The rich amount of information generated by nowaday’s electronic trading systems, and how it can be exploited e.g. in volatility modeling and in developing profitable trading strategies, have been discussed in the literature in many circumstances. Yet, still little is known about such complex systems, driven by the interaction among different agents, and featuring high aleatoriety. In his dissertation, Martin Magris provides an insight on the effective applicability of machine learning methods for mid-price movements’ prediction, on the nature of long-range autocorrelations in financial time-series, and on the econometric modeling and forecasting of volatility dynamics with high-frequency financial data.
The research articles presented in the dissertation unveil that the detection of complex data-patterns for successfully predicting mid-price movements is achievable by standard machine learning methods, such as ridge regression, simple neural networks, and discriminant analysis. However, the analytics of limit-order book datasets naturally scales also to methods and analyses proper to different fields. In this regard, a central aspect of modern high-frequency financial data is the ubiquitous presence of the so-called long-range autocorrelation in a variety of time-series. By exploiting methods commonly used in physics, but applied to market data, the nature of long-range autocorrelations is analyzed and characterized, by extending the current knowledge on this phenomenon in different directions. On the other hand, long-memory effects are still a challenge for econometricians. Both theoretically and practically very complex econometric models for volatility dynamics have quickly been abandoned in favor of today's most used alternative. This alternative model reads as a simple regression and works quite well on real time-series data. However, there are aspects such as non-linearities, positivity issues of the volatility estimates, and lack of certain modeling approaches in the outstanding literature that allow for improvements and new modeling opportunities that the public discussion will address.
The doctoral dissertation of M.Sc. Martin Magris in the field of high-frequency econometrics titled Volatility modeling and limit-order book analytics with high-frequency data will be publicly examined in the Faculty of Engineering and Natural Sciences at Tampere University at 12 o’clock on Tuesday 27th of August 2019 in Konetalo building, room K1702 (Korkeakoulunkatu 6, Tampere). The Opponent will be Professor Michael McAleer, affiliated with Asia University (Taiwan), Sydney University (Australia), Erasmus University of Rotterdam (The Netherlands), Complutense University of Madrid (Spain) and Yokohama University (Japan). The Custos will be Professor Juho Kanniainen, from Tampere Univerisity’s Faculty of Information Technology and Communication Sciences.
Read the dissertation online: https://trepo.tuni.fi/handle/10024/116373