mrince/CBRT-RoBERTa-Large-HawkishDovish-Classifier

Model Details

This model is a RoBERTa-based classifier fine-tuned on 7,200 sentences extracted from the Central Bank of the Republic of Türkiye's (CBRT) Monetary Policy Committee (MPC) Summary Reports. The goal is to classify monetary policy tone as hawkish, dovish, or neutral.

Dataset: mrince/CBRT-MPC-SummaryReports

PERFORMANCE

Label	Precision	Recall	F1-score	Accuracy
neutral	0.8970	0.8235	0.8587	0.8235
hawkish	0.9226	0.9245	0.9235	0.9245
dovish	0.8926	0.9641	0.9270	0.9641
Overall Accuracy	-	-	-	0.9042

Model Description

• Developed by: Muhammet Rıdvan İNCE
• Model type: RoBERTa-large text classifier
• Language(s) (NLP): English
• License: MIT
• Finetuned from model: FacebookAI/roberta-large

---

Model Sources

• Repository: CBRT-RoBERTa-Large-HawkishDovish-Classifier
• Paper: In preparation

---

Direct Use

This model can be used to analyze the tone of central bank policy statements or similar macroeconomic texts in English, especially in academic research or financial market monitoring.

Downstream Use

It may be incorporated into monetary policy forecasting tools, financial sentiment dashboards, or market reaction models.

Out-of-Scope Use

This model is not intended for real-time trading decisions, legal interpretation of monetary policy, or classification of unrelated financial texts.

---

Bias, Risks, and Limitations

Limitations

• The model does not incorporate numerical macroeconomic indicators or forward guidance intensities.

Recommendations

Use in conjunction with economic domain knowledge. Model performance should be monitored when applied to new types of central bank communication.

---

How to Get Started with the Model

from transformers import pipeline

# Load the classifier pipeline
classifier = pipeline("text-classification", model="mrince/CBRT-RoBERTa-Large-HawkishDovish-Classifier")

# Example sentence from a monetary policy context
sentence = "On the other hand, the recent deceleration in economic activity may curb services inflation."

# Perform classification
result = classifier(sentence)
print(result)

# Output example:
# [{'label': 'Label_2', 'score': 0.9983660578727722}]

# Label meanings:
# - 'hawkish [Label_1]': Tightening bias, upward rate signal. Indicates prioritization of inflation control.
# - 'dovish [Label_2]' : Easing bias, downward/inflation-tolerant tone. Indicates support for growth/stimulus.
# - 'neutral [Label_0]': Informational or balanced tone without a clear policy stance.