README.md

---
base_model: EleutherAI/gpt-neo-125M
library_name: peft
license: mit
metrics:
- accuracy
- precision
- recall
- f1
- Pattern Detection Rate
- Cross-Entropy Loss
tags:
- finance, IndianStocks, Technical Analysis, Chartless Trading
language:
- en
pipeline_tag: text-generation
---

---
base_model: EleutherAI/gpt-neo-125M
library_name: peft
---

# Model Description

**IndicFinGPT** is a specialized transformer model, re-engineered from **EleutherAI's GPT-Neo-125M** architecture, which is a GPT-3 class architecture, designed specifically for the **Indian financial market**. The model has undergone **retraining on its top layers** to enhance its performance in providing insights into the **top 100 companies listed in the NIFTY50 Index, BSE, and NSE exchanges**. 
The primary objective of this model is to **serve the unique needs of Indian stock markets** and **investors engaged in chartless trading**. IndicFinGPT aims to provide insights that could **minimize capital loss and drawdowns** while **maximizing financial ratios** such as the **Sharpe, Sortino, Calmar, Omega, and Treynor Ratios**. Additionally, the model is designed to help in **reducing maximum drawdowns** in financial portfolios, offering a robust AI solution tailored to **India’s dynamic financial landscape**.

## First Indic-Stock Small Language Model Focused Top 100 Companies Listed in NSE and BSE Stock Exchanges 

<p align="center">
  <img src="https://huggingface.co/bhaskartripathi/GPT_Neo_Market_Analysis/resolve/main/indicBull.JPG" alt="IndicFinGPT Logo" width="400" height="300">
  <strong>भारतीय बाजार की शीर्ष 100 कंपनियों का वित्तीय विश्लेषण करने वाला पहला Small Language Model</strong>
</p>

## Training Data and Procedure

**IndicFinGPT 125M** utilizes the **Pile dataset** created by EleutherAI and includes the **top 100 tickers** (by volume and liquidity) from Indian stock markets, covering data from **January 1, 2018, to October 30, 2024**. This dataset encompasses diverse market periods, including **pre-COVID-19 (stable), COVID-19 (volatile), and post-COVID-19 (recovery phase)**. Such comprehensive data exposure allows the model to recognize **problem-solution patterns across various bull and bear runs**.
The training data also incorporates **local influences** such as cultural factors and **market-specific volatility**, enhancing its ability to perform **automated technical analysis** for chartless trading. Key capabilities include identifying **classical chart patterns** using technical analysis, conducting **earnings analysis**, interpreting **market sentiment** from multiple sources, and **assessing risks**, all aimed at **improving decision-making for Indian investors**.
This model weights were obtained after **310 billion tokens over 692,380 steps**. It utilized 4-bit Quantized Low-Rank Adoption (PEFT) method on top of  the masked autoregressive language model architecture of Neo, utilizing cross-entropy loss, F1, Accuracy, Precision, recall,Pattern Detection Rate, and Cross-Entropy Loss as performance metrics.


## Key Highlights

1. Trading Patterns: Specialized in recognizing BSE/NSE-specific patterns and cycles
2. Market Sentiment: Built-in understanding of Indian market sentiment and cultural influences
3. Macro-Economic Indicators: Adapted to domestic economic and financial metrics
4. Indian Economic Influences: Awareness of timing, festival impacts, and market-specific volatility


## Implementation

### Quick Start
```python
from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("bhaskartripathi/GPT_Neo_Market_Analysis")
tokenizer = AutoTokenizer.from_pretrained("bhaskartripathi/GPT_Neo_Market_Analysis")

input_text = '''[INST] Given the following stock market data and technical analysis:
Stock: EXAMPLE
Date: 2024-01-01
Technical Analysis:
Current Price: ₹100
Daily Range: ₹98 - ₹102
Trading Volume: 1,000,000
RSI: 55
MACD: Bullish
Based on this technical analysis, what is the likely price movement for tomorrow and why? [/INST]'''

inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)
result = tokenizer.decode(outputs[0], skip_special_tokens=True)
```

## Training Details

### Dataset and Fine-tuning
- **Dataset**: Comprehensive dataset featuring 6 years of Indian market data.
- **Method**: Fine-tuned using QLoRA (4-bit quantization) for optimal efficiency.
- **Training Infrastructure**: Utilized an Nvidia T4 GPU, trained for ~6 hours with PEFT framework version 0.13.2.

## Performance Metrics
- **Pattern Recognition**: High accuracy in classical and advanced pattern detection in Indian markets.
- **Sentiment Correlation**: Strong alignment with local market movements.
- **Risk & Volatility Handling**: Reliable risk analysis in volatile market conditions.

## Market Understanding

### Technical Analysis Expertise
The model is adept at identifying crucial market formations including:
- **Classical Patterns**: Head & Shoulders, Double Top/Bottom, Triangle, Flag, Wedge, Cup and Handle.
- **Advanced Techniques**: Local support and resistance levels, volume analysis, and momentum indicators specifically tailored to Indian volatility.

### Market Intelligence
IndicFinGPT includes:
- **Comprehensive Financial Reports**: Analysis of quarterly and annual earnings.
- **Risk Metrics**: Indian-adapted VaR, Beta, and volatility models.

### Cultural Context in Trading
Culturally aware strategies include:
- **Indian Market Timing**: Recommendations tailored to pre-market, regular, and post-market phases.
- **Festival & Cultural Factors**: Insights into events like Diwali (Muhurat Trading), budget announcements, and investor sentiment.
- **FII/DII Flow and Retail Behavior**: Specific guidance considering both institutional and retail dynamics.

## Social Impact

IndicFinGPT democratizes sophisticated AI-based financial analysis for the Indian stock market, providing affordable and accessible tools for both seasoned investors and new traders.

