NLP for Fundamental Analysis

# Install required NLP libraries (run once)
pip install nltk textblob scikit-learn pandas numpy matplotlib seaborn

# Download NLTK data
python -c "import nltk; nltk.download('punkt'); nltk.download('stopwords'); nltk.download('vader_lexicon'); nltk.download('averaged_perceptron_tagger')"

import pandas as pd
import nltk
from nltk.tokenize import word_tokenize, sent_tokenize
from nltk.corpus import stopwords
from nltk.probability import FreqDist
import re
from collections import Counter

# ==========================================
# Simulated TCS Q3 FY24 Earnings Call Excerpt
# ==========================================
tcs_call_text = """
Good morning everyone. I am pleased to report that TCS has delivered strong Q3 results
with revenue of 60,583 crore rupees, representing a year-over-year growth of 8.2%.
Our operating margin improved to 24.5%, driven by strong execution and operational
efficiencies. The digital transformation pipeline continues to grow robustly, with
digital revenues now constituting 58.2% of our total revenue.

Our order book remains strong at $8.1 billion in TCV for the quarter. We are seeing
significant traction in cloud migration, cybersecurity, and AI-driven services.
The banking and financial services vertical showed resilience despite global headwinds.

Looking ahead, we are cautiously optimistic about Q4. The demand environment remains
stable though we are monitoring macroeconomic uncertainty in key markets. Our attrition
rate has declined to 13.3%, and we have added 2,667 employees this quarter. We continue
to invest in upskilling our workforce in generative AI and cloud technologies.
"""

# Step 1: Sentence Tokenization
sentences = sent_tokenize(tcs_call_text)
print(f"📊 Total Sentences: {len(sentences)}")
print("=" * 50)

# Step 2: Word Tokenization & Cleaning
stop_words = set(stopwords.words('english'))
# Add financial-specific stopwords
financial_stopwords = {'crore', 'rupees', 'quarter', 'also', 'would', 'shall'}
stop_words.update(financial_stopwords)

words = word_tokenize(tcs_call_text.lower())
clean_words = [
    w for w in words
    if w.isalpha() and w not in stop_words and len(w) > 2
]

# Step 3: Frequency Distribution — Most Important Words
freq_dist = FreqDist(clean_words)
print("🔑 Top 15 Keywords (by frequency):")
print("=" * 50)
for word, count in freq_dist.most_common(15):
    bar = '█' * count
    print(f"  {word:>18} {bar} ({count})")

# Step 4: Extract Financial Key Phrases
financial_phrases = []
phrase_patterns = [
    r'revenue of [\w, ]+',
    r'growth of [\d.]+%',
    r'margin[\w ]* \d+[\.\d]*%',
    r'order book[\w ]*',
    r'attrition rate[\w ]*',
    r'digital revenues[\w ]*',
]
for pattern in phrase_patterns:
    matches = re.findall(pattern, tcs_call_text.lower())
    financial_phrases.extend(matches)

print("\n💰 Extracted Financial Key Phrases:")
print("=" * 50)
for phrase in financial_phrases:
    print(f"  → {phrase.strip()}")

from sklearn.feature_extraction.text import TfidfVectorizer
import pandas as pd

# Simulated earnings call excerpts for 5 Indian IT companies
earnings_calls = {
    'TCS': "Revenue grew 8.2% driven by digital transformation cloud migration and AI services. Operating margin improved to 24.5%.",
    'Infosys': "We raised FY24 guidance to 4-4.5% growth. Digital and cloud services contributed 62% of revenue. Large deal TCV was $3.2 billion.",
    'Wipro': "Revenue declined 2.1% qoq due to consulting weakness. We are restructuring our leadership team and focusing on AI-first strategy.",
    'HCL Tech': "IT services revenue grew 5.3%. Our Mode 2 and Mode 3 strategy continues to deliver. Cloud native offerings growing at 35%.",
    'Tech Mahindra': "5G and telecom vertical showed strong momentum. AI and automation pipeline grew 40%. Enterprise digital transformation deals accelerating.",
}

