There is no Cloud Parallel Computing for Home Users ?

I have a copula fitting computing problem with 21 stocks, which with 4 cores an 8 hours running it cannot compute. Therefore I tried to look for a solution in the cloud, however I found out that Home Users cannot have access to then....therefore no solution to my problem, is that correct ?

[Stephan]

Maybe, may also be your code could be optimzed to run faster. Since we dont know your code it can not be said.

[Flavio]

Thanks Stephan. Below is the code:

 lowerTail = 0.07;
upperTail = 0.93;

% K in the code is 21 (stocks)

 for idx = k:-1:1
    currentRet = returns(:,idx);
    marginal{idx} = paretotails(currentRet, lowerTail, upperTail, 'kernel');
    U(:,idx) = cdf(marginal{idx}, currentRet); 
end

 % THE PART BELOW IS THE ONE THAT DOES NOT END CALCULATING (EVEN AFTER 8 HOURS)

 [rho, nu] = copulafit('t', U);

[Stephan]

If this is your whole code you do not calculate in parallel. Or is there some more code?

[Flavio]

Only that last line that needed to use parallel computing, the test of the code is computed in a good manner.

I changed the method of the copulafit function to ‘ApproximateML’ and it ran fine now.

[Stephan]

Glad to hear. Please accept helpful answers in order to help people with similar Problems find helpful answers.

Answer 1

[Stephan]

Hi,

your code is not written in a way that it performs the calculations in parallel - have a look at statset. This is the way you tell the copulafit function to do parallel and other options. First find out the standard options for copulafit:

>> oldopts = statset('copulafit')

oldopts = 

    struct with fields:

            Display: 'off'
        MaxFunEvals: 200
            MaxIter: 100
             TolBnd: 1.0000e-06
             TolFun: []
         TolTypeFun: []
               TolX: 1.0000e-06
           TolTypeX: []
            GradObj: []
           Jacobian: []
          DerivStep: []
        FunValCheck: []
             Robust: []
       RobustWgtFun: []
             WgtFun: []
               Tune: []
        UseParallel: []
      UseSubstreams: []
            Streams: {}
          OutputFcn: []

Parallel is not active by standard. To set parallel, start a parallel pool first by parpool command:

>> parpool
Starting parallel pool (parpool) using the 'local' profile ...
connected to 4 workers.

Then use:

>> options = statset(oldopts,'UseParallel',true)

options = 

    struct with fields:

            Display: 'off'
        MaxFunEvals: 200
            MaxIter: 100
             TolBnd: 1.0000e-06
             TolFun: []
         TolTypeFun: []
               TolX: 1.0000e-06
           TolTypeX: []
            GradObj: []
           Jacobian: []
          DerivStep: []
        FunValCheck: []
             Robust: []
       RobustWgtFun: []
             WgtFun: []
               Tune: []
        UseParallel: 1
      UseSubstreams: []
            Streams: {}
          OutputFcn: []

Of course you can do the same in your script and supress outputs with ';' - but for showing hot the difference is, it is better this way i guess.

Then change the call of copulafit so that it uses those options above:

[rho, nu] = copulafit('t', U, options);

Now the calculation should run in parallel mode.

If this is still to slow consider:

[rho,nu] = copulafit('t',[u v],'Method','ApproximateML', options);

This method can be faster but less accuracy on small or middle data - resd this link of the documentation to learn more:

https://de.mathworks.com/help/stats/copulafit.html#bumtt42-Method

Best regards

Stephan

[Flavio]

Thanks a lot Stephan ! I will try it out and let you know

[Flavio]

Stephan, just a correction in the last line:

[rho, nu] = copulafit('t', U, 'Options',options);

The code activate Parallel Computing mode, however, not to run this function. A technical lady from Matlab told me PC does not run in the copulafit function...

I have another question posted regarding how to setup 'Conditional' lower bounds in the PortfolioCVaR object in the same way we have for the Portfolio object. Ex: lower bound could be 0 or 0.02. Do you have this answer ? Thanks a lot!

[Stephan]

I tried to answer here.