Chapter 11: Performance Evaluation and Incentive Design for Policy-Guided Green Finance

Chapter 10 closed on a conditional claim: government-guided venture capital works when the fund has a mandate it can actually pursue, a duration long enough to reach the outcome, and a signal credible enough that private co-investors will follow. That claim pushes the problem one step downstream. If the mandate is the condition, someone has to write it, measure whether it was met, and pay the people who run the fund according to whether it was met. This chapter is about that machinery. It is the least glamorous part of green finance and probably the most decisive.

The thesis, stated early so the rest of the chapter can qualify it: performance evaluation and incentive design for policy-guided green finance fail, when they fail, because they collapse two objectives into one metric. The policy objective (emissions avoided, transition speed, additionality) and the commercial objective (risk-adjusted return, capital preservation, liquidity) are both real, both binding, and measured on different clocks. A single scoreboard that tries to serve both will be gamed against whichever side is easier to manipulate. The design problem is not to find the one true metric. It is to build a contract that pays for the policy objective without destroying the commercial discipline that makes the policy objective achievable at scale.

Three sections follow. The first lays out the objectives and the structural tensions between them. The second builds the evaluation framework, separating environmental from financial performance and insisting that the two be reported side by side rather than blended. The third turns to contract design: how to pay fund managers and enterprises so that incentives are compatible with both objectives, drawing on the principal-agent literature that runs from Holmström and Milgrom (1991) through Lerner's studies of public venture capital.

11.1 Objectives and Challenges: Two Clocks, One Balance Sheet

11.1.1 What policy-guided green finance is actually for

A policy-guided green fund exists because the private market, left alone, underprovides a specific thing. That thing is not "green investment" in general. Private capital already flows to solar farms with contracted offtake and to efficiency retrofits with three-year paybacks. The gap is in the segments where the risk is real but the information to price it is missing, or where the payoff is real but arrives on a horizon longer than a standard fund life, or where the social return exceeds the private return by a margin the sponsor cannot capture.

Consider a concrete case. A provincial government in eastern China establishes a RMB 5 billion green industry guidance fund to support hydrogen electrolyzer manufacturing. The fund is capitalized 30 percent by the provincial finance bureau and 70 percent by commercial limited partners (LPs), mostly insurers and wealth-management subsidiaries. The policy intent is to build a domestic electrolyzer supply chain within seven years. The commercial LPs signed up because the government tranche is subordinated and because they expect a 7 percent net IRR. Now ask: what does success look like at year seven? The finance bureau wants installed electrolyzer capacity and a domestic supply chain. The LPs want their 7 percent. These are not the same target and they do not resolve on the same date. The electrolyzer plants may be built and running while the equity is still under water, or the fund may return capital on schedule by exiting early to a strategic buyer who then relocates the capacity. Both outcomes are "success" on one scoreboard and "failure" on the other.

This is the central tension. It is not a tension between good and bad incentives. It is a tension between two legitimate principals with different utility functions and different time horizons, sharing one balance sheet.

11.1.2 The three frictions

Three frictions stand out in the design of policy-guided green finance. I will name them and then take them one at a time.

The first is metric substitution: the agent optimizes the measured proxy rather than the unmeasured objective. The second is horizon mismatch: the policy objective matures on a decade scale, the commercial evaluation on a three-to-five-year fund cycle. The third is additionality opacity: it is hard to establish what would have happened without the public money, so the fund cannot easily prove it caused the transition rather than financed it.

Metric substitution. Every evaluation system creates a proxy, and every proxy is a target once it enters a contract. Goodhart's law is not a curiosity here; it is the operating constraint. If the fund manager is paid on "green assets under management," the manager will classify assets as green. The EU Taxonomy Regulation (Regulation (EU) 2020/852) was written precisely because the classification problem is severe enough to need a legal definition, and even with the taxonomy the boundary cases multiply. A fund that counts a gas-fired peaker plant as "transition-aligned" because it displaces coal is not lying under a loose definition; it is responding to the metric. The design lesson is not to write a tighter definition forever. It is to accept that any single metric will be gamed and to build redundancy into the scoreboard so that gaming one dimension is visible in another.

Horizon mismatch. A green venture fund has a ten-year life, often extendable by two. The environmental outcome it is supposed to produce—a commercial-scale supply chain, a cost curve that crosses parity, a forestry rotation that reaches harvest—matures on a fifteen-to-thirty-year clock. The fund's LPs evaluate it at year five and year seven. The government evaluates it at the end of the political cycle. Neither evaluation date matches the outcome date. This is not a problem that better metrics solve. It is a problem that contract design has to absorb, either by lengthening the evaluation window (hard, because LPs have their own liabilities) or by decoupling the policy payment from the commercial exit (feasible, and I will return to it).

Additionality opacity. The question "would this have happened anyway?" is counterfactual and therefore unanswerable with certainty. A fund can show that it invested in a first-of-kind electrolyzer plant. It cannot show that no one else would have. The best it can do is argue from market structure: no commercial fund was willing to take the first-loss position, the technology readiness level was below the threshold at which private capital participates, the deal size was below the minimum ticket of the large funds. These are suggestive, not dispositive. Lerner's (2002) critique of public venture capital programs turns on exactly this point: many programs finance firms that would have been financed anyway, and the measured "impact" is selection, not causation.

