Forward Guidance Through  Newspaper Articles : Explaining Movement in Bond Yields Through Text Analysis

Sammanfattning: Forward guidance policies are necessarily difficult to asses through their nature as verbal publishing. This thesis aims at estimating a forward guidance indicator through measuring public perception of the central bank’s interest rate policy. For the case of the Swedish Riksbank, public perception is measured through automatic classification of 29 250 Swedish newspaper articles into categories of Hawkish (expecting an interest rate increase), Dovish (expecting an interest rate decrease) and Neutral articles. As a classification mechanism, a Support Vector Mechanism is trained on manual classification data and used to predict the indicator category based on information on feature frequency. The resulting article level information is aggregated on a monthly level and used in a Vector Autoregressive Model to predict yield information of government benchmarks with one month, 2 years and 10 years maturity. Through this I find that public perception measured through categorized newspaper articles is useful in predicting short term yields but insignificant in predicting the movement in yields of longer maturity benchmarks.