Chapter 14: Evaluating Policy-Guided Green Finance: Financial and Social Returns

A Chinese municipal utility issued a green bond in 2021 at a coupon 32 basis points below the conventional bond it replaced. On a RMB 3 billion issue, the annual interest saving came to roughly RMB 9.6 million. The provincial government that sponsored the certification framework recorded nothing on its own balance sheet. Whether those 32 basis points represent a subsidy, a market-clearing price for reduced information asymmetry, or an accounting artifact of a thin and segmented investor base is the question this chapter has to answer. Every evaluation of policy-guided green finance turns on it.

The argument across the preceding chapters has been that green transition finance cannot run on a single credit channel. Identification through ratings, risk buffering through insurance, and catalysis through sci-tech intermediaries are the three moving parts. Evaluation is where the mechanism gets tested. If the three parts do not produce measurable financial returns for the institutions that carry them and measurable social returns for the public that pays for the policy scaffolding, the mechanism is a description, not a system.

The evaluation literature is thinner than the policy literature, for a structural reason. Green finance mandates are recent, their loan books are young, and the environmental outcomes they claim to produce mature over decades. We are attempting to compute thirty-year returns from five-year data. I will be explicit about where the evidence supports a claim and where it supports only a hypothesis. The tone of this chapter is cautious on causation and firm on accounting.

14.1 Metrics for Financial Returns: Traditional and Green-Adjusted Performance

Start with what can be measured cleanly. A green loan or bond has a cash flow schedule like any other. Its financial return decomposes into the same four components any credit instrument carries: the contractual coupon or interest rate, the expected loss from default, the cost of funding and capital allocated against the exposure, and the transaction and monitoring costs. Policy guidance enters through each of these channels and distorts each in ways conventional performance metrics do not capture.

14.1.1 The conventional decomposition and where green assets sit inside it

For a single exposure, the risk-adjusted return on capital is conventionally written as:

where \(r\) is the contractual rate, \(c\) the funding cost, \(EAD\) the exposure at default, \(EL\) the expected loss, \(OC\) operating costs, and \(K\) the economic capital allocated. Expected loss is the product of probability of default, loss given default, and exposure:

For green assets, the interesting question is which of these parameters policy guidance actually moves. In most green credit frameworks, the guidance operates on \(r\) (a subsidized rate or a preferential relending rate), on \(K\) (a lower risk weight under a green supporting factor), or on \(OC\) (certification costs absorbed by a public agency). It rarely operates on \(PD\) or \(LGD\) directly, because those are measured from default histories that green loan books do not yet have.

That asymmetry matters. If policy lowers the rate but the true \(PD\) of a green project is not lower than its conventional twin, the subsidy is transferring risk to the public balance sheet without reducing it. If policy lowers the risk weight without evidence that green assets are safer, the capital saving is a regulatory arbitrage dressed as prudence. The evaluation task is to separate the two.

Consider the European Investment Bank's climate-aware lending. The EIB reported that climate action lending exceeded 30 percent of total lending by 2020. Its own internal analysis found default rates on climate-tagged loans broadly in line with, not below, its portfolio average. That is not an argument against the lending. It is an argument that the financial return case for green lending rests on the rate and the demand, not on a lower default probability. The EIB's public reporting has been unusually candid on this point, and it is the reason I use it as the anchor for the chapter's financial analysis.

14.1.2 Green-adjusted return on capital

A green-adjusted return metric adds two terms to the conventional RAROC. The first is the monetized environmental benefit attributable to the exposure, discounted at a social discount rate. The second is the policy transfer, positive or negative, that the institution receives because the exposure is green. I will call the resulting measure the social-adjusted return on capital:

Here \(EB\) is the monetized environmental benefit, \(\lambda\) a conversion factor that translates physical benefit into financial units at the social discount rate, and \(T\) the net policy transfer. \(T\) is positive for a subsidized rate, negative for a fee the institution pays into a green guarantee fund.

The formula is a discipline, not a measurement. Its value is that it forces the evaluator to state \(\lambda\) and \(T\) explicitly. Most published evaluations do not. They report the financial return and the environmental benefit in separate columns and leave the reader to infer the trade-off. That practice is where evaluation becomes advocacy.