11.1.3 Why balancing is harder than it sounds

The naive response to these frictions is to say: set a blended objective, weight the environmental and financial components, and optimize. This is what most guidance funds do in practice, and it is where they get into trouble. Blending assumes the two objectives are commensurable—that a unit of emissions avoided can be traded against a unit of IRR. They are not commensurable in any principled way, and the exchange rate is a political choice dressed as a technical one.

There is a second, subtler problem. The two objectives are measured with different reliability. Financial returns are observable, auditable, and benchmarked against a market. Environmental outcomes are estimated, modeled, and frequently self-reported. When you blend a reliable metric with an unreliable one, the optimization gravitates to the reliable one. The manager who is paid on a blend of IRR and emissions will manage the IRR, because that is the number the LP can check. The emissions number will be whatever the model says it is. This is not cynicism; it is the predictable response to an asymmetric information structure.

What follows from this is a design principle I will use throughout the chapter: do not blend, separate. Report environmental and financial performance on parallel tracks, evaluate them against separate benchmarks, and pay against them through separate instruments. The financial track disciplines the fund. The environmental track disciplines the policy. The two are reconciled at the level of the fund's mandate and the sponsor's overall portfolio, not inside a single score.

11.1.4 The commercial sustainability constraint

A policy-guided fund that loses money is not a success, however green its portfolio. This sounds obvious and is routinely violated. The reason is that public sponsors can absorb losses that private sponsors cannot, and the temptation to absorb them—to keep a failing green project alive because shutting it down would be a political embarrassment—is strong. When that happens, the fund stops being a market-building instrument and becomes a subsidy channel with extra steps.

The commercial sustainability constraint has a specific meaning in this context. It does not mean the fund must earn a market rate of return on every deal. It means the fund must earn enough, across its portfolio, to (a) return the private LPs' capital with a risk-appropriate return, and (b) preserve the sponsor's ability to recapitalize for the next vintage. A fund that fails (a) cannot raise private money again. A fund that fails (b) cannot raise public money again. Both failures end the program.

There is a boundary condition here that I want to state carefully. The commercial constraint binds hardest in funds structured as limited partnerships with private co-investors. It binds more loosely in funds that are wholly government-capitalized and explicitly structured as grant-like instruments. The design of the evaluation system must match the capital structure. A wholly public fund can accept a lower financial return if the environmental return is high and the additionality is clear; a blended fund cannot, because the private LPs will exit. Confusing the two structures is one of the more common errors in program design.

11.2 Performance Metrics and the Evaluation Framework

11.2.1 The structure of the framework

The evaluation framework I propose has three layers, and it is worth being explicit about why three rather than one.

The first layer is fund-level financial performance: the standard metrics that any LP would apply—net IRR, TVPI (total value to paid-in capital), DPI (distributions to paid-in capital), and the J-curve. These are necessary and well understood. They discipline the manager.

The second layer is portfolio-level environmental performance: the physical outcomes the fund was created to produce—installed capacity, emissions avoided, hectares under improved management, cost reductions achieved. These are the policy metrics. They are harder to measure and easier to game.

The third layer is additionality and system-level effects: what the fund changed in the market beyond its own portfolio—the cost of capital for the sector, the entry of follow-on private capital, the diffusion of the technology, the policy learning. This is the layer that justifies the public money, and it is the layer that almost no fund measures well.

The three layers are reported separately. They are not summed. The reconciliation happens at the level of the sponsor's overall program evaluation, where the sponsor decides how much system-level effect it is willing to buy with how much financial concession. That decision is political and should be made explicitly, not buried in a weighting scheme.

flowchart TD
    A[Policy Mandate] --> B[Fund-Level Financial Metrics]
    A --> C[Portfolio Environmental Metrics]
    A --> D[Additionality and System Effects]
    B --> E[LP Reporting and Manager Compensation]
    C --> F[Policy Reporting and Sponsor Compensation]
    D --> G[Program Evaluation and Recapitalization Decision]
    E --> H[Separate Scorecards, Reconciled at Sponsor Level]
    F --> H
    G --> H

11.2.2 Financial metrics: what to measure and what to ignore

The fund-level financial metrics are the least controversial part of the framework, so I will be brief and focus on the choices that matter for green funds specifically.

Net IRR is the standard, and it is the right headline number for a fund with private LPs. It has known defects—it assumes reinvestment at the IRR, it is sensitive to the timing of cash flows, it can be inflated by early distributions from a single successful exit—but the defects are well understood by the LPs who use it, and substituting a "better" metric creates comparability problems. Use net IRR, and use it alongside TVPI and DPI so that the LPs can see whether the IRR is being carried by paper marks or by realized cash.

For green funds, two adjustments matter. The first is the treatment of concessional capital. If the government tranche is subordinated or takes a below-market return, the fund's blended IRR overstates the private LPs' experience and understates the public subsidy. Report the IRR separately for the government tranche and the private tranches. The gap between them is the price of the policy objective, and it should be visible.

The second is the treatment of environmental liabilities. A fund that invests in a battery recycler inherits the obligation to manage the waste stream. If that obligation is off-balance-sheet or under-provisioned, the reported IRR is overstated. The standard remedy is to require the fund to mark environmental liabilities to a conservative estimate and to disclose the provision. This is not standard practice in venture funds, and it should be. The credit-rating logic from Chapter 3 applies here: environmental performance is a credit factor, and a fund that ignores it is mispricing its own portfolio.

A word on what to ignore. Do not use "green AUM" as a financial metric. It is a classification count, not a performance measure, and it invites the manager to reclassify rather than to perform. Do not use "number of green projects" for the same reason. Do not use a single "impact score" that aggregates environmental and financial dimensions; that is the blending error I described in section 11.1.3.

