Question

I’m looking to implement an exponentially weighted moving average with a sliding window using data that I am pulling from a large data set.

The code works but the results are definitely not what they should be and I can’t seem to figure out why. Here is my code and please give me good detail as to what I am exactly doing wrong:

for(unsigned int i = window; i< close_price.size(); i++)
{
     double tmp3;
     double tmp4;
     for(int j = 0; j < window; j++) 
     {
          tmp3 += pow(lambda,j) * pow(close_price[i-j], 2);
          tmp4 += pow(close_price[i-j], 2); 
          if(j == window-1)
          {
              double temp = (1-lambda) * (pow(close_price[window], 2) + tmp3);
              ewma.push_back( sqrt(temp) );
              sma.push_back( tmp4/window );
          }
          tmp3 = 0;
          tmp4 = 0;
    }
}

Answer 1

Basically the problem is that tmp3 and tmp4 are not initialized, so tmp3+=blah has undefined result.

The way I see your code now, it should look something like this:

for(size_t i = window; i< close_price.size(); i++)
{
     double tmp3 = 0.0;
     double tmp4 = 0.0;
     for(size_t j = 0; j < window; j++) 
     {
          tmp3 += pow(lambda,j) * pow(close_price[i-j], 2);
          tmp4 += pow(close_price[i-j], 2); 
     }
     double temp = (1-lambda) * (pow(close_price[window], 2) + tmp3);
     ewma.push_back( sqrt(temp) );
     sma.push_back( tmp4/window );
}

Explanation: there is no need for the extra if within the for loop, as j’s last value will be window-1 anyway, tmp3 and tmp4 are to be initialized at every i-loop start. The type of size() is size_t not unsigned int, if any.