₿ BTC Next-Hour Predictor

Predict Bitcoin's next-hour price range using Monte Carlo simulation.

Model

• GBM (Geometric Brownian Motion) with FIGARCH conditional volatility
• FIGARCH(1,1) volatility estimation (with GARCH(1,1) fallback)
• Student-t distributed shocks for fat tails
• Mean-reverted FIGARCH variance for volatility clustering
• 0.96 volatility calibration scale selected on a pre-test calibration window
• 10,000 Monte Carlo simulations per prediction

Project Structure

File	Description
model.py	Core model: data fetching, volatility estimation, GBM simulation
backtest.py	30-day rolling backtest (720 hourly predictions)
app.py	Live Streamlit dashboard with prediction persistence
backtest_results.jsonl	Backtest output (one prediction per line)
requirements.txt	Pinned Python dependencies

Quick Start

pip install -r requirements.txt

# Run the 30-day backtest
python backtest.py

# Launch the dashboard
streamlit run app.py

Backtest Metrics

Run python backtest.py to generate backtest_results.jsonl and print:

• Coverage : fraction of predictions containing the actual price (target: ~0.95)
• Mean Width : average prediction range width (narrower = better)
• Winkler Score : combined accuracy + tightness metric (lower = better)

Current saved backtest_results.jsonl metrics:

Metric	Value
coverage_95	0.9513888889
mean_width	$1,188.3203891551
mean_winkler_95	$1,680.9890754666
Predictions	720
Backtest window	2026-04-02 17:00 UTC to 2026-05-02 16:00 UTC
Volatility scale	0.96

Live Dashboard

URL: btc-predictor-abhiraj.streamlit.app

The dashboard shows:

• Backtest metrics as headline cards
• Current BTC price + 95% prediction range for the next hour
• Candlestick chart of last 50 bars with shaded prediction band
• Prediction history with hit/miss tracking

Persistence saves each generated dashboard prediction through isolated load_history() / save_prediction() functions backed by local JSON storage. On hosted platforms, local files may reset after app sleep or redeploy. The storage code is intentionally small so it can be swapped to an external backend (Supabase, S3, GitHub Gist, etc.).

Data Source

BTCUSDT 1-hour klines from Binance public API:

https://data-api.binance.vision/api/v3/klines?symbol=BTCUSDT&interval=1h

No API key required.