SEC 10K Filing Analysis

import kagglehub
import pandas as pd

# Download the Financial Q&A 10K dataset
path = kagglehub.dataset_download('yousefsaeedian/financial-q-and-a-10k')

# Load the CSV file
df = pd.read_csv(f'{path}/Financial-QA-10k.csv')

# View the dataset structure
print(f"Dataset shape: {df.shape}")
print(f"Columns: {df.columns.tolist()}")
print(f"Unique companies: {df['ticker'].nunique()}")

from keyneg import KeyNeg

# Initialize the analyzer
kn = KeyNeg()

# KeyNeg includes financial-relevant sentiment labels:
# - compliance issues
# - technical debt
# - lack of transparency
# - safety concerns
# - ethical violations
# - downsizing
# - bureaucracy
# - and 90+ more...

print("KeyNeg initialized for financial document analysis")

# Sample context from a bank's 10K filing
context = """We are subject to extensive regulation and supervision
under federal and state banking laws. Failure to comply with
applicable regulatory requirements could result in significant
penalties, restrictions on business activities, and reputational harm."""

# Analyze the context
result = kn.analyze(context)

# View the results
print("Top Sentiment:", result['top_sentiment'])
print("Negativity Score:", result['negativity_score'])
print("All Sentiments:", [s[0] for s in result['sentiments']])
print("Categories:", result['categories'])

from collections import Counter

# Sample 500 contexts for analysis
sample_df = df.sample(n=500, random_state=42)
contexts = sample_df['context'].tolist()

# Analyze all contexts in batch
results = kn.analyze_batch(contexts)

# Count contexts with negative sentiment
negative_contexts = [r for r in results if r['sentiments']]
print(f"Contexts with negative sentiment: {len(negative_contexts)}/500")

# Aggregate all sentiments
all_sentiments = []
for r in results:
    if r['sentiments']:
        sentiment_names = [s[0] for s in r['sentiments']]
        all_sentiments.extend(sentiment_names)

# Display top sentiments
sentiment_counts = Counter(all_sentiments)
print("\\nTop 10 Negative Sentiments in 10K Filings:")
for sentiment, count in sentiment_counts.most_common(10):
    print(f"  {sentiment}: {count}")

# Track negativity by company
tickers = sample_df['ticker'].tolist()
company_data = {}

for i, r in enumerate(results):
    ticker = tickers[i]
    if r['sentiments']:
        if ticker not in company_data:
            company_data[ticker] = {'count': 0, 'sentiments': [], 'total_neg': 0}

        company_data[ticker]['count'] += 1
        company_data[ticker]['total_neg'] += r['negativity_score']
        sentiment_names = [s[0] for s in r['sentiments']]
        company_data[ticker]['sentiments'].extend(sentiment_names)

# Calculate average negativity and sort
company_scores = []
for ticker, data in company_data.items():
    if data['count'] >= 3:
        avg_neg = data['total_neg'] / data['count']
        top_issues = Counter(data['sentiments']).most_common(3)
        company_scores.append((ticker, avg_neg, data['count'], top_issues))

company_scores.sort(key=lambda x: x[1], reverse=True)

# Display top companies
print("Companies with Most Negative 10K Language:")
for ticker, avg_neg, count, top_issues in company_scores[:10]:
    issues = ', '.join([f'{i[0]}({i[1]})' for i in top_issues])
    print(f"  {ticker}: {avg_neg:.2f} ({count} contexts) - {issues}")