Scenarios and simulation in venture funds

Two ways to put a confidence interval around a fund forecast. Named cases for narrative side-by-side comparisons; Monte Carlo simulation for the per-investment distribution.

A venture fund forecast that returns a single number is overconfident. Outcomes vary because tier mixes vary, multiples vary, hold periods vary, and any one company's exit can dominate the math. The honest framing is a distribution, and there are two complementary ways to communicate one.

This page covers the venture-specific shape of both. For the more general framing across model types (financial models, cap tables, runway), see scenarios.

Named cases (discrete scenarios)

A small number of explicit columns, each varying one or two assumptions, with the same metrics rendered across all columns. Most LP decks use this format because it's narrative and easy to scan.

Three columns is the de-facto convention: Conservative / Base / High, sometimes Pessimistic / Likely / Optimistic. Each column tells a story:

Conservative: what if your large-tier outcomes don't materialize? Lower hit rate or smaller multiple on the upper tier; everything else held constant.
Base: your central forecast. The numbers you'd defend in an LP meeting.
High: what if you catch one of the truly outsized exits? Higher hit rate or higher multiple on the upper tier.

The constraint that makes the columns comparable is varying one or two knobs at a time. If the Conservative column also has different fees, fund size, and recycling, you're not comparing scenarios; you're comparing three different funds.

A common pattern is to vary only the largest return tier (multiple and share of capital) and let the writeoff tier auto-plug to keep the distribution at 100%. That isolates the effect of the long tail, which is what dominates venture returns anyway.

What named cases are good at:

Pitching strategy. Three-column tables tell a clean story.
Talking to LPs about portfolio shape. "Even in our Conservative case we return 1.8x net" is a concrete claim.
Highlighting one or two key sensitivities. The discrete framing forces you to pick what you're testing.

What named cases aren't:

A sensitivity tool. They don't tell you how net returns move with fees, fund size, or check sizes. For those, edit the inputs directly.
A risk model. Three columns doesn't tell you the probability of any of them. The High column might be a 5% outcome or a 50% outcome; the format gives no answer.

Monte Carlo simulation

Rather than picking three columns, simulate many funds. Each iteration:

Draws an outcome per investment from the tier distribution. If your tiers are 60/20/10/10 (writeoff/small/medium/large), each company has those odds.
Draws a multiple per investment from a distribution centered on the tier's stated multiple. Lognormal is the conventional choice because it's right-skewed (matches how venture returns actually behave) and because its mean reconciles to the deterministic average.

After enough iterations (usually 10,000 or more), the average outcome across all simulated funds equals the deterministic forecast, and you have a distribution to look at.

A few parameters that matter:

σ per tier (the spread). The bigger the multiple, the wider the distribution should be. A 32x large-tier exit could realistically be 16x, 50x, or 100x; a 1.5x small-tier exit is much narrower. A common parameterization scales σ with log(1 + multiple), which gives writeoffs near-zero spread and unicorns wide tails without you tuning a knob explicitly.
Hold period. Most simulations don't draw hold per investment because doing so inflates IRR variance without changing the central tendency, and decouples MOIC from IRR. Using the deterministic weighted hold from the return-tier table keeps the two metrics coherent within an iteration.
What you headline. Lognormal distributions are right-skewed, so the median sits below the mean. The mean reconciles to the deterministic Outputs; the median doesn't. Reporting the median implicitly says "half my LPs do worse than this," which is true but misleading because the right tail is meaningful. Mean is the honest headline.

The histogram usually truncates at p99 with a footnote when the right tail is meaningful, so the chart isn't crushed against a few extreme outliers. The percentile table shows the full distribution.

The most useful single number from a simulation is P(LP net MOIC < 1), the share of simulated funds where LPs got back less than they put in. It's the answer to "what's the chance my LPs lose money?" and it captures both fee drag and the chance that the upper tiers don't fire enough to overcome writeoffs.

A 10% loss-of-capital probability is not the same as "10% of LPs lose money in a real fund." It's "10% of the simulated funds don't return capital." For a single LP in a single fund, the realized outcome is one draw from this distribution. The simulation gives you the shape of that draw before you commit.

When to use which

Pitching strategy → Named cases. Tells a clean three-column story.
Talking to a quant LP → Simulation. Show them the loss-of-capital number and the percentile table.
Sanity-checking your assumptions → Both. If your Base named case looks great but the simulation shows 25% loss-of-capital probability, the upper-tier σ is hiding more risk than the headline implies.
Stress-testing fund structure (fees, fund size, recycling, carry) → Edit the inputs directly. Both scenario formats vary return outcomes, not the structural assumptions.

What neither replaces

Neither view replaces a real fund forecast over time. Neither models J-curve, capital call timing, NAV, distributions per quarter. For those, use a fund forecast model with quarterly cash flows.

Both also assume your tier distribution is right. If you've miscalibrated the share of capital that goes into writeoffs vs the upper tier, both views inherit the error.

In Hemrock's tools

The hosted Fund Economics Tool renders both views on a single Scenarios tab, toggled at the top:

Named cases: three columns (Conservative / Base / High by default; column headers are editable). Two knobs per column: large-tier multiple and large-tier share. Everything else is held constant from your live inputs; the writeoff tier auto-plugs the gap. Four metrics per row (Gross MOIC, Net MOIC, Gross IRR, Net IRR), with a toggle to switch Net between Total Fund and LP Only.
Simulation: Monte Carlo over your live inputs, with the σ-scales-with-multiple parameterization described above. Histograms for each headline metric, percentile bands, and the loss-of-capital readout.

Both views run on top of your live inputs and reconcile back to the deterministic Outputs tab. The tool is also published as the @tdavidson/fund-economics-tool npm package, with applyScenario and runMonteCarlo exposed for scripted use.

The spreadsheet Fund Economics Tool ships with three named-case sheets (Forecast, Forecast_1, Forecast_2). The spreadsheet Venture Capital Model also has three independent scenario sheets. Monte Carlo is hosted-only.