Chapter 15: Policy Implications, International Cooperation, and a Research Agenda

Eight million yuan a year. That is what it costs to run the credit-rating, insurance-origination, and technology-vetting infrastructure for a single agricultural county—the figure that surfaced from the pilot described in Chapter 5, and the one that should anchor any honest policy discussion. A county government working from a 400 million yuan fiscal budget cannot absorb it. A province can, but only by treating the platform as shared infrastructure rather than a one-off project. No private rating agency will build it: the revenue from scoring a few hundred technology firms and a few thousand farms does not cover the fixed cost, and the information the platform produces is non-excludable once it exists.

That arithmetic is the policy problem in miniature. The identification–buffering–catalysis chain is not a market that clears on its own. Each link carries a public-good component that private capital will underprovide—rating infrastructure because information leaks, insurance because correlated tail risk in weather and fire exceeds what any single underwriter can hold without reinsurance or state backing, catalytic sci-tech finance because early-stage green technology is too uncertain and too slow for private venture capital at the scale the transition requires. Policy is not an add-on to the chain. It is a load-bearing wall.

The rest of this chapter works through three problems: which policy instruments make the chain function and what we know about their cost and effectiveness; how cross-border cooperation should be structured when carbon is a global externality but credit and insurance remain national; and what research would close the largest evidentiary gaps. The evidence is thin in places. I say so where it is.

15.1 Policy Recommendations: Building the Chain Deliberately

15.1.1 The Policy Problem Is Coordination, Not Subsidy

The reflex response to green finance is a subsidy—lower the interest rate on green loans, subsidize the insurance premium, capitalize the venture fund. China has tried all three. Its green credit balance reached roughly 30 trillion yuan by 2023, and a substantial share carries some implicit or explicit subsidy: lower risk weights under the PBOC's green relending facility, preferential treatment in macroprudential assessment, local government interest rebates. The results are real. The mechanism is blunt. A subsidized green loan identifies nothing. It lowers the price of capital for anyone who can claim the green label, and the label is cheap to claim.

The chain framework implies a different logic: identify first, buffer second, catalyze third. Subsidy should attach to the information and risk infrastructure that makes identification and buffering possible, not to the loan itself. Paying farmers to plant trees is a transfer. Paying for the fire insurance that makes the plantation bankable is a market-making investment that can, in principle, pay for itself.

Run the arithmetic. A green loan of 10 million yuan at a 2-percentage-point subsidy costs the fisc 200,000 yuan a year. The same 200,000 yuan, spent on a county rating platform that scores 200 firms, costs 1,000 yuan per firm-year and produces information every lender in the county can reuse. If the platform raises the share of green loans that perform—because the scoring model screens out firms whose environmental claims do not survive fundamental analysis—the fiscal return shows up as a reduction in non-performing loans rather than as a transfer. That is the logic behind the credit-rating pilot in Chapter 5, and it is why the fixed cost matters more than the spread.

Stated plainly: fund the infrastructure, not the spread. Where a spread subsidy is used, tie it to a verification standard produced by the rating platform, so the subsidy cannot be claimed without identification.

15.1.2 Four Instruments and What Each Is For

Four policy instruments correspond to four distinct failures in the chain. Table 15-1 maps them; the prose that follows takes up the design trade-offs the table cannot carry.

Table 15-1: Policy instruments mapped to chain failures

Rating infrastructure grants are the least tested instrument, and the 8 million yuan figure comes from a single pilot that should not be generalized. What the pilot does show is that marginal cost per firm falls sharply with scale: software and model development are fixed, per-firm scoring is variable and small. A province operating one platform across 40 counties spreads the fixed cost to 200,000 yuan per county, within reach of county budgets. Regional consolidation, not county-by-county replication, is the implication. This is an unglamorous recommendation and it will meet resistance from county governments that want their own platform for reasons of visibility. The cost arithmetic does not care about visibility.

