经济研究
經濟研究
경제연구
Economic Research Journal
2008年
6期
145~156
,共null页
不完美信息 贝叶斯模型 实验经济学 羊群行为
不完美信息 貝葉斯模型 實驗經濟學 羊群行為
불완미신식 패협사모형 실험경제학 양군행위
Imperfect Information; Bayesian Model; Experimental Economics; Herd Behavior
Celen和Kanv(2004a)针对不完美信息下序贯决策的社会学习场景,建立连续信号.离散行动的贝叶斯学习模型,但该模型不能很好地解释全部样本的实验结果。本文尝试通过实验进一步考察不完美信息下序贯决策的个体行为是否符合贝叶斯推断。通过设计两种不完美信息(“仅见行动”和“仅闻建议”)下序贯决策实验,对广义贝叶斯模型进行了估计检验,检验结果表明在仅有决策者的前一个人的行动信息下,序贯决策者不可能依贝叶斯模型推断自己的最优解;决策行为大致可以用正态分布描述;处于前后不同决策顺序位置的个体行动选择的分布没有显著差异,他们只能简单地进行一步贝叶斯推断。
Celen和Kanv(2004a)針對不完美信息下序貫決策的社會學習場景,建立連續信號.離散行動的貝葉斯學習模型,但該模型不能很好地解釋全部樣本的實驗結果。本文嘗試通過實驗進一步攷察不完美信息下序貫決策的箇體行為是否符閤貝葉斯推斷。通過設計兩種不完美信息(“僅見行動”和“僅聞建議”)下序貫決策實驗,對廣義貝葉斯模型進行瞭估計檢驗,檢驗結果錶明在僅有決策者的前一箇人的行動信息下,序貫決策者不可能依貝葉斯模型推斷自己的最優解;決策行為大緻可以用正態分佈描述;處于前後不同決策順序位置的箇體行動選擇的分佈沒有顯著差異,他們隻能簡單地進行一步貝葉斯推斷。
Celen화Kanv(2004a)침대불완미신식하서관결책적사회학습장경,건립련속신호.리산행동적패협사학습모형,단해모형불능흔호지해석전부양본적실험결과。본문상시통과실험진일보고찰불완미신식하서관결책적개체행위시부부합패협사추단。통과설계량충불완미신식(“부견행동”화“부문건의”)하서관결책실험,대엄의패협사모형진행료고계검험,검험결과표명재부유결책자적전일개인적행동신식하,서관결책자불가능의패협사모형추단자기적최우해;결책행위대치가이용정태분포묘술;처우전후불동결책순서위치적개체행동선택적분포몰유현저차이,타문지능간단지진행일보패협사추단。
Celen and Kariv (2004a) established the Bayesian Learning Model adding a setup with continuous signal and discrete action, which is in allusion to the social learning setting with sequential decision-making under imperfect information. The Bayesian Model, however, can't be used to explain the experiments' results well. This paper attempts to study whether sequential individuals' behavior under imperfect information accords with the prediction of the Bayesian Model by doing economic experiments. By designing and organizing two kinds of sequential decision-making experiments under imperfect information, we imitate and test the general Bayesian Model. The results show that decision-maker isn't able to calculate his own optimal decision when he can only get the information of his immediate predecessor' s action. Individuals ' behavior can be described by normal distribution approximately. The distribution of individuals' behavior in different decision-turns doesn't have significant differences, and they are only able to do the Bayesian calculation in one step simply.