Portfolio Management For Asset Forecasting Using Recurrent Neural Network

Abstract

This thesis examines the use of emerging neural networks to predict future financial asset price movements in a set of futures contracts. To help with our research, we compare ourselves to a simple set Feed Network. We do more research on different networks by considering the different functions that lose purpose and how they affect the performance of our networks. This discussion is expanded by considering the Mass Loss Network. The use of different law functions highlights the importance of feature selection. We learn about a set of simple and complex features and how they affect our model. This will enable us to take a closer look at the differences between our networks. Finally, we analyse our model gradients to provide more information about the features of our features. Our results show that repetitive networks offer higher specification performance than relay networks when considering sharpening ratings and accuracy. General features show better results when it comes to accuracy. While the goal of the network is to expand to shards, complex features are selected. Using high-loss networks is successful because we consider achieving high Sharp ratings as our main goal. Our results show better performance than the usual set of benchmarks.