The (Daily?) Forecaster

Using Richard Russell's PTI with BioComp Profit

The market appears to be entering into a phase in which fundamental data -- related to the economic climate and corporate earnings -- is becoming more of a prominent factor in stock prices. This realignment is occuring after a period in which prices and price fluctuations drifted far away from fundamental realities.

One of the most interesting fundamental indicators that I follow is called the PTI, a proprietary indicator developed by stock market analyst Richard Russell, author of the reknowned Dow Theory Letters. Russell definites the PTI as follows:

My PTI is a composite of eight "action of the market" indices. When the majority of these eight indicators rise persistently, it's almost certain that the market will head the same way. Sometime the PTI leads, sometimes it's coincident and sometimes it lags. But I've learned not to go against the PTI. If the PTI is in a rising trend above its moving average, I grade it bullish. I may not buy stocks if the PTI is above its MA, but I sure won't short stocks.

Conversely, when the PTI is below its MA, I may not shorts stocks, that's optional -- but I most definitely do not want to hold stocks. When the PTI is below its MA, and the MA is also declining, that's the sign of a liquidating market, and I want to part of it. I've learned that lesson over and over again. In the old days I use to tell subscribers that "My PTI is a lot smarter than I am." And that's as true today as it was in 1970 or 1980 or 1990.

The moving average to which Russell refers is an 89-day simple, moving average.

Fortunately, Richard Russell now offers his website subscribers (in my opinion, well worth the annual fee of $250) access to the daily PTI values going back about four years. This is about as much data as is needed for the development of a neural "mesh" system using BioComp Profit.

I found that, for the PTI and a variety of other indicators as well, the most promising neural systems were those in which training ran on time series data from early 1999 to January 26, 2003. This finding seems to have been replicated by others and definitely deserves further discussion.

The chart below shows the out-of-sample test results for a system based primarily on the PTI, trading the S&P futures contract from January 26 through today:

Out-of-sample testing, PTI-based, BioComp Profit neural "mesh" system

The red and green dots on the price chart show the buy and sell signals. All trades were to be made at the close of the regular trading session. The statistics for this out-of-sample test are shown below:

Statistics for the PTI-Based neural "mesh" system

It is worth noting that the BioComp neural "mesh" system actually generates more trades than the basic moving average crossover system used by Richard Russell. The chart below, from Richard Russell's website, shows the crossovers during the six months or so.

The question of interest is, of course, whether the trading signals provided by this neural "mesh" system will continue to function, in the future, as a level anywhere near the highly profitable level of "27% of perfect." The 80 day out-of-sample test period is much shorter than is customary. And, even signals that appear viable in out-of-sample tests of much greater length (such as those I reported in The Forecasting Systems Letter on October 19) have a penchant for breaking down just when one has developed sufficient confidence in them to begin actual trading!!

Interestingly, the PTI system shown above actually performed better in out-of-sample testing than it did during the in-sample training period. This is rather rare -- and suggests some interesting lines of thinking regarding the whole notion of timing in the training of neural systems. After all, these systems are supposed to do better during the training phase, because that is when they are able to actually learn from future data -- a feat that is not possible in out-of-sample testing, and certainly not possible (precognition excepted) in real-time trading.

During the training period, this same system performed at 18.29% of perfect overall. That included a drawdown of over $58,000 that lasted for 137 bars. So, I suppose, it would not be unreasonable to expect that this system might revert back to a level of performance similar to that attained during the training phase.

The 137 bar drawdown, interestingly, occurred during the second half of 2000, just as the market experienced a major phase shift from bull to bear. I, personally, do not expect that we are about to experience another such event in the near future. Even if we see a rally over the next several months, I do not think it will ultimately mean that the bear market is over. So, for the time being, I remain confident that this PTI system has "legs."

Another reason why I generally feel confident is that a similar, and even more successful, pattern of successful out-of-sample testing is exhibited in additional systems that I shall present soon.