Who is Robert? Beyond what I've told you, you'll need to contact McLean directly. He's working on the 2.0 release so hey may not respond right away as he's trying to keep is word on the release....In any case hope it works out for you....
Hey jcy, no offense man but so many of your assumptions about SAP were wrong, I dont know where to begin. And you took a "couple" last week, and they went into DD, so you say the system has no real edge? Again, I dont even know where to begin there was so much in your post that was plain ill-informed. For example, how do YOU know that SAP doesnt use co-integration? SAP states explicitly that it uses co-integration as ONE of the qualifying characteristics for 2 pairs to be a valid potential arb. And we ALREADY KNOW that the bands are fairly run-of-the-mill volatility bands based on the spread values - the point of what I personally posted was simply that the bands themselves were too jumpy and could use a smoothing setting, pretty simple concept - also, the lot sizing is NOT stationary nor fixed, it very clearly is weighted based on the calculations of the indicator ....Really, read the manual before posting all that nonsense, its all there.
well first off to answer JCY, yes, I HAVE studied cointegration extensively.My personal feeling is that the tests Brett mentions arent in fact the best, though they are extensively used.Dickey-Fuller is the most comprehensive test IMO.But that is not really relevant to this thread.All we know about SAP is that it uses a "proprietary" blend of tests that *includes* cointegration testing. We dont know any more than that, so how can you pick it apart? You really cant. You want to use your own method of selecting the arbs, go right ahead.Secondly, all I said was that in your post you made a lot of invalid and inaccurate assumptions and statements.Mclean himself has said that the volatility bands are nothing extra special, the "special sauce" is the arb selections. You were talking about timeframes and their differences, and while thats true its also 100% irrelevant, SAP uses daily TF only. Period.My own observation was that the volatility bands that are used would be better as triggers if they were SMOOTHER ( I realize why they are jumpy, but again that isnt relevant), and yes, as anyone who knows how smoothing of indicators works knows, that will of course introduce lag. My postulation was that I would rather have that entry lag than have the unsmoothed indicator bounce back out of the bands (which means I had an early entry that is now in DD sucking up margin unnecessarily). Any good double-smoothing snippet could be added to the indicator itself, and make the smoothing value user settable, so that XOden or whoever else feels that it is "fine" now can just set it to zero and get what we have now. This is a very common concept in indicators (smoothing), it isnt anything radical, and doesnt even address the other parts of the system. I remain a happy customer and so far the arb selection seems pretty good, but as its a daily system it will of course be a few months before we know what is what. I agree with XOden that risk management and proper sizing are important here, and I meant no offense to you at all JCY.
Hi guys,First of all, I would kindly ask you guys to google up statistical arbitrage to get the slightest idea on how it's suppose to work. You should at least get to know why your trades fail when they do, and not do something over and over until you get a substantial amount of your account blown.I was actually hoping for someone with a fair bit of knowledge in this area of statistics to comment on this system, but I guess they are all busy making their own algorithms or working in hedge funds/ on wall street. I do not claim to be an expert, as I'm merely a college student working on my own stat arb system, got stuck, and ran across this system hoping it would shed some light. There's alot of information I want to share, but I'll make it really concise.First, despite what the website says, this system does not implement the use of cointegration, which is the meat of stat arb trading. The indicator on your chart is merely the spread ratio of the two pairs you entered, plotted as a time series along with some standard devation bands (I actually did not look into the code, it is not something of my interest). So when you trade the indicator, you are essentially trading another non-stationary time series. Any trades you take off those volatility bands are as effective as taking trades off the bollinger band on a daily/weekly chart. Granted I would not be surprised if you make some killing trades, but I think we can all agree that the band itself does not give you any real edge. (a lot of information here that I didn't go into detail, for example, you should notice the position size is fixed, a cointegrated pair on a certain time frame does not guarentee cointegration on a different time frame, the weight of the basket should definitely not be the same from the 1M to the weekly chart. Sorry I cannot explain this better, you're welcome to do some research, it's no magic.)The point of this post is not trying to bash this system, I just cannot stand to see so many people paying money for a system only to lose money, and not know what's wrong. There are lots of posts here discussing watching price action, drawing trendlines, and the EUR problems etc etc. which IMO is not the heart of the problem, hence my post here. I took a couple of arbs last week with a crossing of the extreme volatility band, and still suffering a DD of -4%. In the meantime, I've close 2 arbs from my own system with +2.02% with another +2% floating, all with similar position sizes. Just some food for thought, I honestly believe it's still far from being a CONSISTENT system, but I think (and hope) I'm on the right track. In conclusion, I believe this system, like many other systems out there, can be profitable with good MM, but it does not provide any real edge. Best of luck to you guys.
Hi guys,First of all, I would kindly ask you guys to google up statistical arbitrage to get the slightest idea on how it's suppose to work. You should at least get to know why your trades fail when they do, and not do something over and over until you get a substantial amount of your account blown.
