System, Rate of Change, System Analysis, Filter At OASIS 2008, I used a simple Rate of Change [ROC] trading system to serve as a base for evaluating results over a six-year backtest period and improving those results with different filters. A portion of the session is provided here as background to look at the current performance for one system variation.

Base System

The base system identified at OASIS used an ROC crossing above or below its simple moving average [SMA], with two different ROC settings for entry and exit. Table 1 provides details for the initial backtest performed.

Table 1: Base ROC System Details

Most traders will find the system results pretty unappealing if they look solely at net returns or even average gains—actually losses—for the system. However the system stability provided a nice starting point for adding filters to remove losing trades. By using this base, traders can add and test filters that allow them to make the system their own. Trading in a way that suits your style is absolutely one key to trading success.

Figure 1 provides a sample trade for the parameters provided. Since only one ROC can be displayed, this chart uses the entry parameter of 34-days with a 13-day SMA.

Figure 1: Daily Chart for PDCO with 34-13 ROC Entry
click here for larger view

As mentioned, the trade results look pretty unpalatable at first. Note that MFE = Maximum Favorable Excursion, a John Sweeny term which describes the greatest unrealized profits for the position.

Table 2: Base System Performance

Judging system stability using mean and standard deviation (std dev) alone can be challenging—consider comparing the mean and median results as a first step. When the mean and median are relatively close, you may have uncovered a stable system. A mean value that is much higher (lower) than the median suggests one or a few trades may be contributing significantly to system profits (losses).

Although this system is relatively stable, there are a ton of false starts as seen by the Zero MFE information. A whopping 40% of the trades resulted in losses from the first day the position was established. At no point were there unrealized profits in these trades. That’s an extremely high value, and in this case, represented a possible opportunity to improve results with filters.

In addition to comparing mean and median results when evaluating system performance and stability, consider using a scatter plot for all of the trades generated. Figure 2 provides the scatter plot for the base system with trade profits/losses on the y-axis. Note the nice concentration of trades with a small portion of results displaying greater profits. Reasonable filters attempt to reduce the total number of trades and shift the remaining concentration upward above the zero profits line.

Figure 2: Scatter Plot of All Base System Trades

The main goals for a filter included:

1. Sufficiently reduce false starts (% of 0 MFE trades)
2. Identify logical stops
3. Improve profitability while maintaining system stability

A few filters were examined during the session, focusing on those that identified bullish trends (MAs and directional movement) and those that identified short-term bullish momentum (price over an exponential moving average, EMA). A total of 14 more system backtests were run using different combinations of these technical indicators and/or trade allocation, as well as two additional stock lists. Although not completed as part of the OASIS session, trades can also be filtered by delaying entry or requiring consecutive entry signals.

A second series of tests were run that varied entry speed and timed exits (i.e. exit after 10 days). The filter parameters selected for the second batch were based upon a detailed review of the first group of results. The filters were not optimized; a small sampling of speeds was tested for the different indicators.

Table 3 provides collective data for the 15 systems that were run, including the initial system. Table 4 provides collective data for the additional 13 systems that were run and exclude the initial system.

Table 3: Base System + 15 Filtered Backtests

Table 4: Second Round of 13 Filtered Backtests

The results remained relatively stable with a majority of the systems being profitable. During the analysis of results, it was noted that a timed exit of less than ten days yielded extremely poor performance. This is not surprising since the systems averaged approximately 19 days in winning trades and 11 days in losing trades. Some of the most consistent results were obtained by requiring price to be above its 10-day EMA.

A final scatter plot displays the risk adjusted returns for the second batch of 13 systems tested. Three of the four unprofitable systems had a timed exit of 10 days or less, and the fourth used a short-term bearish momentum filter (price crosses below the 10-day EMA).

Figure 3: Scatter Plot for Risk Adjusted Performance of 13 Filtered Systems

Next week a specific system highlighted at OASIS will be reviewed, along with more current results for an options-based approach. For more information on basic statistics, see the Technical Toolbox article: Basic Statistics for Technical Analysis from July 2005 and the Analytical Toolbox series for different system testing series. Diversifying Trade Systems from June 2007 provides multi-system considerations.

To access other articles written by Clare White, please click here.

Clare White, CMT
Contributing Writer and Options Strategist
Optionetics.com ~ Your Options Education Site
Questions for Clare? Visit the Optionetics.com Discussion Board