My own small "revelations" about volatility | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Broker Challenge "NDD/STP" vs. True "ECN" and "Mkt Mkr" broker talk 46 replies

Dealing with "Invalid Price", "Requote", "Server Busy" in MT4 3 replies

Small bells, small whistles. 2 replies

Small time relative newbie taking small steps 177 replies

Reality check - do small FX accounts inevitably stay small? 14 replies

My own small "revelations" about volatility

  • #21
  • Sep 1, 2008 11:42am Sep 1, 2008 11:42am
he uses daily data... not hourly.
  • #22
  • Sep 1, 2008 11:43am Sep 1, 2008 11:43am
Quoting tdion
if you found something that seems to work, good for you!
Typical reply I was expecting from you.

... because you just hate it when you see people finding "something that works for them" and you're stuck with your BS about randomness and judging from your other posts, measly profits or large losses, otherwise I can't explain why you're always bickering, very arrogant and negativist.

A smart thing to say would've been to apologize for your previous comments when you over-reacted and you could've noticed if you read carefully my follow-up posts that I wasn't referring to volatility expressed in pips.It's my fault too that I didn't express this from the start as well.

You could've actually learned a useful thing if you managed to get past your illusions of grandeur and arrogance how to easily asses how strong is the support in a certain pair for instance.

But, obviously, no, my opinions don't coincide with yours, therefore I must be a noob and not a "real" trader like the other 99 you mentioned.

You're the biggest troll of this forum, I just don't know how come you still get to keep your account...the admins must be very pro free speech and against any form of speech censorship.
  • #23
  • Sep 1, 2008 11:46am Sep 1, 2008 11:46am
Quoting Rabid
Using your link... I get:

Did you use a normal distribution for the deviation mean? I've worked quite a bit lately on market stats and found that normal is completely off.
Yes I did, but I also used daily data.I'll try with a log normal as well, good idea. I already mentioned this a few times, on Mataf, look for volatility expressed in %, the 4th column, not the one if pips, I am not blind lol, I can see the one in pips is bigger for GBPUSD.
  • #24
  • Sep 1, 2008 11:48am Sep 1, 2008 11:48am
haha apparently you want to go to war with me on your own thread.

unfortunately i was being sincere when i said "GOOD FOR YOU." what you do with your own money and account is strictly on your shoulders.... if you found a way to make money, it is a good thing (whether you want to dispute or not)


i'm glad you've let the little man in you come out and say what needed to be said. it doesn't change anything. i will ALWAYS post what i think. enjoy.
  • #25
  • Sep 1, 2008 12:01pm Sep 1, 2008 12:01pm
Quote
Yes I did, but I also used daily data.I'll try with a log normal as well, good idea. I already mentioned this a few times, on Mataf, look for volatility expressed in %, the 4th column, not the one if pips, I am not blind lol, I can see the one in pips is bigger for GBPUSD.

Hrm. I told it to use daily data too, altho the little pulldown keeps reverting back to hourly. 170 pips tho looks like daily data.

A lot comes down to how you define volatility. If you define it as absolute pips or if you define it as percentage of average (which is what that % looks like, altho it doesn't say... so I could be wrong there. So % of what?).

If you define it as percentage of average than you're plotting increases/decreases in relative volatility, so while the EURUSD may be moving slightly more of it's average than the GBPUSD, it's overall movement is still less. A higher percentage of a smaller movement is still a smaller movement, hehe. At least if I'm reading that right.

That's why I normally use a combination of ATR indicators. A long term (100 day) average compared to a short term 10-day average. If the 10 day > 100 day, then volatility is increasing, and vice-versa. I run the 10-day math for the last 4 bars (daily, hourly, whatever) so I can clearly see NR3 and NR4 setups and how they compare to the long term. I've found that changing average volatility can tell me a lot about the trend phase.

Either way, it'll come down to how you define your risk and stop loss. If you're using a mechanically fixed stop loss then a higher absolute volatility will be very useful for profits, but might also mean a higher percentage of stopped trades. Perry Kaufman recommends using a standard deviation for stop loss, which would account for both.
  • #26
  • Sep 1, 2008 12:03pm Sep 1, 2008 12:03pm
this is a followup to my % explanation given in #3 in my first post of this page.

i think the % is derived like this:

hypothetical example: (looking back 22 days)
GBPUSD high 2.0210, low 1.9342
EURUSD high 1.5643, low 1.4722

divide the low by the high, and that is the %.
  • #27
  • Sep 1, 2008 12:16pm Sep 1, 2008 12:16pm
Quote
it isn't expressely defined, but it makes the most sense to me.

Ahhh, ok. So it's saying the average daily move is around 1% of price? That makes way more sense.

But doesn't that also mean that higher prices should naturally have lower numbers?

.0097 * 1.8 = 0.01746

versus

.0099 * 1.46 = 0.01445

I wonder how that scaled when the euro was closer to 1.6. Was the percentage lower? If so, then what we're really measuring w/ the percentage isn't volatility as much as the bearish trend.

