Hey Real Money, are you Neo Ultra? https://quant.stackexchange.com/questions/51277/process-transforms-fractional-difference If not, you should contact him, he's doing the same thing you are.
Can you replicate the results? Usually closing and reopening excel fixes any broken data that shows values of N/A for me but I've never had any misprints where the data is just incorrect as your picture suggests.
No, I couldn't get rid of the incorrect value. I did open and close the Excel file to no avail. This makes it completely unusable with these sort of bugs, as of course trading decisions will be made based on these data. I can deal with n/a or something, but not with random errors like this. I will not be using this. Never seen anything like this with TWS DDE. It actually gets worse. It actually is switching between that 300 something and correct last randomly as the last changes in the market
To clarify, this period's value minus last period's EMA of that value is strictly a form of fractional differencing as the ema weights on past (t[-1 .. -inf]) sum to one asymptotically. Today - yesterday's EMA: x = P[t] - EMA[t-1] Today - today's EMA is just a re-scaled version of that: x = P[t] - a * P[t] + (1 - a) * EMA[t-1] = (1 - a) * ( P[t] - EMA[t-1] ) Not strictly fractional differencing, but pretty close. If your ema coefficient (a) is small, it will be practically the same thing.
This is interesting. I'm wondering if I should take another look at these methods. Is this related with Autoregressive fractionally integrated moving averages (ARFIMA)?