A Rust desktop application for analyzing market structure by modeling sector volatility, cross-sector correlations, bond spreads, and training a local neural network for volatility regime prediction.
- Sector Volatility Analysis: Rolling historical and Parkinson volatility for 11 SPDR sector ETFs (XLK, XLF, XLE, XLV, XLI, XLP, XLY, XLU, XLRE, XLC, XLB)
- Cross-Sector Correlation Matrix: Pairwise Pearson correlation heatmap with average cross-correlation tracking
- Bond Spread Analysis: Treasury yield curve visualization, 10Y-2Y term spread, curve slope (30Y-3M), and inversion detection
- Neural Network: LSTM-based volatility regime prediction model trained locally using the Burn deep learning framework
- Interactive Dashboard: egui-powered GUI with tabbed navigation, interactive charts (pan/zoom), and real-time data refresh
- Yahoo Finance API — Historical OHLCV data for sector ETFs and SPY benchmark
- Financial Modeling Prep API — Treasury rates across all maturities, sector performance snapshots
| Component | Technology |
|---|---|
| Language | Rust (2021 edition) |
| GUI | eframe + egui + egui_plot |
| Charts | egui_plot (line, bar, heatmap) |
| ML Framework | Burn (NdArray backend with autodiff) |
| Async | Tokio |
| HTTP | reqwest + yahoo_finance_api |
- Rust toolchain (1.75+): https://rustup.rs
- Internet connection for API data fetching
cargo run --releaseThe first build will take several minutes to compile all dependencies. Subsequent builds are fast.
- Launch the application
- Click Refresh Data to fetch market data from Yahoo Finance and FMP
- Navigate between tabs:
- Dashboard — Overview heatmap with sector metrics
- Sector Vol — Per-sector volatility charts with selector dropdown
- Correlations — Cross-sector correlation matrix heatmap
- Bonds — Yield curve and term spread analysis
- Neural Net — Train the LSTM model and view predictions
- On the Neural Net tab, click Train Model to start training (runs in a background thread)
API responses are cached locally in the cache/ directory to avoid redundant network requests. Cache freshness is 12 hours for market data and 1 hour for sector performance snapshots.
src/
main.rs — Entry point, launches eframe window
app.rs — Application state, tab routing, data fetch orchestration
config.rs — API keys, sector symbols, NN hyperparameters
data/
yahoo.rs — Yahoo Finance async client
fmp.rs — Financial Modeling Prep async client
models.rs — Core data structures (OHLCV, TreasuryRate, etc.)
cache.rs — Local JSON cache layer
analysis/
volatility.rs — Historical vol, Parkinson vol, vol ratio
cross_sector.rs — Correlation matrix, rolling correlation
bond_spreads.rs — Term spread, yield curve, inversion detection
nn/
model.rs — Burn LSTM model definition
dataset.rs — Feature engineering and sliding window dataset
training.rs — Training loop with progress reporting
ui/
dashboard.rs — Sector heatmap and key metrics
sector_view.rs — Per-sector volatility charts
correlation_view.rs — Correlation matrix heatmap
bond_view.rs — Yield curve and spread charts
nn_view.rs — Model training controls and predictions
- Rolling Historical Vol: Annualized standard deviation of log returns (21-day and 63-day windows)
- Parkinson Volatility: Range-based estimator using High/Low prices (more efficient than close-to-close)
- Volatility Ratio: Short-term / long-term vol ratio as a regime indicator (>1.0 = rising vol)
- Architecture: LSTM (hidden=64) -> Linear (output=1)
- Features (26): 11 sector rolling vols, 11 sector returns, avg cross-correlation, 10Y-2Y spread, curve slope, VIX proxy
- Training: Sliding 60-day windows, MSE loss, Adam optimizer, 80/20 chronological split
- Output: 5-day forward realized volatility prediction
