Semantic Analysis of the Central Bank of the Republic of Turkey Communication Reports and Forecasting Model with LSTM

Abstract

In this research, a statistical system is designed to understand, interpret, and quantify the reports issued by the Central Bank of the Republic of Turkey (CBRT), an institution that also drives expectations and shapes the market to all agents are related to them. The corpora are CBRT's summaries of monetary policy committee meetings (SMPCM hereafter) published as official press releases. These summaries have three parts: inflation developments, factors affecting inflation, monetary policy, and risks. The graphical representation of items counted and words used shows they ripple together with business cycles. Later, the corpora under these three categories are evaluated according to their sentiment scores. After obtaining them, a Long-Short Term Memory Network is established to derive a quantitative model in order to forecast the sentiment score of each SMPCM, which will be issued in the near future. The LSTM model provides 93% accuracy to estimate semantic scores of SMPCMs.