by Larry Swing - July 17, 2005

Education: 

Trading Systems Development

Now that you have an idea and know how to measure trading system success, you can create a model.

The Model

The model typically will consist of two parts:

1. Data preprocessing
2. Trading Rules

Data Preprocessing

Data preprocessing is the translation of raw stock data into a more meaningful form.  This step may also include cleaning errors from data.

Cleaning Errors

Errors may be found through three methods:  Logical checks, comparison checks, and anomaly checks.

Logical checks ensure that the data does not contradict itself.  For example:

• If there is a trade (the price changes), volume must be greater than zero
• The high must be greater than or equal to both the open and close
• The low must be less than or equal to both the open and close

Comparison checks ensure that the database you are using matches up reasonably well with another database.

Anomaly checks automatically search for data points that are several standard deviations away from the mean.  These data points may then be inspected manually.  For example, a search may be done for 1-day percentage returns in excess of 200%, or below -70%.  Unusual data points are more likely to be erroneous, and correction of these data points is especially important since their magnitude will make them important points in your testing.

Data Transformations

Data transformations turn raw data into something that is more meaningful from a trading standpoint.  Instead of using percentage changes in a stock to predict future percentage changes, one may instead wish to use the MACD, a moving average, stochastics, or a host of other data transformations. 

Data transformations may have multiple steps involved.  A simple example:

1. Generate a moving average over a six year period
2. Subtract the stock price from the moving average each day
3. Find the standard deviation of (2)
4. Set all values that are more than four standard deviations above the mean equal to 1, and all values that are more than three standard deviations below the mean equal to -1.
5. Scale the remaining values between -1 and 1, such that they are normally distributed with mean of zero.

The above data transformation will generate a signal line and ensure that it does not take on very extreme values that may skew the interpretation of the indicator.  The downside is that extreme positive values are all treated the same, when in fact they may not be.

Trading Rules

Trading rules may be based on predetermined statements, expert systems, or they may be generated automatically.  The choice between the three is based on the differences between development cost, the type of system, and the knowledge available.

Predetermined Statements

A trader may develop a set of rules that states exactly how he wants to trade.  This makes sense when:

1. It is based on arbitrage
2. The relationship between the indicator(s) and profits is complex, and not amenable to optimization
3. Developing rules using other methods is too expensive

Expert Systems

Similar to a complex set of predetermined statements, expert systems take codified knowledge by and generate inferences.  These are different from predetermined statements in that they generally consist of two separate parts:

1. A knowledge base
2. An inference engine

Expert systems are usually more expensive to develop than predetermined statements, but may be more successful when it is difficult to create explicit trading rules.

Automatic Rule Generation

Trading rules may be automatically generated using neural networks, genetic algorithms, or other algorithms.  This method has the advantage of being able to detect complex nonlinear patterns and develop optimal strategies to profit from these nonlinear patterns.  The disadvantages are:

1. The costs of building the system
2. The requirements for a large amount of data
3. The possible lack of transparency in how the system the interprets data

When generating trading rules automatically, one should have an idea of how variables will interact before inputting those variables into the neural network or other fitting algorithm.

Analyses of HOLDRs

We are neutral on all Holdrs this week; the majority of Holdrs are overbought and positive momentum.  None are within any convincing technical patterns.

Analyses of Individual Stocks

Bullish Stocks

HLTH

• Consolidation breakout
• Strong volume
• Bullish money flow
• MACD crossover

SwingTracker
MrSwings Real-Time Stock Charts RISK-FREE TRIAL featuring one-click access to Larry Swing's profit-generating indicators - Force Index, EquiVolume, True Strength Index

Bearish Stocks

HOTT

• Double top
• Bearish volume