November 04, 2011
In my preceding test I mentioned that my methods bulldozed over the charts. So I decided to add a little finesse. Most of the trades are still random entries. My intent was more to provide a smoother transition from trade to trade and at the same time extract a bit more profits.
Naturally, it implied executing more trades; if you want more profits and the price series does not change there are not that many choices offered: either you increase your profit spread, the number of trades or both. For an automated system, that would not be a problem as long as there was cash in the account. And there, there was a lot of cash sitting idle doing nothing in the first place. Using random entries is a rough way to trade.
So here is the result using the improved procedures on AAPL:
The performance compares favorably to the script's previous version as seen in the following chart.
Only AAPL has been tested (the need to debug the script) using these improved procedures, some of which might not even get executed due to the randomness of the other competing procedures. But that is not important, it is not because you have a procedure that it has to be executed. I suspect that some form of procedure hierarchy will come out of this, but there too, I do not care since on the next test the hierarchy would be different due to randomness again.
The modified procedures would similarly impact both data sets tested making the modifications to the trading procedures even more valuable. I do not expect to have just a stock here and there showing profit improvements. I expect to see most of the stocks in either data set to show more profits; not only that, I expect to see the bottom line: the total portfolio value to about double its profits.
After making such a statement, I was almost obligated to do a test at least on one data set. So, I used data set 2, this way, 43 out of 43 stocks will not have seen the modified procedures making it an out of sample test for those procedures. And this way any portfolio profit improvement over the last test will be attributed to the modified procedures.
Random Test: Improved Script
The table below summarize the outcome of my latest test. The added strategies and the refinement brought to existing ones did in fact improve overall performance. It should have been expected that not all stocks would improve their performances with over 95% of trades still depending on a random function; however, none produced negative results. And the bottom line, where it counts, more than doubled.
The system generated some 912,171 trades over its 5.83 years life span. It maintained a 62% hit rate which is remarkable having used mostly random entries. The previous test had a 60% hit rate on 498,743 trades. The improved procedures added 366,200 trades at the expense of a slight decrease in the average profit per trade.
|Random Test: Improved Script|
Here are all the charts generated using the Wealth-Lab 4 simulator:
Created on ... November 04, 2011 © Guy R. Fleury. All rights reserved.