| | --- |
| | 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) |
