Reading return and variation together

What these two indicators add

The Composite Efficiency Index combines five checks of a fund's investment plan, capital and team. Return per Unit of Variance and the Consistency Score are separate model-generated indicators. They bring return and variation into the discussion using recorded fund and portfolio inputs. They do not measure your realized performance.

Return per Unit of Variance

The result is displayed as a multiple. It is a normalized model indicator, not a return multiple or a comparison with an observed median fund. A value above 1.00x should not be read as a promise to outperform peers.

Dispersion means how widely outcomes vary. Comparing return with that spread asks a useful question, but the answer depends on the measure and its assumptions. These product indicators use an existing calibration that remains subject to methodology review. They should support investigation, not stand in for underwriting.

Why a risk-adjusted reading rather than a return reading

The research gives a good reason to examine return and risk together. Korteweg and Sørensen show how the data available to researchers can shape the answer. Because a venture-backed company's value is only observed when it raises again or exits, and both are likelier for companies doing well, the observed data is selected. Once they correct for that selection, estimated alpha falls by about 40 percent, market beta rises by about 20 percent, and estimated idiosyncratic volatility rises from 36 percent to 41 percent per month. In their company-level model, the uncorrected observations paint a more favorable return-risk picture. That finding explains a measurement problem; it does not validate this product's calibration.

Korteweg's later review shows how much the answer varies across studies. Risk-adjusted return estimates for venture vary substantially by method, time period, and data source, with published alphas ranging from negative 6 percent to positive 13.2 percent a year and no consensus on the right approach. His own summary is that average venture funds earned positive risk-adjusted returns before roughly 2000, and that net-of-fee risk-adjusted returns have been at or below zero since. A reading that expresses return against variance is the right shape of question. It is not a shortcut past a question the literature has not settled.

The shape of both readings is older than venture. Markowitz's 1952 paper, which introduced reading return against variance, names the coefficient of variation, dispersion over expected return, as a measure closely related to variance, and observes that an investor who cares about that ratio rather than variance alone is still choosing from the same efficient set. He also makes the point the pair rests on, that the portfolio with the highest expected return is generally not the one with the lowest variance. That is theory about portfolios of securities rather than evidence about venture funds, and it is where the frame comes from rather than a reason to trust it here.

The Consistency Score

This indicator is shown on a 0 to 100 scale. Higher values represent greater consistency under the current model. It uses recorded inputs, including portfolio characteristics; it is not a measured probability or a guarantee that future returns will stay within a particular range.

A higher score is not automatically a better investment proposition. Venture portfolios can depend on a few unusually strong outcomes. Review the strategy and the underlying inputs before treating a narrower modeled spread as an advantage.

Reading them together

Use the pair to frame questions, rather than to classify a fund as good or bad.

If the return-risk indicator is relatively high while consistency is lower, ask how much the investment plan depends on a few large outcomes. If both are higher, check which inputs produce that combination. Neither pattern establishes the quality of company selection or the returns a fund will achieve.

A lower return-risk indicator is a reason to inspect the assumptions and recorded portfolio, whatever the consistency score says. The practical value is in the questions it raises about company count, check size and reserves. The model cannot settle those questions without your context.

What these two readings do not separate

A dispersion measure treats a fund that overshot and a fund that undershot as equally inconsistent, and there is good evidence those are not the same event. Buchner, Mohamed and Schwienbacher decompose dispersion into its upside and downside components across 308 venture funds holding 10,131 portfolio companies. Fund internal rate of return correlates with upside volatility at 0.4707, significant at the 1 percent level, and with downside volatility at 0.0073, which is not significant. Upside dispersion in their sample averages roughly 2.6 times downside dispersion, which is another way of saying venture returns are strongly right-skewed and are not well described by a mean and a spread.

That skew is not a venture quirk. Harvey and Siddique state the general case directly, that unconditional return distributions cannot be adequately characterized by mean and variance alone, and find that systematic skewness commands a risk premium averaging 3.60 percent a year in United States equities. Kraus and Litzenberger established the theoretical version a quarter century earlier: investors are simultaneously averse to variance and attracted to positive skewness, which are two distinct preferences over two distinct properties of a distribution. Korteweg makes the venture-specific version of the point, that private equity payoffs resemble option payoffs and that standard factor models work poorly on option-like payoffs because their risk loadings move over time.

The distinction is useful when reading any summary of variation: unusually strong outcomes and disappointing outcomes can both widen a spread. The cited research concerns observed outcomes; the product indicator is a separate model output. Review the underlying portfolio before drawing a conclusion from either.

What these are not

Neither indicator forecasts realized IRR, TVPI or another return metric. Recorded portfolio activity can affect their inputs, so they need not stay unchanged after the fund begins investing. Their calibration is distinct from the five-check Composite Efficiency Index and should not be described as an empirically verified peer benchmark.

They also cannot establish investment quality. Access, company selection, terms and execution need their own assessment.

The calibration is proprietary to Colibrí Strategies. The purpose of this page is to explain what you are seeing and its limits, so you can decide how much weight to give it.

Sources

• Risk and Return Characteristics of Venture Capital-Backed Entrepreneurial CompaniesArthur Korteweg and Morten Sørensen, Review of Financial Studies 23(10), 2010Shows how much observed venture returns overstate performance and understate risk once selection is corrected for.
• Risk Adjustment in Private Equity ReturnsArthur Korteweg, Annual Review of Financial Economics 11, 2019A survey of how the field measures risk-adjusted private equity returns, and where the estimates disagree.
• Diversification, risk, and returns in venture capitalAxel Buchner, Abdulkadir Mohamed, and Armin Schwienbacher, Journal of Business Venturing 32(5), 2017Separates upside from downside dispersion across 308 venture funds and finds only one of the two relates to returns.
• Conditional Skewness in Asset Pricing TestsCampbell R. Harvey and Akhtar Siddique, Journal of Finance 55(3), 2000Establishes that return distributions are not adequately described by mean and variance, on United States public equities.
• Skewness Preference and the Valuation of Risk AssetsAlan Kraus and Robert H. Litzenberger, Journal of Finance 31(4), 1976The theoretical basis for treating variance aversion and skewness preference as separate.
• Portfolio SelectionHarry Markowitz, Journal of Finance 7(1), 1952The origin of the return-against-variance frame, and of the coefficient of variation as a related measure.