Discussing the article: "Self Optimizing Expert Advisors in MQL5 (Part 8): Multiple Strategy Analysis (3) — Weighted Voting Policy"
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Check out the new article: Self Optimizing Expert Advisors in MQL5 (Part 8): Multiple Strategy Analysis (3) — Weighted Voting Policy.
This article explores how determining the optimal number of strategies in an ensemble can be a complex task that is easier to solve through the use of the MetaTrader 5 genetic optimizer. The MQL5 Cloud is also employed as a key resource for accelerating backtesting and optimization. All in all, our discussion here sets the stage for developing statistical models to evaluate and improve trading strategies based on our initial ensemble results.
When building an ensemble of strategies, naturally, a question that follows is, how can we prove that all the strategies we've selected are necessary? How can we be reasonably sure that we wouldn’t perform better with just a few of them? How can we prove any of this to ourselves?
Fortunately for us, the genetic optimizer can help answer such challenging questions, provided we carefully frame the question for it.
To achieve this, we will allow our strategies to collaborate through a democracy, where each strategy is allowed only one vote. The weight of the vote that each strategy casts can be a tuning parameter, adjusted by the genetic optimizer. If the optimizer determines that one of our strategies isn’t positively contributing to overall performance, it will set the weight of that strategy’s vote close to zero. Likewise, it will add more weight to the strategies that are profitable.
Therefore, we present this framework as a weighted voting policy, in which we initially set a benchmark performance level by giving each of our strategies uniformly distributed vote weights. In our example, we start with each strategy having a vote weight of 0.5, on a scale that ranges from 0 to 1.
From there, we allow the genetic optimizer to adjust these weights to maximize profitability and determine whether all three strategies are truly helpful.
It turns out this procedure returns a wide array of different configurations, each showing how the usefulness of a strategy can change depending on the particular strategy settings. In each unique configuration, the weight of each strategy shifts. So, there may be a setup where only one strategy proves useful, while in another configuration, all three are contributing positively.
This makes the question "Are all three strategies necessary?" a genuinely challenging one to answer. Our findings suggest that the answer is sensitive to the configuration the application employed in the first place. Let us begin.
Author: Gamuchirai Zororo Ndawana