How to calculate EWMA?

I have a basic question. I hope its not too technical.
I wanted to calculate EWMA. I trade using FXCM's MArketscope 2.0

What I don't understand is how the system calculates the exact values of EWMA using the previous 25 periods.

 

1. Does it use open or close prices?
2. What is the value of the smoothing parameter?
3. Can anyone suggest a way to choose the optimal smoothing parameter?

Any input/help will be appreciated.

What does EWMA stand for? Is that Exponential Weighted Moving Average or what????


Regards,

Nader

yes, that is correct

here's how you calculate it.


http://en.wikipedia.org/wiki/Moving_...moving_average

My only question so far is regarding the smoothing parameter.

Anyone???

Is it something that people never really bothered calculating?

Steve

edit:
http://stockcharts.com/school/doku.p...oving_averages

Sure, we calculate them all the time. The eWma term threw me, but you're pointing at the wiki entry for standard old ema's.

It's really as simple as that formula:

1. Chose your period. EMA's are quoted in TF and "length", like D200 being a 200 day. But EMA's don't really have length in the same sense as, say, a simple MA. So, the calculation of equivalent volatility of the ema and an SMA is done, which gives the value of the exponential ("smoothing") parameter as: a = 2/(N+1) where N is the length. That doesn't mean an ema and sma of the same N will be identical; it means that they will both tend to be equivalently "wriggly" and responsive as they pass through price data. It's worth noting, however, that an ema of N will tend to have lower lag than an sma of N when there is a "surprise" move up or down.

2. Begin calculating the average (say, E for EMA) by E[0] = P[0], the first observation.

Thereafter, E[i] = (1-a)*E[i-1] + a*P[i]. P[i] means the price at the i-th bar from the start.

3. Don't use the thing until it is "tuned up." It's completely "tuned" by between 3 and 4 times the length ("N").

That's all there is to it.

OldQuant,
It's been a while since I've performed this calculation, but yes, that is the standard. I think it approximates an exponential weighted rolloff for a given equivalent simple average period?

I haven't done the calculations for some time as I still find the applications and metrics pretty subjective.

Am I delusional? Didn't you just post that you stood by your assertions of "subjective?" If you did or didn't, choice of parameter is very subjective.

And I think you asked about "equivalent volatility." That's where the rule of thumb of 2/(N+1) = a comes from. Basically, what's done is that, if the underlying "bar difference" metric (like close, or midrange, or average quote, or whatever) has an assumed constant value (sigma), then an sma of N bars has a volatility that is 1/sqrt(N) times the sigma of the metric itself. If you equate that to the ema (using the expanded exponential, infinite series form rather than the shorthand form), then the alpha that gives you the same vol is the 2/(N+1).

But you're right in that the performance profiles of ema's are very dependent on their frequency.

Nope. Yes, I wiped the trail to avoid muddying my original intention. I haven't seen it expressed as equivalent volatility (or at least that I recall), I thought it was 'objectively' derived as an approximation to an exponential rolloff decay basis function convolved with the original signal over an N bar window (while still preserving the exponential decay rate).

Nevertheless, I tend to look at the application as subjective, and the parameter choice as well, hence my initial rant. I'm going to check back on your explanation. thanks.

That's true of all moving averages, regardless of their complexity, though.

For a while I was using some of John Ehlers' Gaussian filters, which, after you cut through the signal processing talk, are basically triangular moving averages. Depending on the frequency, and the number of times they were iterated, we could get them to lead and then overshoot a market top or bottom by a phenomenal amount. That sounds good, but actually made them next to worthless.

And then there's the whole issue of lag....

The bottom line, though, is that one way to characterize functions of past prices (and what in technical analysis or market dynamics isn't a function of past prices) is this:

If the weights assigned to the past prices add up to one, then the thing is a moving average, for all the good and bad that will entail.

If the weights assigned add up to zero, then the thing is an oscillator, for all those entail.

Beyond, and within, those distinctions, it's pretty much a "choose your poison" decision.

Has somebody the indi for the EWMA from the link below? :

http://www.investopedia.com/articles/07/EWMA.asp

if not any thoughts how to implement it?

Thanks!