[R-SIG-Finance] Howto test for true stationarity with adfTest()

[Matthieu Stigler]

I would suggest:

1 test for unit root:
     if unit root: stochastic trend
     if not: 2

2 test for trend in normal AR(MA) model
     if not trend: "pure stationary" as you call
     if trend: trend stationary

Note if just the question of trend is of interest for you, you might be 
interested in using a test for trend indeopendeant of I(1)/I(0). Such a 
test has been developed in:

David I. Harvey, Stephen J. Leybourne, A.M. Robert Taylor
A simple, robust and powerful test of the trend hypothesis, J 
Econometrics 2007, 1302-1330

Mat



Le 20. 01. 11 07:48, Mark Breman a écrit :
> Thanks Paul, you are right; if the trend is stronger it's revealed by the
> adfTest.
>
> As a follow-up question:
> Does anyone know if a test exists that would reveal a weaker trend in the
> data?
>
> Regards,
>
> -Mark-
>
> 2011/1/19 Paul Teetor<paulteetor at yahoo.com>
>
>> Mark,
>>
>> The adfTest with type="nc" works for me. I think your trend time series is
>> not "trendy" enough. When I change the assignment statement to this:
>>
>>      trend=notrend+(row(notrend)*.4)         # Change 0.04 to 0.4
>>
>> and execute this:
>>
>>      adfTest(coredata(trend), lag=0, type='nc')@test$p.value
>>
>> I get a p-value of 0.98.
>>
>>
>> Paul Teetor, Elgin, IL USA
>> http://quanttrader.info/public
>>
>>
>> ------------------------------
>> *From:* Mark Breman<breman.mark at gmail.com>
>> *To:* r-sig-finance at r-project.org
>> *Sent:* Wed, January 19, 2011 4:38:56 AM
>> *Subject:* [R-SIG-Finance] Howto test for true stationarity with adfTest()
>>
>> Hi everyone,
>>
>> I want to test for true stationarity (not trend-stationarity) in my
>> timeseries. I understand that i can't use the adf.test(), because it
>> de-trends the series.
>> Indeed, i can verify this:
>>
>>> notrend=xts(rnorm(100), Sys.Date()-100:1)
>>> trend=notrend+(row(notrend)*.04)
>>> adf.test(coredata(notrend), alternative="stationary", k=0)$p.value
>> [1] 0.01
>>> adf.test(coredata(trend), alternative="stationary", k=0)$p.value
>> [1] 0.01
>>
>> The p-values are the same so it's not usable for my purpose.
>>
>> I'm told i can use the adfTest() from package fUnitRoots instead because
>> this test does not de-trend the series.
>> So i try:
>>
>> library(fUnitRoots)
>>
>>> adfTest(coredata(notrend), lag=0, type='c')@test$p.value
>> 0.01
>>> adfTest(coredata(trend), lag=0, type='c')@test$p.value
>> 0.01
>>
>> Again, the p-values are the same!? Changing the type parameter (to 'ct' or
>> 'nc') does not help either (it gives the same p-values).
>> What am i doing wrong here?
>>
>> Thanks,
>>
>> -Mark-
>>
>>      [[alternative HTML version deleted]]
>>
>> _______________________________________________
>> R-SIG-Finance at r-project.org 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]]
>
> _______________________________________________
> R-SIG-Finance at r-project.org 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.