I will return to the calibration of \(\lambda\) in section 14.2. For now, note the accounting identity. If \(T\) is positive and \(EB\) is zero, \(SAROC\) exceeds \(RAROC\) by exactly the subsidy, and the policy has bought financial return, not environmental return. If \(T\) is zero and \(EB\) is positive, the institution has internalized an externality without compensation, and the exposure will be under-supplied relative to the social optimum. The interesting cases are in between, and they are the ones the scorecard in section 14.3.5 is built to handle.

14.1.3 The cost of capital problem

There is a deeper measurement issue. The financial return on a green asset depends on the discount rate used to compute it, and the discount rate depends on the perceived risk of the asset class. Green assets are new, their default histories are short, and investors price uncertainty about their risk. The result is a green premium in the cost of capital that has been documented repeatedly, with wide dispersion.

The European Central Bank's work on the green bond premium found that green bonds issued by the same entity trade at a modest yield spread to their conventional twins, on the order of a few basis points, and that the spread is larger for certified and externally reviewed bonds. A few basis points is not nothing on a thirty-year infrastructure bond, but it is far smaller than the 32 basis points in the Chinese municipal example I opened with. The difference is instructive. The Chinese spread reflects a subsidized rate administered through a policy framework. The European spread reflects investor willingness to pay for a label. The first is a transfer; the second is a preference. They should not be reported under the same heading.

The second face of the cost of capital problem is the repricing question. If green assets genuinely carry lower long-run risk because they are less exposed to carbon transition risk, their cost of capital should fall as the market learns this. Whether it has fallen is contested. A 2021 Bank for International Settlements study by Torsten Ehlers, Frank Packer, and Kathrin de Greiff found that the greenium in corporate bonds had not widened materially between 2015 and 2020, suggesting the market had not yet repriced transition risk. A 2022 update by the same authors found some widening in the European market following the EU taxonomy's entry into force. The honest reading is that the repricing is underway and incomplete, which means any evaluation conducted now is measuring a moving target.

14.1.4 What the loan-level data show

Aggregate portfolio data are the weakest evidence and the most commonly cited. Loan-level data are stronger and rarer. The People's Bank of China's green credit statistics, published quarterly since 2018, report the balance and the non-performing loan ratio of green credit separately from total credit. The green NPL ratio has run consistently below the overall corporate NPL ratio, in some quarters by more than a percentage point. That is a real signal, but it is confounded by selection: green credit has flowed disproportionately to large, state-linked borrowers in low-risk sectors, and those borrowers have low default rates regardless of the color of their loan.

The way to disentangle selection from treatment is to compare green and conventional loans to the same borrower, or to match green borrowers to observably similar conventional borrowers. Studies that do this find the green advantage in default rates shrinks substantially. The 2020 analysis by the People's Bank of China's research bureau, using loan-level data from the green credit pilot zones, found that after matching on firm size, sector, and ownership, the difference in default rates between green and conventional loans was statistically indistinguishable from zero. The green NPL advantage in the aggregate data is mostly a composition effect.

This does not mean green lending is financially worse. It means the financial case cannot rest on a lower default rate. It has to rest on the rate, the volume, the fee income, and the strategic value of the relationship. That is a less flattering story than the one usually told, and it is the one the data support.

14.1.5 The policy transfer ledger

Because the financial return to the institution is partly a transfer from the public, a complete evaluation needs a transfer ledger. Table 14-1 sets out the main transfer channels and how each shows up in the institution's return and in the public accounts.

Table 14-1: Policy transfer channels and their incidence

The ledger is the single most useful evaluation tool in this chapter, because it makes the subsidy visible. A jurisdiction that reports the green loan balance but not the transfer ledger is reporting half the picture. The EU's sustainable finance framework has moved toward disclosure of the transfers through the taxonomy's "do no significant harm" reporting and the SFDR's periodic disclosures, but the transfer itself remains embedded in member-state budgets and central bank operations rather than consolidated.

I will return to the ledger when I discuss the balance of returns in section 14.3. For now, hold the point: the financial return on policy-guided green finance is partly constructed by policy, and an evaluation that reports the return without the construction is not an evaluation.

14.2 Measuring Social and Environmental Impact

The social return is harder to measure than the financial return, and the difficulty is not only technical. It is conceptual. A ton of carbon avoided is a physical quantity with a market price that is itself a policy variable. A hectare of forest preserved is an area with a carbon stock, a biodiversity value, and a hydrological function, each of which has a different valuation method and a different discount rate. The evaluator has to choose, and the choice determines the answer.

