Developing Machine Learning Technique for Measuring Central Bank Credibility

The formulation of monetary policy by the central bank in order to achieve the target will only be effective with the presence of credibility. The role of credibility in supporting policy effectiveness has long been believed by economists and central bank practitioners, including determine the ability of central bank to manage the expectation of economic actors.

Empirically, credibility is a qualitative concept, which is not easy to measure. There are several approaches have been used in measuring credibility, including using survey and constructing composite index from several indicators. For Indonesia’s case Bank Indonesia regularly conducts survey to external stakeholders to measure the policy credibility. The policy credibility survey is prepared based on 6 (six) aspects of credibility, i.e. formulation, independence, communication, accountability, coordination, and effectiveness. But, in practice, the survey method has several weaknesses for measuring policy credibility.

In this research, we aim to utilize Big Data Analytics method, particularly text mining and machine learning, as an alternative method for measuring stakeholders' perceptions on Bank Indonesia’s credibility from news data. Measurement of credibility is done by utilizing machine learning models that are built based on annotated news sentences. From the out-of-sample evaluation results, we achieve an average F1-score of 61.1%. The resulting credibility index also moves in-line with the result of the policy credibility survey, with a correlation of 79.7%.