In kwojdalski/rpm2:

Research Objective

The primary research goal is to evaluate the Reinforcement Learning-based algorithm for multiasset trading. The main idea behind the algorithm deployment is that it can systematically outperform benchmarks in terms of both risk and return. The trading system will be able to spot non-trivial patterns, faster than human, and exploit them. Research will The goal of this project is to assess the possibility of using Reinforcement Learn- ing to create a trading agent which is capable of nding persistent similarities in nancial time series and which learns how to dene and exploit deviations from the expected, prevalent behaviour. We design and compare two approaches, a basic approach based on Monte Carlo Control and an extended approach based on Qlearn- ing and value function approximation. The rst approach is aimed to provide more interpretability, the second approach is to provide better performance. We assess the outcome by the trading performance of the two agents on two weeks of out-of- sample fx market test data with one minute granulation. We set two benchmarks by which we measure trading performance. The percentage return and the Sharpe ratio of the trades the agent engages in should be higher than the return of a buy-and- hold strategy of either of the underlying cointegrated assets. A secondary goal is to draw conclusions about the interactions in parameter settings of the pair trading frame work and how they in uence protability of the pair trade. The objectives are summarized in the following list: The objectives are as follows:

Design and deployment - this part

kwojdalski/rpm2 documentation built on May 29, 2019, 3:40 a.m.