Stock Market Movement Prediction Using 8-K Filings

class: center, middle, inverse, title-slide

.title[
# Stock Market Movement Prediction Using 8-K Filings
]
.author[
### Team Members: Abhishek Varma, Kevin Roy, Saurabh Shete
]
.date[
### December 11, 2024
]

---

## Original Data Analytics Plan
 
.large[
- **Objective**: Stock movement prediction using 8-K filings
  - Predicting next-day stock movements in three categories:
     * Strong Positive (>2% increase)
     * Small Positive (0-2% increase)
     * Negative (any decrease)
 
- **Dataset**: 22,490 rows, 5 columns
  - Including filing text and stock price data
 
- **Time Period**: 2010-2024
  - Covering multiple market cycles
 
- **Companies**: Major corporations (PFIZER, MICROSOFT, APPLE, AMAZON, etc)
  - Focus on large-cap stocks with regular filings
]

---

## Data Source and Structure

.large[
**Dataset Details**:

- **Source**: 
   - SEC EDGAR database
   
   - Yahoo Finance

- **Key Variables**:

- CIK (Company Identifier)
  
  - Company Name
  
  - Filing Date
  
  - Event Information
  
  - Stock Movement Labels
]

---

## Top 10 Companies Analysis
 
![](data:image/png;base64,#8K_Filings_files/figure-html/top_10-1.png)
 
---
## Data Preprocessing
 
.large[
1. **Text Processing**:
 
   - Tokenization & Normalization
   
   - Stop-word Removal
   
   - Special Character Cleaning
 
2. **Feature Engineering**:
 
   - Financial metrics extraction
   
   - Technical indicators
   
   - Text-based features
]
 
---
## Model Architecture
 
.pull-left[
### Deep Learning Model (H2O)
- 3 hidden layers
 
  - Layer 1: 200 neurons
  
  - Layer 2: 100 neurons
  
  - Layer 3: 50 neurons
  
  - RectifierWithDropout activation
]
 
.pull-right[
### Training Features
 
- Adaptive learning rate
 
- Class balancing
 
- Early stopping
 
- L1/L2 regularization
]
 
---
## Company-wise Performance
 
![](data:image/png;base64,#8K_Filings_files/figure-html/company_perf-1.png)
 
---
## Key Findings
 
.large[
1. **Overall Performance**:
 
   - 74.27% prediction accuracy 
   
   - Root Mean Squared Error (RMSE) (0.4585)
   
   - Consistent performance across companies
   
   - Strong financial sector predictions
 
2. **Technical Achievements**:
 
   - Successful text-market data integration
   
   - Effective deep learning implementation
   
   - Robust feature engineering
]
 
---
 
## Response to Peer Comments
 
.large[
### Addressing Key Feedback:
 
1. **Performance Metrics**: Implemented comprehensive evaluation metrics through confusion matrix
 
2. **Time Analysis**: Added temporal pattern analysis across 14 years of data and also stock market data
 
3. **Enhanced Features**: Enhanced model with advanced text processing and technical indicators including moving averages, and volume-price ratios
]

---

class: inverse, center, middle

# Thank You!