We have been very busy this last month with programming in order to complete a project based on an insight. As you might be aware, our Cannon150 system had a couple of losing months during Aug06 and Sept06 which totaled close to a 27% drawdown. This was very significant for us and it came completely unexpected. The FX markets had formed a deadly range bound pattern for too long a time, and our 150 strategy system had found itself caught and lacerated within too many apparent breakout and trending plays that ended being stopped out on whipsaws and reversals.

The next table shows actual results from live trading of C150 for a period of 10 months.  Hypotheticals matched live trading very closely with C150.  It was not until Aug & Sept that a weakness was exposed in C150. The blue highlights show the new performance of C25 after changes and consolidation of C150 strategies was complete.  The core C150 was a very strong system but no one wanted to see a 27% draw down even if we were still up 47% for the year.

C150 Live Results Compared to the New C25 Hypothtical Results 

 It might be worth noting that though we had suffered a 27% draw down during two months, we are still up over 47% for the year, which is better than can be said for most fx managers. Nevertheless, the drawdown hurt us and our clients, and it called for a re-examination of our strategies and how they work together.  We had learned from our examination that there was a serious weakness in the macro design of how the 150 strategies were supposed to operate in concert. While the 150 strategies did operate very well together under most market conditions, there was a potential for them during an extended period of consolidation (such as the last few months) to generate too many trades in both directions of the choppy market. While some of the trades could withstand the market choppiness with large enough stops, others could not and over time those losing trades piled up as a heap of dead bodies in a war of attrition that the market seemed to be winning.

 The insight received from seeing this aforementioned weakness was that we needed to consolidate the 150 strategies into fewer strategies that could work smarter together. We had 15 core strategies worked out on 7 time frames (120min, 240min, 300min, 360min, 480min, 780min, and 960min), which in turned created many stem strategies for one market (150 stem strategies in total for 2 markets). What we needed to do was regroup the 15 core strategies into 3 core strategies corresponding to the three primary market behaviors: trending strategies, breakout strategies and counter-trend strategies. Thus, we decided to take each of the 5 core strategies from each of three classes and fuse them together for each time frame. For example, we took the 5 trend based strategies (each of a different algorithm from the other but loosely based on identifying a trend behavior) of the 240 minute time frame and weaved them together into one page of code in order to create one composite trading strategy for the 240 minute time frame The composite strategy would in turn generate signals from all 5 sub strategies within its code but preference would only be given to the first signal to arrive on the scene. All other signals generated after the first would be ignored as being redundant. This absence of redundancy is important. There would be no more need for a gang of trades of the same class (and of the same time frame) to appear together on the market when one of them was sufficient to do the job. Having the extra trades appear could potentially increase the leverage and exposure within a particular spot/ behavior in the market that could potentially reverse against the whole gang, leading to a bigger drawdown than otherwise. Having only the one trade appear on that particular spot/behavior in the market could ensure that the situation is being exploited for its potential upside benefit without excessive aggregate trade leverage and risk-exposure.

Besides minimizing the risk exposure and trade redundancy, the building of composite strategies lead us to build more statistically viable strategies.

Of the 150 strategy set, we had many singular strategies that looked powerful in hypothetical statistics when in actuality they were statistically weaker because the sample size of generated trades was not enough. For instance, we had a 240 minute EURUSD strategy that looked powerful over 10 years (profit factor > 5), but because it only generated 30 trades in that 10 year span, its sample size was not sufficient to be able to lend sufficient credence to its own statistics. However, after we hybridized 5 strategies of a 240 minute strategy together and produced 187 trades over 10 years, then the 5 fold increase in trades on the now compounded strategy builds a far greater degree of credence in terms of the sample size for the statistics generated. Caveat: While it might be assumed that we compounded 5 statistically unviable strategies into one and just further confused the matter, that assumption is not the case here. Each of the 5 strategies were set within the same page of code but now they had to compete amongst themselves for first positioning (see the above point about first trade preference), a point which greatly alters the matrix: now only the first and arguably the strongest of the trades of each class (and time frame) would enter the scene and dominate it. Furthermore, each of the 5 strategies compounded together could not be equipped with its own exit logic adapted for its own distinct behavior; rather, all 5 strategies must now work with the same exit logic adapted for all behaviors. This puts allot of pressure now on the exit logic to be much more adaptable than otherwise. It is more of a challenge to see how the exit logic works on 187 trades versus 30 trades. Not surprisingly the exit logic evolved under pressure (combined with a bit of programming insight) to become much more responsive to all different challenges and pressures within the market. 

Now there are 25 composite strategies as opposed to 150 singular strategies and each is more powerful in terms of unique trade positioning, sample trade size and exit logic than the first generation. The hypothetical figures over 10 years are stellar to behold, and it will be exciting to see how they play out over the next 10 years trading forwards.

Below is a the Fixed Capital Return Analysis of the 25 composite strategies set on a 100K account WITHOUT compounding for the past 11 years (from Jan 1994 till present), and the 14% average monthly return with less than 13% intramonth drawdown is interesting.

More stats can be seen from the combined Rina Excel Report (1.35MB)

(note: $50 commission has been deducted from each trade).