October 18, 2011
Since last April, I have done numerous trading simulations looking for which of several trading methods I would like to adopt. Its a search not only for the best performer but also for the one I'll be most comfortable with; and which would become my trading platform for the coming years.
That is a strange statement as if when developing trading strategies there was something else than profits that mattered. For instance, even with its numbers, I would not choose the modified Turtles V3.2 as I found it mostly a horrible script, first for its metrics and then for its stressful environment generated. There are much better scripts around.
I am a research team of one. Everything takes time. If I was looking to improve on usual trading methods, my search would not take so long; and I certainly would not even consider writing about it.
Improving the Buy & Hold strategy by 50% is not enough. This would generate only 5 alpha points over average long-term expected returns. I wanted more, a lot more. Any trading methodology generating less than 50 alpha points would be considered as having missed its primary objective.
Aiming for 50 alpha points or more is unheard of. This means that a portfolio needs a compounded annual growth rate (CAGR) of 60% or more. It's like “pushing” the limits a “little” bit too far. But my research was already saying it was not only possible, it would be relatively easy.
Because you are operating in a compounded universe, anything above the 50 alpha points has sufficient value to warrant the search for additional points; especially if you already know that they are obtainable. From my first paper (2007): Alpha Power, all the equations were provided to build and design trading strategies that would live by equation16 which is just the original form of the ITRADE matrix.
Therefore, since last April, progressively, test after test, various scripts found on the old Wealth-Lab 4 site were used as a foundation on which to build, structure, improve to such a point as to provide the performance I was looking for. Each script had a basic trading philosophy which was extensively modified to better define trend definitions and trading procedures to which was added my own methods in line with my objectives and equation set.
I had to be creative in designing my trading procedures knowing quite well that traditional methods had a hard time extracting profits from almost random price movements. So I've opted to go partly random as well. I would design some of the trading functions to trigger at random and maybe land on a good spot on the price line. I was to accumulate shares for the long term anyway and I could not predict short-term price movements, so why not. This way, the triggered orders would not miss their long-term objectives. The belief was simply that it was better to be in the trade than not being there due to other considerations.
My latest simulation: base on the “One Minute Bollinger Band System”, (it has been modified so much that I should change the name) is provided below. It has a CAGR of over 200%; and yet, it is still not quite what I am looking for. I feel unease, not that the script is wrong or anything like that. It's just that it holds on to losing trades too long; it's like wasting time and resources, lacking some little finesse (see chart below). I would find it more stressful than the Momentum Trader script for instance. Yet, it has extraordinary properties including interesting side effects. It should be noted that most probably, the loss distribution might be due to the prolong recent decline in prices (I do not see all the data).
This is almost the same script (still many modifications) as the one used in the articles: On Back-Testing II, and Trade Acceleration; where only 5 of the charts were presented. I needed to compare with earlier simulations as those provided in the simulations menu. Therefore, I tested once on data set 1 with my Model 1.9 Level 1.6 of the script.
The table below summarizes the data. The strategy generated 127,706 trades over the 5.83 years test thereby screaming loud and clear for trade automation. Some 103,675 trades are listed as profitable; an amazing 81%. We should see this just as a side effect. The trading methodology tends to hold on to its positions for a long time. The high percentage includes trades showing a profit that are still on hand and the profitable closed positions. At a new market high, the percentage of profitable trades might even be higher due to having still opened positions showing a profit.
One Minute Bollinger Band System (modified)
(click to enlarge)
The cumulative sum of all the stop losses and the shares still under-water amount to 3.31% of total profits generated. Some 24,034 trades suffered losses, either by stop loss or are still on hand and in the red.
Trading Losses (sorted by Losses)
(click to enlarge)
- BHH is at such a high level, that the trading procedures have acquired almost all, if not, all the shares. Nonetheless, the Override in the ITRADE matrix should have been triggered months ago after the fund announced it would be closed.
- TBL has been acquired. There too, the Override should have kicked in.
- DIT should have been removed, there is not sufficient trading volume. Here too the Override function should have forced an exit.
- Nonetheless, all 3 would have resulted in positive exits.
I will need to make some changes to dataset 1 and doing so will start to affect my comparison basis.
The table above is over the same data set as the first Gyro Trend Checker test last April. Both charts span 1,500 trading days and over-lap trading intervals (220 trading days apart). Most charts have suffered from recent price declines and provide further insight into the relative strength of the trading methodology used. In some cases, even after severe declines, the stocks remained more than highly profitable. The ability to play short-term market swings provides the means to acquire and hold greater amounts of shares than by using the Buy & Hold alone; or by only trading over market swings. It might be a case where the sum of the parts is greater than the whole.
(Click to enlarge)
|Gyro's Trend Checker II r= 47%||Gyro's Trend Checker II Improved r= 54%|
|QQQ and QID Trader (2007) V2 r= 91%||Livermore Challenge Response r= 86%|
|After Livermore Challenge r= 109%||Turtles V 3.2 r= 127%|
|ADD3 V 1.7 r= 137%||Trend Study II r= 120%|
|Myst's XDev Modified r= 133%||Momentum Trader r= 195%|
( click to enlarge))
The above table shows the progression in performance from script to script as I moved from a rigid trend definition to a non-existing one as in the One Minute BB System. As I advanced in performance, I was gradually relaxing the notion that a trend was necessary to improve performance to the point of not needing it at all. That is a strange outcome: a trend-following methodology not needing a trend definition to make its trading decisions.
I am still looking for which trading strategy I will implement. As I increased performance levels, I could not help but put a value on compounding for longer periods at higher rates. It meant, to me, that the time used to find the strategy that I would like best was well invested.
(click to enlarge)
Created on ... October 18, 2011, © Guy R. Fleury. All rights reserved.