Thursday, September 24, 2009
Live Test - Performance Summary #3
Each week, the system's predictive model selects up to 10 stocks for the ASM-10 portfolio. Many stocks will be held over from the previous week, while a few lower ranking stocks will be replaced with higher ranking stocks. The selection criteria are based on growth characteristics that are common to the "big winners" in the model database. The ASM-10 portfolio is the system's "alpha engine." These stocks are expected to generate above market returns, giving us a winning edge.
The table shows the model's win / loss ratio along with other performance data for three time periods. Data from 2003-06 were used to develop the predictive model. I then used out-of-sample data from 2007 to validate that the model's strong performance wasn't due to over-fitting the data. Over fitting can result in a system that performs well in simulation mode, but fails miserably in the real world. Upon validation it was time for the ultimate stress test.
In January 2008 I began a live system test on my public blog, posting each trade in advance. What I didn't know then was that market conditions were due for a seismic shift, as the economy entered into what some now call The Great Recession.
Over the past 87 weeks of the live test, the system executed over 130 round-trip trades in the ASM-10 portfolio. This sample is large enough to have statistical merit. And when combined with the additional 52 weeks of 2007 out-of-sample data, it is even more significant.
To summarize the performance data shown in the table here, the model's performance has held up very well under the stress of real world conditions. It performed nearly as well in bear market conditions as it did during the preceeding bull market. While its win / loss ratio is close to 50%, nothing spectacular, what makes the model so profitable is that the average percentage gain on each winning trade is more than twice as great as the average loss on each losing trade. This epitomizes the trader's adage, "cut your losses short and let your winners run."
The system's "Tradeable Edge" (TE) Factor is the most meaningful performance measurement shown in the table. This factor multiplies the win / loss ratio and the %gain / %loss ratio to quantify the model's overall statistical edge. Anything greater than 1.0 implies profit potential. The system's TE Factor has maintained above 2.0 for all test periods, implying greater than 2:1 odds of gaining profits on any given trade. This is a powerful edge!
More system performance measures, such as Maximum Drawdown (interim loss), Beta and Sharpe Ratio will be discussed in a future post.