June 16, 2015
There is sufficient data to start connecting the dots. What follows are explanations given to tests performed over the last few weeks to answer some questions on a LinkedIn forum. The last two tests have not been presented yet, but they will shortly. The point was to show that the trading method used mattered more than the stock selection that could be made.
The trading strategy reviewed in the past few days was designed some 4 years ago. Improvements have been added to the initial version simply because they were kind of evident and easy to do. But still, from its infancy to its latest version, the strategy was greatly transformed.
The strategy started as a modification to an existing public trading script (circa 2002) that was available free on the old Wealth-Lab v4 site. It had for name: Myst's Xdev, Myst being the author's handle. Since then, it has been changed so extensively that its name was changed to DEVX to keep track of its roots. But very little remains of the original version.
My first viewable iteration was in July 2011, where its excuse for existence was to answer someone's questions while at the same time answering my own search to find better trading methods.
The dataset used was the same selection as for the first test done in April of that year using this trading methodology on market data. Stocks had been selected from charts viewed by other members. A weird kind of selection; half, the randomness of the draw, and half viewer popularity at that time. Certainly not the best way to select stocks for the long term. But I just wanted stock tickers, and it appeared as good a method as any for my purpose.
All I was looking for were answers concerning a trading method; stock-picking methods were secondary. The 43 stock selection was enough to make the point (it was also the most I could fit on a single screen). Does anyone see how intricate this "scientific" stock selection method was?
This enabled comparing results with previous strategies using the same data set over the same time interval (1,500 trading days, about 6 years). The sought-after answer to the question in payoff matrix notation was: Σ(H(strat_A).*ΔP) >? Σ(H(strat_B).*ΔP)? Is strategy A better than strategy B?
To show that the selection might not be the main problem for a long-term trading strategy, I opted to select another 43 stocks using the same selection method and keep one of the stocks the same for verification purposes. That test is available here.
What's interesting about those 2 tests was the low stop loss levels. Each test did more than 100,000 trades, more than statistically significant to make any of their averages relevant. As I've concluded in one of those tests: "... It might not be an orthodox method, it misbehaves at times, but then I do like the numbers ".
Since then, the program has continued to improve. Designing trading procedures that made more and more demands at the portfolio level. One thing that most certainly should be taken from all those tests is that it might not be the stock selection that is the most important thing, but the trading strategy itself that matters most.
The DEVX program has been tested under several scenarios:
two 43 stock datasets over a 6-year period (1,500 trading days);
one 30-stock portfolio over 25 years (6,250 trading days) at 3 levels;
13 stocks over 10 years (2,500 trading days);
10 stocks over 20 years (5,000 trading days);
Overall, tests were done on 6, 10, 20, and 25-year trading intervals over 10, 13, 30, and 3 groups of 43 stocks. All stocks tested positive, all 123 stocks, and this from over 6 to 25 years in duration. A sufficient sample to say: there is something there.
Further explanations can be found here and here.
Hope it can help.
Created... June 16, 2015, © Guy R. Fleury. All rights reserved.