Building a portfolio around the power law.

The situation

A fund with rules, a pipeline and a hurdle

The fund has $220M. For every dollar of initial cheque it keeps 0.6 in reserve for follow-on rounds, so about $137M goes out as initial tickets and $83M is held back. Its limited partnership agreement caps any one company at 12% of the fund, any sub-sector at 30% and any stage at 50%. It must back at least nine companies with tickets of at least $2.5M, and it is judged against a 16% hurdle and a 2.3× target, after 2% management fees, 20% carry and an 8% preferred return.

The pipeline is 18 invented companies across green hydrogen, EV charging, battery recycling and storage, circular plastics, bioenergy, sustainable aviation fuel, carbon removal, electric mobility and green steel, from seed to growth stage, spread across Asia, Europe and North America.

What Capibud did

Model the book the way venture returns behave

1. A stage chain for every company. Each company moves from its current stage towards growth, an exit or failure, with transition odds, time in stage, dilution and step-ups anchored on public venture-funnel data and shifted by sub-sector. Capital-intensive climate hardware graduates less often and more slowly.
2. Correlated scenarios. 2,500 scenarios draw shocks to the carbon price, policy support, the venture funding cycle, lithium, oil and power. A funding winter hits every company in the same scenario.
3. Power-law exits through the preference stack. Exit values mix a heavy tail of home runs with soft landings and trade sales, and every exit or failure is paid through each company's preferred-stock stack rather than pro rata.
4. Diligence as evidence. Scorecards for team, technology, market, unit economics and fit shift each company's odds of exit, weighted by how deep the diligence went. In the example, deeper diligence on one company moved its chance of exit from 37% to 51%.
5. Three portfolios on identical scenarios. The optimiser maximised expected value with a penalty on the tail, inside every limit. It was compared with conviction-weighted tickets and with spray-and-pray: equal tickets across the whole pipeline.
6. Fund outcomes. Cash flows by year were turned into TVPI, DPI and IRR distributions net of fees and carry, with a reserve plan and a J-curve fan.

Results

More expected value, a modest median and a worse tail

The optimiser backs 10 companies with minority tickets between about $5M and $18M. It earns about $39M more expected value than spreading the money evenly, but its chance of losing money is higher (41% against 34%) and its bad year is about $13M worse. The conviction-weighted portfolio has the best chance of clearing the hurdle. There is no free lunch here, and the page does not pretend there is.

The power law, measured

Reserves and the preference stack

Follow-on demand at P80 is about $61M against $83M held, and there is a 9% chance the pro-rata ask exceeds the reserve. The reserve looks about right; a committee could still test a ratio of 0.5 on the same scenarios.

Paying exits and failures through each company's preferred stock, rather than a flat recovery fraction, lowers the chance of losing money from 45% to 41%, moves the bad year from −$124M to −$112M and lifts TVPI at P10 from 0.24× to 0.33×. Earlier-round terms are reconstructed estimates until a company's real terms are entered.

What it missed and limits

What a model of a venture book can and cannot do

• It cannot pick the outlier. The model sizes exposure to outliers; it does not know which company becomes one. Nearly half the proceeds from one position is a feature of the scenarios, not a forecast about a name.
• Priors are public. Stage odds and exit multiples are anchored on public venture data and tuned to published fund-level results. A fund's own track record should replace them.
• The fund is synthetic. Its size, rules, fees, pipeline and every figure on this page are invented for illustration. No real company or Capibud demo sits behind them.
• Terms are reconstructed. Preference stacks for earlier rounds are estimates, and they move the downside materially, as the comparison above shows.
• One run. In a real run, near-equal companies can swap places between seeds; the committee sees that spread.

What this means for you

Make the trade-off explicit at the committee