Predicting change in data using a probabilistic method...

Ok guys / girls - Very coarse way of predicting a change in data; the idea of the algorithm is to register a counter value when it thinks there will be a change in the direction of data. The code is below;

function [output] = analyseTrends(data,threshold)
%%Check for a threshold
if (isempty(threshold) == 1)
  threshold = 0.3;
end
%%Generate information about the data
L = length(data);
dfdt = gradient(data);
dd = sign(dfdt);
%%Initialie Variables
store            = [];
test             = 'true';
i                = 1;
succession_count = 0;
probability      = 0;
%%Calculate a probability of change in the direction of data
for i = 2 : L;
  if     (dd(i) ~= dd(i-1))
      probability = 0;
      succession_count = 0;
  else
      succession_count = succession_count + 1;
      if     (succession_count == 1);
          probability = 0.25;
      elseif (succession_count == 2);
          probability = 0.85;
      elseif (succession_count == 3);
          probability = 0.95;
      elseif (succession_count == 4);
          probability = 0.20;
      else
          ...
      end
  end

    if (rand(1) <= probability)
        store = [store;i];
        probability = 0;
        succession_count = 0;
    else
        ...
    end
end
output{1} = store;
output{2} = dd;
end

If you then look at output{1} and output{2} we can see output{1} is the point at which the counter is registered, and output{2} is the sign of the gradients between each point. Okay my question is this: is the value being stored in output{1} the index before the change in sign of the gradients e.g:

    output{1} = [3 5]
    output{2} = [1 1 1 -1 -1 1]

I hope you understand the question!

[Andrew Sanderson]

Ha! Just tested it; turns out on say 20 items of data (taken from google finance) its got an average accuracy of 60% but for larger data sets its down at 35%... Also; any ideas on improving it but keeping the idea the same would be much appreciated! Cheers