Enhancing Stock Investment Returns with Learning Aggressiveness and Trust Metrics
expectation of future market moves . Self-confidence is assigned randomly in the simulation setup stage , and changes based on the agent ’ s learning aggressiveness during the simulation . This mechanism makes it possible for agents to explore the best degree of confidence in their decision rule sets . At the same time , best performers ’ self-confidence values will be tracked and available to all agents in the learning stage of the simulation .
4.5 Learning and Interaction
In agent-based models , agents interact with each other . Learning is one of such
interactions . In our model , learning offers agents the opportunity to check and compare their decision rules with those of the best performers , thus making it possible for them to refine their rules and secure more profits in the future similar market trends ( Cui , Wang , Ye , & Yan , 2012 ).
Unlike the degree of trust , which only changes the agents ’ decision rule temporarily based on the latest market trends , learning has a permanent impact on agents ’ trading strategies . The learning mechanism makes it possible to investigate alternative strategies that have not yet been discovered in the market ( Outkin , 2012 ). In this implementation , to preserve computational time and to maximally alleviate the constraints of the limits of available computing power , a radius is introduced in the simulation . Agents can only see other agents within their own radius while moving around , which helps with avoiding the homogenization of agents . Agents have the opportunity to learn from the best performers within their radius , thus within their neighborhood , which helps with improving the learning efficiency as well . With the ability to learn and to decide how much they want to learn from their neighbors ( i . e ., how close they want to get to their neighbor ’ s value on a particular variable ), the size of the exploration space of optimal decision rules is reduced from the size of trillions of combinations into a much smaller one .
The variable aggressiveness defines the extent of neighborhood best performers ’ behavioral structure the agent wants to adopt . It ranges from 100 % to nothing , with most cases being somewhere in the middle . Prior to the learning phase , agents ’ original decision rules will be given sufficient time to evaluate their performance . The learning process starts only after this evaluation period .
4.6 Search Space and Mutation
Given the number of variables in the model used in this research , the size
of the search space of all possible combinations is in trillions of space states . In order to run a simulation with a much small search space and to still make sure that the simulation can theoretically explore all possible states , a
57