A Typical Technical Analysis Investment System

• Enter a long order when the closing price of the stock is greater than the 30-day simple moving average of the closing price.
• Close a long order when the closing price of the stock falls below the 10-day weighted moving average of the low price.

The construction of an successful investment system is often time-consuming and must be performed periodically as market conditions change. Furthermore, a system which works well for one stock may not be profitable for another stock.
Because of the number of possible indicator terms multiplied by the number of variations of each indicator, a genetic programming search may be efficiently employed to quickly find investment systems.

Remove the Blinders

Do not fixate on one indicator to the exclusion of others. It is impossible to know which indicators will produce a good system for a particular security and time period so cast a wide net. Investors often have ‘pet’ indicators to which they’ve grown fond. The indicator, like the market, doesn’t know that you love it.
Each trading rule should be required to periodically prove its worth. Most investors resist performing this critical task due to the effort demanded. Now the process is significantly easier with modern evolutionary computation techniques.

Genetic Programming Versatility

A genetic programming search will use the given components at random to find novel systems. Occasionally it will derive trading systems already given a name but usually the system is original. Would you like a trading system with your name on it? I say only half in jest, give Genetic System Search for Technical Analysis a set of data and let it churn for a few minutes. Voilà! You have an indicator to which you can attach your name.
Furthermore, terms can be added to the algorithm’s fitness function to direct the solution. It is possible to automatically evolve trading systems that do not exhibit large drawdowns and long runs of losing trades.

Crossover

In one area of the program it selects two population members with good fitness and therefore, the potential to be even better. As an example, the following two signals might be selected to participate in a crossover:

1. The MACD is greater than 0 AND the 120-period Simple Moving Average of the low is greater than the low
2. The 9-period Relative Strength Index is greater than 10 AND the 9-period ADXR is greater than 25

In a crossover operation one term in the first rule is exchanged for a term in the second rule. The resulting pair might then be:

1. The MACD is greater than 0 AND the 9-period ADXR is greater than 25
2. The 9-period Relative Strength Index is greater than 10 AND the 120-period Simple Moving Average of the low is greater than the low

Mutation

The genetic mutation operator modifies just one rule at a time. The program selects a rule and randomly makes a change to one of its elements. Let’s use the first unmodified rule above as an example:

The MACD is greater than 0 AND the 120-period Simple Moving Average of the low is greater than the low

After a mutation that rule might be changed to:

The MACD is greater than 0 AND the 120-period Simple Moving Average of the low is greater than the close.

The change of the comparison term from the low to the closing price is subtle but may be enough to push the rule set to a higher fitness number.

Numerical Modification

A variation of mutation is the adjustment of numerical constants. Use the revised rule from the mutation operation as an example:

The MACD is greater than 0 AND the 120-period Simple Moving Average of the low is greater than the close

The program may change it to:

The MACD is greater than 0 AND the 130-period Simple Moving Average of the low is greater than the close

Cascading Changes

There is no limit to the number of revisions each rule may undergo. During its life a rule appearing in a trading system may be transformed once, many times or never.

Reevaluation

Changing any of the individual rules in a trading system affects the performance of the system. After each genetic operation, the set of rules for the population members involved are reevaluated and a new fitness number is assigned. The population is sorted according to the fitness number. Low fitness elements eventually die off while members with a high fitness score continue to live and undergo additional genetic operations. The changes don’t always result in improved rules and lower-ranked components are culled from the population.

The individual changes to trading rules at each generation may not be drastic but the program performs hundreds, even thousands of these operations in a typical genetic search. Eventually a population member that scores well in the fitness calculation will appear.

In the construction of technical analysis trading systems one might use a brute-force or exhaustive search for the optimized set of rules. You select a list of indicators and, for each indicator, the range of possible values. As an example, one indicator would be the simple moving average. There are versions of that average for the open, high, low, close and volume values at each trading bar. Each of those versions must be tested with a range of periods. One might build tests for each period from 5 to 300 bars.

All indicators must be tested with each of the possible variables. That’s everything from the 5-bar, simple moving average of the open to the 300-bar simple moving average of the volume. Now multiply that by the number of bars in your stock’s data file. So far you have built the tests for just one indicator. The number of test combinations quickly explodes.

You might reduce the size of the computing project by limiting the number of indicators or the range of the variables but you can’t predict which terms are beneficial to a particular stock’s investment system. If you don’t evaluate the exponential moving average, for instance, that may be just the indicator which can turn an average system into a great performer for a particular security.

Exhaustive searches are simple to implement but at a price. A computer can be programmed to step through every combination of investment rules and variables to find the best system but the number of calculations is massive. Modern computers are fast but exponentially increasing the demands on them results in a test suite that will take an unreasonable amount of time to complete.

Evolutionary Computation and Genetic Programming

More efficient methods have been developed. To significantly reduce the time required to arrive at a set of technical trading rules you may employ one of the forms of evolutionary computation such as genetic algorithms or genetic programming. The Genetic System Search for Technical Analysis program uses the latter technology.

The algorithm randomly creates a large population of candidate systems. Then by a Darwinian process, the program searches through several generations of evolution for the most fit trading rules. The program ends when no improvements can be made to the solution.

As an additional benefit, further enhancements are possible by adding terms to the fitness formula of the genetic programming algorithm. You can tune the computer program to minimize drawdown or the volatility of trade returns. That’s not possible with an exhaustive search. The technique of evolutionary computation combined with modern computer programming results in a clever solution to an unwieldy problem.

See the full explanation of this discrepancy in the Equis MetaStock Knowledge Base.

Consider the case of a huge drawdown during an open trade. Your profit has turned into a loss. Will you stay with the current rules or jump ship? Can you hold on for a vague promise of a profitable outcome in a few days or weeks? That’s unlikely.

The Genetic System Search for Technical Analysis program is designed to produce a system of rules that will encourage you to stay with them. There are factors in the fitness formula that promote rules that avoid confidence-undermining trades. Each candidate system is reduced in fitness according to the number of consecutive losing trades and the amount of the greatest drawdown during the simulation period. Systems exhibiting those characteristics will rank lower than an otherwise equivalent set of rules that don’t include confidence-crumbling properties. The set of investment rules labeled the best of the population at the end of the run has minimized detrimental characteristics.

The pitfalls in the use of the Enhanced System Tester have been documented and it is wise to be familiar with the available cautionary publications. Foremost is Sprunger’s Guide to the Enhanced System Tester. Retrieve it from the Files section of the MetaStock Online Community Website. http://forum.equis.com/files/3015/metastock_files/entry19676.aspx

There can be calculation differences between them but if you plug investment rules created by GSSTA into the MetaStock System Tester, you should see agreement in equity and profit values. The differences will amount to less than one-half of one percent.

The Genetic System Search for Technical Analysis program, Equis MetaStock and others use the following rules to resolve any signal conflicts.

Initial Position

All system tests begin with the investor out of the market.

Entry to Long or Short Trades

In the case of conflicting Enter signals at the same bar, the Enter Long signal will be executed and the Enter Short signal ignored.

Signals Received During an Active Trade

All redundant Enter Long signals are ignored while the investor is in a long trade. All Enter Short signals are ignored while in a short trade.

If the user has specified Long Trades Only, all short trade signals are ignored. If the user specifies Short Trades Only, all long trade signals are ignored.

If the system is in a short position and an Enter Long signal is generated, the short position is exited and a long trade is entered. An Enter Short trade signal will close an active long trade and the short trade will be entered.