Perform Cluster Analysis (Mulitple Walk-Forward Analysis)

Run the TradeStation Walk-Forward Optimzer to perform a walk-forward analysis on the strategy that was optimized in TradeStation:

When using Exhaustive or Genetic sample walk-forward:

Select File > Open Walk-Forward Test and choose an data set from the Select a Walk-Forward Test drop-down list that you wish to perform the Walk-forward analysis on.

Press the Open button to automatically select all trading list files that form part of this Walk-forward analysis.

Cluster Analysis

To determine if a strategy is truly robust, a Cluster Analysis of Multiple Walk-Forward Analyses (WFA's) is recommended. The Cluster Analysis will also assist in determining the optimal trading window, after which re-optimization should take place.  While a single Walk-Forward Analysis may give a preliminary indication as to whether a strategy is robust, a Cluster Analysis has the ability to prove or disprove the validity of a trading strategy with greater assurance.

Before you can start a Cluster Analysis, you first need to identify the criteria for measuring the success of your strategy.

Use the Setup > Test Criteria Settings menu sequence to access the Setup Test Criteria dialog.  Then review each of the five Criterion settings and change their values if desired.  If you are uncertain, the default values are a good place to start and are sufficient is most cases.  Click the OK button to save the settings and return to the WFO.

Next, use the Setup > Optimization Settings menu sequence to access the Setup Optimization Settings dialog.  

A check mark in the Prescribe # of Walk-forward Runs box indicates that you will specify the percentage of out-of-sample data and the number of walk-forward runs to use.  If the box is unchecked, WFO will automatically adjust the number of runs to based on the values specified for In-Sample and Out-Of-Sample bars.

Set an appropriate Out Of Sample% and number of Walk-forward Runs.  

The default settings under Cluster Analysis: Matrix Multiple Walk-Forward Analyses are:

Cluster Analysis Settings	Start	End	Inc
Out-Of-Sample %	10	30	5
Walk Forward Runs	5	30	5

The total number of cluster combinations tested is number of Out-Of-Sample % iterations (ie 10, 15, 20 25, 30) times the number of walk-forward Runs iterations (ie 5, 10, 15, 20, 25, 30) or 5x6=30 possible cluster test combinations.

These default settings are highly recommended unless you are an experienced user of WFO.  A minimum of 5 Walk-forward runs are recommended to overcome random results, while specifying a number too high, will result in the trading history file to be divided into out-of-sample windows that are too short on their own, with too little trades to have any statistical significance. The Out Of Sample% is that portion of the trading window which is walked forward with each new run.

The total trading window length for each run is automatically calculated as a function of the number of walk-forward runs, except when Anchored WFO is selected. With anchored walk-forward optimization, the starting point of the trading window is fixed (as the beginning of the trade history file) and the out-of-sample length remains the fixed based on the first anchored run.

(Refer FAQ: How many walk forward runs should be done to validate a WF analysis? and FAQ: What is the Out-of-sample % and what % should be used? for more detail)

Place a check mark in the Anchored check box to select anchored walk-forward optimization. (If Anchored is not ticked, WFO uses rolling walk-forward optimization)

   Cluster Analysis results for each checked/unchecked setting of Prescribe # of walk-forward runs and Anchored are saved as separate data sets that can be accessed using the Type drop-down on the Cluster Analysis tab of the WFO for the specified Walk-Forward Test Name.

Click the OK button once you are satisfied with the Cluster Analysis optimization settings.

Finally, use the Walk-Forward Analysis > Start Cluster Analysis (Multiple WFA's) menu sequence to perform the cluster analysis.