Quantitative and Algorithmic Trading | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Algorithmic/automated trading and emotions 13 replies

Learn Computer and Data Science with Algorithmic Trading 5 replies

Attachments: Quantitative and Algorithmic Trading
Exit Attachments

Quantitative and Algorithmic Trading

  • #21
  • Jun 20, 2013 3:01pm Jun 20, 2013 3:01pm
Important Paper

A Mean-Reversion Theory of Stock-Market Crashes. Eric Hillebrand


http://www.bus.lsu.edu/economics/papers/pap03_10.pdf
  • #22
  • Edited Jun 21, 2013 1:02am Jun 20, 2013 3:02pm | Edited Jun 21, 2013 1:02am
Visual Depiction of SMA vs EMA Weighting

http://marketsci.wordpress.com/2013/04/28/visual-depiction-of-sma-vs-ema-weighting/
  • #23
  • Jun 20, 2013 10:43pm Jun 20, 2013 10:43pm
Thanks, nice. Unfortunately I seem to be unable to read most of the links - like the ones in posts4,7,12

Thanks
  • #24
  • Jun 21, 2013 1:03am Jun 21, 2013 1:03am
Hi, thanks for bringing that to my attention. There has been a problem with shortening URL's. I have repaired that, so that the links are fully accessible now


Quoting Forex37
Thanks, nice. Unfortunately I seem to be unable to read most of the links - like the ones in posts4,7,12 Thanks
  • #25
  • Jun 21, 2013 1:13pm Jun 21, 2013 1:13pm
Quite a lot of material to read! Thank you for that.

Quoting infinitus
Does Averaging-In Work? http://epchan.blogspot.co.uk/2010/01...g-in-work.html read the comment section, too
I thought the link was about averaging up (aka adding in) but it's about averaging down (aka adding on a loser). The example is yet quite biaised as the author (almost) assumes the price will go down to $1. His argument is only valid under this condition. It's obvious that if one knew in advance the price will decrease one wouldn't buy now!

If you have that, I'm interested in getting the same kind of document (but more objective!) for adding-in as I feel like it is a hidden over-leverage (http://www.forexfactory.com/showthread.php?t=411923 p9) but couldn't really prove it...

Quoting infinitus
Find The Right Fit With Probability Distributions http://www.investopedia.com/articles...stribution.asp
Sadly this article forgets about the Cauchy distribution. This distribution is quite important IMO because it has no mean and no variance and it fits the FX tooooo well (http://www.forexfactory.com/showthre...58#post6646358). Hard to think about a mean reversion strategy at second or third standard deviation when you have no mean to reverse to and no std.dev defined anyway! Fitting this distribution to the market gives a good reason to never trade without a SL.
No greed. No fear. Just maths.
  • #26
  • Jun 22, 2013 12:35am Jun 22, 2013 12:35am
Hi Pipmeup,

Reverting to mean strategies are a bit tricky. As you noted there are sometimes
Distributions that do not follow textbook Gaussian curves.

Add to this, that in Forex we do fairly often see really fat tails.

A general problem is, that mean reverting strategies are not non-fragile
but fragile. That means, in betting on mean reverting we have the situation
of limited gains (the mean) by facing unlimited losses eventually.

What can we do to protect us?

A) build the profiles on multiple TF

On loeer TF you see more often fat tails (my experience) but also more
mean reverting

B) look for situations where more than one TF screams for mean reverting

C) start with lower leverage and scale in; sooner or later you will see mean reversion or partial reversion; stay solvent

D) now where your pain sets in and adjust money management accordingly

E) mean reverting strategies are best to be used with other strategies like Elliott wave, pivots, Murrey Math or for examble S/R techniques
  • #27
  • Jun 22, 2013 1:07am Jun 22, 2013 1:07am
Hello , good to have this new discussion.
Let´s take a look at http://www.grupoempresarialer.com, and their SDS sperad software, for statistical arbitrage.
  • #28
  • Jun 22, 2013 1:14am Jun 22, 2013 1:14am
Hi Pipmeup,

Regardind adding to loser/winner:

I think it all comes down to your money management and the strategie you use:

A) trend following - adding to your winner

B) anti- trend - adding to your loser


Personally I am no big fan of adding to winners, because you lever up in a maturing move where chance of mean reverting gets bigger and bigger

You lever up into a situation where risk grows
  • #29
  • Jun 22, 2013 3:17am Jun 22, 2013 3:17am
Great thread, high level
unfortunately not covered by my simple brain

I respectfully
S.C.
if there is something you should keep secret,simply just save it,, so, just
  • #30
  • Jun 22, 2013 5:03am Jun 22, 2013 5:03am
What do you mean by ""That means, in betting on mean reverting we have the situation of limited gains (the mean) by facing unlimited losses eventually.""?
If the price reaches say the 2nd std dev you enter against the move and target the mean. You're doing so under the mean reverting assumption. Understand: you consider the market is currently "Gaussian-enough" to take this risk. The price has (is supposed to have) 95% chance of not being here therefore it is supposed to move back with high probability. Say price keeps running away and reaching the 3rd std dev. Under Normal assumption the price has 99.6% chance of not being here. Yet it is here. Therefore the good reasoning to me is to challenge the mean reversion assumption itself, not adding another position. That is take the loss now. You bet 1 std dev of risk to target 2 std dev of reward. Not really good, I can only agree with you, but the losses aren't unlimited.

