Abduction in Econometrics

DELBIANCO, FERNANDO; TOHMÉ, FERNANDO

Handbook of Abductive Cognition

Springer

Año: 2022;

We analyze in this chapter the nature and definition of abduction in Econometrics. Unlike abduction in Economics in general, here we address the question of how surprising results in empirical analyses arise and may be treated. We cover different ways in which traditional econometric methods as well as machine learning tools handle abductive processes. While usually this is not made explicit, the methods we cover here proceed by generating new assumptions of conceptual or methodological nature to manage surprising outcomes. In the case of traditional econometrics, we discuss the unexpected results that may arise in the estimation of regression coefficients. Failures in capturing the actual functional forms, in including omitted variables or due to the violation of other assumptions, may lead to wrong estimations. Different methods like usinginformation criteria or using dummy saturation may help to correct the causes of those failures. Machine learning methods like LASSO or unsupervised learning may help, as well as heterodox econometric methods like autometrics, address the question of obtaining the right elements of a model by just analyzing data, without a very precise initial theoretical model. Other methods covered in this chapter consist in the use of meta-analytic and transitional inference to either use or to obtain different answers to the same empirical question. We also discuss different methods to assess assumed causal relations or to detect them in data. In any case, all the rich information provided by the methods discussed here is obtained through abduction processes.