Extract Trading Signals from Fed Announcements in 20 Minutes

# Personal note: Learned this after wasting time on generic models
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import pandas as pd
import numpy as np
from datetime import datetime

# Use FinBERT - crucial for financial context
MODEL_NAME = "ProsusAI/finbert"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)

# Watch out: Default device is CPU, use GPU if available
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
model.eval()

print(f"Model loaded on {device}")
print(f"FinBERT vocabulary size: {len(tokenizer)}")

# Personal note: Fed statements average 800 words - too long for BERT's 512 token limit
import re

def split_into_sentences(text):
    """
    Smart sentence splitter that handles Fed-specific formatting.
    Preserves context by keeping section headers with their content.
    """
    # Clean up common Fed statement artifacts
    text = re.sub(r'\s+', ' ', text)  # Normalize whitespace
    text = re.sub(r'\.{2,}', '.', text)  # Fix multiple periods
    
    # Split on periods, question marks, exclamation points
    # But preserve common Fed abbreviations
    sentences = re.split(r'(?<!\bU\.S)(?<!\bMr)(?<!\bMs)(?<!\bet al)[.!?]\s+', text)
    
    # Filter out very short fragments (likely artifacts)
    sentences = [s.strip() for s in sentences if len(s.split()) > 3]
    
    return sentences

# Test with sample Fed text
sample_text = """
The Committee seeks to achieve maximum employment and inflation 
at the rate of 2 percent over the longer run. In support of these goals, 
the Committee decided to raise the target range for the federal funds rate 
to 5-1/4 to 5-1/2 percent. The Committee will continue to assess additional 
information and its implications for monetary policy.
"""

sentences = split_into_sentences(sample_text)
print(f"Split into {len(sentences)} sentences")
for i, sent in enumerate(sentences, 1):
    print(f"{i}. {sent[:80]}...")

def analyze_fed_sentiment(text, return_details=False):
    """
    Analyzes Fed statement sentiment with financial context.
    
    Returns:
        - hawkish_score: 0-1 (higher = more hawkish/tightening)
        - dovish_score: 0-1 (higher = more dovish/easing)
        - confidence: Model's certainty level
    """
    sentences = split_into_sentences(text)
    
    results = []
    
    for sentence in sentences:
        # Tokenize with proper truncation
        inputs = tokenizer(
            sentence, 
            return_tensors="pt", 
            truncation=True, 
            max_length=512,
            padding=True
        )
        inputs = {k: v.to(device) for k, v in inputs.items()}
        
        # Get predictions
        with torch.no_grad():
            outputs = model(**inputs)
            probs = torch.nn.functional.softmax(outputs.logits, dim=-1)
        
        # FinBERT outputs: [negative, neutral, positive]
        # For Fed: negative = dovish, positive = hawkish
        scores = probs[0].cpu().numpy()
        
        results.append({
            'sentence': sentence,
            'dovish': float(scores[0]),  # negative sentiment
            'neutral': float(scores[1]),
            'hawkish': float(scores[2]),  # positive sentiment
            'confidence': float(max(scores))
        })
    
    # Aggregate scores weighted by confidence
    total_weight = sum(r['confidence'] for r in results)
    
    hawkish_score = sum(r['hawkish'] * r['confidence'] for r in results) / total_weight
    dovish_score = sum(r['dovish'] * r['confidence'] for r in results) / total_weight
    avg_confidence = total_weight / len(results)
    
    output = {
        'hawkish_score': round(hawkish_score, 3),
        'dovish_score': round(dovish_score, 3),
        'net_stance': round(hawkish_score - dovish_score, 3),
        'confidence': round(avg_confidence, 3),
        'sentences_analyzed': len(results)
    }
    
    if return_details:
        output['sentence_details'] = results
    
    return output

# Test with real Fed language
test_statement = """
Recent indicators suggest that economic activity has been expanding at a solid pace. 
Job gains have remained strong, and the unemployment rate has remained low. 
Inflation remains elevated. The Committee remains highly attentive to inflation risks.
"""

result = analyze_fed_sentiment(test_statement)
print("Fed Statement Analysis:")
print(f"Hawkish Score: {result['hawkish_score']} (higher = tightening bias)")
print(f"Dovish Score: {result['dovish_score']} (higher = easing bias)")
print(f"Net Stance: {result['net_stance']} (positive = hawkish, negative = dovish)")
print(f"Confidence: {result['confidence']}")

# Real FOMC statement excerpts from 2023
fed_statements = {
    "2023-02-01": """
    Recent indicators point to modest growth in spending and production. 
    Job gains have been robust in recent months, and the unemployment rate 
    has remained low. Inflation has eased somewhat but remains elevated. 
    The Committee decided to raise the target range for the federal funds rate 
    to 4-1/2 to 4-3/4 percent.
    """,
    "2023-05-03": """
    Economic activity has continued to expand at a modest pace. Job gains have 
    been robust, and the unemployment rate has remained low. Inflation remains 
    elevated. The Committee decided to raise the target range for the federal 
    funds rate to 5 to 5-1/4 percent and will continue to assess additional information.
    """,
    "2023-09-20": """
    Recent indicators suggest that economic activity has been expanding at a solid pace. 
    Job gains have moderated since earlier in the year but remain strong. Inflation remains 
    elevated. The Committee decided to maintain the target range for the federal funds rate 
    at 5-1/4 to 5-1/2 percent.
    """
}

# Analyze each statement
results_df = []

for date, statement in fed_statements.items():
    result = analyze_fed_sentiment(statement)
    result['date'] = date
    results_df.append(result)

df = pd.DataFrame(results_df)
df = df.sort_values('date')

print("\nFed Sentiment Timeline:")
print(df[['date', 'net_stance', 'hawkish_score', 'dovish_score', 'confidence']].to_string(index=False))

# Calculate month-over-month changes
df['stance_change'] = df['net_stance'].diff()
print("\nKey Shifts:")
for _, row in df.iterrows():
    if pd.notna(row['stance_change']):
        direction = "more hawkish" if row['stance_change'] > 0 else "more dovish"
        print(f"{row['date']}: {direction} by {abs(row['stance_change']):.3f}")