Enhancing Stock Investment Returns with Learning Aggressiveness and Trust Metrics
4.9 Global Trading Environment
The CAS stock-trading model is a two-dimensional square world with both
X- and Y-axes ranging from – 10 to + 10 . A variable called radius is used to define the maximal view range for an agent ( a neighborhood ). In order to make sure that every agent has the same opportunity to discover and learn from other agents , the radius is the same for all agents . Also , agents know their location and the identity of agents within their radius .
5 . Implementation
The stock-trading CAS model used in this project is implemented using the
Netlogo 5.1.0 integrated programmable modeling environment ( Wilensky , 1999 ). Netlogo offers a user-defined grid and the possibility of defining agents , commonly called turtles .
In the stock-trading model , variables are created to capture many of the properties investors evaluate in the real world . Also , each variable has a different value range and the increment step size , thus pushing the overall exploration space to the size of trillions . Theoretically , a simulation of this complexity is doable but it requires a & tremendous computational power . For most computing platforms it may take decades before we are able to obtain final results of running a model that uses trillions of agents . Therefore , the agent number was set to 1,000 as a trade-off between the available computing speed and the minimum size of the exploration space . Table 2 shows the settings for the parameters used in the stock-trading model . All transaction decision rules are randomized within the [ -0.4,0.4 ] range for required returns , and within [ 0,100 ] range for the trading periods . Aggressiveness is randomized from 0.0001 to 0.1 , with step size of 0.0001 . Self-confidence is randomized from 0 to 1 , with the step of 0.01 . The small range used in the simulation is chosen with the purpose of decreasing the search space and increasing the coverage for each run . Because mutation is enabled in the simulation , the mutated agents are initialized with a full search range . The initial capital allocated to agents is $ 50,000 , and the transaction cost is fixed at $ 10 per transaction . The mutation rate is fixed at 0.1 , which allows 10 % of all agents to get buy / sell threshold and buy / sell period generated in the whole range [ -1,1 ] and [ 1,1,000 ], respectively .
The price of the BAC stock used in the simulation is adjusted for the effect of dividends . The interest is distributed at the end of each tick , based on the amount of cash held on hand by each agent . As a result , investors have to decide between two goods , BAC shares and money with a fixed interest rate .
59