BTW since you seem to like the MR idea keep in mind FX is not I(0) or I(1) but something in-between (fractionally integrated). ADF test and his friends totally fail in this situation. => You can not even test the MR assumption with a satisfying confidence level... Fragile indeed!

""On lower TF you see more often fat tails (my experience) but also more mean reverting""
What you witness is the decreasing of the signal-to-noise ratio. The mean reverting effect comes from the "whitening" of the time series: the lower the TF the more the white noise component, which is MR, dominates. The amplitude of the price move reduces down to the amplitude of the noise. You may read The Optimum Time Frame for Trading if it is not already done.

""E) mean reverting strategies are best to be used with other strategies like Elliott wave, pivots, Murrey Math or for examble S/R techniques""
The context free grammar generated by Elliott's rules is not LALR parsable. It is only explanative. You can always find a set of future swings which will invalidate the current counting arbitrarily far backward.

""I think it all comes down to your money management and the strategie you use:""
No. It all comes down to risk management. MM cannot turn a loser into a winning system (http://www.forexfactory.com/showthre...87#post6379487). Your system must be a winner pip-wise. If it fails with a fixed lot size MM, it will fails with any other MM. That's why a backtest must be executed with fixed lot size to be meaningful.

""Personally I am no big fan of adding to winners, because you lever up in a maturing move where chance of mean reverting gets bigger and bigger""
That's exactly what I write in my document. Even for trend following it's not good. Each additional entry has a lower RR and a lower probability. "Set and forget" seems to be the optimal solution to the trade management. But I cannot prove it. I could only disprove some of the others...
No greed. No fear. Just maths.
  • #31
  • Jun 22, 2013 9:01am Jun 22, 2013 9:01am
Hello man, nice research after you disappeared.

I will study your stuffs.

Is Quantitative analysis a different thing to the algorithmic trading ?
  • #32
  • Jun 22, 2013 11:26am Jun 22, 2013 11:26am
Quoting Jack168
Hello man, nice research after you disappeared. I will study your stuffs. Is Quantitative analysis a different thing to the algorithmic trading ?

Hi Jack,


here is some good definition about quant and algorithmic:


Trade carried out using algorithms is known as algorithmic trading. Algorithmic trading can be defined as “placing a buy or sell order of a defined quantity into a quantitative model that automatically generates the timing of orders and the size of orders based on goals specified by the parameters and constraints of the algorithm”. The rules built into the model attempt to determine the optimal time for an order to be placed that will cause the least amount of impact on the price of the financial instrument. Algorithmic trading is a way to codify a trader’s execution strategy.


http://www.tcs.com/SiteCollectionDoc...mictrading.pdf
  • #33
  • Jun 22, 2013 11:27am Jun 22, 2013 11:27am
Quoting PipMeUp
What do you mean by ""That means, in betting on mean reverting we have the situation of limited gains (the mean) by facing unlimited losses eventually.""? If the price reaches say the 2nd std dev you enter against the move and target the mean. You're doing so under the mean reverting assumption. Understand: you consider the market is currently "Gaussian-enough" to take this risk. The price has (is supposed to have) 95% chance of not being here therefore it is supposed to move back with high probability. Say price keeps running away and reaching...

Hi Pipmeup,


will answer the questions later in detail.


Thanks for the nice discussion :-)


Greatly appreciated
  • #34
  • Jun 23, 2013 1:20am Jun 23, 2013 1:20am
What to choose?

normal, distribution, cauchy distribution, student-t-distribution?


http://www.slideshare.net/gaetanlion...fat-tail-issue
  • #35
  • Jun 23, 2013 2:16am Jun 23, 2013 2:16am
Quoting PipMeUp
What do you mean by ""That means, in betting on mean reverting we have the situation of limited gains (the mean) by facing unlimited losses eventually.""? If the price reaches say the 2nd std dev you enter against the move and target the mean. You're doing so under the mean reverting assumption. Understand: you consider the market is currently "Gaussian-enough" to take this risk. The price has (is supposed to have) 95% chance of not being here therefore it is supposed to move back with high probability. Say price keeps running away and reaching...