# Compute TF-IDF
vectorizer = TfidfVectorizer(stop_words='english', max_features=20)
tfidf_matrix = vectorizer.fit_transform(earnings_calls.values())

# Create DataFrame
tfidf_df = pd.DataFrame(
    tfidf_matrix.toarray(),
    index=earnings_calls.keys(),
    columns=vectorizer.get_feature_names_out()
)

print("📊 TF-IDF Scores — Top Terms per Company:")
print("=" * 55)
for company in tfidf_df.index:
    top3 = tfidf_df.loc[company].nlargest(3)
    terms = ', '.join([f"{t}({s:.2f})" for t, s in top3.items() if s > 0])
    print(f"  {company:>14}: {terms}")

from textblob import TextBlob
import pandas as pd

# ============================================
# Management commentary from Indian companies
# ============================================
management_quotes = {
    'TCS (CEO)': "We are pleased with our strong performance this quarter. Revenue growth has been broad-based across all verticals and geographies. Our digital transformation pipeline is at an all-time high and we remain confident about the demand environment.",

    'Infosys (CEO)': "We have raised our FY24 revenue guidance to 4-4.5%. Our largest deal wins this quarter demonstrate the strength of our capabilities. We continue to see robust demand for cloud, data analytics, and AI services.",

    'Wipro (CEO)': "While we faced headwinds in our consulting business this quarter, we are taking decisive steps to restructure and reposition our portfolio. The macro environment remains challenging and we are seeing delayed decision-making from some clients.",

    'Reliance (Chairman)': "Jio has achieved the milestone of 450 million subscribers. Our new commerce business is scaling rapidly. We are committed to creating world-class digital ecosystems. The energy business delivered record earnings despite volatile markets.",

    'HDFC Bank (CEO)': "We have successfully completed the merger integration. Our deposit franchise remains strong. Asset quality is stable with gross NPA at 1.24%. We are well-capitalized and positioned for sustainable growth in the evolving regulatory environment.",
}

# Analyze sentiment for each
results = []
for speaker, text in management_quotes.items():
    blob = TextBlob(text)
    polarity = blob.sentiment.polarity
    subjectivity = blob.sentiment.subjectivity

    if polarity > 0.3:
        signal = "🟢 BULLISH"
    elif polarity between 0.1 and 0.3:
        signal = "🟡 MODERATELY POSITIVE"
    elif polarity between -0.1 and 0.1:
        signal = "⚪ NEUTRAL"
    else:
        signal = "🔴 BEARISH"

    results.append({
        'Speaker': speaker,
        'Polarity': round(polarity, 3),
        'Subjectivity': round(subjectivity, 3),
        'Signal': signal,
    })

# Display results
df = pd.DataFrame(results)
print("📊 SENTIMENT ANALYSIS — Management Commentary")
print("=" * 65)
print(df.to_string(index=False))

print(f"\n📈 Average Sentiment: {df['Polarity'].mean():.3f}")
print(f"📊 Most Bullish: {df.loc[df['Polarity'].idxmax(), 'Speaker']}")
print(f"📊 Most Bearish: {df.loc[df['Polarity'].idxmin(), 'Speaker']}")

from textblob import TextBlob
from nltk.tokenize import sent_tokenize

# Deep analysis of Wipro CEO's commentary
wipro_text = """While we faced headwinds in our consulting business this quarter,
we are taking decisive steps to restructure and reposition our portfolio.
The macro environment remains challenging and we are seeing delayed
decision-making from some clients. However, our AI-first strategy is
gaining traction with enterprise customers. We remain committed to
delivering long-term value for our shareholders despite near-term volatility."""