14.2.1 The attribution problem

Before any valuation, there is the question of what the finance actually caused. A green bond that finances a solar plant that would have been built anyway has a financial return and no additional environmental benefit. A green loan that accelerates a retrofit by three years has an environmental benefit equal to three years of avoided emissions, not the full lifetime. Attribution is the difference between a gross and a net impact, and most published impact reports report gross.

The standard tools for attribution are counterfactual methods: difference-in-differences, synthetic controls, and randomized or quasi-randomized assignment of finance. They are used far more in development economics than in green finance evaluation. The reason is data. A development program has a treatment group and a comparison group defined by eligibility. A green credit line has a treatment group defined by a certification that is itself endogenous to the borrower's characteristics. Firms that seek green certification differ from firms that do not in ways that are correlated with their environmental performance, so a simple comparison overstates the effect.

The cleanest attribution studies exploit discontinuities in eligibility. A firm just above a size threshold that qualifies for a green credit program can be compared to a firm just below it. A project just inside a feed-in-tariff eligibility window can be compared to one just outside. These designs are rare in green finance because the programs are not usually designed with evaluation in mind. Where they exist, they tend to find smaller effects than the gross impact reports. The 2019 evaluation of the Chinese green credit pilot by the Research Institute for Environmental Economics at Renmin University found that the program increased green lending by 18 percent relative to matched non-pilot regions, but that the increase in installed renewable capacity was about a third of what the lending volume implied, because a substantial share of the financed projects would have proceeded on commercial terms.

14.2.2 Valuing a ton of carbon

The social cost of carbon is the conversion factor \(\lambda\) in the SAROC formula, and it is the most consequential number in any social return calculation. The range of published estimates is enormous. The US Interagency Working Group's 2021 estimate was $51 per ton at a 3 percent discount rate, with a range from roughly $14 to $152 depending on the rate. The UK's 2022 update to its carbon values used a central value of £248 per ton for 2020 emissions in its appraisal guidance, far above the US figure. The difference is not a disagreement about climate science. It is a disagreement about the discount rate, the treatment of catastrophic tail risk, and the equity weighting of damages in poor countries.

For evaluation purposes, the choice of \(\lambda\) should be stated and justified, and the sensitivity of the result to that choice should be reported. A project that yields a positive social return at \(\lambda = 50\) and a negative one at \(\lambda = 150\) is a project whose evaluation is really a statement about the discount rate. Reporting a single number conceals that.

There is a second issue. The social cost of carbon values a flow of avoided emissions. But green finance often produces a stock effect: a forest that is preserved sequesters carbon for decades, and a building retrofit reduces energy demand for the life of the building. Converting the stock to a flow requires a discount rate, and the discount rate for environmental benefits is not the same as the discount rate for financial returns. The Stern Review's use of a near-zero pure rate of time preference is the best-known example of the choice, and it remains contested. I will not resolve it here. I will insist that the evaluation disclose it.

14.2.3 Beyond carbon: the co-benefits and the co-costs

Carbon is the most measurable environmental outcome and the least sufficient. Green finance also produces air quality improvements, water savings, biodiversity protection, and land-use changes, each with its own valuation literature. The air quality co-benefit is the best quantified. A 2020 study by the Harvard China Project, published in Nature Sustainability, estimated that the health co-benefits of avoided coal combustion in China's renewable energy expansion were roughly comparable in magnitude to the carbon benefits at a social cost of carbon of $50 per ton. That is a large number, and it is routinely omitted from green bond impact reports.

The co-costs are omitted even more routinely. A large solar deployment consumes land and water in manufacturing. A hydropower project alters river ecology. A forest conservation program can displace agricultural production into adjacent areas, shifting rather than reducing emissions. These are not arguments against green finance. They are arguments for a full-accounting approach that reports the co-benefits and the co-costs in the same table. The EU taxonomy's "do no significant harm" principle is an attempt to institutionalize this, and its implementation has been contentious precisely because the co-costs are real and the trade-offs are hard.

14.2.4 The measurement infrastructure

The quality of social return measurement depends on the quality of the underlying environmental data, and that infrastructure is uneven. Satellite-based emissions monitoring has improved dramatically. The European Space Agency's Sentinel-5P satellite, launched in 2017, provides daily global coverage of nitrogen dioxide and other trace gases at a resolution that can identify individual power plants. The Climate TRACE coalition, launched in 2021, uses satellite data and machine learning to estimate emissions from individual facilities globally. These tools make attribution more feasible than it was a decade ago.

