the reference correlation: population Pearson correlation of two series over
period. Each window takes two passes, oldest first: both means are finished
before any deviation is formed, then
(cross / period) / (sqrt(squaresA / period) * sqrt(squaresB / period)).
A window with a missing value or an overflowing step is NaN, and so is one
whose deviations are all exactly zero, such as three bars of 5. A flat window
whose mean is inexact is not caught by that: three bars of 0.1 average to
slightly more than 0.1, the deviations are a few units in the last place, and
the window reads what that arithmetic gives, as the reference does. Against a
bar index that is exactly 0; against another series it is within rounding of
0. A period that is not a whole number above 1 gives NaN throughout, as it
did before.
the reference
correlation: population Pearson correlation of two series overperiod. Each window takes two passes, oldest first: both means are finished before any deviation is formed, then(cross / period) / (sqrt(squaresA / period) * sqrt(squaresB / period)). A window with a missing value or an overflowing step is NaN, and so is one whose deviations are all exactly zero, such as three bars of 5. A flat window whose mean is inexact is not caught by that: three bars of 0.1 average to slightly more than 0.1, the deviations are a few units in the last place, and the window reads what that arithmetic gives, as the reference does. Against a bar index that is exactly 0; against another series it is within rounding of 0. A period that is not a whole number above 1 gives NaN throughout, as it did before.