11.2.3 Environmental metrics: from physical outcomes to verified claims

The environmental metrics are where the design work is. I will organize them by the type of outcome they measure, because different outcome types require different verification regimes.

Physical output metrics are the most reliable. Installed megawatts, tonnes of CO2 avoided per year, hectares under improved forest management, cubic meters of water saved. These are measurable by inspection or by meter. The verification regime is engineering, not modeling. A fund that reports 200 MW of installed electrolyzer capacity can be audited by a site visit. The weakness is that physical output is not the same as environmental outcome: a 200 MW electrolyzer running on grid electricity in a coal-heavy province avoids less CO2 than the same unit running on curtailed renewables. So physical output metrics need to be paired with an emissions factor.

Emissions-avoided metrics are the policy headline and the hardest to verify. The standard approach is a counterfactual baseline: emissions with the project minus emissions without it. The baseline is modeled, and the model is where the gaming happens. A fund can inflate avoided emissions by choosing a high baseline (compare the project to the oldest, dirtiest alternative) or a low additionality threshold (count projects that would have been built anyway). The remedy is to fix the baseline methodology in advance, publish it, and have it verified by an independent assessor. The EU Taxonomy's technical screening criteria and the TCFD's scenario disclosure framework are the two most developed attempts to standardize this, and both are still works in progress.

Cost-curve metrics are the ones that matter for innovation funds. The policy objective for a green venture fund is usually not emissions avoided in the near term; it is the cost reduction that makes the technology self-sustaining. A fund investing in solid-state batteries should be measured on the cost per kilowatt-hour of the cells it helped develop, benchmarked against the incumbent technology's cost curve. This metric is forward-looking and therefore speculative, but it is the right metric for the mandate. It also has the useful property that it is hard to fake: either the cost curve moved or it did not, and the market will tell you.

Transition metrics apply to funds that invest in brown-to-green conversion—a steel mill installing electric-arc capacity, a cement plant switching to alternative fuels. Here the relevant metric is the emissions intensity per unit of output, tracked over time against a sectoral pathway. The advantage of intensity metrics is that they cannot be gamed by reducing output. The disadvantage is that they require a sectoral pathway, which requires a policy decision about what "aligned" means for that sector. The EU Taxonomy's delegated acts for steel and cement are the reference points; they are contested, and the contest is the point.

11.2.4 A worked example: the electrolyzer fund

Let me make this concrete with the hydrogen fund from section 11.1.1. Suppose the fund has committed RMB 5 billion and has made twelve investments over five years. The evaluation at year five might look like Table 11-1.

Table 11-1: Illustrative Year-Five Evaluation, Electrolyzer Guidance Fund

The table tells a story that no single number could. The fund is behind on every dimension, but the pattern matters. The financial shortfall (4.2 percent versus 6.0 percent) is concentrated in the private tranche, which means the government tranche is absorbing losses as designed. The environmental shortfall (620 MW versus 800 MW) is smaller in proportional terms than the financial shortfall, which suggests the fund is building capacity but not capturing value—a common pattern in early-stage hardware. The cost-curve metric (−22 percent versus −35 percent) is the most worrying, because it is the leading indicator of whether the technology will become self-sustaining. The additionality metrics are weak, but five years is early for follow-on capital in a hardware sector.

The evaluation does not produce a verdict. It produces a set of signals that the sponsor can read. The sponsor might decide to extend the fund's life, to inject additional capital into the cost-reduction work, or to wind down the private tranche and convert the fund to a wholly public instrument. Each of those decisions is a judgment, and the framework's job is to make the judgment informed rather than to make it automatic.

11.2.5 Additionality: measuring what would have happened anyway

Additionality deserves its own treatment because it is the metric that justifies the public money and the one most often fudged.

There are three usable approaches, and they should be used together.

The first is market-gap evidence. Document, at the time of investment, that no commercial fund was willing to lead the deal at the relevant stage. This is a factual claim that can be verified from the deal record: who was approached, who declined, at what terms. It is not proof of additionality, but it is evidence that the fund was not crowding out private capital. Lerner's (2002) work on public venture capital emphasizes exactly this: the test is whether the public fund is financing at the frontier where private capital will not go, or whether it is competing with private capital for the same deals.

The second is follow-on tracking. Measure the private capital that enters the portfolio companies after the fund's investment. If the fund's companies attract private co-investors at later stages, the fund has served its market-building function. If they do not, either the companies are failing or the fund is not building the market. The metric is noisy—follow-on capital depends on the macro cycle—but over a full fund life it is informative.

The third is counterfactual benchmarking. Compare the fund's portfolio companies to a matched set of companies that did not receive the fund's investment, on the outcomes the fund cares about (cost reduction, capacity growth, survival). This is the most rigorous approach and the hardest to execute, because the matched set is hard to construct and the selection problem is severe. The best one can do is a difference-in-differences design with a credible comparison group, and even then the result is suggestive. I have seen this done well in evaluations of the Israeli Yozma program and the U.S. SBIR program, and the lesson from both is that the counterfactual has to be built before the program starts, not reconstructed afterward.

A note on what additionality measurement cannot do. It cannot establish causation with the confidence of a randomized trial, and it should not pretend to. The honest framing is: "we can show that the fund invested where private capital was not investing, and that private capital followed; we cannot show that the follow-on would not have happened without us." That framing is defensible to a legislature and honest to a researcher. The dishonest framing—claiming a precise "jobs created" or "emissions avoided" multiplier—is what gets programs into trouble when the auditor arrives.

