Institutional-grade quantitative analytics — C++20 numerical core, Python orchestration, live market data
Eight live pipeline stages: market data → Kalman filtering → volatility surfaces → derivative pricing → portfolio construction → backtesting → walk-forward optimization → probabilistic forecasting. Every formula documented, every number cross-verified.
▶ Try the live demo · Screenshots · Quickstart · Architecture · Methodology · License
Quantitative finance tooling is either locked inside institutions (Bloomberg, in-house desk libraries) or scattered across academic notebooks that never touch live data. QuantSphere Terminal closes that gap: a single auditable platform where the models of the standard quant curriculum — Black-Scholes PDEs, Heston, jump-diffusions, Kalman filters, GARCH, Markowitz — run natively in C++ against real market data, with the mathematics of every displayed number documented in-app.
| Stage | What it does | The hard part underneath |
|---|---|---|
| 📡 Market Data | 10y daily + intraday bars (1m–1h), VWAP, drift/vol/skew/kurtosis, historical VaR | Session-aware axes (no phantom overnight moves), vendor-failure retries |
| 🛰 Kalman Filter | Latent price & drift extraction with ±2σ bands, h-step-ahead prediction | Exact linear-Gaussian filter + RTS smoother in C++; innovation whiteness diagnostics |
| 🌋 Vol Surface | Live options chain → 3-D implied volatility surface + smiles by expiry | Newton–Raphson IV with analytic vega and guaranteed bisection fallback; no-arbitrage quote rejection |
| ⚛ Pricing Lab | European / American / barrier options under 4 dynamics, Greeks, VaR | Multithreaded Monte Carlo (GBM, Heston, Merton, Variance Gamma) vs Crank–Nicolson PDE (Thomas / PSOR) vs closed form — three engines, one screen |
| 🧭 Portfolio | Efficient frontier, tangent & min-variance portfolios, capital market line | SLSQP over ridge-regularized covariance; realized performance curves of the weights and a deploy-ready $-allocation table |
| 📈 Backtest | 5 causal strategies × 5 timeframes, costs, lot sizing, trend & volume filters | Strict 1-bar execution lag; a clairvoyant-signal test proves zero look-ahead mechanically |
| 🎯 Optimizer | Hyperparameter sweeps with 3-D objective landscapes and overfitting diagnostics | Chronological walk-forward split — configurations are ranked only on data they never saw |
| 🔮 Forecast | Probabilistic price cones, GARCH(1,1) vol term structure, 3-D forward density | Exact MLE GARCH; circular block bootstrap preserving fat tails; honest quantiles, not point predictions |
Plus 📚 in-app documentation (⚙ Settings): 40+ LaTeX formulas — model equations, estimators, numerical schemes, assumptions and model risk — the full methodology the platform can be audited against.
 |
 |
| Live market ingestion & statistics | 3-D implied volatility surface from the live chain |
 |
 |
| Monte Carlo vs PDE vs closed form | Walk-forward optimization landscape |
 |
 |
| Causal backtesting with exposure tracking | Every formula documented in-app (LaTeX) |
git clone https://github.com/<you>/quantsphere-terminal.git
cd quantsphere-terminal
pip install -r requirements.txt
streamlit run app.pyThat's it — the platform boots on the NumPy engine everywhere (including Streamlit Community Cloud / Hugging Face Spaces). No API keys: market data is Yahoo Finance.
pip install pybind11
cmake -S engine -B engine/build -DPYBIND11_FINDPYTHON=ON \
-DPython_EXECUTABLE="$(which python)"
# Windows: add -G "Visual Studio 17 2022" -A x64
cmake --build engine/build --config Release
# copy the built qsengine.* module next to quantsphere/__init__.pyBoth engines expose an identical API and are cross-verified against each other to 10⁻¹⁴ in the test suite — the app transparently uses whichever is available.
┌──────────────────────────────────────────────────────────────┐
│ app.py — Streamlit terminal (institutional dark theme) │
├──────────────────────────────────────────────────────────────┤
│ quantsphere/ — Python orchestration │
│ data.py ingestion, cleansing, stats (retry-hardened)│
│ backtest.py causal vectorized backtester │
│ optimize.py walk-forward hyperparameter search │
│ models.py GARCH(1,1) MLE, EWMA, forecast cones │
│ portfolio.py Markowitz MVO (SLSQP, long-only) │
│ engine.py dispatch: native C++ ⇄ NumPy fallback │
│ _fallback.py API-identical NumPy mirror │
├──────────────────────────────────────────────────────────────┤
│ engine/ — C++20 core (pybind11, GIL released, multithreaded)│
│ mc.hpp Monte Carlo: GBM · Heston · Merton · VG │
│ pde.hpp Crank–Nicolson + Thomas O(N) / PSOR (American)│
│ kalman.hpp Kalman filter + RTS smoother │
│ iv.hpp implied volatility: Newton + bisection │
└──────────────────────────────────────────────────────────────┘
Design decisions that matter:
- Deterministic parallelism — Monte Carlo work is split into 64 fixed units seeded via splitmix64: results are bit-identical on any machine, any thread count.
