Rolling Hinich Bicorrelation & Rolling Hurst Exponent Tests

[vmadhavan]

I understand that @Bicorrtest and @hurst procedures in RATS can be utilized to execute Hinich Bicorrelation test and Classical R/S test respectively. I have come across many papers in journals such as International review of Financial Analysis, Physica, Journal of International Financial Markets Institutions & Money, that employ the rolling hurst exponent and rolling hinich bicorrelation test so as to explore the evolving efficiency of global stock markets. Are there any routines available in RATS that could be used to execute Rolling Hurst Exponent and Rolling Hinich Bicorrelation tests?

[TomDoan]

There's nothing I can see about either algorithm that would benefit from any optimization when done using rolling samples. So it's basically a matter of running a loop over the existing procedures as described in Section 11.2 of the RATS v8 User's Guide.

[vmadhavan]

Dear Tom

I have been using RATS for quite some time now. Having said so, I am not all that good in programming.

I tried writing a code for rolling hurst technique, wherein

a) my input prices for an instrument are converted into logarithmic returns

b) a rolling Ar(1)-GARCH(1,1) of fixed length (1000) is estimated and

c)standardized residuals from rolling GARCH is then subjected to @hurst procedure to estimate the hurst exponent for every rolling iteration

I have disabled the ln r/s vs/ ln n graph of hurst exponent when I run it.

But for every iteration, @hurst brings up this dialog box...I understand that I can disable that by commenting out the lines 107 to 115 of the @hurst code

My input time series (daily prices) length is 2127; and hence logarithmic returns length is 2126.
As said earlier fixed length of each rolling ar(1)-garch(1,1) estimation is 1000. This leaves me with a total 1126 iterations.
Standardized AR-GARCH residuals of each of these 1126 iterations is subjected to @hurst(nograph) proc to estimate hurst exponent.

I encounter the following problem, and have been struggling with this for more than a day now.

After 29 rolling AR(1)-garch(1,1) followed by @hurst iterations, RATS 8.0 gives the following message
*******************************************************
## M4. A memory request for an additional 32768 bytes cannot be satisfied
The Error Occurred At Location 742, Line 40 of HURST
Called From Location 254, Line 5 of loop/block
*****************************************************

How do I overcome this problem? After a lot of experimentation, I was able to run the rolling iterations of 25 at a time. Then I save the output. Close RATS, reopen RATS, read the dataset again, and then fine-tune the coding and run the next 25 iterations. And I see that I have to do this about 45 times for completing all rolling 1126 iterations. And I plan to do this for 14 time series.

How do I overcome this error and make it (ar(1)-garch(1,1) followed by @ hurst) run for all 1126 rolling iterations in one shot? Please help.

P.S.: I tried @rsstatistic(classical) proc, but it yields R/S statistic and not hurst exponent. I need hurst exponent.

I hereby request for your help and guidance

I am attaching the program file(tom.rpf) and the dataset(SB1.xlsx)

Hoping to hear from you

Sincerely

Dr. Vinodh Madhavan

[TomDoan]

I've posted a revised version of @Hurst at http://www.estima.com/forum/viewtopic.php?f=7&t=2060 which has NOPRINT and NODIALOG options to get rid of the output and the dialog for choosing the entry range for the R/S regressions. It looks like if you use those options, it will work.

BTW, why do you think there will be anything interesting in looking at the Hurst on the standardized residuals from a GARCH?