Breakthroughs in artificial intelligence make the headlines every day; therefore, it is no surprise that the finance industry doesn’t want to stay behind. Machine learning is a programming technique which helps the computer to learn by using its past results to get an improved outcome. The first part of the machine learning algorithm is running the program through training dataset to learn patterns and characteristics of the data for the second part. There, the program is making predictions or actions based on past results on training data.
Especially for trading, the machine learning technique can be used to create quantitative, algorithmic or other data-based trading strategies. This segment of trading could be straightforward if you are provided with the code which you can simply run through your broker software. On the other hand, if you want to build the code which uses machine learning techniques or only adjust it, you’ll need to possess the high-level skill of programming.
Machine learning methodologies can be used for every asset class. Those methodologies allow creating strategies with better precision of predictions, robustness or also new abilities. Moreover, there is a possibility to build strategies that weren’t able to be made without the abilities of this technique.
Let’s explore some examples. Han and Kong in their paper: The Serial Dependence of the Commodity Futures Returns: A Machine Learning Approach shows how ML methodology can help with selecting only significant predictors which wouldn’t cause overfitting, which is incompatible with out of sample testing. The exact Machine Learning technique, in this case, is LASSO regression which is able to select only essential variables, thus avoiding overfitting and creating meaningful predictions that can be utilized as a base of a profitable trading strategy.
Another example, the research paper that is written by Corbacho, Huerta and Elkan present an enhanced approach of traditional stock-picking anomalies that are based on various technical and fundamental factors that are well-examined in the academic literature. This example uses a nonlinear support vector Machine Learning model to learn the correlations between features, and the profitability of stocks in the training dataset to make and improve out of sample decisions. In this case, it is much more important to choose appropriate training dataset than learning algorithm because each sector could be impacted differently by various fundamental factors which could be misleading in the learning process and ruin the whole strategy. But also, usage of the reduced data set used to train the classiﬁer can improve the resulting performance of the entire strategy. Therefore, training dataset should be based on the quality of the data, not quantity which would lead to redundant computation time.