sentences = sent_tokenize(wipro_text)

print("🔬 WIPRO CEO — Sentence-by-Sentence Sentiment:")
print("=" * 65)

for i, sent in enumerate(sentences, 1):
    blob = TextBlob(sent)
    pol = blob.sentiment.polarity
    icon = '🟢' if pol > 0.1 else ('🔴' if pol < -0.1 else '⚪')
    bar_len = int(abs(pol) * 30)
    bar = '█' * bar_len

    print(f"\n{icon} Sentence {i} [Polarity: {pol:+.3f}]")
    print(f"   {bar}")
    print(f"   \"{sent[:80]}...\"")

import re
from textblob import TextBlob
from nltk.tokenize import sent_tokenize

# Simulated MD&A section from an annual report
mda_text = """
Management Discussion and Analysis — FY 2023-24

The company achieved record revenue of ₹85,000 crore during the fiscal year,
representing a growth of 12.3% over the previous year. The digital services
segment was the key growth driver, contributing 62% of total revenue.

Going forward, we expect the demand environment to remain stable. We anticipate
that our AI-powered solutions will drive significant revenue growth in FY25.
We plan to expand our presence in Southeast Asian markets and aim to achieve
$2 billion in cloud revenue by FY26.

However, there are certain risk factors that investors should consider.
Global macroeconomic uncertainty may impact client spending in the near term.
Currency volatility, particularly the rupee-dollar exchange rate, could affect
our margins. The rising competition in the AI services space may pressure pricing.
Geopolitical tensions in certain regions pose operational risks.

We believe our strong balance sheet, robust order pipeline, and continued
investment in talent development position us well for sustainable growth.
The board has recommended a dividend of ₹28 per share, subject to shareholder
approval. We are targeting an operating margin of 24-26% for FY25.
"""

sentences = sent_tokenize(mda_text)

# Classification patterns
forward_patterns = [
    r'\b(expect|anticipate|plan|aim|target|will|going forward|believe|project)\b',
]
risk_patterns = [
    r'\b(risk|uncertainty|volatility|threat|challenge|concern|pressure|may impact)\b',
]
opportunity_patterns = [
    r'\b(growth|opportunity|strong|robust|record|expand|sustainable|investment)\b',
]

results = []
for sent in sentences:
    if len(sent) < 20: continue

    is_forward = any(re.search(p, sent.lower()) for p in forward_patterns)
    is_risk = any(re.search(p, sent.lower()) for p in risk_patterns)
    is_opportunity = any(re.search(p, sent.lower()) for p in opportunity_patterns)

    polarity = TextBlob(sent).sentiment.polarity

    category = []
    if is_forward: category.append('🔮 Forward-Looking')
    if is_risk: category.append('⚠️ Risk')
    if is_opportunity: category.append('💡 Opportunity')
    if not category: category = ['📄 Factual']

    results.append({
        'Sentence': sent[:70] + '...',
        'Category': ' | '.join(category),
        'Polarity': round(polarity, 3),
    })

# Print categorized results
print("📋 MD&A ANALYSIS — Sentence Classification:")
print("=" * 70)
for r in results:
    print(f"\n{r['Category']:>35}  [Polarity: {r['Polarity']:+.3f}]")
    print(f"  → {r['Sentence']}")

# Summary statistics
forward_count = sum(1 for r in results if 'Forward' in r['Category'])
risk_count = sum(1 for r in results if 'Risk' in r['Category'])
opp_count = sum(1 for r in results if 'Opportunity' in r['Category'])

print(f"\n📊 SUMMARY:")
print(f"  Total Sentences: {len(results)}")
print(f"  🔮 Forward-Looking: {forward_count} ({forward_count/len(results)*100:.0f}%)")
print(f"  ⚠️ Risk Sentences: {risk_count} ({risk_count/len(results)*100:.0f}%)")
print(f"  💡 Opportunity: {opp_count} ({opp_count/len(results)*100:.0f}%)")

import re

# Financial text with entities to extract
financial_text = """
TCS reported revenue of ₹60,583 crore for Q3 FY24, up 8.2% year-over-year.
Infosys raised its FY24 revenue guidance to 4-4.5%. Wipro's revenue declined
2.1% to ₹22,200 crore. Reliance Industries invested $5.7 billion in Jio.
HDFC Bank's gross NPA stood at 1.24% as of December 2023.
Bajaj Finance disbursed 76 lakh loans in Q3 FY24.
"""

