||In the first part of this thesis, we consider a simplified version of the Wealth Game, which is an agent-based financial market model with many interesting features resembling the real stock market. Market makers are not present in the game so that the majority traders are forced to reduce the amount of stocks they trade, in order to have a balance in the supply and demand. The strategy space is also simplified so that the market is only left with strategies resembling the decisions of optimistic or pessimistic fundamentalists and trend-followers in the real stock market. A phase transition between a trendsetters' phase and a bouncing phase is discovered in the space of price sensitivity and market impact. In the second part, analysis based on a semi-empirical approach is carried out to explain the phase transition and locate the phase boundary. Another phase transition is also observed when the fraction of trend-following strategies increases, which can be explained macroscopically by matching the supply and demand of stocks. Finally, we examine the evaluation sensitivity investment scheme, which is based on local evaluation of simple strategies. The dependence on ways to switch positions, market impact and multiple-period evaluation are studied and the performance of the best combination of methods of this scheme is judged by the random agent benchmark. The study on the probability distribution of decision switching according to this scheme reveals power law characteristics of the first-passage time of market price and supports the existence of market trends instead of pure random walk.