Impact analysis - financial correlation matrices

Hello everyone,

I am quite new with Matlab so please bare with me. I am doind an impact analysis for financial correlation matrices and I cannot ger the code right. Could you please take a look and advise?

The basic idea is that the accuracy of the financial matrices will vary with the training period used in the estimation. Therefore, I will work with different lengts of the training period in order to minimise the error to the correlation matrix of the investment perios (test period). The data I am using is the S&P 500 stock prices. Here is the basic code:

>> P = COPY;
R = price2ret(P);
lengthInvestmentPeriod = 5*size(R,2);  
InvestmentPeriodReturns = R(end - lengthInvestmentPeriod:end,:);
OracleInvestmentPeriodCorrelationMatrix = corr(InvestmentPeriodReturns); 
TrainingPeriodReturns = R(1:end - lengthInvestmentPeriod - 1,:); 

>> for i = size(TrainingPeriodReturns,2):size(TrainingPeriodReturns,1)  
tempReturns = TrainingPeriodReturns(1:I,:);  % I chose I=1.5 but I am not sure what the range can be 
tempCorrelation(:,:,i) = corr(tempReturns);

end
>> 
>> for i = 1:size(tempCorrelation,3)  
Error(i) = sum(sum(abs(tempCorrelation(:,:,i)-OracleInvestmentPeriodCorrelationMatrix))); 

end
>> plot(Error); 
Undefined function or variable 'tempCorrelation'. 

So in the end I get the above error message and I am not sure why the variable tempCorrelation is not identified.

Could you please advise?

Thank you in advance!

Nadya

[David Goodmanson]

Hi Nadya,

I'm guessing that

size(TrainingPeriodReturns,2) > size(TrainingPeriodReturns,1)

so the loop that defines tempCorrelation(:,:,i) never gets run at all. Just a guess.

The best thing you can do is get familar with the debug option which is a highly valuable tool.

[Nadezhda Dimitrova]

Hi David,

I made it work, thanks for the reply.

I have another question: for the investment (test) period or the crossvalidation I used the first 20 assets but I would like to write another function that randomly samples some stocks, not just the first X assets. Do you have an idea what function could that be?

Currently, my approach is as follows:

FirstXAssets = 20;
R=R(:,1:FirstXAssets);
MinimumConcentrationRatio = 5;
lengthInvestmentPeriod = MinimumConcentrationRatio*size(R,2);  
InvestmentPeriodReturns = R(end - lengthInvestmentPeriod+1:end,:);
OracleInvestmentPeriodCorrelationMatrix = corr(InvestmentPeriodReturns); 

Thanks a lot!

[David Goodmanson]

Hi Nadezhda

randperm(n) supplies a random permutation of the integers 1:n. So, e.g.

a = randperm(100);
b = a(1:20);

gives 20 items randomly chosen from 100 items.

[Nadezhda Dimitrova]

Hi David,

Perfect, thank you very much! It is a great idea and Ive used it in my algorithm :)

I have another problem with the code now but will post another question.

Many thanks again!

Nadya