The Colour of Finance Words

We study a standard machine learning algorithm (Taddy, 2013) to measure sentiment in financial documents. Our empirical approach relies on stock price reactions to colour words, providing as output dictionaries of positive and negative words. We compare head-to-head the performance of the standard bag-of-words approach (Loughran and McDonald, 2011) relative to our machine learning implementation. Our new dictionaries significantly outperform existing techniques when predicting stock price movements out-of-sample. By comparing their composition, word-by-word, our method refines and expands the sentiment dictionaries in the literature.