Hi Pipmeup,

I should have said "literally" unlimited losses :-)

If we get a really fat tail (think of the huge drop in Gold a few weeks ago, that was, when I remember correctly, a sigma 7 or 8 event), it can proof to be difficult to stay solvent and to not get a margin call.

Since we have to live with this RISK (how to control extrinsic risk?), we either can stay away from trading, or adjust our money management accordingly. For me that means not to bet the farm on 2 SD or 3 SD, rather than having firepower left for prices beyond of them.

Trading for me, is taking opportunities, taking risk.

What I do often see, is, that quant trader try to factor all the things into a formula, trying to calculate everything with high precision.

For me, the goal is not to find the perfect formula, aka the holy mathematical grail, but to make money in the market. Thus I can live with "models", knowing their shortcomings and benefits, knowing how they behave under market conditions.

Since my trading inherit Elliott Wave to a large extent in my decision making for a trade, I have given up on backtesting. And to be honest, backtesting of system leads often enough to optimizing the system for the past, and thus survivorship bias comes into play.


But I am not to say that backtesting is entirely worthless :-)


Yes, ADF-tests fail on most time-frames. When remembering correctly, only for the bigger TF's like weekly you get supporting results. Strangely enough, this goes against that, what we can observe day in day out.


Regarding MR white noise: from a traders perspective it may be not so much of importance if we trade MR white noise or MR true signals. All we need is mean reverting :-)

Hadn't read the article about the right time frame to trade. Very interesting. As I have often stated in my other threads, my feeling is, that the 4h TF seems to be best for Murrey Math judgement as well as MR. I love that really, that this hypothesis of mine found mathematical proof.
  • #36
  • Jun 23, 2013 2:27am Jun 23, 2013 2:27am
http://www.tcs.com/SiteCollectionDoc...mictrading.pdf


Algorithmic Trading: Pros and Cons
  • #37
  • Jun 23, 2013 8:52am Jun 23, 2013 8:52am
hi Markus! so Shangri-La is no more. I did try to post in The club but not able to do. is it any restriction. I totally agreed with your gold view. going to close my short (@1463). and looking for long in next 1-2 week for retesting of 14XXS.
regards.
don't predict direction, but dance with flow of water
  • #38
  • Jun 23, 2013 9:12am Jun 23, 2013 9:12am
Quoting infinitus
Since we have to live with this RISK (how to control extrinsic risk?), we either can stay away from trading, or adjust our money management accordingly. For me that means not to bet the farm on 2 SD or 3 SD, rather than having firepower left for prices beyond of them.
I have another point of view. But that's the whole idea of an open discussion

I read somewhere that the market are ranging 80% of the time. I still don't understand if that means 8 minutes every 10 minutes or 8 months out of 10 months (makes a difference!). But for the sake of the explanation let's imagine that it implies the market is MR 80% of the time. Well, let's even go as far as pretending that the market is MR 90% of the time. I feel generous today. Let's use Bayes' theorem and let's define
N the event "we are under a mean reverting market" and
S the event "We are at a price which corresponds to the 3 std dev".

We know
p(N) = 0.9 (we generously pretent it)
p(S|N) = 0.004, per definition of the 3rd SD under Normal assumption. (You can use another distribution that's won't change things too much.)

Using Bayes' theorem let's see what the probability p(N|S), the probability of being under Normal market given the fact that we are at the 3 sigma level.

p(N|S) = p(S|N) . p(N) / p(S)

We need the marginal probability of S. We can write it p(S) = p(S|N) + p(S|!N). Where !N denotes "we are NOT under a MR market"
p(S|!N) can be set to 50% for the FX because this market is symmetric and price is now free to go where ever it wants now we removed the MR assumption.


So we have p(N|S) = p(S|N) . p(N) / p(S) = 0.004 * 0.9 / (0.004 + 0.5) ~ 0.007143

My point is then that there is 99.28% chance that the assumption of mean reverting market simply doesn't hold any longer if we reach 3 sigma.
No greed. No fear. Just maths.
  • #39
  • Jun 23, 2013 2:10pm Jun 23, 2013 2:10pm
Quoting priceaction1
hi Markus! so Shangri-La is no more. I did try to post in The club but not able to do. is it any restriction. I totally agreed with your gold view. going to close my short (@1463). and looking for long in next 1-2 week for retesting of 14XXS. regards.


Hi Priceaction,

I have removed the restrictions in "The Club" now. Feel free to post there :-)
  • #40
  • Jun 24, 2013 6:34am Jun 24, 2013 6:34am
Quoting infinitus
{quote} Hi Priceaction, I have removed the restrictions in "The Club" now. Feel free to post there :-)
thanks,
don't predict direction, but dance with flow of water
0 traders viewing now