Weather index insurance premium support has the most evidence behind it. China's pilots in Anhui, Heilongjiang, and Xinjiang between 2016 and 2022 showed that premium subsidies at 40–60 percent raised adoption among smallholders from single digits to 30–50 percent of eligible area, depending on crop and basis risk. The binding constraint after subsidy was not price but trust: farmers who had received a payout that did not match their actual loss were less likely to renew. That is the basis-risk problem, and it is a design problem, not a subsidy problem. A policy that subsidizes a badly designed index buys adoption and then loses it.

Forest insurance with a reinsurance backstop addresses the correlated tail. One wildfire can destroy a county's entire insured forest area. No primary insurer holds that risk without a reinsurance treaty, and the reinsurance market prices it at a level that puts the premium out of reach for smallholder forest cooperatives. The policy choice is between a state reinsurance facility and a parametric trigger that pays on a measurable index—fire radiative power, rainfall deficit—rather than on adjusted loss. Parametric design reduces moral hazard and adjustment cost but introduces basis risk. State reinsurance preserves indemnity but concentrates fiscal risk. China has used both in different provinces, and the comparative evidence is not yet in. Without one of the two, forest insurance does not scale.

Government guiding funds with follow-on mandates are the most contested instrument, and the crowding-out literature is large and inconsistent. Cumming and MacIntosh (2006) found crowding out in Canadian data; Leleux and Surlemont (2003) found seeding rather than crowding out in a pan-European sample; Da Rin, Nicodano, and Sembenelli (2006) found that public policy can create active venture capital markets, but the effect depends on design. The design lesson across this literature: guiding funds work when they are small relative to the private market, take minority positions, and invest in segments the private market will not—not when they compete with it. A fund that is large, majority-positioned, and bidding on the same deals as private VCs will crowd out. A fund that is small, minority, and concentrated in pre-seed green technology may seed.

China's experience illustrates both patterns. The National Integrated Circuit Industry Investment Fund, at 138.7 billion yuan in its first phase, was large enough to move prices and overlapped heavily with private capital. Provincial green technology guiding funds have been smaller and more targeted, and the early evidence is more favorable. I would not overstate it: selection effects are severe and the evaluation horizon is short.

15.1.3 The Sequencing Problem

The chain has a natural order—identify, buffer, catalyze—and policy that ignores the order wastes money. Catalytic capital deployed before rating infrastructure exists flows to firms that can tell a good green story, not to firms that can survive a credit review. Insurance deployed before risk data exists will be mispriced, and will either lose money or fail to pay claims.

One constraint is non-skippable: stage 1, the rating and verification infrastructure. Without identification, the other two stages allocate capital and risk on the basis of unverified claims, which is the failure mode the chain was designed to correct. Stages 2 and 3 can run in parallel once stage 1 is in place—a province with existing agricultural insurance infrastructure can build the index and the catalytic fund simultaneously, and many will.

There is a fiscal argument for the sequence as well. The rating platform is the cheapest of the three stages and carries the highest information return per yuan. A province that spends its first 20 million yuan on a regional rating platform and its next 100 million on insurance subsidy will get more green transition per yuan than one that reverses the order. That is a claim about marginal returns, and it is falsifiable. I would like to see it tested.

15.1.4 What Not to Do

Three approaches should be avoided on the evidence reviewed in this book.

Do not use the green label as a substitute for credit analysis. Green bond market growth has come with a proliferation of verification schemes of varying rigor, and the correlation between a green label and a firm's environmental performance is weaker than the label implies. A rating model that treats the label as an input without checking it against emissions data, energy intensity, and capital expenditure plans will misclassify. The scoring models in Chapter 6 are built to check the label, not to accept it.

Do not subsidize insurance premiums without addressing basis risk. The Chinese pilot data are clear: farmers who experience a payout that does not match their loss exit the program. A subsidy that buys adoption and then loses it is worse than no subsidy, because it consumes fiscal resources and leaves the risk uninsured.

Do not deploy catalytic capital at a scale that dominates the private market. The crowding-out evidence is contested, but the mechanism is simple. If the government fund is the marginal buyer in every deal, private investors will price the government's participation into their own terms, and the market will not develop independent pricing capacity. The guiding fund should be a minority participant with a segment mandate, not a price-setter.