11.2.6 Institutionalizing the evaluation: who measures, and when

The evaluation framework is only as good as the institution that runs it. Three design choices matter.

Independence. The evaluator should not report to the fund manager. In a government-guided fund, the evaluator should report to the sponsor's oversight body, not to the fund's general partner. This is the same logic as an independent credit rating: the entity being evaluated should not control the evaluator. The cost is some friction in the reporting relationship; the benefit is that the environmental metrics are not negotiated with the manager who is paid on them.

Timing. Evaluation should happen at three points: at investment (baseline documentation), annually (tracking), and at exit or fund close (final assessment). The baseline documentation is the most important and the most neglected. A fund that does not document the counterfactual at investment cannot measure additionality at exit. The baseline should include the deal-sourcing record, the emissions baseline, the cost benchmark, and the market-gap evidence.

Publication. The evaluation should be published, in a form that protects confidential business information but discloses the methodology and the results. Publication is the cheapest discipline available. A fund that knows its environmental metrics will be public is a fund that thinks harder about how it reports them. The EU's SFDR (Regulation (EU) 2019/2088) and the TCFD framework are the reference points for disclosure design; both have weaknesses, but both establish the principle that green finance claims are public claims.

11.3 Incentive-Compatible Contract Design

11.3.1 The principal-agent problem in policy-guided green finance

The contract design problem has a standard structure. The principal (the government sponsor, and behind it the taxpayer) wants the agent (the fund manager) to pursue the policy objective. The agent has private information about effort and deal quality, and the principal cannot observe effort directly. The agent's payoff is a function of the contract, and the agent will allocate effort across the tasks the contract rewards.

Holmström and Milgrom (1991) is the canonical reference, and its central result is the one that matters here: when an agent has multiple tasks and the principal can measure only some of them, the agent will allocate effort toward the measured tasks and away from the unmeasured ones. If the contract rewards financial return and does not reward environmental outcome, the agent will pursue financial return. If the contract rewards environmental outcome and does not reward financial return, the agent will pursue environmental outcome, including by financing projects that produce the outcome at a loss. The multi-task problem is not solved by adding more metrics; it is managed by choosing which tasks to reward and accepting that the unrewarded tasks will be underprovided.

The green fund context adds two complications. First, the environmental outcome is measured with error and with a lag, so the contract has to pay on a noisy, delayed signal. Second, the policy objective is not the agent's only task: the agent also has to preserve capital, manage the fund's operations, and maintain relationships with LPs. The contract has to reward the policy objective without letting the agent neglect the operational tasks that make the fund viable.

11.3.2 The compensation structure: carry, hurdle, and clawback

The standard venture fund compensation structure is a management fee (typically 2 percent of committed capital) plus carried interest (typically 20 percent of profits above a hurdle rate, usually 8 percent). The structure exists to align the manager with the LPs: the manager earns carry only if the LPs earn their hurdle. For a policy-guided green fund, the structure needs modification on three dimensions.

The hurdle rate. A green fund with a long horizon and early-stage technology risk will have a lower expected return than a conventional venture fund. If the hurdle is set at the conventional 8 percent, the manager may never earn carry, and the fund will struggle to attract talent. If the hurdle is set too low, the manager earns carry on mediocre performance. The right approach is to set the hurdle at the fund's cost of capital, which for a blended fund is a weighted average of the government tranche's opportunity cost and the private LPs' required return. In the electrolyzer example, if the government tranche's opportunity cost is 3 percent and the private tranche requires 8 percent, and the capital is split 30/70, the blended hurdle is roughly \(0.3 \times 3\% + 0.7 \times 8\% = 6.5\%\). Setting the hurdle at 6.5 percent rather than 8 percent is defensible and transparent. Setting it at 4 percent is a subsidy to the manager.

The carry base. Conventional carry is calculated on the fund's total profits. For a green fund, the carry base should be adjusted for the concessional capital. If the government tranche is subordinated, the manager should not earn carry on the returns that accrue to the government tranche's subordination. The cleanest structure is a tiered carry: a lower carry rate on returns up to the hurdle, a standard rate above it, and a separate "impact carry" that vests only if the environmental metrics are met. The impact carry is the mechanism that pays for the policy objective without blending it into the financial return.

The clawback. Conventional fund agreements include a clawback provision that requires the manager to return carry if the fund's later losses reverse earlier gains. For a green fund, the clawback should extend to the environmental metrics: if the fund's verified emissions reductions fall below a threshold at fund close, a portion of the impact carry is clawed back. The clawback is the mechanism that handles the horizon mismatch. The manager cannot know at year three whether the environmental outcome will materialize at year ten, so the contract pays the impact carry into escrow and releases it only on verification. This is the same logic as a credit default swap's collateral posting: the payment is contingent on a verified state of the world.

11.3.3 A formal sketch

Let me put the compensation design in a simple formal frame, because the algebra clarifies the trade-offs.

Let \(e\) be the manager's effort allocated to the financial task and \(g\) be effort allocated to the environmental task, with total effort normalized to \(e + g \leq 1\). The fund's financial return is \(R = r(e) + \varepsilon\), where \(r\) is increasing and concave in \(e\) and \(\varepsilon\) is noise. The environmental outcome is \(E = q(g) + \eta\), where \(q\) is increasing and concave in \(g\) and \(\eta\) is noise with a longer lag. The manager's compensation is

where \(\alpha\) is the fixed fee, \(\beta_R\) and \(\beta_E\) are the incentive intensities on the financial and environmental metrics, \(\lambda\) is the manager's risk aversion, and \(\sigma_R^2\) and \(\sigma_E^2\) are the variances of the two signals. The manager chooses \(e\) and \(g\) to maximize \(C\) subject to the effort constraint.

