English · Español
Open your Excel model, mark the uncertain inputs, and run thousands of scenarios.
OpenRiskSim shows the distribution of your NPV, the probability of losing money and which variables matter most —
a free, open-source alternative to Risk Simulator, @RISK and Crystal Ball that runs on any computer.
🌐 Open the app · ⬇️ Download for desktop · 📖 Documentation · 🎓 Tutorial · 🛠️ Develop
| 33 distributions | Up to 120 M trials/s | 20 tools | 563 tests | Any OS |
|---|---|---|---|---|
| continuous and discrete, truncation | optional GPU (WebGPU) and compiled formulas | analysis · forecasting · optimization · finance | against Excel, R and SciPy | macOS · Linux · Windows · Chromebook |
Your data never leaves your computer. Everything — the formula engine, the simulation, the charts — runs locally in your browser. No server, no account, no tracking. Install it as an app and it works offline.
- Works with your spreadsheet. Open
.xlsxfiles from Excel, LibreOffice, WPS or Google Sheets (or.csv). An Excel-compatible formula engine recalculates your model — including Spanish function names (=SUMA,=VNA,=TIR). The risk model is saved inside the.xlsx, so you can keep editing it in Excel. - Risk Simulator workflow, nothing new to learn. Define assumption, Define forecast, Run, Step, Reset — the same ribbon, the same green/blue cells, the same forecast chart with two-tail / left-tail / right-tail certainty.
- Serious statistics. 33 distributions with truncation, Monte Carlo and Latin Hypercube sampling, rank correlations (Iman–Conover with nearest-correlation repair), reproducible seeds, precision control, and a full statistics panel (percentiles, skewness, kurtosis, confidence interval of the mean…). Thousands of trials per second in a Web Worker.
- Optional GPU acceleration. Your formulas are compiled into exact native code or into a WebGPU compute shader that runs on your graphics card (Metal on Macs, Direct3D 12 / Vulkan on NVIDIA, AMD and Intel). Also multi-core CPU mode. Every fast run is automatically validated against the spreadsheet engine and falls back on its own if anything doesn't match. Benchmarks ↓
- Configurable confidence levels. Choose the certainty band and the confidence of the mean for each forecast, and up to three prediction-interval levels in time series, regression and stochastic processes.
- Find what matters. Tornado and spider charts, sensitivity (rank correlation and contribution to variance), scenario tables, bootstrap and hypothesis tests.
- Fit distributions to data. Kolmogorov–Smirnov, Anderson–Darling and χ² tests with p-values, AIC/BIC ranking, P–P / Q–Q plots — then turn the best fit into an assumption with one click.
- Forecasting. Moving averages, exponential smoothing, Holt, Holt–Winters, ARIMA and auto-ARIMA, trend models, multiple and stepwise regression, and stochastic processes (GBM, mean reversion, jump diffusion).
- Optimization under uncertainty. Continuous, integer, binary and discrete decision variables, constraints, static or stochastic optimization (a simulation per candidate), Nelder–Mead, genetic algorithm, simulated annealing and efficient frontiers.
- Project finance toolkit. A project evaluator that builds the classic cash-flow statement (project and equity cash flows, taxes with loss carry-forward, working capital, salvage value, loans) and computes NPV, IRR, MIRR, payback, discounted payback, profitability index and EAA — and exports it to a sheet with live formulas ready to simulate. Plus NPV/IRR (with multiple-IRR detection), loan schedules, depreciation, WACC/CAPM, break-even and scenarios.
- Reports. Export a report workbook (statistics, percentiles, sensitivity, chart images) or a printable PDF.
- Pleasant to use. English/Spanish interface (auto-detected), light and dark themes, keyboard shortcuts, autosave, and five ready-to-run example models.
Simulation → Settings → Acceleration (the default, Auto, picks the fastest mode your model supports):
| Mode | 100,000 trials* | Results |
|---|---|---|
| Standard (spreadsheet engine) | 4.31 s | reference |
| Multi-core CPU (14 workers) | 1.05 s | identical |
| Compiled formulas | 0.31 s | identical |
| GPU (WebGPU, Metal) | 0.31 s | f32, mean differs by 2·10⁻⁶ · validated |
*Project-evaluation example, Apple M5 Pro, Chrome; wall-clock time including sampling, statistics and sensitivity. 1,000,000 trials take 1.5 s on the GPU and 5,000,000 take 8 s. Raw evaluation throughput: spreadsheet ≈15k trials/s, compiled ≈1.5M/s, GPU 64–120M/s.
Browsers cannot call CUDA or MPS directly; WebGPU is the portable layer on top of them and uses the same GPU. See the acceleration guide.
- Open the app (or install it: browser menu → Install OpenRiskSim).
- File → Examples → Project evaluation.