15.2 International Cooperation: Carbon Is Global, Credit Is National

15.2.1 The Mismatch

A ton of carbon emitted in Hebei warms the atmosphere in Hamburg. The instruments the chain uses to address the transition—credit ratings, insurance contracts, venture funds—are national or subnational. They are governed by national regulators, priced in national currencies, backed by national fiscal capacity. This mismatch is the central problem of international cooperation in green finance, and none of the existing institutional arrangements resolves it.

The Paris Agreement sets a global target and relies on nationally determined contributions. The financial system that must fund those contributions is organized around national balance sheets. The consequence: the countries with the largest transition financing needs are often the countries with the weakest financial infrastructure and the smallest fiscal capacity to build it. This is why the Green Climate Fund exists. But the Fund's disbursements—roughly $12 billion committed since 2010—are small against the need. The IEA's World Energy Investment 2023 estimated that emerging and developing economies will need $4–5 trillion per year in clean energy investment by 2030, roughly triple the current level.

The chain framework suggests a different role for international cooperation than gap-filling capital transfers. Cooperation should transfer the institutional infrastructure that makes the chain work—the rating methodology, the insurance design, the fund governance. That is slower and less visible than a capital transfer. It is also more likely to be sustainable, because it builds capacity that remains after the cooperation ends.

15.2.2 Three Channels of Cooperation

International cooperation in green finance runs through three channels, each with a different logic and a different record.

Harmonization of disclosure and taxonomy. The EU taxonomy, the ISSB sustainability disclosure standards, and China's green bond catalog are three attempts to define what counts as green, and they do not agree. A bond that qualifies under the Chinese catalog may not qualify under the EU taxonomy, and vice versa. This creates friction for cross-border flows and opportunities for arbitrage. Harmonization is desirable but slow, and the political economy is difficult: each jurisdiction's taxonomy reflects its own industrial structure and transition priorities. China's catalog includes clean coal technology, which the EU taxonomy excludes. The EU's includes nuclear, which some member states and some third countries contest.

The practical question is not whether harmonization is desirable—it is—but whether it should proceed through convergence or mutual recognition. Convergence means a single global standard, unlikely in the near term. Mutual recognition means each jurisdiction accepts the other's standard for cross-border purposes, which is feasible but requires trust in the other's verification infrastructure. The chain framework favors mutual recognition with a verification floor: a minimum set of data and audit requirements that any recognized standard must meet. Less ambitious than convergence, more achievable.

Cross-border insurance and reinsurance. Weather and catastrophe risk does not stop at borders. A drought in the North China Plain and a drought in the U.S. Corn Belt are correlated through global atmospheric patterns, and a reinsurer holding both exposures diversifies its portfolio. The international reinsurance market is the most developed cross-border risk-sharing mechanism in existence. The policy question is how to extend it to the agricultural and forestry risks that the chain's buffering function must cover in developing countries.

The obstacle is not the absence of a market but the absence of data and legal infrastructure. A reinsurer will not underwrite a weather index in a country without a reliable meteorological record, and it will not enforce a contract in a jurisdiction without a functioning insurance regulator. International cooperation can address both. The World Bank's catastrophe risk pools—the Caribbean Catastrophe Risk Insurance Facility, the African Risk Capacity—are multilateral institutions providing the data and the legal wrapper that private reinsurers require. They are small relative to the need, but the model is sound, and the chain framework would extend it to the forest insurance and eco-agriculture risks discussed in Chapters 8 and 9.

Co-investment and blended finance. The third channel is direct co-investment of international public capital alongside domestic private capital in green transition projects, through concessional loans, guarantees, and first-loss tranches. The logic is to use public capital to absorb risks private capital will not take at an early stage of a market, then exit as the market matures.

The record is mixed. The multilateral development banks have co-invested in green infrastructure for two decades, and leverage ratios—private capital mobilized per dollar of public capital—have generally fallen below targets. The reasons are familiar: projects are small, transaction costs are high, and the private investors who do participate often require a risk-adjusted return that the public capital's concession does not fully provide. Blended finance works best when the project has a clear revenue stream and the public capital absorbs a specific, quantifiable risk—a first-loss tranche on a portfolio of green loans, for example. It works least well when public capital is used to subsidize a project with no commercial return.

