Pairs Trading Code: Further Guidance Please

Good afternoon.

I am trying to design a Pairs Trading code: I am using Matlab R2011b version. The dataset I am using is the S&P500, 1982 – 2012 Daily.

The procedure includes:

1. A distance measure on each 12 month period (formation) sliding by 6 month period (Trading)
2. Normalize
3. Do pair wise distance measure (basically a 500, 500 matrix)
4. Find pairs which minimize the distance measure (the sum of squared deviations between the two normalized price series.)
5. Filter and identify top ranking pairs formation.

I would like to know:

1. What are the next coding steps needed to find the pair wise distance measure criterion,
2. What are the next coding steps to minimise the distance measure?
3. How do I filter and identify the top ranking pairs integrating the time period sliding?
4. How would I include a restriction on stocks being matched in the same industry sector according to the relevant SIC codes?

The code completed so far is below inclusive of note headers (%) for simplification.

Any help, guidance or advice would be very much appreciated.

Kind regards

Tomasz Mlynowski

% Import data
[data,text] = xlsread('G:\me\Desktop\Pairs Trading\sp500 price data.xlsm',1);
% Dates in numeric format
dates       = datenum(text(4:end,1),'dd/mm/yyyy');
% Keep names (for reference)
names       = text(1,2:end);
% Free memory from text (more than 80 MB)
clear text

% Retrieve year and month
[y,m]       = datevec(dates);

% Unique pairs of year-month
ym          = unique([y,m],'rows');

% Distance measure on each 12 month block sliding by 6 (except last block is 5)
% For reference, appendix: http://www.tinbergen.nl/discussionpapers/11150.pdf
for r = 12:6:numel(y)
    % Select the data for the formation period
    tmp    = data(r-11:r,:);

      % 1. Normalization

      % Find columns with at least one non NAN value
      idxnan = isnan(tmp);
      cols   = find(~(all(idxnan)));

      % LOOP through columns
      for c = cols
          % First non NaN value 
          first    = find(~idxnan(:,c),1,'first');
          tmp(:,c) = tmp(:,c) / tmp(first,c);
      end

      % Now, don't loop by single column because you need to normalize taking
      % into consideration pairs.

%         LOOP through columns
%     for c = cols
%         Now select a specific column and loop against all others
%         for cpair = setdiff(cols,c)
%             first1 = find(~idxnan(:,c    ),1,'first');
%             first2 = find(~idxnan(:,cpair),1,'first');
%             
%             tmp(:,c) = tmp(:,c) / tmp(first,c);
%         end
%         
%         
%         First non NaN value 
%         first    = find(~idxnan(:,c),1,'first');
%         tmp(:,c) = tmp(:,c) / tmp(first,c);
%         
%     end

end

%Pairwise distance measure
% Minimum distance criterion identification