January 30, 2013
Recently, I was asked about showing some forward tests on systems based on my alpha research. Having opted to go live instead of waiting for further testing, and that most likely the request would have required showing some of my programs, I declined. Nonetheless, forward testing could be considered as having been done in parallel to being live. Here was my reply:
Technically, forward testing has been done in parallel as a side effect of all the tests performed since April 2011, most chronicled live in the Wealth-Lab Alpha Power forum. I needed some basis for strategy comparison purposes, and tests were performed using the same 3 data sets, same price series, and same time durations. Deviation from expected average portfolio returns would have to be the result of the trading strategies used. And furthermore, I could continue to test, improve my methods while at the same time managing my live accounts.
As I progressed in my market data testing phase, I was looking for only one answer: what were the limits of my mathematical trading models? I had set these objectives in my very first post in the Alpha Power thread on Wealth-Lab in 2008. I think I have succeeded in showing, in the progression of all the tests performed, that the concept of trading over a stock accumulation program has merit. At least in my view.
Prior to April 2011, all the testing was done using randomly generated prices and it is only after that that I started using real market data to demonstrate my trading methods. Each of the simulations I have done since used the Wealth-Lab 4 simulator. And each simulation had its own set of goals to prove; had its own set of Wealth-Lab time stamped charts as corroborative evidence. I had this need to see how far I could go on the concept of trading over an accumulative process. What would be the requirements, what type of technical background would do best or how could I increase efficiency? Each new test would try to go a little bit further, some starting with known and published trading strategies available on Wealth-Lab and then followed by my attempts to improve on their design (meaning pushing performance higher).
So forward testing: yes, since the same datasets were used and recycled on the 1,500 trading day windows spanning over a year. I was always testing the same thing using various methods in order to show that: trading over a stock accumulation program can generate better profits than just trading or Buy & Holding alone.
I preferred to go live in Sept. 2010, having sufficient data to convince myself that indeed trading over a stock accumulation process worked. I currently manage 3 small accounts which over the past 2 ½ years have gradually increased to an average 40%+ CAGR. I expect the CAGR to continue to increase slowly in time. We all know portfolios go up and down, suffer drawdowns and that it is not that easy to out-perform. And therefore only time will tell if my expectations will continue to be fulfilled, I am of the patient type.
The lessons learned during this adventure is that even if you have a simple concept, like trading over an accumulative process, does not mean that it is necessarily easy to produce a program that will do the job over the long run. It does take time to design, debug, test and structure a trading program to deliver worthwhile results. The trading procedures mattered but not necessarily the entry or exit methods as if being in the game was the first step to consider.
I have simulations with strong trend definitions as if set in stone, some with fuzzy semi-trend definitions where you did not know if there was a trend or not, and even some simulations with no trend definition at all. But all could produce exceptional results (at least in my book). I did simulations using technical indicators of all sorts to using none at all and still provided impressive performance levels. I even dabbled in random entries which technically showed that about any entry mechanism could have been used since random entries could produce about the same or better results than using any other entry methods. One funny side effects, on random entries, was that I could have up to 95% random entries but if I tried the 100% level, performance would be cut in half.
It is not by doing what everybody else is doing that one will be different.
Hoping that this narrative can help others design better systems. At least, my simulations can serve as an example of what can be done. I now know that my methods can piggyback on other program structures and not only survive but thrive. The trading methods I've presented are not the only ones that can out-perform. IMHO, and based on my research, there is out there a multitude of solutions.
Created... January 30, 2013, © Guy R. Fleury. All rights reserved.