Machine and deep learning for effective time series forecasting
The increasing complexity of financial markets revealed the need for frameworks and methods development that facilitate traders, market makers, and modelers in their trading activities. Yet, still little is known about what drives stocks’ market price behaviour. In his dissertation, Adamantios Ntakaris provides an insight on how feature engineering, combined with machine and deep learning methods, facilitates the task of financial time series forecasting. The forecasting task is related to the prediction of mid-price movements of high-frequency financial limit order book data.
The research articles presented in the thesis tackle the challenging task of mid-price movement prediction by combining machine and deep learning models together with hand-crafted features inspired by technical and quantitative analysis, econometrics and deep learning. This combination, which was part of several extensive experimental protocols, was tested on Nordic and US stocks and proved to be very efficient in forecasting mid-price movements.
These articles expand the feature engineering literature since they set the basis for every future work on advanced feature engineering development for financial time series. The conducted work reveals that specific feature sets and machine (or deep) learning models react differently according to stock selection. These experimental protocols, suitable for online learning, suggest that a data-driven approach for the task of mid-price movement prediction is an efficient way to improve the forecasting ability of both classifiers and regressors.
The doctoral dissertation of M.Sc. Adamantios Ntakaris in the field of feature engineering and machine learning titled Mid-Price Movement Prediction in Limit Order Books Using Feature Engineering and Machine Learning will be publicly examined at the Faculty of Information Technology and Communication Sciences at Tampere University at 12 o’clock on Friday 25th of October 2019 in Tietotalo building, room TB109, Korkeakoulunkatu 1, Tampere. The Opponent will be Professor of Practice Peter Sarlin, affiliated with Hanken School of Economics (Finland). The Custos will be Professor Moncef Gabbouj, from Tampere University’s Faculty of Information Technology and Communication Sciences.
The dissertation is available online at the http://urn.fi/URN:ISBN:978-952-03-1288-6