📈 Bitcoin Price Regression Analysis (Time-Series ML)

A regression-based time-series project to predict Bitcoin short-term returns using historical market signals and engineered features.
The project focuses on time-series-safe validation, interpretable modeling, and realistic evaluation.

---

🎯 Project Goal

This project explores the question:

Can Bitcoin’s next-day movement be predicted using regression models and time-series features?

Instead of chasing unrealistic “perfect price prediction”, the project emphasizes:

• Time-series safe validation (preventing data leakage)
• Feature engineering from market signals
• Model comparison with baselines
• Explainable modeling and interpretable results

---

🧩 Problem Definition

Target Variable

The model predicts next-day log return.

log_return(t+1) = log( Price(t+1) / Price(t) )

Why returns instead of price?

• Bitcoin prices are non-stationary
• Returns provide more stable statistical properties
• Better suited for regression modeling

---

📂 Dataset

This project uses Bitcoin historical market data.

Data Sources

• Bitcoin price and trading volume
• Optional macro indicators such as:
  • S&P 500 return
  • VIX volatility index
  • USD index (DXY)
  • Interest rate indicators

Raw datasets are not included in this repository.
Results can be reproduced by downloading the same data sources and running the notebook.

---

🔧 Feature Engineering

Features are designed to capture momentum, trend, and volatility.

Price-based Features

• Lagged log returns
• Moving averages (MA7, MA30)
• Rolling volatility (7-day and 30-day standard deviation)
• Momentum indicators

Volume-based Features

• Volume change rate
• Rolling volume Z-score

Optional Market Signals

• S&P 500 return
• VIX change
• USD index movement
• Interest rate proxy

These signals help capture market dynamics influencing short-term Bitcoin movement.

---

🧠 Modeling Workflow

The modeling pipeline follows a quantitative ML workflow:

Raw Data
→ Feature Engineering
→ Time-Series Train/Test Split
→ Model Training
→ Prediction
→ Evaluation

Key steps include:

• log-return calculation
• rolling statistical features
• chronological data splitting
• regression model comparison
• performance evaluation

---

🤖 Models Used

Baselines

• Naive baseline (Tomorrow = Today)
• Linear Regression

Regularized Regression

• Ridge Regression
• Lasso Regression

These models evaluate:

• linear relationships
• regularization effects
• robustness against overfitting

---

✅ Evaluation Strategy

Random train/test splits cause data leakage in time-series problems.

Therefore the project uses:

Time-Series Validation

• chronological train/test split
• no future information in training data

Evaluation Metrics

Regression metrics:

• MAE (Mean Absolute Error)
• RMSE (Root Mean Squared Error)
• MAPE (optional)

Directional metric:

• Directional Accuracy (correct prediction of up/down movement)

---

📊 Key Results

(Replace this section with final results)

Model	MAE	RMSE	Direction Accuracy
Naive Baseline	-	-	-
Linear Regression	-	-	-
Ridge Regression	-	-	-
Lasso Regression	-	-	-

Key takeaway:

The objective is not perfect prediction but:

• outperforming naive baselines
• identifying useful predictive signals
• maintaining realistic evaluation

---

📉 Example Prediction Visualization

Example evaluation visualization:

Actual Return vs Predicted Return

This chart helps illustrate how closely the model captures short-term market movements.

You can generate prediction charts from the notebook and place them in:

reports/figures/

Then reference them in this README.

---

🔎 Feature Importance (Interpretability)

Linear and regularized regression models allow us to interpret the influence of features.

Important predictive signals may include:

• Lagged returns
• Rolling volatility
• Volume changes
• Market momentum indicators

These signals help explain what drives short-term Bitcoin movement.

---

▶️ How to Run the Project

1. Install dependencies

pip install -r requirements.txt

2. Launch Jupyter Notebook

jupyter notebook

3. Run the notebook

Open:

notebook/Bitcoin Price Regression Analysis.ipynb

and execute all cells to reproduce the analysis.

---

📌 Business / Financial Interpretation

Although simplified, similar techniques are used in:

• quantitative trading research
• crypto market analysis
• risk monitoring systems
• volatility forecasting

Even simple regression models can reveal meaningful predictive signals in financial time-series data.

---

⚠️ Limitations

Bitcoin markets are challenging due to:

• high volatility
• regime shifts
• macro shocks
• non-linear market behavior

Prediction accuracy may degrade over time.

This project is intended for learning and portfolio demonstration, not financial advice.

---

🗂️ Repository Structure

Bitcoin-Price-Regression-Analysis/

notebook/
Bitcoin Price Regression Analysis.ipynb

src/ (future modularization)
data.py
features.py
train.py
evaluate.py

reports/
figures/

requirements.txt
README.md

---

🛠 Tools & Libraries

• Python
• Pandas
• NumPy
• Scikit-Learn
• Jupyter Notebook

---

📚 What This Project Demonstrates

This repository demonstrates skills in:

• time-series regression modeling
• financial feature engineering
• machine learning experimentation
• model evaluation
• interpretable modeling
• reproducible analytical workflows

---

🚀 Future Improvements

Possible extensions include:

• walk-forward validation
• XGBoost / Gradient Boosting models
• LSTM deep learning models
• crypto-specific indicators
• trading strategy backtesting