Forward-testing Price Patterns Discovered by APS

by Michael Harris

One frequently asked question about APS Automatic Pattern Search concerns the future performance of the price patterns discovered by it: Will these patterns stay profitable?  I cannot give a general answer to this question but what I can do is to give an example of price patterns discovered by APS that have survive the test of time.

The example in this short article demonstrates the robustness of price patterns discovered by APS Automatic Pattern Search. It is an important example because it refers to results published almost two and a half years ago and not to results made up for the purpose of demonstrating a point. Thus, an answer to the fundamental question raised about the robustness of APS results is provided, i.e. whether the price pattern discovered by it maintain their performance. This is also related to forward testing and the capability of trading systems developed using out-of-sample historical data to maintain their level of performance when tested forward in time.

Specifically, in an article published in September 2002 (See Ref. 1) and  in the course of demonstrating the concept of the automatic discovery of price patterns, APS Automatic Pattern Search was used to discover such patterns for the NASDAQ-100 Index Tracking Stock (Q's). The results obtained at that time for price patterns satisfying the parameters listed in Table 1 are shown in Figure 1 (original figure from the article in Ref. 1).

APS discovers price patterns that fulfill user-defined performance criteria and risk/reward parameters (See Ref. 2 below) and does that automatically. Historical price data from 07/11/1990 to 05/07/2002 was used for the Q's. A total of eight (8) long price patterns were found by APS that satisfied the parameters shown in Table 1. Each line in the results shown in the Figure 1 corresponds to a price pattern and its performance parameters as calculated by APS.

Figure 1. Search results for QQQ price patterns (see Ref. 2).
Legend: Trade on designates whether the trade entry executed on the open or close PL is the percent profitability of patterns suitable for long positions. In this case PS = 100 - PL. PS is the percent profitability of patterns suitable for short positions. In this case PL = 100 - PS. Trades is the number of trades. CL is the number of maximum consecutive losers. Type is either Long or short. Target shows the profit target value used in the search. Stop shows the stop-loss value used in the search. C indicates the type of profit target and stop-loss; "%" stands for percentages of entry price. Source: Automatic Pattern Search (APS)

We determine next how the patterns in Figure 1 performed since the time of their discovery by APS and the publication of the results in Ref 2. This is achieved by back-testing each pattern in the time period from 05/07/2002 to 02/28/2005. The back-testing function of APS was used for this purpose and the results for each pattern in Figure 1 are shown (in the same order) in Table 2.

It is evident from Table 2 that seven out of the eight patterns remained profitable during the test period considered while just one, the fifth in the list, ended up just breaking even (recall that the profit-target and the stop-loss were both set at 7%). None of the patterns turned into a net loser during the test period considered. That is a remarkable performance. It means that a trading system model developed in 2002 by incorporating price patterns discovered by APS remained profitable after two and a half years. More importantly, that each pattern used also remained profitable.

Figure 2. Search results for QQQQ price patterns for the period 07/11/1990 to 02/28/2005
Legend: Trade on designates whether the trade entry executed on the open or close PL is the percent profitability of patterns suitable for long positions. In this case PS = 100 - PL. PS is the percent profitability of patterns suitable for short positions. In this case PL = 100 - PS. Trades is the number of trades. CL is the number of maximum consecutive losers. Type is either Long or short. Target shows the profit target value used in the search. Stop shows the stop-loss value used in the search. C indicates the type of profit target and stop-loss; "%" stands for percentages of entry price. Source: Automatic Pattern Search (APS)

Next, let us see what results APS Automatic Pattern Search generates if the search for price patterns is repeated with the same parameters as those shown in Table 1 but over the whole data range, i.e. for the period 07/11/1990 to 02/28/2005. There are now ten (10) patterns found by APS as shown in Figure 2. The majority of these patterns are identical to those in the search results shown in Figure 1 but some new patterns have emerged in this search while a few of the original ones did not make the performance criteria and were dropped. These details are not so important for the purpose of this study. What is important is the fact that although the historical data range was significantly increased, the number of patterns that met or exceeded the performance criteria set in Table 1 also increased. Thus, we have falsified the claim that the passing of time adversely affects the performance of all price patterns forcing them towards break-even returns and as a consequence it is hard to get any significant bias towards profitability. We have showed here with a counterexample that the universality of such claim is false.

The results presented in this article can be confirmed by anyone with access to QQQQ historical data and a copy of the demo version of APS v4.1 (The search and back-test option are fully functional in the demo).

References

1. "Price Pattern Autopilot", by Michael Harris, in Active Trader Magazine, Sep. 2002, Vol. 3, No 9, pp. 70-74

2. APS Automatic Pattern Search software by Tradingpatterns.com, http://www.tradingpatterns.com