Algorithmic Trading with Deep Learning: A Scoping Review

Background. Despite the increasing number of studies in the field of algorithmic trading and deep learning, a very small number of reviews have been conducted. The objective of this scoping review is to summarise the literature on algorithmic trading and to extract the information about DL methods used in algorithmic trading, DL architectures, training sets, and popular metrics used to evaluate models. Methods. This review was conducted in accordance with the PRISMA-ScR recommendations for the search and selection of studies. The search was conducted on the popular scientific database Scopus. A related grey literature research was conducted October, 2022. Eligible studies were those published in English and Russian, that discussed DL in algorithmic trading. Results. 60 records were obtained from the search. 30 of these records met the inclusion criteria. I extracted 10 characteristics from these 30 studies and produced statistics on the following characteristics: demographics, DL methods, network architectures, types of training data, and performance criteria. The most common DL approach in algo-trading was reinforcement learning and LSTM neural networks. The present work has also shown the growing interest in applying alternative types of training data such as text and images in this finance field. Conclusion. This work identified the trends in algorithmic trading, and gaps in the research field. Limitations of the work include our restriction to one database and non-standardized designation of RL models used in studies. The results of this scoping review can be used by the professionals interested in the subject as a supportive material.