Journal on Policy and Complex Systems
Table 2 . Parameter Ranges
Learning from other agents is disabled in the first 1,000 days . This will allow agents sufficient time to evaluate the profitability and durability of their randomly initialized trading strategies . Afterward , agents learn until the end of the simulation . In this way , agents have enough time to optimize their strategies in different phases of the market , which has different volatilities at different times , that is , financial crises , bull markets , and bubble development periods .
6 . Results
Two underlying assets , S & P 500 Index and BAC , with the buy-and-hold
strategy were used as the benchmark for evaluating the performance of this stock-trading model . The timeframe of the simulation starts at 2 January , 1987 and ends at 31 December , 2014 . In this period , the S & P 500 Index increased from $ 246.45 to $ 2059.9 , and BAC increased from $ 2.37 to $ 17.89 . With the buyand-hold strategy , the return for both benchmarks is 735.82 % and 664.53 %, respectively . Figure 3 shows the cumulative return of different agents in one of the simulations .
60