Compute result appears slightly "biased"

[Lawrence]

Hi,

I am taking a timeseries of forex prices.

I process the data to get an array of differences: y[i] = X[i]-x[i-1]

And I use this as my input and output!

The solution results are off center from 0, and have more spikes to top-side than the bottom. The data itself is more centered. That feels wrong to me.

I wanted to upload a picture but I get "could not upload attachment" error.

Code I am using is based on your timeseries demo.

Any thoughts?

[Lawrence]

Here are the graphs to illustrate the issue.

This is the Deltas (differences) - you can see red dots look to be centered around zero. But the blue line has a definite "upside" bias.

I have Sigmoid Alpha set as 1.

DeltaPrice.PNG (106.97 KiB) Viewed 5885 times

Just for good measure I have included a graph of the original data, before taking differences

RawPrice.PNG (35.15 KiB) Viewed 5885 times