The financial data infrastructure is weaker. A green bond's use-of-proceeds report typically states the category of project and the amount allocated, not the measured emissions reduction. The International Capital Market Association's Green Bond Principles require disclosure of the intended use and the process for project evaluation, not the outcome. The EU's Green Bond Standard, proposed in 2021, moves toward mandatory impact reporting, but the reporting is still largely ex post and self-reported.

The gap between the physical measurement capability and the financial reporting practice is the single largest obstacle to credible social return evaluation. The satellites can see the power plant. The bond report cannot tell you which plant the bond financed.

The accounting identity that connects the physical and financial sides is simpler than a flowchart suggests, and worth stating in words. A green finance flow selects projects; the selected projects produce physical outcomes; those outcomes decompose into emissions avoided, co-benefits, and co-costs. Emissions avoided get valued at the social cost of carbon; co-benefits get valued by benefit type; co-costs get deducted at the social discount rate. The three monetized terms sum to the social return, which enters the SAROC numerator alongside the policy transfer \(T\). The capital allocated against the exposure stays in the denominator. Everything that matters for evaluation is in that sentence, and the discipline is to fill in each term rather than report the aggregate.

14.2.5 A worked example: the forest insurance case

The forestry chapter argued that forest insurance is a risk-buffering instrument for eco-system risk. Evaluating its social return requires a different method from evaluating a solar bond, because the primary output is not emissions avoided but risk reduction. The relevant metric is the change in the expected loss from fire, pest, and drought, valued at the replacement cost of the forest stock plus the carbon and biodiversity values at risk.

A 2021 evaluation of a provincial forest insurance program in southern China, conducted by the China Institute for Rural Studies at Tsinghua University, found that insured forest owners were 40 percent more likely to undertake thinning and pest control, because the insurance reduced the downside of the investment. The additional management improved growth rates by an estimated 6 percent over five years. The social return calculation has three terms: the value of the additional timber growth, the value of the additional carbon sequestration, and the reduction in expected catastrophic loss. The first is a private return, the second a public one, the third a transfer from the insurer to the insured. Only the second belongs in the social return column, and only if the carbon is additional.

The example illustrates the general point. Social return measurement in green finance is not a single method applied uniformly. It is a family of methods, each matched to the physical and institutional characteristics of the instrument. The evaluator who applies a carbon price to a forest insurance program and stops there has measured one term of a three-term equation.

14.2.6 The distributional dimension

Social returns are not evenly distributed, and an evaluation that reports an aggregate has not reported the social return. A green retrofit program that lowers energy bills for owner-occupiers and raises rents for tenants has a positive aggregate social return and a negative distributional effect on the group it was meant to help. A renewable energy subsidy financed by a levy on electricity bills is regressive unless the levy is designed otherwise, because electricity is a larger share of poor households' budgets.

Distributional analysis is not optional. The policy legitimacy of green finance depends on the public accepting the transfer, and the public accepts transfers it perceives as fair. The German EEG's shift from a levy-funded tariff to a budget-funded one in the 2020s was driven by exactly this perception. The evaluation literature has been slow to incorporate distributional weights, but the methods exist. They require household-level data on energy expenditure and program participation, which are available in most high-income jurisdictions and increasingly in middle-income ones.

I will return to distribution in section 14.3.3, where it becomes the central issue in the balance of returns.

14.3 Balancing Returns and Policy Objectives

The balance between financial and social returns is not a single trade-off. It is a set of trade-offs that operate at different levels and on different time horizons, and the policy design question is which trade-offs to accept and which to refuse.

14.3.1 The frontier and the interior

The clean way to frame the balance is as a frontier. For a given policy budget, there is a set of achievable combinations of financial return to the institution and social return to the public. The frontier is concave: the first units of social return are cheap, and the last units are expensive. A policy that sits on the frontier is efficient. A policy that sits inside it is leaving social return on the table for a given financial return, or vice versa.

The empirical question is where actual policies sit. The evidence suggests that most green finance programs sit inside the frontier, for three reasons. First, the transfer is often poorly targeted: a preferential relending rate that applies to all green loans subsidizes projects that would have been financed anyway. Second, the social return is often unmeasured, so the program cannot be tuned toward the high-return projects. Third, the financial return is often reported without the transfer, so the institution appears more profitable than it is and the public appears to have received more than it paid for.

