Directed acyclic graph causal inference framework and its applications in econometrics

Abstract

This thesis aims to first provide basic understanding of directed acyclic graph causal inference and then apply this knowledge on how it is affecting its applicability to econometric research. First part is going through theoretical literature on graphs and statistical graphical models. This thesis will especially concentrate on model developed largely by Judea Pearl with its identification machinery. Then the theory is used to show some specific features considering the framework and taking a look at pre-existing literature in economics regarding this kind of model. Finally, some benefits and possibilities currently as well as in future and some possible developments and their effects for possibilities in economics are discussed.

This part concludes that there is quite a lot of challenges considering these models such as problems with instrument variables as well as with other applications requiring shape restrictions for functions used. However there also already exist use cases such as finding good control variables and techniques that allow more flexibility to used data, like the methods based on selection nodes.