ATR values keep changing for same bar

[richthomas]

I originally posted something similar to this in the back testing forum but have not had a satisfactory response.

I have a simple EFS study that looks at the atr(14) value for all the 10 minute bars from 1st Sept 2008 to 31st October 2008 and writes them to a file together with the run date. There are 1716 bars in total for this time frame.

At various times I run this for the same symbol and same dates and find that the atr values for EXACTLY the same bar keep changing.

For example I ran the test against EEV on 20th November once at 10:08 and again at 14:04 GMT. I copied the results to an excel spreasheet and comapared the values for each run for the same bar. EVERY bar had a different value ATR. If I round the values to 3 decimal places and do a comparison there are still 832 bars that have different values between the two runs.

How can this be happening?

[Dave180]

If Wilder smoothing is used by eSignal (I think some people use sma, not ema, don't know with eSig) then there is always some difference unless you start with the same bar (exponential smoothing keeps ALL the historical values in the calc).

ATRi = (ATR(i-1) * (n - 1) + TR) / n

See definitin on, say, Stockcharts.