Monetary policy document analysis for prediction of monetary policy board decision

In terms of market capitalization, the bond market is larger than the stock market, and the bond market is affected by macroeconomic indicators. Despite this, there has been relatively little research, making it a good candidate for the use of data mining techniques. In this paper, a novel approach designed to predict the vote results of the Korean Monetary Policy Committee regarding the base interest rate was proposed. To predict sentence sentiment, prior monetary policy decision text was used as input for classification models. The sentence sentiment prediction model showed 83.7% performance when using a support vector machine. In addition, it was observed that the bigrams extracted from documents provided important descriptions of the Korean economy at the time. Finally, the document sentiment of monetary policy decision was calculated using aggregating sentence sentiment, and the vote results were predicted using this sentiment. As a result, when using the support vector machine to predict the Monetary Policy Committee vote results, the performance improved by 29.5% over the baseline model. Statistical tests confirmed whether there is a difference in document sentiments between unanimous and non-unanimous, and the null hypothesis was rejected at a significance level of 5%.