The first-order conditions give the standard result: the incentive intensity on each task is decreasing in the noise of its signal and in the manager's risk aversion. Because \(\sigma_E^2 > \sigma_R^2\) (the environmental signal is noisier and lagged), the optimal \(\beta_E\) is lower than the optimal \(\beta_R\) for a risk-averse manager. This is the formal version of the intuition from section 11.1.3: the noisy metric gets less weight, not because it matters less but because paying on it is expensive in risk premium.

The multi-task problem appears when the effort constraint binds. If the manager's total effort is fixed, the cross-partial effect of \(\beta_E\) on \(e\) is negative: raising the environmental incentive intensity draws effort away from the financial task. The principal's problem is to choose \(\beta_R\) and \(\beta_E\) to maximize the expected policy objective net of the cost of the compensation, subject to the manager's participation constraint. The solution depends on the curvature of \(r\) and \(q\) and on the principal's relative valuation of the two objectives.

Two implications follow. First, the impact carry should be structured as a bonus on a threshold, not as a continuous share of the environmental outcome. A threshold bonus avoids paying the manager for environmental outcomes that would have happened anyway and concentrates the incentive on the margin where the policy objective is uncertain. Second, the financial carry and the impact carry should be separately escrowed, so that the manager cannot use a strong financial year to offset a weak environmental year. The separation is the contract-level implementation of the "do not blend" principle.

11.3.4 Incentives for enterprises: the other side of the contract

The fund manager is one agent. The portfolio companies are another, and their incentives matter as much. A green fund that pays its manager for environmental outcomes but does not transmit those incentives to the companies it finances will find that the outcomes do not materialize.

There are three transmission mechanisms.

Milestone-based disbursement. The fund commits capital in tranches, with each tranche released on the achievement of a technical or environmental milestone. A battery company might receive its second tranche on demonstrating a cycle life of 2,000 cycles, its third on reaching a cost per kilowatt-hour below a threshold. This is standard venture practice for technical milestones and is easily extended to environmental ones. The advantage is that it aligns the company's incentives with the fund's evaluation metrics without requiring the company to accept an environmental share of its equity.

Environmental ratchets in the investment agreement. The fund's equity stake or liquidation preference adjusts based on the company's environmental performance. If the company meets its emissions-intensity target, the founders' dilution is lower; if it misses, the fund's stake increases. This is a contingent contract, and it is the enterprise-level analogue of the impact carry. It is used in some climate-tech deals and is underused in policy-guided funds.

Technical assistance and shared services. Many green portfolio companies fail not because of incentive problems but because they lack the capability to measure and manage their environmental performance. A fund that provides shared measurement infrastructure—a common MRV (monitoring, reporting, verification) platform, a shared life-cycle assessment team—reduces the cost of the environmental metric and improves its quality. This is not an incentive mechanism in the strict sense, but it changes the production function for the environmental outcome, and it is often the highest-return investment a policy fund can make.

11.3.5 The ratchet problem and the case for pre-commitment

A recurring failure mode in policy-guided funds is the ratchet: the sponsor sets a target, the fund misses it, and the sponsor lowers the target rather than accepting the miss. The ratchet is rational for the sponsor in the short run (it avoids a political embarrassment) and corrosive in the long run (it teaches the manager that targets are negotiable).

The remedy is pre-commitment. The target and the consequence of missing it should be written into the fund's constitutional documents before the first investment, and the consequence should be automatic rather than discretionary. If the fund misses its environmental target at year five, the impact carry is clawed back by a formula, not by a renegotiation. If the fund misses its financial target, the sponsor's recapitalization commitment is reduced by a formula. The formulas can be imperfect; the pre-commitment is what matters. A formula that is written down and applied is worth more than a perfect formula that is renegotiated every year.

There is a boundary condition. Pre-commitment is credible only if the sponsor can resist the ex-post temptation to renegotiate. In a political system where the sponsor changes every few years, the pre-commitment may not survive a change of government. The design response is to embed the pre-commitment in a legal instrument that is harder to change than an administrative decision—a fund statute, a multi-year budget line, a contract with the private LPs that the government cannot unilaterally amend. The EU's Sustainable Finance Action Plan and the associated regulatory technical standards are an example of embedding commitments in law; the trade-off is rigidity, and the rigidity is the point.

11.3.6 The government tranche as a loss-absorbing buffer

The most powerful incentive instrument in a blended green fund is not the carry; it is the subordination of the government tranche. If the government takes the first loss, the private LPs' risk is reduced, and the fund can attract private capital at a lower expected return. The subordination is a subsidy, and it should be priced and disclosed as such.

The design question is how much subordination. Too little and the private capital does not come. Too much and the fund becomes a grant program with a private-equity wrapper, and the private LPs have no skin in the game. The right amount is the amount that makes the private LPs indifferent between the green fund and a conventional fund of the same risk, which is an empirical question that can be answered by the terms on which the private capital actually subscribes. In the electrolyzer example, if the private LPs subscribe at a 7 percent expected return with 30 percent government subordination, the subordination is calibrated correctly. If they subscribe at 9 percent, the subordination is too small. If they subscribe at 5 percent, it is too large.

