Beige Book Sentiment Falls Closer to March Lows

Summary

We use machine learning to convert the Fed’s text-heavy Beige Book into a sentiment score. The latest Beige Book report released on 7 September shows a decrease in sentiment from the last report on 13 July (Chart 1). The sentiment creeps closer to the year lows of the 2 March release.

The drop in sentiment is in line with the concluding sentence of the overall economic activity section of the national summary: ‘The outlook for future economic growth remained generally weak, with contacts noting expectations for further softening of demand over the next six to twelve months.’

Here is a summary of the latest developments:

• Breaking down the sentiment by regional district, we find Minneapolis, Philadelphia, and the National Summary to be the only ones showing a net positive sentiment (Chart 2).
• Philadelphia and the National Summary have flipped from negative to positive sentiment since the last report.
• Boston replaces Cleveland as the district showing the most negative sentiment followed by St Louis and New York.
• Dallas finally moves out of the bottom three districts by sentiment after remaining there for all reports since we released our sentiment index (2 March, 20 April, 1 June, 13 July).
• We will update this report just after the release of the next Beige Book on 19 October.

What Is the Macro Hive Beige Book Sentiment Score?

In machine learning, one way to navigate a sea of text and audio-based information is with natural language processing (NLP) techniques. The goal of NLP is to understand textual data to contextualise and extract useful information within it. One application of NLP is sentiment analysis. Sentiment analysis aims to classify whether the opinion expressed in a text is positive or negative (or neutral).

We focus on the Beige Bookand derive a sentiment score by looking at the proportion of positive and negative words in each report. We calculate a raw sentiment score at a district level. Then we aggregate (equally weighted district level average followed by smoothing and detrending) these into an overall sentiment index. We can do this in real-time as soon as the report is released. Charts 4 and 5 show a comparison of our sentiment score to the US ISM PMI and the University of Michigan Consumer Sentiment Index for the US.

Dalvir Mandara is a Quantitative Researcher at Macro Hive. Dalvir has a BSc Mathematics and Computer Science and an MSc Mathematical Finance both from the University of Birmingham. His areas of interest are in the applications of machine learning, deep learning and alternative data for predictive modelling of financial markets.

(The commentary contained in the above article does not constitute an offer or a solicitation, or a recommendation to implement or liquidate an investment or to carry out any other transaction. It should not be used as a basis for any investment decision or other decision. Any investment decision should be based on appropriate professional advice specific to your needs.)