NLP Results Discussion

📊 BLS Employment Sentiment Analysis

🎯 Key Finding

The model successfully detected the COVID-19 labor market shock with unprecedented precision, showing a sentiment plunge from 0 to -9 in March-April 2020.

Technical Approach: Multi-Indicator Aggregation

Method: Weighted composite scoring across 6 labor market indicators

Data Sources: BLS API (no authentication required)

Update Frequency: Monthly

📈 Sentiment Components & Weights

def get_sentiment_components_weights(df):
    # Core indicators with dynamic weighting
    sentiment_components = []
    weights = []
    
    # Unemployment (25% weight - negative change is positive)
    sentiment_components.append(-df['unemployment_change'])
    weights.append(0.25)
    
    # Payroll changes (20% weight)
    if 'payroll_change' in df.columns:
        payroll_scaled = df['payroll_change'] / 1000
        sentiment_components.append(payroll_scaled)
        weights.append(0.20)
    
    # Additional indicators: wages, participation, etc.
    return sentiment_components, weights

📊 Visualizations

Employment sentiment showing COVID-19 impact in early 2020

Continuous scores revealing sharp pandemic decline to -9.0 and recovery

🔍 Results Interpretation

Pre-Pandemic (2014-2019)

Consistently positive sentiment with minimal volatility, indicating stable labor market expansion

Healthy Baseline

COVID Shock (March 2020)

Dramatic negative spike to -9.0, the most significant labor market disruption in modern history

Crisis Detection

Recovery (2020-2021)

V-shaped recovery pattern as labor markets rebounded with fiscal stimulus and reopening

Rapid Rebound

Normalization (2022-2024)

Return to neutral/positive territory, indicating labor market stabilization

Stable Recovery

💡 Workforce Intelligence Applications

Anomaly Detection: Identify unusual labor market patterns (mass layoffs, hiring surges)
Temporal Trend Analysis: Track long-term workforce composition shifts
Multi-Source Integration: Combine employment data from multiple government agencies
Real-Time Monitoring: Monthly updates for proactive workforce planning

🏦 FOMC Monetary Policy Sentiment

🎯 Key Finding

FinBERT-FOMC accurately tracked the Fed's dovish pivot in 2019-2020 and subsequent hawkish turn in 2021-2022 as inflation emerged.

Technical Approach: Transformer-Based NLP

Model: FinBERT-FOMC (pre-trained on Federal Reserve corpus)

Architecture: BERT with financial domain fine-tuning

Confidence Scores: Most predictions 0.7+ (high confidence)

🤖 Model Implementation

from transformers import BertForSequenceClassification, pipeline

def load_bert_model():
    # Load FOMC-specific FinBERT model
    model = BertForSequenceClassification.from_pretrained(
        'ZiweiChen/FinBERT-FOMC', 
        num_labels=3
    )
    
    # Create sentiment classification pipeline
    sentiment_model = pipeline(
        "text-classification", 
        model=model,
        tokenizer=tokenizer
    )
    
    return sentiment_model

📊 Visualizations

📊 Policy Evolution Timeline

2014-2015: Uncertainty

Negative sentiment (0.3-0.6) during taper tantrum and rate liftoff debates

2019-2020: Dovish Pivot

Positive sentiment (0.9+) as Fed cuts rates and responds to pandemic

2021-2022: Hawkish Turn

Negative sentiment clusters as inflation concerns emerge

2023-2024: Stabilization

Return to neutral/positive as policy normalizes

🔬 Technical Advantages

Domain-Specific Training: Model trained specifically on Fed language and FOMC statements
Contextual Understanding: BERT architecture captures nuanced policy signals
High Confidence: Most predictions above 70% confidence threshold
Automated Scraping: Real-time analysis of new FOMC statements

📄 SEC 8-K Corporate Filing Analysis

🎯 Key Finding

Apple's 8-K sentiment shows 53% positive, 33% negative, 14% neutral filings over 2015-2024, reflecting generally strong corporate trajectory with notable volatility during iPhone transition periods.

Technical Approach: Custom Embedding-Based Analysis

Method: TF-IDF vectorization + custom sentiment axis

Vocabulary: 100+ corporate event terms (positive/negative)

Data Source: SEC EDGAR API with CIK resolution

🎨 Custom Sentiment Axis Construction

def create_corporate_axis(vectorizer):
    # Positive terms: growth, acquisition, synergies, etc.
    POSITIVE_TERMS = [
        'growth', 'expansion', 'acquisition', 
        'synergies', 'breakthrough', 'dividend'
    ]
    
    # Negative terms: restructuring, layoff, litigation, etc.
    NEGATIVE_TERMS = [
        'restructuring', 'layoff', 'litigation', 
        'impairment', 'writedown', 'loss'
    ]
    
    # Create weighted axis vector
    axis = np.zeros(len(feature_names))
    for idx in positive_indices:
        axis[idx] = 1.0
    for idx in negative_indices:
        axis[idx] = -1.0
    
    return axis / np.linalg.norm(axis)

📊 Visualizations

📈 Apple's Corporate Event Timeline

2015-2016: Volatility

Mixed sentiment during iPhone 6S/7 era with varied product reception

2017-2018: Positive Cluster

Strong positive sentiment during iPhone X launch and services growth

2020-2021: COVID Impact

Scattered negative events alongside resilient product launches

2022-2024: Maturation

Return to balanced sentiment with mature product cycles

💼 HR Analytics Applications

M&A Sentiment Analysis: Track workforce integration during acquisitions
Restructuring Detection: Identify layoff announcements and organizational changes
Compensation Events: Monitor dividend/buyback announcements affecting employee stock
Corporate Health: Early warning system for financial distress impacting employment

🏛️ Treasury Fiscal Policy Analysis

🎯 Key Finding

Treasury policy shows overwhelming expansionary stance across three Secretaries (Lew, Mnuchin, Yellen), with the highest spike (score ~3.0) during 2020 COVID relief measures.

Technical Approach: Semantic Vector Analysis

Method: TF-IDF embeddings projected onto fiscal policy axis

Lexicon: 60+ expansionary/contractionary terms

Coverage: 237 press releases across 2014-2025

⚖️ Fiscal Policy Axis Design

def create_policy_axis(vectorizer):
    # Expansionary terms
    EXPANSIONARY = [
        'stimulus', 'spending', 'investment', 
        'infrastructure', 'jobs', 'relief'
    ]
    
    # Contractionary terms
    CONTRACTIONARY = [
        'deficit', 'debt', 'reduction', 
        'austerity', 'consolidation', 'restraint'
    ]
    
    # Calculate sentiment scores with projection
    sentiment_scores = np.dot(document_embeddings, axis)
    
    return sentiment_scores

📊 Visualizations

🌊 Policy Evolution Patterns

2014-2019: Recovery Era

Predominantly expansionary during post-recession recovery (Lew/Mnuchin)

Consistent Growth

2020: COVID Response

Massive expansionary spike (score ~3.0) during CARES Act implementation

Historic Intervention

2021-2022: Infrastructure

Sustained expansion under Yellen with American Rescue Plan

Policy Continuity

2023-2024: Moderation

More balanced stance as focus shifts to fiscal sustainability

Stabilization

🔍 Workforce Policy Implications

Job Creation Programs: Expansionary periods correlate with workforce development initiatives
Infrastructure Hiring: Construction/public works employment tied to fiscal spending
Public Sector Employment: Government hiring patterns follow fiscal stance
Skills Training: Investment in workforce development during expansion phases