# Rule-based entity extraction
entities = {
    '💰 Money': re.findall(r'[₹$][\d,.]+(?:\s*(?:crore|lakh|billion|million))?', financial_text),
    '📈 Percentages': re.findall(r'[\d.]+%', financial_text),
    '📅 Dates/Periods': re.findall(r'(?:Q[1-4]\s*FY\d{2}|FY\d{2}|\w+\s*\d{4})', financial_text),
    '🏢 Companies': re.findall(r'(?:TCS|Infosys|Wipro|Reliance Industries?|HDFC Bank|Bajaj Finance)', financial_text),
}

print("🏷️ EXTRACTED ENTITIES:")
print("=" * 50)
for entity_type, values in entities.items():
    if values:
        print(f"\n{entity_type}:")
        for v in set(values):
            print(f"  → {v}")

# Build structured data from unstructured text
company_data = []
company_data.append({'Company': 'TCS', 'Revenue': '₹60,583 crore', 'YoY Growth': '8.2%', 'Period': 'Q3 FY24'})
company_data.append({'Company': 'Infosys', 'Guidance': '4-4.5%', 'Period': 'FY24'})
company_data.append({'Company': 'Wipro', 'Revenue': '₹22,200 crore', 'QoQ Change': '-2.1%'})

print("\n📊 Structured Data from Unstructured Text:")
for d in company_data:
    print(f"  {d}")

from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.decomposition import NMF
import numpy as np

# Collection of financial documents
documents = [
    "TCS reported strong revenue growth driven by digital transformation and cloud migration services.",
    "Infosys raised guidance on strong deal wins in cloud computing and data analytics space.",
    "HDFC Bank reported stable asset quality with gross NPA at 1.24% and strong deposit growth.",
    "RBI kept repo rate unchanged at 6.5% citing food inflation concerns and global uncertainty.",
    "Reliance invested heavily in 5G infrastructure and digital commerce platforms this quarter.",
    "Bajaj Finance saw 30% jump in loan disbursements driven by consumer credit demand.",
    "IT sector faces headwinds from banking crisis in US and reduced discretionary spending.",
    "India's GDP growth projected at 7.2% for FY24 driven by government capex and services exports.",
    "Wipro restructured its consulting division and appointed new leadership for AI practice.",
    "SBI reported highest ever quarterly profit driven by treasury gains and lower provisions.",
]

# TF-IDF Vectorization
vectorizer = TfidfVectorizer(stop_words='english', max_features=50)
tfidf = vectorizer.fit_transform(documents)

# NMF Topic Modeling — extract 3 topics
nmf = NMF(n_components=3, random_state=42)
nmf_features = nmf.fit_transform(tfidf)

# Display top words per topic
feature_names = vectorizer.get_feature_names_out()
print("📚 DISCOVERED TOPICS:")
print("=" * 50)

topic_labels = ["Tech & Digital Services", "Banking & Finance", "Macro & Strategy"]
for idx, topic in enumerate(nmf.components_):
    top_words = [feature_names[i] for i in topic.argsort()[-5:]]
    print(f"\n  📌 Topic {idx+1}: {topic_labels[idx]}")
    print(f"     Keywords: {', '.join(top_words)}")

# Assign documents to topics
print(f"\n📄 Document-Topic Assignment:")
for i, doc in enumerate(documents):
    topic_idx = nmf_features[i].argmax()
    print(f"  Doc {i+1} → {topic_labels[topic_idx]}: \"{doc[:60]}...\"")