Exploration of a trading strategy system based on meta-labeling and hybrid modeling using the SigTechPlatform.
Nousiainen, Petri (2021)
Nousiainen, Petri
2021
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2021060313890
https://urn.fi/URN:NBN:fi:amk-2021060313890
Tiivistelmä
The thesis aims to study a machine learning (ML) supported trading system. The methodology is based on a process that is utilizing meta-labeling, thus provides labels for a secondary model, where losses and gains are labeled outcomes. The secondary model provides a prediction based on the primary model output correctness. The secondary model predicts whether the primary model succeeds or fails at a particular prediction (a meta-prediction). A probability correctness measure of the direction prediction of the secondary model is used to size the position. They are "meta" labels because the original simple trading strategy predicts the ups and downs of the market(base predictions or labels). The metalabels predict whether those base predictions are correct or not.
The research question is finding a trading strategy process that can predict the next trading opportunity. The system constructed is a binary classification that aims to determine profitable trades, either buying or selling opportunities. The proposed system is a hybrid model setup of a primary and secondary model.
The process starts with processing raw price data given to a primary model, a trend following Donchian Channel technical analysis model. The primary model output is binary labeling used for the secondary model. The secondary model is the Random Forest machine learning algorithm that predicts if the next trade is a profit or not and with what probability. The prediction probability is used for order sizing. Finally, the rule for executing a trade is that both the primary and secondary models agree on the decision.
The result shows that the study supports the hypothesis that machine learning can improve any trading strategy. The coded trading strategy system successfully predicts the profitable trades with probabilities utilizing meta-labeling and the designed hybrid model. Furthermore, the machine learning prediction of the secondary model significantly improves the performance of the primary model trading strategy.
The research question is finding a trading strategy process that can predict the next trading opportunity. The system constructed is a binary classification that aims to determine profitable trades, either buying or selling opportunities. The proposed system is a hybrid model setup of a primary and secondary model.
The process starts with processing raw price data given to a primary model, a trend following Donchian Channel technical analysis model. The primary model output is binary labeling used for the secondary model. The secondary model is the Random Forest machine learning algorithm that predicts if the next trade is a profit or not and with what probability. The prediction probability is used for order sizing. Finally, the rule for executing a trade is that both the primary and secondary models agree on the decision.
The result shows that the study supports the hypothesis that machine learning can improve any trading strategy. The coded trading strategy system successfully predicts the profitable trades with probabilities utilizing meta-labeling and the designed hybrid model. Furthermore, the machine learning prediction of the secondary model significantly improves the performance of the primary model trading strategy.