Denis Peskoff,Adam Visokay,Sander Schulhoff,Benjamin Wachspress,Alan Blinder,Brandon M. Stewart
Abstract:Markets and policymakers around the world hang on the consequential monetary policy decisions made by the Federal Open Market Committee (FOMC). Publicly available textual documentation of their meetings provides insight into members' attitudes about the economy. We use GPT-4 to quantify dissent among members on the topic of inflation. We find that transcripts and minutes reflect the diversity of member views about the macroeconomic outlook in a way that is lost or omitted from the public statements. In fact, diverging opinions that shed light upon the committee's "true" attitudes are almost entirely omitted from the final statements. Hence, we argue that forecasting FOMC sentiment based solely on statements will not sufficiently reflect dissent among the hawks and doves.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is to quantify the differences among members of the Federal Open Market Committee (FOMC) in the process of monetary policy decision - making, especially the differences in views on inflation. The author uses GPT - 4 to analyze the public text records of FOMC meetings, including meeting minutes, statements and complete verbatim transcripts, in order to identify the differences of opinion among members. The study found that, compared with the finally released statements, the meeting minutes and verbatim transcripts can better reflect the diverse views of members on the macro - economic outlook. Specifically, the paper points out that the divergent opinions that reveal the "true" attitude of the committee are almost completely omitted in the final statements. Therefore, predicting the attitude of the FOMC based only on the statements is not sufficient to reflect the differences between hawks and doves.
### Main problems:
1. **Quantifying the differences among FOMC members**: The paper quantifies the differences among members by using GPT - 4 to analyze the text records of FOMC meetings, especially the views on inflation.
2. **Comparing different types of documents**: The paper compares meeting minutes, statements and verbatim transcripts, and finds that verbatim transcripts and minutes can better reflect the real views and differences of members, while the final statements tend to simplify and unify these views.
3. **Improving prediction accuracy**: The paper proposes that relying only on the final statements to predict the attitude of the FOMC is not enough, because many important difference - related information is omitted in the statements.
### Methods:
- **Data sources**: The paper uses the FOMC meeting minutes, statements and verbatim transcripts from 1994 to 2016.
- **Tools**: Use GPT - 4 to analyze the text and quantify the views of hawks and doves.
- **Indicators**: Five labels (dove, dovish - leaning, neutral, hawkish - leaning, hawk) are defined and used to classify each document or sentence.
- **Analysis methods**: Measure the differences by calculating the number of hawkish and dovish views in each meeting.
### Conclusions:
- **Differences are widespread**: The paper finds that differences exist in 82% of the verbatim transcripts and 47% of the statements.
- **Limitations of statements**: The final statements often overlook the differences among members, resulting in their inability to fully reflect the true attitude of the committee.
- **Effectiveness of GPT - 4**: GPT - 4 performs well in identifying similar content in different document types, especially when dealing with long texts.
### Significance:
- **Policy - making**: Understanding the differences among FOMC members is helpful for better understanding the process of monetary policy - making.
- **Market prediction**: For financial market participants, identifying the differences within the FOMC can help more accurately predict future policy trends.
Through these analyses, the paper provides a new perspective for understanding the decision - making process of the FOMC and shows the application potential of large - language models in economic text analysis.