jo(cor2,type="eigen",ecdet="const",K=2,season=12,spec="transitory")
m2co1 = ca.jo(cor2,type="eigen",ecdet="const",K=2,spec="transitory")
(both resulted in order of cointegration =1 at 5% significance level)
Further, I used conversion command to utilize the capability of VAR for
prediction:
vecm.lvl.m1 = vec2var(m1co1,r=1)
and finally used predict(vecm.lvl.m1) command to get the forecast for which I
found out the MAPE manually for both the m1co1 and m2co1 models and got the
MAPE to be lesser for m2co1 as compared to m1co1. So the question is whether to
actually use season parm or not..
Finally, 1.2:
m1co1 = ca.jo(cor2,type="eigen",ecdet="const",K=2,season=2,spec="transitory")
Error in while (nrow(dums) < N) { : argument is of length zero
Now, I dont know whether theoratically 2 is a "correct" value to be passed or
not, but it does give above error.
I'm sorry if above doesn't help in improving understanding of my doubts. Thanks
again for the explanation, it is just enough helpful so as to allow me to carry
on with the work.
Gunjan Narulkar,
Ist Year M. Mgmt., DOMS,
Indian Institute of Science
Contact: +91-99007-40404
LinkedIn: in.linkedin.com/pub/gunjan-narulkar/19/a3b/521
________________________________
From: Matthieu Stigler <[email protected]>
Cc: "[email protected]" <[email protected]>
Sent: Wednesday, June 5, 2013 1:22 PM
Subject: Re: [R-SIG-Finance] Regarding significance of "Season" parameter
Hi
This is the right forum to ask this, not sure though it is the right form ;-)
You are asking here many questions, some of which cannot be answered without
reproducible code. So let me just answer 1.1 and 2:
1.1: Not the seasonal components refer to deterministic seasonality, not
stochastic. So this is nto the same as ARIMA vs SARIMA, since SARIMA concerns
stochastic seasonality. There have been some papers on seasonal cointegration
(what corresponds to SARIMA), but not very popular, and not implemented in R
afaik.
1.2/1.3: not clear or no code to answer
2: yes, you are right, ecdet="const", restricts the constant to enter the coint
relationship.
Best
Matthieu
Hi,
>
>I'm trying to learn about cointegration, specifically about how to use "ca.jo"
>for finding the cointegration basis. The data is that of FX rates and M1
>supply difference. I need help understanding the below two points:
>
>1. Seasonal variables:
>
>-> What is the importance of season parameter apart from seemingly obvious
>explanation in the documentation; in other words, should it be understood as
>equivalent of Seasonal ARIMA vs ARIMA where we take care of the seasonal unit
>roots?
>
>-> Should the parameter be set to the value at which the Y_t under question is
>sampled? Or should it be based on some common frequency derived from
>individual Y_t component series's periodicity as found from their periodogram
>(spectrum command in R)?
>
>-> Should the season paramater must be greater then 2? As by spectrum of M1
>and FX rate series, I was getting the prominent frequencies for both variables
>as 2 and its multiples, but got below error, which got resolved as soon as I
>used anything >2:
>
>"Error in while (nrow(dums) < N) { : argument is of length zero"
>
>Background: I tried checking for cointegration between two monthly series,
>taking the "season" parameter as 12 (as I had monthly data) first time and
>without having any season parameter the second. The order of cointegration in
>both the cases was 1. But further, when I tried fitting VECM and using vec2var
>created 6 months ahead forecasts and calculated MAPE (Mean Absolute Percentage
>Error), the MAPE for ca.jo output without season parm specified was better
>then with season parameter - which lead me to the above confusions.
>f
>
>2. ecdet paramter:
>The awesome book as well as documentation describe this parameter nicely. But
>when I use it, the message that comes in the output is a bit confusing:
>
>> cv1.m1.bop = ca.jo(cor2,type="trace",ecdet="const",K=2,season=12,dumvar=bop)
>> summary(cv1.m1.bop)
>
>
>######################
># Johansen-Procedure #
>######################
>
>Test type: trace statistic , without linear trend and constant in
>cointegration
>.
>.
>.
>
>The confusion is that I'm interested in finding out presence of "restricted
>constant", so I used "ecdet='const'". Am I correct in doing so?
>
>Apologies if this is not the right forum for asking these questions and also
>for the long mail.
>
>Thanks & Regards,
>
>Gunjan Narulkar,
>Ist Year M. Mgmt., DOMS,
>Indian Institute of Science
>Contact: +91-99007-40404
>LinkedIn: in.linkedin.com/pub/gunjan-narulkar/19/a3b/521
>
>_______________________________________________
>[email protected] 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.
>
[[alternative HTML version deleted]]
_______________________________________________
[email protected] 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.