The subordination also has an incentive effect on the manager. A manager whose government tranche absorbs the first loss has less incentive to avoid risk, because the downside is borne by the government. The remedy is to make the manager's carry sensitive to the government tranche's losses: the impact carry is forfeited if the government tranche's loss exceeds a threshold. This is a standard feature of the "public-private partnership" fund structures used in the UK's Green Investment Bank and in several EU structural funds, and it is the right template.

11.4 Putting the Pieces Together

11.4.1 The evaluation and incentive architecture

Let me assemble the chapter's argument into a single architecture, because the pieces are easier to see together.

The architecture has four components:

1. Separate scorecards for financial performance, environmental performance, and additionality, reported on parallel tracks and reconciled only at the sponsor level.
2. A tiered compensation structure with a conventional carry on financial returns, an escrowed impact carry on verified environmental outcomes, and a clawback that extends to both.
3. A pre-committed evaluation schedule with baseline documentation at investment, annual tracking, and final assessment at exit, run by an evaluator independent of the manager.
4. A calibrated government tranche that absorbs first loss up to the amount needed to attract private capital, with the manager's impact carry exposed to the government tranche's losses.

flowchart LR
    A[Sponsor: Policy Objective] --> B[Fund Mandate]
    B --> C[Manager: Financial and Environmental Tasks]
    C --> D[Portfolio Companies]
    D --> E[Verified Outcomes]
    E --> F[Separate Scorecards]
    F --> G[Financial Carry]
    F --> H[Impact Carry in Escrow]
    F --> I[Clawback]
    G --> C
    H --> C
    I --> C
    A --> J[Government Tranche: First Loss]
    J --> C

11.4.2 What the architecture does not solve

I want to be candid about the limits. The architecture does not solve the additionality problem; it only makes the evidence better. It does not solve the political ratchet; it only makes the ratchet more costly. It does not solve the horizon mismatch; it only decouples the policy payment from the commercial exit so that the mismatch does not corrupt the manager's incentives. And it does not solve the fundamental problem that the environmental outcome is measured with error, which means that any contract paying on it will pay for some outcomes that did not happen and miss some that did.

What it does solve is the blending error. By separating the scorecards and the payment instruments, it prevents the manager from trading a reliable metric against an unreliable one. That is a modest achievement, and it is the one that matters most, because the blending error is the failure mode that turns a green fund into a conventional fund with a green logo.

11.4.3 The connection to the three-pillar framework

This chapter sits inside the book's three-pillar framework—rating-based identification, insurance-based risk buffering, and sci-tech finance catalysis—and it is worth being explicit about how.

The evaluation framework is the rating logic applied to a fund. Chapter 3 argued that credit ratings can incorporate environmental performance if the environmental factor is material to default risk. The same logic applies here: a fund's environmental performance can be evaluated if the environmental factor is material to the policy objective, and the evaluation should use the same discipline—defined metrics, independent verification, disclosure. The rating agencies' experience with environmental factors is instructive: the factors that are material are the ones that show up in cash flows, and the factors that do not show up in cash flows are the ones that get ignored. The evaluation framework shouldlearn from this and design the environmental metrics to sit as close to cash-flow-relevant as the policy objective allows. A cost-curve metric is closer to cash-flow-relevant than an emissions-avoided metric; an emissions-intensity metric is closer than an absolute emissions metric. The closer the metric is to the fund's economics, the less the manager can game it.

The insurance logic enters through the risk-buffering function. Chapter 6 argued that weather index insurance and forest insurance can absorb the downside risk that makes green investment unattractive to risk-averse operators. The same function can be performed inside a fund's contract. A revenue-contingent impact carry—one that pays less when the fund's portfolio is under financial stress—is a form of insurance for the manager, and it reduces the risk premium the manager demands for taking on the environmental task. The design is not identical to a weather derivative, but the logic is: transfer the risk to the party best able to bear it, and price the transfer explicitly.

The sci-tech finance logic enters through the catalysis function. Chapter 9 argued that sci-tech financial intermediaries can channel capital toward green innovation when the intermediaries have the technical capability to evaluate the innovation. The evaluation framework is where that capability is exercised. A fund that can measure a battery's cycle life, an electrolyzer's stack efficiency, or a forestry rotation's carbon stock is a fund that can evaluate the technology, and the evaluation capability is the scarce input. The contract design should pay for the capability, not just for the outcome. This is why the technical assistance component in section 11.3.4 matters: it is the capability-building investment that makes the evaluation possible.

11.4.4 Why the Chinese guidance fund system is the right stress test

The framework I have described is general. Before applying it to a specific institutional setting, it is worth explaining why the Chinese guidance fund system is the appropriate place to test it, rather than an arbitrary case chosen for geographic convenience.

Three features of that system make it the hardest available test. First, scale: China's government guidance funds number in the thousands and hold committed capital measured in trillions of RMB, so a design flaw propagates across a portfolio far larger than any single OECD program. Second, layering: the funds are nested—national funds invest in provincial funds, provincial funds in municipal funds, municipal funds in sectoral vehicles—and every layer adds a manager, a fee, and an aggregation step between the underlying project and the ultimate sponsor. Third, dual mandates: most of these funds are expected to earn commercial returns and deliver policy outcomes with the same capital pool. A framework that survives these three conditions will survive almost anywhere; a framework that only works in a single-layer, single-mandate, small-scale fund is not a framework but a special case.

