LLMs in Finance & Trading

1. Financial NLP and Sentiment Analysis

Financial sentiment analysis differs from general sentiment analysis in important ways. The word "liability" is negative in general text but neutral in finance. Phrases like "above expectations" or "revised guidance" carry specific quantitative implications. Financial NLP models must understand these domain-specific nuances to produce reliable signals.

from transformers import pipeline

# FinBERT: finance-specific sentiment model
fin_sentiment = pipeline(
    "sentiment-analysis",
    model="ProsusAI/finbert",
)

headlines = [
    "Company reports Q3 earnings above analyst expectations",
    "Fed signals potential rate cuts amid cooling inflation",
    "Tech giant announces major layoffs, restructuring plan",
    "Supply chain disruptions continue to pressure margins",
]

for headline in headlines:
    result = fin_sentiment(headline)[0]
    print(f"{result['label']:>10} ({result['score']:.3f}): {headline}")

Domain-Specific Financial Models

2. Automated Report Generation

LLMs can generate financial reports by combining structured data (financial statements, KPIs) with natural language analysis. Investment banks, asset managers, and corporate finance teams use these systems to produce first drafts of earnings summaries, market commentaries, and client reports, reducing the time from data availability to published analysis from hours to minutes.

from openai import OpenAI

client = OpenAI()

# Financial data as context
financial_data = """
Q3 2025 Results for TechCorp Inc:
Revenue: $4.2B (vs $3.8B est.), +18% YoY
EPS: $2.15 (vs $1.90 est.)
Cloud segment: $1.8B (+32% YoY)
Operating margin: 28.5% (vs 26.1% prior year)
Guidance: Q4 revenue $4.4B-$4.6B (est. $4.3B)
"""

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": """You are a financial analyst. Write concise, factual earnings
summaries. Include key beats/misses, segment highlights, and forward
guidance. Use professional financial language. No speculation."""},
        {"role": "user", "content": f"Write an earnings summary:\n{financial_data}"},
    ],
)

print(response.choices[0].message.content)

3. Trading Signals and Risk Analysis

LLMs can extract trading signals from news, social media, and regulatory filings. The pipeline typically involves: ingesting text streams in real time, extracting entities and events (earnings surprises, M&A activity, regulatory actions), scoring sentiment and magnitude, and generating structured signals that quantitative systems can act on. The challenge is latency, because in financial markets, information decays rapidly and milliseconds matter.

import json
from openai import OpenAI

client = OpenAI()

def extract_trading_signal(news_text: str) -> dict:
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{
            "role": "system",
            "content": """Extract structured trading signals from financial news.
Return JSON with: ticker, event_type, sentiment (-1 to 1),
magnitude (low/medium/high), time_horizon (immediate/short/long),
confidence (0 to 1), and reasoning."""
        }, {
            "role": "user",
            "content": news_text
        }],
        response_format={"type": "json_object"},
    )
    return json.loads(response.choices[0].message.content)

signal = extract_trading_signal(
    "Apple announces $100B share buyback program, largest in history"
)
print(json.dumps(signal, indent=2))

4. Fraud Detection and KYC/AML

LLMs assist in fraud detection by analyzing transaction narratives, customer communications, and account patterns. For Know Your Customer (KYC) and Anti-Money Laundering (AML), LLMs process adverse media screening, analyze complex corporate structures, and generate investigation summaries. They excel at reducing false positive rates in traditional rule-based systems by understanding the contextual nuances that distinguish legitimate transactions from suspicious activity.