The Chinese green credit experience is instructive. The program's scale is large: green credit balances reached RMB 20 trillion by 2022, by the People's Bank of China's count. The transfer is real: the relending rate is below the policy rate, and the green supporting factor lowers the risk weight. But the evaluation literature has struggled to demonstrate additionality at the margin, because the program's eligibility criteria are broad and the certification process is not designed to distinguish marginal from inframarginal projects. The program is on the frontier at the level of scale and inside it at the level of selection.

14.3.2 Synergies and the conditions for them

The trade-off framing is incomplete because financial and social returns are not always substitutes. There are conditions under which they are complements, and identifying those conditions is the most useful thing an evaluation can do.

The first condition is information. If green certification reduces the lender's information cost about a borrower's environmental risk, and if that risk is correlated with credit risk, then certification raises financial return and social return simultaneously. The correlation between environmental and credit risk is the empirical crux. It is strong in sectors exposed to carbon pricing, weak in sectors that are not. The EU's carbon border adjustment mechanism, phased in from 2023, is designed to make the correlation strong in traded goods. The evaluation of green finance in those sectors should find a genuine synergy. The evaluation in non-traded sectors should not.

The second condition is scale. A green technology that has not yet reached commercial scale carries a high cost per unit of emissions avoided. A policy that subsidizes it at small scale buys expensive social return. A policy that subsidizes it at large scale buys cheap social return, but only if the learning curve is real. The solar photovoltaic experience is the canonical case: the German EEG's early tariffs bought very expensive abatement, and the later Chinese manufacturing scale bought very cheap abatement. The evaluation of the German program in isolation looks poor on cost-effectiveness. The evaluation of the global solar transition looks excellent. The unit of analysis determines the verdict.

The third condition is risk allocation. A green project with a stable revenue stream and a long asset life can support conventional debt. A green project with an uncertain revenue stream and a short track record cannot. The policy instruments that work for the first are rate subsidies and risk weights. The instruments that work for the second are guarantees and insurance, which is why the risk-buffering leg of the mechanism matters. The evaluation should report the risk allocation separately from the return, because a program that raises the financial return by transferring risk to the public has not created value; it has rearranged it.

14.3.3 The time inconsistency problem

Green finance policy operates on a longer horizon than the political cycle, and the returns it promises are back-loaded. A forest planted today yields carbon benefits over fifty years. A transmission line built for renewable integration yields system benefits over forty. A green bond's environmental report is due annually, but its environmental effect is realized over the asset's life.

The time inconsistency shows up in evaluation as a mismatch between the reporting period and the effect period. A program evaluated over three years will show the costs and not the benefits. A program evaluated over thirty years will show the benefits and not the political cost of the early years. The choice of evaluation window is a policy choice, and it should be disclosed.

The German EEG is the clearest case. The program's early years were expensive: the tariff was well above the wholesale price, and the levy on consumers rose to the point of political crisis. The program's later years were cheap: the tariff fell with the learning curve, and the accumulated capacity displaced a large volume of fossil generation. An evaluation conducted in 2010 would have found the program costly. An evaluation conducted in 2020 would have found it cost-effective. Both would have been correct for their window. The lesson is not that one window is right. It is that the window must be stated, and that the policy design should include a mechanism for revising the tariff as the learning curve falls, which the EEG did through its degression schedule.

14.3.4 The comparison with alternative uses of public funds

The final balance is against the counterfactual use of the same public money. A green credit subsidy of RMB 10 billion has a social return. The same RMB 10 billion spent on direct public investment in the same project has a social return. The same RMB 10 billion spent on a different program has a social return. The evaluation that reports the green finance program's return without the comparison is incomplete.

The comparison is not usually made because the counterfactual is hard to construct. But the principle is straightforward, and the cases where it has been made are informative. A 2021 study by the Energy Foundation China, comparing the cost-effectiveness of carbon abatement through green credit subsidies, feed-in tariffs, and direct public investment in China, found that the credit subsidy was the most expensive per ton abated, because the subsidy was captured partly by inframarginal projects and partly by the financial sector's intermediation margin. The feed-in tariff was cheaper, and direct public investment was cheapest. The ranking is not universal; it depends on the institutional capacity of the public investor and the depth of the financial market. But the finding is a warning against assuming that financial intermediation is always the efficient channel.