The system, in other words, concentrates every friction described in section 11.1.2 into one institutional form. Metric substitution is amplified by the layering, because each layer has an incentive to report the classification most favorable to itself. Horizon mismatch is amplified by the political cycle, which turns over faster than the fund's investment period. Additionality opacity is amplified by the sheer volume of deals, which makes deal-by-deal market-gap documentation administratively expensive and therefore rare. If the design principles in this chapter can be made to bind here, they can be made to bind in the smaller and simpler programs of the EU and the United States. If they cannot bind here, the failure will be instructive about which of the principles is doing real work and which is decorative.

With that justification in place, the specific features of the system can be described.

The first is the layered fund structure. A provincial guidance fund may invest in a municipal fund, which invests in a sectoral fund, which invests in companies. Each layer has its own manager, its own fee, and its own evaluation. The environmental metrics can be diluted at each layer: the sectoral fund reports to the municipal fund, which reports to the provincial fund, which reports to the finance bureau. By the time the metric reaches the top, it has been aggregated and adjusted three times. The design response is to require look-through reporting: the top-level sponsor should be able to see the underlying project metrics, not just the aggregated fund metrics. This is technically demanding and administratively costly, but it is the only way to prevent the dilution. The practical implementation is a shared data standard at the project level—each portfolio company reports a fixed set of physical and financial fields into a common schema—so that aggregation is a query rather than a negotiation.

The second is the political cycle. Provincial governors and finance bureau heads rotate on a three-to-five-year cycle, which is shorter than the fund's investment horizon. The pre-commitment problem is therefore acute: the official who signs the mandate may not be the official who evaluates it. The design response is to embed the evaluation schedule in a document that survives the rotation—a fund charter approved by the standing committee, a contract with the private LPs that the incoming official cannot unilaterally amend. The Chinese practice of "government guidance, market operation" (政府引导、市场化运作) is the right principle; the implementation challenge is to make the "market operation" part credible when the "government guidance" part changes hands. A charter that specifies the formula for the impact carry, the identity of the independent evaluator, and the publication requirement does more work than a charter that specifies the policy goal in prose.

The third is the coexistence of policy and commercial objectives within a single fund. Many Chinese guidance funds are expected to earn a commercial return and to serve a policy objective simultaneously, with the same capital. The blending error is therefore built into the fund's mandate. The design response is the one I have argued for throughout: separate the capital tranches, separate the scorecards, and separate the payment instruments. A fund that must serve both objectives with one capital pool and one scorecard will serve the commercial objective, because that is the one the market can check. Splitting the capital into a policy tranche and a commercial tranche is not a cosmetic accounting change; it changes which principal bears which risk, and therefore which objective the manager is actually paid to pursue.

There is a fourth feature that deserves mention because it cuts against the framework. The local government financing vehicle structure through which many guidance funds are capitalized introduces a creditor with its own claims on the fund's cash flows. When the financing vehicle has debt service obligations, the fund's distributions are diverted to the creditor before they reach the LPs, and the evaluation framework's careful separation of financial and environmental tracks collapses into a single question: is there enough cash this quarter? The design response is to ring-fence the fund's capital from the financing vehicle's balance sheet, which is a legal structuring problem rather than an evaluation problem, but it determines whether the evaluation framework has anything to evaluate.

11.5 Evaluation in Practice: Three Failure Modes and Their Remedies

The framework is easier to state than to implement. I want to close the chapter by describing three failure modes I have observed in policy-guided green funds, and the remedies that follow from the analysis. Each failure mode has a distinct mechanism, and each remedy addresses the mechanism rather than the symptom.

11.5.1 The metric drift

The first failure mode is metric drift: the fund's environmental metrics gradually loosen over the fund's life, as the manager and the sponsor negotiate what counts. A fund that starts with a strict definition of "green" ends with a permissive one, because each individual loosening is small and defensible, and the cumulative effect is large.

The mechanism is worth spelling out, because it is not simple corruption. At fund inception, the definition is written by the sponsor's policy staff, who have an incentive to be strict. At the first annual review, the manager presents a deal that is green in substance but falls outside the written categories—a waste-heat recovery project, say, that reduces emissions but does not appear in the taxonomy's list. The sponsor's reviewers face a choice between excluding a genuinely useful project and granting a narrow exception. They grant the exception. The exception becomes precedent. By year four, the exceptions have accumulated into a de facto definition that no one wrote and no one defends, and the annual report's "green" classification no longer means what the charter says it means.

The remedy is to fix the metric definitions in the fund's constitutional documents and to require a supermajority to change them. The definitions should be specific enough to be auditable: not "green projects" but "projects meeting the technical screening criteria of the EU Taxonomy for the relevant sector, as verified by an independent assessor." The specificity is the defense against drift. A definition that requires judgment invites negotiation; a definition that requires verification does not.

There is a trade-off. A rigid definition may exclude genuinely green projects that fall outside its categories, and the exclusion may be costly. The trade-off is real, and the right balance depends on the fund's mandate. A fund with a narrow technology focus can afford a rigid definition. A fund with a broad mandate needs a definition that is specific in its verification requirements but flexible in its categories. The design principle is: rigid on verification, flexible on scope.

11.5.2 The carry cliff

The second failure mode is the carry cliff: the compensation structure is discontinuous, so the manager's payoff drops sharply if a target is missed by a small margin. A manager facing a carry cliff will take excessive risk near the threshold, because the marginal payoff of crossing the threshold is large. This is the same logic as a bank's behavior near a regulatory capital threshold, and the consequences are similar: risk-taking is concentrated at the point where the system is most fragile.

