Murray, I wasnât meaning to imply that everybody should use 80 markets. That is just the number I use. Your suggestion of 30-40 sounds about right. In my experience the Swissie and the Yen trade very differently. The Swissie, Euro and Dollar index however do trade very similarly. Therefore they offer no diversification for testing or trading purposes. The other currency majors, Cable, Aussie, Canadian etc are all quite different markets. My research on intermarket analysis has led me to conclude that each business cycle is unqiue and consequently the rules keep changing. Intermarket analysis is intellectually quite compelling but as with any fundamental analysis timing your trades is problematic. I would lump seasonality into the same category. Iâd loved to be proven wrong because as with any spread strategy you can throw some real size at these trades â and this is important when youâre managing OPM. Anybody else have experience with trading intermarket systems that they can share? As I said I use WealthLab for my initial prototyping. After that I recode the models into C++. I looked at TradersStudio but experience has taught me not to rely on Excel for serious applications. I hope this thread survives the inevitable heckling because I believe it is an important topic. bolter
In terms of Intermarket analysis, I will share some of it with you in a different thread soon. There are many Intermarket relationships, which are reliable. You can also use correlation analysis to see when they have decoupled. In terms of TradersStudio, you said that you have learned that you should not use Excel for anything serious, why did you say that. TradersStudio does have an Excel OLE link but that is an option so you can do more things with global macro's language. TradersStudio does not require Excel, The reports are displayed in a spreadsheet form because it is a nice format and we allow them to be exported to Excel because some people want to do that. If you though TradersStudio was not for you because it was an Excel based product, you need to rethink things because it is not. It is a stand-alone trading platform with external links to Excel to make it more powerful, not less. TradersStudio has custom reports at the portfolio and system level, built in. We allow global macro's to write to Excel so that people can create multiple report, charts ect TradersStudio does not have built in.
Hello Murray, from a quick glance at your site I gather that your platform is for EOD data only. Are you planning a RT version? Also, it's my understanding that most of the time only one system is not enough. The best way would be to have several systems with a low correlation. Of course this requires much more starting capital... What's your opinion?
I think you are substantially overestimating the correlation between markets. The correlation between many markets remains quite low, and this adds substantial power to tests of significance for trading systems. The primary danger in testing across multiple markets is from not compensating for changing levels of portfolio variance (which may be induced by conditional market covariances (although changing levels of portfolio variance also exist in individual markets)). In several of my tests, the independence between markets has increased the significance of my findings by orders of magnitude. Examples: Correlation Market 1 Market 2 .35 S&P 500 FTSE all-share .19 S&P 500 EMC
Yes I understand your point. There are (relatively) uncorellated markets. However they are (in my opinon) disappearing rapidly. Also there is the issue of stationarity. I like to test for stationarity each weekend. I find that on an intraday basis, markets cycle through periods, exhibiting stationarity then going totally random. What I find is that there is a tidal quality to this cycling that can be captured using fractionally integrated GARCH. Otherwise it is very difficult to detect as it is masked by residual long memory effects. Translating to English, that means that markets change their character periodically, and that tendency to change is difficult to model, but seems to happen on a regular basis. This cyclic quality is probably the reason why mechanical systems tend to start out profitable and then blow up. Most developers just don't have the skills to overcome that problem. Although I would not trade a single system on a portfolio, I would trade several well integrated systems on a list of carefully selected markets. Thanks for your comment Steve
In order to avoid any confusion (or perhaps introduce some more): 1). A stationary process may or may not be random. Of course, when discussing such entities, referring to a non-random stationary process is a pathological case. Usually when one uses the term stationary one is referring to random processes. 2). A process is referred to as Strict-Sense-Stationary if its statistics (mean, variance) are invariant with respect to temporal displacements (shifts about the origin). 3). A process is referred to as Wide-Sense-Stationary if its mean is constant and its autocorrelation is only dependent on the temporal displacement used in its calculation. If I understand it correctly, by "going totally random", you must be referring to a process regime shift that occurs as the process goes from having nearly constant mean and variance, to only constant mean, to none being constant.
We have a real time version which is currently in beta, we expect it will be ready to ship this spring. We have decided that we will upgraded our orginal customers to real time for $500.00. We plan on selling the real time version for 1495.00 to 1995.00 so you can see buying now would save you at least 500.00. In TradersStudio you can use a Trading Plan and combine multiple system both with and without money management. I agree with you. I think the best strategy is multiple systems in which each was developed to trade multiple markets.
Equalizer: I am sorry to be sloppy in the way I form my comment. I agree with your points. The issues of stationarity and dependence are (in my opinion) the central concepts that developers must be aware of. As you may know, there are several kinds of dependence that can be found in a data series. If like most developers, one simply tries to write code that captures what seems to be a non-random event, without testing the background data for stationarity and dependence, it is possible (even likely) that the "non-random event" is simply spurious. I would guess that most developers have no clue about A. the existence of regime shift, and B. how to determine when it is occurring. Nor would they have the programming skills to deal with the obstacles that present as a result. What I am saying is that developers either don't know what they are doing, or they hope that customers are naive (as many system buyers often are) and by the time they get a system in place, it is then a matter of chance (in the literal sense) how long the system performs. The "tip-off" is the drawdown and "dead time" factors for the system. To me this is analogous to the way manufacturers used to incorporate "designed obsolescence" into their products to insure that customers had to replace them periodically. After all, how viable would a system developer's business be if after buying the system, you never had reason to buy another? I appreciate your thoughtful comments. Steve
@ bolter I agree with you that this is a very interesting topic. I myself tested many systems on like 60 markets and i also come up with the conclusion that silver, and bOil and Live cattle are very difficult. For some time now i try to develop some sort of volatility indicator that tells me which market to trade and which to avoid. Unfortunately I am not very successful so far. I have many ideas but its difficult. I think the indicator should tell in what short term volatility (like 5 days) the market moves from point A to point B -- maybe 5 %. Because silver moves with spikes - while bund future trends nicely and calmer. I am still looking for one universal vola indicator that lets me compare each markets better - because one day it could be that bund future becomes like silver and then i want somthing to warn me about the general change in behaviour. In intraday markets i have the same problem. Before 2003 many market participants had high trading fees. Therefore they only traded longer term and thus trends were with less noise. Now everybody has cheap rates and the intraday vola is a lot higher when compared to daily vola. What happened, I think is that intraday vola went up while daily vola went down. This is not very good for my short term system as i get whipsawed more often while at the same time the daily ranges get smaller and trends intraday are less profitable. I am curious how markets will behave when vola rises again. That could be good again or even worse. Imagine the case that intrady vola rises at the same time - then you have huge daily bars but the markets makes new high and lows four to six times a day - that would hurt me a lot. Maybe somebody has some more experience on a suitable vola indicator that tells us which markets to avoid at what times.