Designing Trading Systems For The Stock Market by Roger Altman, Ph.D.
System designers are aware that the traditional methods used to select parameter values are prone to
overfitting. What, then, can be done about it? Stocks & Commodities Roger Altman presents a method to
reduce this problem by using randomized data to produce parameter values for trading systems that have
the best chance of duplicating real-time results.
Mechanical trading systems attempt to exploit the non random portion of price fluctuations. Generally
speaking, such trading methods have two parts: the setup and the trigger. The setup usually refers to an
indicator that can predict price direction in the longer time span, whereas the trigger has predictive value
in the near term. Good system development requires that these two indicators work independently of one
another and complement each other's predictive power. This requirement implies that the same price
change cannot be used for both setup and trigger indicators because then their separate predictive abilities
would be lost.
SETUP AND TRIGGER INDICATORS
The most practical way to ensure independent forecasting capacity is to formulate setup and trigger
indicators from unrelated sources of raw data. Because the system developer should be aware that the
response from a trigger indicator should be as fast as possible to limit drawdown, calculations based on
recent price change should usually be relied on to provide short-term market predictions.
Having taken into account short-term price fluctuations for the trigger indicator, the system developer
must find unrelated data to determine when the stock market is set up to rise or fall in the longer
timeframe. Sources for this information come from four general areas: interest rate changes (especially
Federal Reserve policy changes), sentiment (always be a contrarian at bullish or bearish extremes), commercial, versus public open interest activity (follow the smart money) and broad-based market
momentum in the longer timeframe (that is, select a trigger that relies on daily price changes of a narrow
index such as the Dow Jones Industrial Average [DJIA], and for the setup use weekly values of a
broad-based index such as the Value Line index).
While interest rate policy changes at the Federal Reserve Board are the single most reliable indicator of
longer-term market direction, such changes are not frequent enough to serve as a practical setup
indicator. Similarly, sentiment extremes — which produce the most reliable forecasts of market direction
— do not occur often enough to satisfy the requirements of an active trading system. On the other hand,
combining signals from weekly stock market momentum and commercial versus public open interest
levels through the Commitment of Traders Report provide a sufficient number of setup signals.