Probability of informed trading (PIN) models characterize trading with certain types of information through a tree structure. Different tree structures with different numbers of groups for market participants have been proposed, with no clear, consistent tree used in the literature. One of the main causes of this inconsistency is that these trees are artificially proposed through a bottom-up approach rather than implied by actual market data. Therefore, in this paper, we propose a method that infers a tree structure directly from empirical data. More precisely, we use hierarchical clustering to construct a tree for each individual firm and then infer an aggregate tree through a voting mechanism. We test this method on US data from January 2002 for 7608 companies, which results in a tree with two layers and four groups. The characteristics of the resulting aggregate tree are between those of several proposed tree structures in the literature, demonstrating that these proposed trees all reflect only part of the market, and one should consider the proposed empirically driven method when seeking a tree representing the whole market.

知情交易（PIN）模型的概率通过树形结构来表征某些类型的信息的交易。已经提出了针对市场参与者的具有不同数目的组的不同树结构，在文献中没有使用清晰，一致的树。造成这种不一致的主要原因之一是，这些树是通过自下而上的方法人为提出的，而不是由实际的市场数据暗示的。因此，在本文中，我们提出了一种直接根据经验数据推断树结构的方法。更准确地说，我们使用层次聚类为每个公司构建一棵树，然后通过投票机制推断出一棵聚合树。我们使用2002年1月以来针对7608家公司的美国数据测试了该方法，结果得出一棵有两层和四组的树。