## Core Capabilities

#### Automated Q&A based Technical Analysis for chartless Trading:
Investors, Traders, Economists, Econometricians and Researchers can ask any types of questions related to the below areas:

- **Head and Shoulders patterns**
  - What are the implications of a Head and Shoulders pattern forming for Tata Consultancy Services (TCS) in the upcoming week?
  - How does the identification of a Head and Shoulders pattern for Reliance Industries influence its potential price movement?

- **Double Top/Bottom patterns**
  - What is the expected market behavior for Infosys if a Double Top pattern has formed over the last two weeks?
  - How does a Double Bottom pattern in Tata Steel indicate a possible upward trend?

- **Triangle formations**
  - What trading opportunities are indicated by a symmetrical triangle formation in Hindustan Unilever?
  - How could an ascending triangle in Tata Motors impact its price performance in the coming days?

- **Flag patterns**
  - What are the implications of a bullish flag pattern for the stock of Infosys in the short term?
  - How can a flag pattern formation in Reliance Industries affect trading strategies for the next three days?

- **Wedge patterns**
  - How does a rising wedge pattern in Tata Steel signal a potential market reversal?
  - What are the likely outcomes of a falling wedge pattern detected in Tata Consultancy Services (TCS)?

- **Cup and Handle patterns**
  - Can you provide an analysis of a Cup and Handle pattern formation in Hindustan Unilever?
  - How could a Cup and Handle pattern affect the price movement of Reliance Industries in the coming week?

Earnings Analysis:

- **Key metrics extraction**
  - What are the key earnings metrics extracted for Infosys for the latest quarter?
  - How do the extracted financial metrics for Tata Motors compare to previous earnings?

- **Historical comparisons**
  - How does the historical earnings performance of Tata Consultancy Services (TCS) compare to the current quarter?
  - What insights can be gained by comparing historical earnings of Hindustan Unilever over the last three years?

- **Red flag identification**
  - Are there any red flags in the latest earnings report of Reliance Industries?
  - What potential risks are identified in Tata Steel's financial report?

- **Positive indicator detection**
  - What are the positive financial indicators in the latest earnings of Tata Motors?
  - How do the positive indicators for Infosys reflect its market position?

Market Sentiment Interpretation:

- **Price-based sentiment analysis**
  - How does the recent price movement of Reliance Industries reflect market sentiment?
  - What sentiment indicators can be derived from the price fluctuations of Tata Steel?

- **News sentiment analysis**
  - How might recent news regarding Tata Consultancy Services (TCS) impact its stock price in the next few days?
  - What is the sentiment derived from the latest business news about Hindustan Unilever?

- **Social media sentiment analysis**
  - How is social media sentiment trending for Infosys, and what impact could this have on its stock price?
  - What does the current social media sentiment indicate about Tata Motors in the upcoming week?

- **Sentiment divergence calculation**
  - How does the divergence between price-based sentiment and news sentiment impact the outlook for Tata Consultancy Services (TCS)?
  - What are the implications of a sentiment divergence for Reliance Industries over the next few days?

Risk Assessment:

- **Volatility analysis**
  - What does the volatility analysis indicate for Tata Steel over the next week?
  - How volatile is the stock of Hindustan Unilever in the current market scenario?

- **Beta calculation**
  - How does the beta of Tata Motors compare to other companies in the Nifty 50 index?
  - What does the beta calculation imply about the risk associated with Infosys?

- **Value at Risk (VaR) computation**
  - What is the VaR for Reliance Industries, considering the current market conditions?
  - How does the VaR for Tata Consultancy Services (TCS) help in understanding the potential risk in the next three days?

- **Risk rating determination**
  - How is the risk rating for Hindustan Unilever determined based on current data?
  - What is the risk rating for Tata Steel, and how could it influence trading strategies?

Trading Strategy Recommendations:

- **Pattern-based analysis**
  - What are the potential trading opportunities for Reliance Industries based on recent flag or wedge pattern formations in the next week?
  - How does the Double Top pattern for Tata Steel indicate a possible trend reversal in the coming days?

- **Sentiment-driven insights**
  - How might recent news and social media sentiment affect the stock price of Infosys over the next three days?
  - What is the current sentiment regarding Tata Consultancy Services (TCS), and how could it impact its performance over the next week?

- **Risk-adjusted recommendations**
  - What are the risk-adjusted trading strategies for Infosys in light of current market volatility?
  - Based on beta calculations and current market sentiment, what are the recommended actions for Tata Steel in the coming days?

- **Historical context integration**
  - How have similar market conditions in the past affected the performance of Hindustan Unilever, and what can be expected this week?
  - Considering past Diwali trading patterns, what is the expected impact on Reliance Industries this year?

## Evaluation Results

#WandB Report: https://wandb.ai/bhaskar-tripathi-indian-institute-of-foreign-trade/indian-market-analysis-system/workspace

<p align="center">
  <img src="https://huggingface.co/bhaskartripathi/GPT_Neo_Market_Analysis/resolve/main/eval_results.JPG" alt="IndicFinGPT Logo" width="1024" height="800">
</p>

## Citation
```bibtex
@misc{tripathi2024indicfin,
  title={IndicFinGPT: Market Analysis Model for Indian Stocks},
  author={Bhaskar Tripathi},
  year={2024},
  url={https://huggingface.co/bhaskartripathi/GPT_Neo_Market_Analysis}
}
```

## Contact
- **Email**: bhaskar.tripathi@volkswagen.co.in
- **HuggingFace**: [@bhaskartripathi](https://huggingface.co/bhaskartripathi)
- **Google Scholar**: [Profile](https://scholar.google.com/citations?user=SCHOLAR_ID)
- **Github**: [Click Here](https://github.com/bhaskatripathi)