Exploring optimal trading rules in a high-frequency portfolio
Given a set of financial instruments with inherent characteristics at different time intervals, we are interested in finding an optimal trading rule in a high-frequency trading context. A trading rule is defined as a combination of indicators as well as an entry threshold (and potentially other trading parameters). The objective function we are trying to maximize is the profits of the strategy based on the trading rule. One impact of the non-linearity of such problems is that the gradient of the objective function is hard to estimate using a black-box approach. High frequency tick data are used so a high volume of samples is available and this fact compounds the problem. In this project, we set to explore different methods to perform an efficient random search such as the use of low discrepancy sequences, simulated annealing and evolutionary procedures.