The bullwhip effect is one of the core problems in supply chain management. Research on it has made a great progress over the years. However, little research has discussed it combining trend demand with order-up-to level replenishment policy in a multi-echelon supply chain. The fluctuation of the bullwhip effect itself is also seldom analyzed. This paper studies the two problems from the view of agent-based simulation. An agent-based model with complex calculation is developed. One highlight is that the players in the supply chain dynamically and intelligently estimate corresponding parameters at every tick. The results indicate that both the bullwhip effect and its fluctuation aggravate with the increase of leadtime. The trend size of customer demand has little influence on the bullwhip effect. The noise of customer demand does not affect the bullwhip effect significantly when it accords with normal distribution. The mean of the bullwhip effect changes differently with the increase of noise autocorrelation coefficient when noise is an AR(1) process. Meanwhile, the standard deviation of the bullwhip effect roughly changes in line with the mean. This paper also presents one bullwhip effect measurement suitable for trend demand.