Myopic Agents in Assessments of Economic Conditions: Application of Weakly Supervised Learning and Text Mining
Masahiro Kato · arXiv (Cornell University) · 2019
We reveal thepsychological bias of economic agents in their judgments of future economic conditions by applying the behavioral economics and weakly supervised learning. In the Economy Watcher Survey, which is a dataset published by the Japanese government, there are assessments of current and future economic conditions by people with various occupations. Although this dataset gives essential insights regarding economic policy to the Japanese government and the central bank of Japan, there is no clear definition of future economic conditions. Hence, in the survey, respondents answer their assessments based on their interpretations of the future. In our research, we classify the text data using learning from positive and unlabeled data (PU learning), which is a method of weakly supervised learning. The dataset is composed of several periods, and we develop a new algorithm of PU learning for efficient training with the dataset. Through empirical analysis, we show the interpretation of the classification results from the viewpoint of behavioral economics.