I was actually hoping for someone with a fair bit of knowledge in this area of statistics to comment on this system, but I guess they are all busy making their own algorithms or working in hedge funds/ on wall street. I do not claim to be an expert, as I'm merely a college student working on my own stat arb system, got stuck, and ran across this system hoping it would shed some light. There's alot of information I want to share, but I'll make it really concise.First, despite what the website says, this system does not implement the use of cointegration,
which is the meat of stat arb trading.
The indicator on your chart is merely the spread ratio of the two pairs you entered, plotted as a time series along with some standard devation bands (I actually did not look into the code, it is not something of my interest).
So when you trade the indicator, you are essentially trading another non-stationary time series. Any trades you take off those volatility bands are as effective as taking trades off the bollinger band on a daily/weekly chart.
Granted I would not be surprised if you make some killing trades,
but I think we can all agree that the band itself does not give you any real edge.
In conclusion, I believe this system, like many other systems out there, can be profitable with good MM, but it does not provide any real edge.
Quote from: odysseus11 on July 10, 2012, 01:28:13 pmHey jcy, no offense man but so many of your assumptions about SAP were wrong, I dont know where to begin. And you took a "couple" last week, and they went into DD, so you say the system has no real edge? Again, I dont even know where to begin there was so much in your post that was plain ill-informed. For example, how do YOU know that SAP doesnt use co-integration? SAP states explicitly that it uses co-integration as ONE of the qualifying characteristics for 2 pairs to be a valid potential arb. And we ALREADY KNOW that the bands are fairly run-of-the-mill volatility bands based on the spread values - the point of what I personally posted was simply that the bands themselves were too jumpy and could use a smoothing setting, pretty simple concept - also, the lot sizing is NOT stationary nor fixed, it very clearly is weighted based on the calculations of the indicator ....Really, read the manual before posting all that nonsense, its all there.No offence taken, but have you done any cointegration tests/balancing in R or Matlab?Do you know how the indicator is coded?You complain about the indicator being too "jumpy", the indicator is taking a FIXED value for each closing bar, so of course its JUMPY, you can smooth it with statistical methods, but is that what you really want given more problems it'll create? such as lag or inaccurate value?And do you know why your trades are going back into DD? look at your volatility band, it fluctuates more rapidly than the middle band itself, and it's been doing so for a good 3 WEEKS. Is that not suppose to cause a loss of stationarity given previously "balanced" weights?
Again, no offence taken man.So tell me then, if you were to model a basket on different time frames, given the distribution is stationary, let it be 10 years, 1 year, 1 month, or 1 day. That the basket would have no difference in weighting?
Quote from: jcy on July 10, 2012, 04:48:18 pmAgain, no offence taken man.So tell me then, if you were to model a basket on different time frames, given the distribution is stationary, let it be 10 years, 1 year, 1 month, or 1 day. That the basket would have no difference in weighting? One last thing. This tells me that you've been on the MT4 to R project website. Good for you. It's a great starting point.The fault in this assumption is that we're trading basket of pairs. 2 pairs is all the arb is. Positions are normalized for the currency pairs and deposit currency and calculated for a 10% fluctuation in price. No additional weights are necessary.
Quote from: jcy on July 10, 2012, 01:50:11 pmQuote from: odysseus11 on July 10, 2012, 01:28:13 pmHey jcy, no offense man but so many of your assumptions about SAP were wrong, I dont know where to begin. And you took a "couple" last week, and they went into DD, so you say the system has no real edge? Again, I dont even know where to begin there was so much in your post that was plain ill-informed. For example, how do YOU know that SAP doesnt use co-integration? SAP states explicitly that it uses co-integration as ONE of the qualifying characteristics for 2 pairs to be a valid potential arb. And we ALREADY KNOW that the bands are fairly run-of-the-mill volatility bands based on the spread values - the point of what I personally posted was simply that the bands themselves were too jumpy and could use a smoothing setting, pretty simple concept - also, the lot sizing is NOT stationary nor fixed, it very clearly is weighted based on the calculations of the indicator ....Really, read the manual before posting all that nonsense, its all there.No offence taken, but have you done any cointegration tests/balancing in R or Matlab?Do you know how the indicator is coded?You complain about the indicator being too "jumpy", the indicator is taking a FIXED value for each closing bar, so of course its JUMPY, you can smooth it with statistical methods, but is that what you really want given more problems it'll create? such as lag or inaccurate value?And do you know why your trades are going back into DD? look at your volatility band, it fluctuates more rapidly than the middle band itself, and it's been doing so for a good 3 WEEKS. Is that not suppose to cause a loss of stationarity given previously "balanced" weights?FYI the ability to add lag is introduced in 2.0. And with that I must get back to the 2.0 rollout which is 48 hours behind schedule at this point.
Subscribe to our weekly newsletter... (we do not send out those spammy email ads, I promise!)
We respect your email privacy
Author
Topic: StatArbPro EA (Read 99667 times)
Recent Posts
