ModelRisk: Risk modeling software for Excel

ModelRisk turns any Excel model into a Monte Carlo simulation. You describe uncertain inputs with its 136 distributions, fit them to data, link them with copulas and time series, run thousands of scenarios, and read the results in a Results Viewer that ranks which inputs drive the outcome. The steps below follow the ModelRisk ribbon in Excel from left to right.

Which probability distributions does ModelRisk include?

ModelRisk includes 136 probability distributions, from everyday choices such as PERT, Normal and Lognormal to specialist families for extreme values, insurance losses and reliability. Each is an Excel function with an interactive guide to its theory and use, so you can see its shape before you commit to it.

• Grouped by what they model, so you search by problem rather than by name.
• Expert-estimate distributions (PERT, Modified PERT, Triangle) for when there is no data.
• The Shaper tool lets you draw a distribution of your own.
• Distribution Objects give exact percentiles and probabilities without simulating.

How does ModelRisk fit distributions to data?

ModelRisk fits distributions, time series and copulas to spreadsheet data by maximum likelihood estimation (MLE). It ranks the candidates on three information criteria (SIC, HQIC and AIC), which penalise unnecessary parameters, and overlays every fit on your data so you can judge the tails by eye as well as by score.

• One function does the whole job: VoseOptimalFit picks the best-ranked distribution.
• Bayesian model averaging blends close candidates instead of forcing a single winner.
• Switch on parameter uncertainty and a small dataset shows up as wider results.

Can ModelRisk forecast how a value changes over time?

Yes. ModelRisk has 34 time series models, from geometric Brownian motion and its seasonal variant to ARMA, ARCH and GARCH families and Yule growth processes, with multivariate versions that keep related series consistent. Fit them to history, or describe a trend from expert estimates when there is no data.

• Not only for finance: demand, failure counts, adoption curves and commodity prices.
• Expert-input series take three-point, range, percentile or saturation estimates for each period.
• Fit to history by MLE, or average several models with Bayesian model averaging.

How does ModelRisk model correlation between risks?

ModelRisk links uncertain inputs with 14 copulas. A copula controls not only how strongly two variables move together but where: some make bad outcomes cluster, others do not. Fit a copula to your data, or use the data copula, which reproduces whatever pattern the data shows. Every link stays visible in the spreadsheet.

• Normal, Student-t, Clayton, Gumbel and Frank, in bivariate and multivariate form.
• The data copula captures patterns no formula describes. NASA uses it for exactly that.
• Right-click a set of cells to correlate them, plus joint-success and target-point tools on the correlation plot.

How does ModelRisk keep large models small?

Distribution Objects and aggregate functions do in one cell what would otherwise take a row per event. A frequency–severity model, for example, becomes one formula that adds up a random number of random losses, by Monte Carlo, Panjer, FFT or de Pril methods. Fewer cells mean faster simulation and fewer places for errors to hide.

• Deductibles and insurance layers applied to a whole distribution.
• Splice a body distribution onto a heavy tail.
• Combine several experts' estimates into one distribution.

How fast is ModelRisk?

Turbo is ModelRisk's optional high-speed simulation engine. Switch Turbo on for a workbook and ModelRisk analyses the model once, compiles only the cells that change between iterations, and runs them on every processor core without Excel recalculating. Across 89 internal test models, the simulation phase ran a median of about 200× faster than ModelRisk's classic engine.

• Large workbooks take a few seconds to prepare the first time, then run at full speed.
• Results are statistically equivalent to the classic engine, with a consistency check built into the first Turbo run.

• Go To Sample jumps to the exact scenario behind any outlier.
• What-if and stress testing: Goal Seek on simulated results, Impact Analysis one input at a time, and Extreme Scenario Analysis.
• A histogram sparkline inside every input and output cell.
• Save results as a .vmrs file and reopen them without the model, or export to SIPMath 3.0.

Can ModelRisk optimize decisions under uncertainty?

Yes. The ModelRisk Optimizer searches for the decision with the best simulated result: the highest mean profit, the highest 5th percentile, or the lowest chance of a loss. Each candidate is scored by running a simulation, so it compares distributions rather than single numbers. Constraints can apply to a statistic or to any worksheet formula.

• Stochastic models, where every trial is a simulation, and deterministic ones, where every trial is one recalculation.
• Adjustable cells can be continuous, integer, binary or a list of allowed values, with bounds that are themselves worksheet cells.
• Groups for budgets, recipes, orderings and schedules; hard or penalised constraints.
• The same answer on any machine: one core, six cores or automatic gave an identical objective, trial count and search trace.
• Coming from RISKOptimizer or Crystal Ball? The converter brings the optimization model across: objectives, adjustable cells, constraints and groups.

Select any ModelRisk cell and click View Function to see exactly what it does. An invalid input returns a plain explanation instead of a bare error, the help file covers more than 1,000 topics, and Excel's own auditing tools work on every ModelRisk function, alongside Excel Tables, Power Query and modern Excel functions.

Can AI assistants build ModelRisk models?

Yes. The open-source ModelRisk MCP server lets AI assistants such as Claude Desktop, Claude Code, Cursor and Zed read, build and run ModelRisk models in your workbook through conversation. Every change is a dry-run preview until you approve it, writes are logged, and Excel's undo still works. Tamara and ModelChoice have servers of their own.

• More than 50 tools for reading, building, simulating, analysing and reporting.
• Never overwrites your own formulas; bulk writes need explicit confirmation.
• MIT licensed. Windows, Excel 2019 or later.