E!12894 An advanced investment tool based on machine learning and big data

Prosjektets hovedmål har vært å utvikle effektive metoder for finansielle porteføljer og forlenge tidshorisonten for investeringer med kontrollert risiko. Avanserte maskinlæringsmetoder kombinerer tradisjonell tidsserieanalyse med tilbakekoblingsmekansimer og resultatene som oppnåes er mye bedre nå en da prosjektet startet. Prosjektet har derfor jobbet med å videreutvikle og tilpasse disse metodene for de deltagende småbedriftene. En bimålsetting har vært å lære opp personell hos bedriftene i de nyeste teknikkene og stimulere til videre forskning på området i Europa. Hovedresultatet er en plattform for porteføljeoptimalisering for globale finansielle marked. Partnerne har utviklet avanserte visualiseringer som hjelper analytikere i beslutningsprosessen. Avanserte maskinlæringsmetoder er implementert og tatt i bruk av småbedriftene mot langsiktige investorer som pensjonsfond, større bedrifter og offentlig virksomhet for å optimalisere deres porteføljer og minimere finansiell risiko ved plutselige justeringer i markedet. Disse metodene har blitt implementert i Robo Advisor plattformen som opereres av prosjektpartner inbestMe.

We have developed a highly innovative solution for investment firms already using, or interested in using, advanced quantitative methods in asset management. It builds on recent advances in machine learning (ML) and reinforcement learning (RL), which have never been applied to solve complex quantitative finance problems. We positively verified the level of accuracy for time series prediction (10% improvement over reference methods), which clearly proves the breakthrough potential of advanced deep learning methods in investment management. The platform has been tested on real investments, proving the level of prediction accuracy and long-term rate of return, which showed low correlation with the underlying markets affected by the COVID-19 crisis. The Forecasting Module component can be used not only for financial time series forecasting, but also for time series forecasting in other industries such as transportation, sales, energy consumption, climate risk analysis and many others.

The AI Investment platform will be new solution offered in the financial market as Software as a Service. The new solution will be created based on the results of the proof of concept prepared in the project. AI Investment will use the state-of-the-art machine learning (ML) algorithms: both in financial time series prediction (multi-head neural networks based on combination of different types of networks: Wavenet, ResNet, Dilated LSTM, SFM, Differentiable Neural Computer and other; and reinforcement learning: differentiable neural trees, MCSnet, Alpha Zero, multi agent cooperation, meta learning and many more. Unlike some available solutions, AI Investment will use the potential of the latest achievements of machine learning and learn optimization methods and their parameters from data instead of deterministic work performed by the analysts that is optimized in the data mining process. Furthermore, the AI Investment platform will be capable of recursive self-improvement.