I guess I'm trying to figure out how a % number translates in terms of trading.
  • #28
  • Sep 1, 2008 12:18pm Sep 1, 2008 12:18pm
i think it deals more with fundamentals (though tenantro may disagree)

if you look back over the past 12 months you could get a feeling for a pair's strength in terms of percentage against itself...
  • #29
  • Sep 1, 2008 12:24pm Sep 1, 2008 12:24pm
Rabid, I just checked, log normal is not a good fit for me.I checked for both small time windows like 22 days and for a very large one, 2500 or so values, from the parametric distributions I get a slightly better log likelihood value for a logistic distribution than for a normal one.

Did you use the log normal distrib for return series as well, or for a price series? I computed the initial standard deviations that I posted previously on a return series.

I could fit a nonparametric distribution in the middle of the CDF and then do a peak over threshold fit with a Generalized Pareto for the tails to get the best fit possible, except I'm not really sure it's worth it.I'll probably try it nevertheless.
  • #30
  • Sep 1, 2008 12:30pm Sep 1, 2008 12:30pm
Quoting tenantro
Typical reply I was expecting from you.

... because you just hate it when you see people finding "something that works for them" and you're stuck with your BS about randomness and judging from your other posts, measly profits or large losses, otherwise I can't explain why you're always bickering, very arrogant and negativist.

A smart thing to say would've been to apologize for your previous comments when you over-reacted and you could've noticed if you read carefully my follow-up posts that I wasn't referring to volatility expressed in pips.It's my fault too that I didn't express this from the start as well.

You could've actually learned a useful thing if you managed to get past your illusions of grandeur and arrogance how to easily asses how strong is the support in a certain pair for instance.

But, obviously, no, my opinions don't coincide with yours, therefore I must be a noob and not a "real" trader like the other 99 you mentioned.

You're the biggest troll of this forum, I just don't know how come you still get to keep your account...the admins must be very pro free speech and against any form of speech censorship.
Lol, Its true that Tdion is a total Cunt. But I enjoy his posts.



Leon
Living the adventure in my head.
  • #31
  • Sep 1, 2008 12:31pm Sep 1, 2008 12:31pm
hey tenantro all you need to do is press the ignore button on my profile.... then you wouldn't have to worry about your feathers ruffling.
  • #32
  • Sep 1, 2008 12:31pm Sep 1, 2008 12:31pm
Quote
Did you use the log normal distrib for return series as well, or for a price series? I computed the initial standard deviations that I posted previously on a return series.

Price series. Altho I'm not all that far into development yet. A logistic distribution may fit better, will depend on the tail obviously.
  • #33
  • Sep 1, 2008 12:46pm Sep 1, 2008 12:46pm
I just realized that fitting a log normal to my return series was out of the question from the beginning because I have negative returns as well, so you must've been using a price series not a return series.
  • #34
  • Sep 1, 2008 1:26pm Sep 1, 2008 1:26pm
hey tenantro, with your DSP background, have you considered downloading the tick data from GainCapital? (available for free...)

in case you don't know, this is each and every tick the pair touches INSIDE of the 1 minute bars.... it is the finest grained data available.

cheers.
  • #35
  • Sep 1, 2008 1:37pm Sep 1, 2008 1:37pm
I did a bootstrap estimation with 500 bootstrapped standard deviations for each of the sample lows and then computed the median of the values which should be a robust estimator. I'll try a trimmed mean as well.

For instance, for NZDUSD, the bootstrapped median standard deviation of the lows for 22 trading days is 0.0097, while the standard deviation computed based on a normal probability distribution is 0.0101.

I tried this on a larger sample size, a 1000 point window, boostrapped median std is 0.0071, standard deviation based on a normal prob dist is exactly the same 0.0071

It seems the larger the sample size, the more "normal-like" the distribution of the returns of the lows is, or at the very least, the computed std based on a normal distribution is not too biased to be taken into consideration.

edit:

tdion, i don't really have any use for high frequency data. the processed daily data i use already expands into matrices of about 100*40000 for each pair.i'm thinking about converting my programs to run on the CUDA-enabled GPU of my Nvidia card.daily data is more than enough.

i'm already very happy with what i have from alpari, also for free.
  • #36
  • Sep 1, 2008 1:55pm Sep 1, 2008 1:55pm
Quote
It seems the larger the sample size, the more "normal-like" the distribution of the returns of the lows is, or at the very least, the computed std based on a normal distribution is not too biased to be taken into consideration.

In "The Hedge Fund Handbook" it's stated that returns follow a normal distribution (or are assumed to follow) while price series data is more fractal. There's some of it on Google books.

I found the overview at:
http://en.wikipedia.org/wiki/Levy_sk...e_distribution

To be a nifty starter. There's a lot of research on financial data statistics.
  • #37
  • Last Post: Sep 1, 2008 2:16pm Sep 1, 2008 2:16pm
Yeah..I knew that to be true at least for returns of the closing prices.I was having doubts about the returns of the lows though, especially since you said lognormal worked better for you...that's why I wanted to test it.Seems bootstrapping was sufficient to do in order to test this.

Though it's not a perfectly normal distribution..I remember estimating a best fit of the returns of the closes some while ago with the method I mentioned earlier, middle estimated nonparametrically, tails with a peaks over threshold using GP distributions.

Anyway, I learned something new today Thank you for mentioning the lognormal earlier and the better fit, I wouldn't have had the incentive to test all of this if it wasn't for that.
0 traders viewing now