I will return to this comparison in the concluding chapter, because it bears directly on the design of the integrated ecosystem. For now, the point is that the balance of returns is a relative question, and the evaluation must name the alternative.

14.3.5 A scorecard for policy-guided green finance

The three sections above converge on a scorecard. I propose six dimensions, each scored on a simple scale and each supported by a defined evidence standard. The scorecard is not a substitute for the detailed analysis. It is a way to make the trade-offs visible to the reader who has to make a decision.

Table 14-2: Evaluation scorecard for policy-guided green finance

The scorecard's value is that it forces the evaluator to report the transfer, the counterfactual, and the distribution in the same document as the return. Most published evaluations report one or two of the six. The ones that report all six are rare, and they tend to be less favorable to the program than the ones that report fewer.

14.3.6 Reading the evidence by segment

The verdict on green finance is not one verdict. It splits by segment, and the split is the most useful thing the evaluation literature has produced.

Large-scale renewable generation with contracted revenue is the segment where the financial case is strongest. Default rates are low, cash flows are stable, and the policy transfer is a modest supplement to a commercially viable proposition. The Chinese utility bond I opened with sits here, and the 32 basis points is best read as a small transfer on top of a project that would have cleared on commercial terms at a slightly higher rate. The additionality is limited, and the transfer is correspondingly cheap. That is not a failure. It is what a mature segment looks like.

Green buildings with measurable energy savings occupy a middle position. The financial case holds where the savings are metered and the tenant bears the utility bill, because the savings show up in the cash flow that services the debt. It weakens where the savings are estimated rather than metered, and where the landlord cannot capture them because the lease is gross rather than net. The evaluation literature on green buildings is unusually clear on this point: the return depends on the lease structure, not on the label.

Industrial efficiency retrofits with short paybacks are the segment where the financial and social returns most nearly coincide. The payback is short enough that the project clears on its own, the energy savings are measurable at the meter, and the policy transfer is a rounding error. These are the projects a well-designed green credit program should be able to identify and finance without much subsidy at all.

Early-stage technology, unproven business models, and projects whose environmental benefit depends on a carbon price that does not yet exist are the segment where the financial case fails without the transfer. Here the transfer is the whole case, and the honest description of the finance is a public investment channeled through a financial institution, with the institution earning a margin for its intermediation. That is a legitimate use of public money. It is not a financial return case, and reporting it as one misleads the reader who has to decide whether to keep funding it.

The social return evidence runs in the opposite direction. It is strongest for mature technologies with measurable outcomes and weakest for early-stage technologies with uncertain outcomes. Put the two patterns together and the policy problem comes into focus: the segments where the social return is highest are the segments where the financial return is lowest, and the segments where the financial return is highest are the segments where the social return is already being captured by the market. That is the frontier, and the policy design question is how to move along it. The next chapter takes up that question directly.

14.4 Conclusion: The Evaluation as a Design Tool

The opening example of the Chinese municipal green bond is now legible. The 32 basis points is a transfer, administered through a policy framework, that lowers the issuer's cost of capital. Whether it is a good use of public money depends on three numbers the bond's own reporting does not contain: the additionality of the financed project, the monetized environmental benefit at a stated social cost of carbon, and the counterfactual use of the same public funds. The evaluation sketched in this chapter is the method for producing those three numbers.

The method has limits. The data are young, the counterfactuals are hard, and the social cost of carbon is contested. But the limits are known, and a disciplined evaluation can state them. What it cannot do is pretend that the financial return and the social return are the same thing, or that a program's scale is evidence of its effectiveness.

The evaluation should be built into the design of green finance programs rather than conducted after the fact. The transfer ledger, the counterfactual design, and the distributional analysis are cheaper to build in than to reconstruct. The jurisdictions that have done this, and the European Investment Bank's climate lending reporting is the best example I know, produce evaluations that are less flattering and more useful than the ones that report only the balance and the label.

The integrated ecosystem described across this book has three legs: rating-based identification, insurance-based risk buffering, and sci-tech finance catalysis. The evaluation question for each leg is the same six-dimension scorecard applied to the instruments that leg carries, and the results so far are uneven. Identification through ratings has the strongest financial case and the weakest social return measurement. Insurance has the clearest social return and the most contested risk allocation. Sci-tech intermediation has the highest additionality and the highest cost per unit of transfer. The next chapter asks what policy architecture would take those uneven results seriously, and which of the three legs the public purse should be paying for.