Walk Forward PatternsINTRO
In Euclidean geometry, every mathematical output has a planar projection. 'Walk Forward Patterns' can be considered a practical example of this concept. On the other hand, this indicator might also be viewed as an experiment in 'how playing with Lego as a child contributes to time series analysis' :)
OVERVIEW
This script dynamically generates the necessary optimization and testing ranges for Walk Forward Analysis based on user-defined bar count and length inputs. It performs automatic calculations for each step, offers 8 different window options depending on the inputs, and visualizes the results dynamically. I should also note that most of the window models consist of original patterns I have created.
ADDITIONAL INFO : WHAT IS WALK FORWARD ANALYSIS?
Although it is not the main focus of this indicator, providing a brief definition of Walk Forward Analysis can be helpful in correctly interpreting the results it generates. Walk Forward Analysis (WFA) is a systematic method for optimizing parameters and validating trading strategies. It involves dividing historical data into variable segments, where a strategy is first optimized on an in-sample period and then tested on an out-of-sample period. This process repeats by shifting the windows forward, ensuring that each test evaluates the strategy on unseen data, helping to assess its robustness and adaptability in real market conditions.
ORIGINALITY
There are very few studies on Walk Forward Analysis in PulseWire. Even worse, there are no any open-source studies available. Someone has to start somewhere, I suppose. And in my personal opinion, determining the optimization and backtest intervals is the most challenging part of WFA. These intervals serve as a prerequisite for automated parameter optimization. I felt the need to publish this pattern module, which I use in my own WFA models, partly due to this gap on community scripts.
INDICATOR MECHANICS
To use the indicator effectively, you only need to perform four simple tasks:
Specify the total number of bars in your chart in the 'Bar Index' parameter.
Define the optimization (In-Sample Test) length.
Define the testing (Out-Of-Sample Test) length.
Finally, select the window type.
The indicator automatically models everything else (including the number of steps) based on your inputs. And the result; you now have a clear idea of which bars to use for your Walk Forward tests!
A COMMONLY USED WINDOW SELECTION METHOD: ROLLING
A more concrete definition of Walk Forward Analysis, specifically for the widely used Rolling method, can be described as follows:
Parameters that have performed well over a certain period are identified (Optimization: In-Sample).
These parameters are then tested on a shorter, subsequent period (Backtest: Out-of-Sample).
The process is repeated forward in time (At each step, the optimization and backtest periods are shifted by the backtest length).
If the cumulative percentage profit obtained from the backtest results is greater than half of the historical optimization profit, the strategy is considered "successful."
If the strategy is successful, the most recent (untested) optimization values are used for live trading.
OTHER WINDOW OPTIONS
ANCHORED: That's a pattern based on progressively expanding optimization ranges at each step. Backtest ranges move forward in a staircase-like manner.
STATIC: Optimization ranges remain fixed, while backtest ranges are shifted forward.
BLOCKED: Optimization ranges are shifted forward in groups of three blocks. Backtest ranges are also shifted in a staircase manner, even at the cost of creating gaps from the optimization end bars.
TRIANGULAR: Optimization ranges are shifted forward in triangular regions, while backtest ranges move in a staircase pattern.
RATIO: The optimization length increases by 25% of the initial step’s fixed length at each step. In other words, the length grows by 25% of the first step's length incrementally. Backtest ranges always start from the bar where the optimization ends.
FIBONACCI: A variation of the Ratio method, where the optimization shift factor is set to 0.618
RANDOM WALK
Unlike the window models explained above, we can also generate optimization and backtest ranges completely randomly—offering almost unlimited variations! When you select the "Random" option in the "Window" parameter on the indicator interface, random intervals are generated based on various trigonometric calculations. By changing the numerical value in the '🐒' parameter, you can create entirely unique patterns.
WHY THE 🐒 EMOJI?
Two reasons.
First, I think that as humanity, we are a species of tailless primates who become happy when we understand things :). At least evolutionarily. The entire history of civilization is built on the effort to express the universe in a scale we can comprehend. 'Knowledge' is an invention born from this effort, which is why we feel happiness when we 'understand'. Second, I can't think of a better metaphor for randomness than a monkey sitting at a keyboard. See: Monkey Test.
Anyway, I’m rambling :)
NOTES
The indicator generates results for up to 100 steps. As the number of steps increases, the table may extend beyond the screen—don’t forget to zoom out!
FINAL WORDS
I haven’t published a Walk Forward script yet . However, there seem to be examples that can perform parameter optimization in the true sense of the word, producing more realistic results without falling into overfitting in my library. Hopefully, I’ll have the chance to publish one in the coming weeks. Sincerely thanks to Kıvanç Özbilgiç, Robert Pardo, Kevin Davey, Ernest P. Chan for their inspiring publishments.
DISCLAIMER
That's just a script, nothing more. I hope it helps everyone. Do not forget to manage your risk. And trade as safely as possible. Best of luck!
© dg_factor
Indicator