- Two implementations of everything numerical — the C++ core and the NumPy fallback agree to 10⁻¹⁴ on deterministic algorithms; disagreement is a test failure, not a shrug.
- Causality as an invariant — every trading signal is verified bit-identical when future data is truncated; the backtester's clairvoyant-signal test earns exactly $0 on information it shouldn't have.
- Risk-neutral ≠ real-world — pricing simulates under ℚ, forecasting under ℙ with an explicit drift choice; the docs explain why you must never read a pricing fan as a prediction.
python tests/test_engine.py # 43 checks — numerical engines
python tests/test_quant.py # 45 checks — backtester, optimizer, forecasting88 network-free checks, including:
- Black-Scholes closed-form values and put-call parity to machine precision
- Implied-vol round-trips at 10⁻¹³; arbitrage-violating quotes rejected
- Monte Carlo within standard error of closed form for all four dynamics
- Exact path-wise barrier in/out parity (
V_KO + V_KI = V_vanilla) - PDE vs closed form at 4×10⁻⁵ relative; American ≥ European ≥ intrinsic
- Kalman/RTS native ≡ fallback at 10⁻¹⁴
- Look-ahead guards: clairvoyant signals earn zero; all signals causal under truncation
- GARCH(1,1) parameter recovery on simulated data (α within 0.05 of truth)
- Optimizer test scores equal an independent out-of-sample replay at 10⁻⁹
Streamlit Community Cloud (free): fork → share.streamlit.io → point at app.py. The NumPy engine activates automatically.
Hugging Face Spaces: create a Streamlit Space, push this repo.
Réalisé par Ismael LADJOHOUNLOU — © 2026.
Licensed under the MIT License (LICENSE):
- ✅ Free to use, study, self-host, modify and ship, including commercially
- 📎 Keep the copyright notice; that is the whole obligation
- 💼 Available for consulting and custom quantitative engineering
QuantSphere Terminal is a research and educational instrument. Market data comes from public sources and may be delayed or inaccurate. Backtests are in-sample research artifacts, not promises. Nothing produced by this software constitutes investment advice.
🔎 Keywords & topics covered by this project
Institutional trading system · quantitative finance platform · quant terminal · algorithmic trading system · quantitative analysis software · financial engineering toolkit · quant research platform
Derivatives & pricing: options pricing engine · Black-Scholes model · Black-Scholes PDE solver · Crank-Nicolson finite differences · American options (PSOR / linear complementarity) · barrier options (knock-in, knock-out) · exotic options · Monte Carlo simulation · Heston stochastic volatility · Merton jump-diffusion · Variance Gamma process · Lévy processes · option Greeks (delta, gamma, theta) · risk-neutral pricing
Volatility & forecasting: implied volatility solver (Newton-Raphson) · 3D volatility surface · volatility smile & skew · options chain analysis · GARCH(1,1) maximum likelihood · EWMA / RiskMetrics · volatility forecasting · probabilistic price forecasting · forecast cones · block bootstrap · Value-at-Risk (VaR) · Expected Shortfall (CVaR)
Signal processing & strategies: Kalman filter trading · RTS smoother · state-space models · latent drift estimation · trend following · mean reversion · moving average crossover (SMA/EMA) · volatility targeting · trading signal filters · VWAP
Backtesting & optimization: vectorized backtesting engine · look-ahead bias prevention · walk-forward analysis · out-of-sample validation · hyperparameter optimization · overfitting diagnostics · Sharpe ratio · Sortino · Calmar · maximum drawdown · transaction cost modeling · multi-timeframe backtesting
Portfolio: Markowitz mean-variance optimization · efficient frontier · maximum Sharpe portfolio · minimum variance portfolio · capital market line · covariance regularization · asset allocation
Technology: C++20 quantitative library · pybind11 bindings · high-performance computing · multithreaded Monte Carlo · NumPy · SciPy · pandas · Streamlit dashboard · Plotly 3D visualization · yfinance market data · Python quant stack