📈 Financial Reasoning CoT R1 (gitmoney313/financial-reasoning-cot-r1)

A high-tier, specialized quantitative finance and market strategy dataset with DeepSeek-R1 style step-by-step <think>...</think> chain-of-thought reasoning.

Designed to train, fine-tune, and evaluate Large Language Models on complex, mathematically rigorous financial decision-making, corporate valuation, options hedging, crypto tokenomics, and portfolio risk analysis.

---

🌟 Why This Dataset?

Most open reasoning datasets in the AI ecosystem focus predominantly on grade-school arithmetic (GSM8K) or introductory Python programming.

financial-reasoning-cot-r1 fills the institutional gap by providing deep, multi-step analytical chains that mirror the mental models of Wall Street quants, investment bankers, and decentralized finance architects.

---

🔬 Core Domains Covered

1. 🏢 Corporate Valuation & Multi-Stage DCF
   • Free Cash Flow to Firm (FCFF) forecasting
   • WACC discounting and terminal value perpetuity models
   • Enterprise Value to Equity Value reconciliation
2. 📉 Options Greeks & Dynamic Hedging
   • Delta-neutral market maker portfolio adjustments
   • Gamma risk and price gap sensitivity
   • Rebalancing trades under volatility shocks
3. 🦄 Crypto AMMs & Impermanent Loss
   • Constant product invariant ($x \cdot y = k$) mechanics
   • Liquidity provider value vs. passive holding (HODL)
   • Closed-form impermanent loss quantification
4. 🛡️ Portfolio Risk & Parametric Value at Risk (VaR)
   • Multi-asset covariance matrix and diversification benefits
   • Daily volatility scaling via square-root-of-time
   • 99% parametric dollar VaR thresholds
5. 🏛️ Macroeconomics & Yield Curve Arbitrage
   • Inverted yield curves (2s10s spread dynamics)
   • Duration weighting and DV01 (Dollar Value of a Basis Point) immunization
   • Parallel shift neutrality vs. curve steepener positioning
6. 🪙 Crypto Tokenomics & Supply Overhang
   • Market Cap vs. Fully Diluted Valuation (FDV) float dynamics
   • Price elasticity of demand under venture lockup expiries
   • Secondary market absorption rates and equilibrium price estimation
7. 📊 Algorithmic Trading & Downside Risk
   • Annualized excess return over risk-free benchmarks
   • Sharpe Ratio vs. Sortino Ratio divergence
   • Semi-variance and downside deviation profiling

---

📋 Data Schema

Each sample in the dataset contains the following fields:

Field	Type	Description
id	string	Unique sample identifier (e.g. fin_r1_001)
category	string	Financial domain classification
prompt	string	The complex quantitative or strategic scenario prompt
thinking	string	Deep step-by-step mathematical reasoning wrapped in <think> ... </think>
response	string	Final structured answer with executive takeaways and formulas

---

🚀 How to Use

Load directly in Python using the Hugging Face datasets library:

from datasets import load_dataset

# Load the dataset directly from the Hugging Face Hub
dataset = load_dataset("gitmoney313/financial-reasoning-cot-r1")

# Inspect the first example
sample = dataset["train"][0]
print("Category:", sample["category"])
print("\n--- Prompt ---")
print(sample["prompt"])
print("\n--- Chain-of-Thought Reasoning ---")
print(sample["thinking"][:500] + "...")
print("\n--- Final Answer ---")
print(sample["response"])

Fine-Tuning with Unsloth / Hugging Face TRL

This dataset is pre-formatted for direct Supervised Fine-Tuning (SFT) or Direct Preference Optimization (DPO) with models like Llama 3.1, Qwen 2.5, and Mistral:

def format_prompt(example):
    return {
        "text": f"<|user|>\n{example['prompt']}\n<|assistant|>\n{example['thinking']}\n{example['response']}"
    }

formatted_dataset = dataset.map(format_prompt)

---

📄 License

This dataset is released under the permissive MIT License. Free for academic and commercial use.