- Simulation → Run. Read the histogram: the certainty band tells you, for example, "there is a 90% chance that the NPV is between −73,851 and 157,669", and the statistics show P(NPV ≥ 0).
- Try Analytical tools → Tornado to see which assumption drives the result.
Then open your own .xlsx, select an input cell and click Define assumption. The
documentation walks through a full project-evaluation
assignment step by step.
Prefer a regular app? Installers for macOS, Windows and Linux are on the Releases page:
| System | File |
|---|---|
| macOS (Apple Silicon / Intel) | OpenRiskSim-x.y.z-mac-arm64.dmg · OpenRiskSim-x.y.z-mac-x64.dmg |
| Windows (x64 / ARM64) | OpenRiskSim-x.y.z-win-x64-setup.exe · …-win-arm64-setup.exe |
| Linux (x64 / ARM64) | OpenRiskSim-x.y.z-linux-x86_64.AppImage · …-linux-amd64.deb (and ARM64) |
Note
The apps are not code-signed yet. macOS: the first time, right-click the app → Open
(or run xattr -cr /Applications/OpenRiskSim.app). Windows: if SmartScreen appears, click More info → Run anyway.
It is the same app as the web version (same files, same results) and works fully offline. Every push also builds the installers in GitHub Actions (Desktop apps workflow → Artifacts).
| Risk Simulator | OpenRiskSim |
|---|---|
| New Simulation Profile | Simulation → New profile |
| Set Input Assumption | Simulation → Define assumption (or right-click a cell) |
| Set Output Forecast | Simulation → Define forecast |
| Run / Step / Reset Simulation | Simulation → Run / Step / Reset |
| Forecast chart (certainty) | Results panel → Histogram |
| Tornado, Sensitivity, Scenario Analysis | Analytical tools |
| Distributional Fitting | Analytical tools → Distribution fitting |
| Forecasting (time series, ARIMA, regression, stochastic processes) | Forecasting |
| Optimization | Optimization |
The full equivalence table — and what is not implemented yet — is in the docs.
Correctness comes first. The numerical packages are tested against reference values from Excel, R and SciPy:
pnpm test| Package | What it checks |
|---|---|
distributions |
pdf/cdf/quantile of all 33 families vs SciPy (≈1e-11), tails to 1e-12, moments of 200k samples, fitting recovers known parameters |
engine |
Latin Hypercube strata, Iman–Conover hits the target rank correlation while preserving marginals, statistics vs Excel (STDEV.S, SKEW, KURT, PERCENTILE.INC) |
finance |
PMT, NPV, IRR, MIRR, XNPV, XIRR, VDB… vs Excel; cash-flow identities, loan balances |
forecast |
LINEST example, Holt–Winters seasonality, ARIMA parameter recovery, MacKinnon p-values |
optimizer |
Rosenbrock, Himmelblau, 0-1 knapsack, constrained and equality problems, efficient frontier |
workbook |
xlsx round-trip, Spanish CSV, formula engine, example models with zero formula errors |
The example workbooks were also opened in Microsoft Excel for Mac with zero formula errors.
Requirements: Node.js 20+ and pnpm (corepack enable).
pnpm install
pnpm dev # web app at http://localhost:5173
pnpm test # 563 unit tests (Vitest)
pnpm typecheck
pnpm docs:dev # documentation site
pnpm screenshots # regenerate icons, banners and screenshots (needs Google Chrome)packages/
core/ shared types (the contracts between packages)
distributions/ RNG, special functions, 33 distributions, distribution fitting
engine/ sampling (MC / LHS), correlations, statistics, simulation runner, sensitivity
finance/ NPV, IRR, MIRR, payback, project cash flow, loans, depreciation…
forecast/ smoothing, ARIMA, regression, stochastic processes
optimizer/ Nelder–Mead, genetic algorithm, simulated annealing, stochastic optimization
workbook/ xlsx/csv I/O, HyperFormula engine, model evaluator, Web Worker, reports, examples
apps/web/ React 19 + Vite + Tailwind + ECharts user interface
docs/ VitePress documentation (Spanish and English)
scripts/ image generation (icons, banners, screenshots)
Architecture and package contracts: developer docs.
Pushing to main deploys the app and the docs to GitHub Pages.
Contributions are welcome — bug reports with an example .xlsx, new distributions or tools, translations and docs.
Issues and pull requests in Spanish or English. See CONTRIBUTING.md and the
Code of Conduct.
GPL-3.0-or-later © 2026 Fernando Daza and contributors.
Built with React, TypeScript, Vite, HyperFormula (GPLv3), ExcelJS, Apache ECharts, Tailwind CSS and VitePress.
Risk Simulator, @RISK and Crystal Ball are trademarks of their respective owners. OpenRiskSim is an independent project and is not affiliated with them.