The mechanism is a convexity in the manager's payoff function. Below the threshold, the manager's marginal return to effort is zero or negative; above it, the return is large. A risk-neutral manager facing this payoff will prefer a gamble that has a fifty percent chance of crossing the threshold over a certain outcome just below it, even if the gamble has a lower expected value. The manager is not being reckless; the contract is paying for the gamble.

The remedy is to smooth the compensation function. Instead of a binary threshold, use a graduated schedule: the impact carry vests at 50 percent if the environmental target is met at 80 percent, at 75 percent if it is met at 90 percent, and at 100 percent if it is met at 100 percent. The graduated schedule reduces the risk-taking incentive at the threshold and preserves the incentive to perform across the range. The cost is that the manager is paid for partial success, which some sponsors find objectionable. The benefit is that the manager does not take a gamble at the threshold, which is the behavior that produces the large failures.

A concrete illustration. Suppose the impact carry is RMB 120 million, vested only if verified emissions reductions reach 1,000 kt CO2 per year. At year six the fund's verified reductions are 940 kt, and the manager can either accept the miss or fund an additional project that has a 40 percent chance of adding 120 kt and a 60 percent chance of adding nothing, at a cost of RMB 30 million to the fund. The binary contract makes the gamble attractive: 40 percent of RMB 120 million is RMB 48 million, which exceeds the RMB 30 million cost. The graduated contract, which would pay roughly RMB 90 million at 940 kt and RMB 120 million at 1,000 kt, reduces the expected gain from the gamble to about RMB 12 million, below the cost. The manager declines the gamble. The fund is better off, and so is the policy objective, because the additional project was a bad project that only the cliff made worth doing.

11.5.3 The evaluation theater

The third failure mode is evaluation theater: the evaluation is performed, the report is written, and nothing changes. The fund's managers learn that the evaluation is a compliance exercise, not a decision input, and they manage accordingly. The evaluation theater is the most corrosive failure mode because it destroys the credibility of the entire framework. Once the managers believe the evaluation does not matter, no metric design can restore the incentive.

The mechanism is a feedback loop between the sponsor's behavior and the manager's expectations. In year one, the sponsor receives the evaluation and files it. In year two, the manager notices that the evaluation had no consequence and allocates less effort to the data quality. In year three, the evaluation's numbers are unreliable, which gives the sponsor a reason not to rely on them, which confirms the manager's expectation. The loop closes within a single political cycle.

The remedy is to tie the evaluation to a consequential decision. The evaluation should be an input to a decision that the sponsor actually makes: the recapitalization of the next vintage, the extension of the fund's life, the replacement of the manager, the adjustment of the impact carry. If the evaluation has no consequence, it will be treated as theater. The consequence does not have to be dramatic; it has to be real. A sponsor that reviews the evaluation and decides to continue the fund on the same terms has made a decision, and the managers know it. A sponsor that reviews the evaluation and files it has not.

The strongest version of the remedy is a pre-committed decision rule: the charter specifies in advance what happens if the evaluation crosses certain thresholds—for example, that a verified shortfall of more than 20 percent against the environmental target triggers an automatic reduction in the sponsor's recapitalization commitment for the next vintage. The rule converts the evaluation from an input into a trigger, and a trigger cannot be filed.

11.6 Conclusion: The Contract Is the Policy

The chapter's argument can be stated in one sentence: the contract is the policy. A government that wants a green transition and writes a mandate without an evaluation framework and an incentive structure has written a press release, not a policy. The transition happens through the contracts that govern capital allocation, and those contracts have to pay for the transition if they are to produce it.

The framework I have described is not novel in its components. Separate scorecards, tiered carry, escrowed impact payments, independent evaluation, calibrated subordination—each of these is used somewhere in the financial system. What is novel, or at least underused, is the combination. The combination is what makes the framework work, because the failure modes are connected: metric substitution is checked by separate scorecards, horizon mismatch is checked by escrowed impact carry, additionality opacity is checked by independent evaluation, and evaluation theater is checked by consequential decisions. No single component is sufficient. Remove the escrow and the impact carry becomes a payment for promises; remove the independent evaluator and the escrow releases on the manager's own numbers; remove the consequential decision and the evaluator's report becomes a document nobody reads. The components hold each other up.

I want to close with a concrete observation from the field. In 2021 I sat in on a review of a provincial green guidance fund in China's Yangtze River Delta. The fund had a mandate to support green manufacturing, a blended capital structure with 40 percent government subordination, and an evaluation framework that consisted of a single annual report to the finance bureau. The report listed the fund's investments, their green classifications, and their financial returns. It did not list the emissions avoided, the cost reductions achieved, or the follow-on capital attracted. The finance bureau official who received the report told me, without irony, that the fund was "performing well" because it had not lost money. The policy objective—the reason the fund existed—was not being measured, and therefore was not being pursued.

That fund is not unusual. The framework in this chapter is a response to that fund and to the many like it. The response is not to demand better intentions from the officials or the managers. It is to write better contracts: contracts that measure what the policy is for, that pay for what is measured, and that make the measurement consequential. The green transition will be financed through contracts, and the contracts will determine whether the financing produces the transition.

The measurement itself has a physical substrate. A verified emissions reduction requires a monitoring system that can observe the plant, a data pipeline that can move the observation to the evaluator without passing through the manager's spreadsheet, and a registry that can hold the claim against double counting. Chapter 12 examines the satellite-based MRV systems and the fund-level data platforms that make look-through reporting feasible, and the question of who owns the data when the fund, the portfolio company, and the sponsor all have claims on it.