[R-SIG-Finance] Optimization of ARFIMA in rugarch - Quick question and context

[alexios ghalanos]

Nicholas,

Email me off list with the details and I may have some time to look at 
this over the next few days.

Alexios

On 04/12/2014 04:56, Nicholas Manganaro wrote:
> Alexios,
> I determined how to apply rugarch:::.nyblomTest, and have pulled the results.
> Thanks, again.
> I do not have a cloud storage locker, so I did not want to send out the data file to everyone on the list, although I'd appreciate someone having a look at my next few steps. Commentary would be welcome.
> -Nick
>
> -----Original Message-----
> From: R-SIG-Finance [mailto:r-sig-finance-bounces at r-project.org] On Behalf Of Nicholas Manganaro
> Sent: Tuesday, December 02, 2014 10:25 PM
> To: alexios ghalanos
> Cc: r-sig-finance at r-project.org
> Subject: Re: [R-SIG-Finance] Optimization of ARFIMA in rugarch - Quick question and context
>
> Alexios,
> Thank you for your concise answer. I had seen the function nyblom available for use with the uCARCHfit objects. It does not seem to work with ARFIMAfit objects in the same way the identified methods do (please see below for a sample).  However, I have not yet found the key to running nyblomTest on the ARFIMAfit object that results from running the following code snippit:
>
> ## optimization methods = "nlminb", "solnp", "gosolnp", "nloptr"
>
> solv <- c("nlminb","solnp","gosolnp","nloptr")
> auto_nloptr <- list()
> for (i in 1:z)
> {
> auto_nlminb[[i]] <- try(autoarfima(Z.std[,i], ar.max = 5, ma.max = 5, criterion = "BIC", method = "full", arfima = FALSE, include.mean = NULL, distribution.model = "sstd", solver = solv[1], return.all = FALSE), silent = TRUE) }
>
> This process eventually provides ARFIMAfit objects as list members for the sectors for which an optimized solution was found. I ran it solver by solver, creating 4 lists of objects. These solver-specific lists were then made members of a list object which I have been querying, as indicated in the following:
>
> AIC.array[j,k] <- infocriteria(varMod[[k]][[j]]$fit)[1]
>
> Where j is the sector, k is the solver and varMod is the list of lists of ARFIMAFit objects mentioned earlier.
>
> Details:
> R 3.1.2, platform = i386-w64-mingw32/i386, OS = Win 7 SP1 Assume rugarch and zoo are loaded. Z.std is a zoo object of subsets of the monthly return series identified in the previous message with start and end points that are uniform across sectors and with no NA entries. They have been standardized to have 0 mean and st. dev = 1. The data are unweighted as they are submitted to the autoarfima function. The variable z represents the number of sectors, equivalent to the number of series in Z.std.
>
> I hope that is enough detail to help you help me with this.
> I look forward to your response.
> Regards,
> Nick
>
> -----Original Message-----
> From: alexios ghalanos [mailto:alexios at 4dscape.com]
> Sent: Tuesday, December 02, 2014 5:27 PM
> To: Nicholas Manganaro; r-sig-finance at r-project.org
> Subject: Re: [R-SIG-Finance] Optimization of ARFIMA in rugarch - Quick question and context
>
> The returned list from autoarfima has 2 slots, 'fit' and 'rank.matrix'.
> The former is of class ARFIMAfit, and though not exported to work with it, you can use
>
>   >rugarch:::.nyblomTest(fit).
>
> If you provide some additional details (i.e. reproducible code and which data you are using...monthly/daily,value-weighted or equally weighted) then we might be able to provide some additional help.
>
> Regards,
>
> Alexios
>
> On 02/12/2014 21:57, Nicholas Manganaro wrote:
>> Question:
>> Is there a way to query the Nyblom joint statistic result out of an ARFIMAfit class object, the way there is for a uGARCHfit class object?
>>
>> Context:
>> I am running an analysis on subsets from 12 return time series from the Fama/French Factors files (source: http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html) which evidence skewed Student t distributions.
>> I initially used autoarfima from the rugarch package using the nlminb solver, which generated results for some of the series, but not some others. I sequentially tried the other available solvers on the entire set of series to see if they would identify ARMA models for more of these sets.
>> As I hope you can see in the table below the results were mixed. Some observations:
>> •	There was no solver that found ARMA coefficients for all sectors (not necessarily a problem);
>> •	For many sectors, the values estimated for the orders of p and q for AR and MA varied across the solvers; and
>> •	None of the solvers consistently provided either the lowest BIC or AIC score for the results across all sectors.
>>
>> 					Solver			
>> 									
>> 		nlminb		solnp		gosolnp	nloptr
>> 		AR	MA	AR	MA	AR	MA	AR	MA
>>           Manuf  	4	4	4	4	5	4	4	4
>> 	Other  	NA	NA	NA	NA	0	1	0	1
>> 	Money  NA	NA	NA	NA	NA	NA	NA	NA
>> 	Chems  0	5	0	5	0	5	0	5
>> 	NoDur  	NA	NA	NA	NA	NA	NA	NA	NA
>> 	Shops  	NA	NA	NA	NA	3	4	NA	NA
>> 	Durbl  	0	1	5	5	5	5	5	3
>> 	BusEq  	5	3	5	3	5	3	5	2
>> 	Hlth   	NA	NA	NA	NA	0	5	NA	NA
>> 	Telcm  	5	0	5	5	5	5	5	0
>> 	Utils  	3	4	3	4	0	5	3	4
>> 	Enrgy  	0	1	5	3	4	3	5	3
>> 	Mkt.RF 	5	3	3	2	NA	NA	NA	NA
>>
>>
>> While I understand that the Nyblom test is an indicator of the stability of the results across a series, within the model results for a sector from the available solvers, a higher Nyblom joint statistic qualitatively seems to indicate higher quality test results across a number of tests, including the coefficient t statistics and the Ljung-Box results per sector. As a result, I was thinking I might use the Nyblom joint statistic as an indicator in automating the selection of the desired model from possible alternatives, but do not see how to query out that result for further testing as an indicator, although I can see it as part of the $fit object.
>>
>> Please let me know if there is either such a capability for the ARFIMAfit class that I may have overlooked, or a function I might call to generate the joint Nyblom statistic and critical values using the zoo time series data and the existing autoarfima results on hand.
>>
>> Thanks.
>>
>> -Nick
>>
>>
>>
>> _______________________________________________
>> 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.
>>
>
>
>