15.2.3 The China Question

Any discussion of international cooperation in green finance has to address China, because China is simultaneously the largest emitter, the largest green credit market, and a major source of transition finance for other developing countries.

The numbers are large. China's green credit balance of roughly 30 trillion yuan is the largest in the world. Its green bond issuance ranks in the top two annually. Its pledge to peak emissions before 2030 and reach neutrality before 2060 is the most consequential single climate commitment outside the Paris Agreement. And its Belt and Road Initiative has financed infrastructure in more than 140 countries, some of it carbon-intensive.

The cooperation question is how to engage China's financial system in a way that supports the global transition without requiring China to adopt standards it does not accept. Mutual recognition is the most promising route. If the EU and China can agree on a verification floor—a minimum data and audit standard both accept—then Chinese green bonds can be recognized in European markets and European green capital can flow into Chinese transition projects without the friction of full taxonomy alignment. This is not a small undertaking. The verification infrastructure on both sides would need to be audited and trusted. It is more feasible than convergence, and it addresses the largest bilateral flow in green finance.

A concrete first step is available: a joint pilot in which a small number of Chinese green bond issuers submit to verification by an EU-accredited auditor under a floor standard agreed by both regulators, with the results published. The pilot would cost little, would test whether the two verification systems can produce comparable outputs, and would generate the trust that mutual recognition requires. I would put this ahead of any broader taxonomy negotiation, because the negotiation will stall without evidence that the verification floor is workable.

The Belt and Road dimension is harder. Chinese overseas energy finance has shifted toward renewables in recent years, but the shift is incomplete and the data are opaque. Cooperation here would require transparency commitments that Chinese financial institutions have not made. The chain framework's logic still applies: the first step is identification, and identification requires disclosure. Without disclosure of the environmental performance of Belt and Road projects, no rating, no insurance, and no catalytic capital can be reliably allocated. The policy recommendation is narrow but real: make project-level environmental disclosure a condition of any co-financing arrangement in which Chinese policy banks participate alongside multilateral institutions. That condition is achievable within existing co-financing structures and does not require China to adopt the EU taxonomy.

15.2.4 What International Cooperation Cannot Do

Two limits deserve to be stated.

International cooperation cannot substitute for domestic financial infrastructure. A country without a functioning credit registry, a reliable meteorological service, or an independent insurance regulator cannot import the chain ready-made. The infrastructure has to be built domestically, with international support, but the support is a complement, not a substitute.

Nor can international cooperation resolve the distributional conflict at the heart of the transition. The countries that have emitted the most have the most capacity to pay for adaptation and mitigation; the countries that have emitted the least are the most exposed. The international financial mechanisms that address this—the Green Climate Fund, the Loss and Damage Fund agreed at COP27—are transfers, and their scale is set by political bargaining, not financial engineering. The chain framework can make the transfers more effective by building the infrastructure through which they are deployed. It cannot determine their size.

15.3 Future Research Directions: The Gaps That Matter

15.3.1 The Additionality Problem

The largest gap in the evidence is additionality. We do not know, with any precision, how much of the green finance already extended would have been extended anyway. A green loan to a firm that was going to install solar panels regardless is not additional. A green bond issued by a utility that was already going to build a wind farm is not additional. The counterfactual is hard to construct and the data are rarely available.

The question is not academic. If the green credit balance grows while additionality is low, the policy is subsidizing a relabeling, not a transition. The Chinese green credit data are consistent with both interpretations: the balance has grown rapidly, the emissions intensity of borrowing firms has fallen, and the causal attribution is unclear.

What would resolve it is a difference-in-differences design comparing firms that received green credit with matched firms that did not, using the rating platform's scores as the assignment variable. This requires firm-level emissions data, which are becoming available through the carbon accounting infrastructure discussed in Chapter 13, and a credible matching strategy. The design is feasible. It has not been done at scale.

I would prioritize this over any other research question. The policy stakes are large and the current evidence is essentially anecdotal.

15.3.2 Cost-Effectiveness Reporting

Very few green finance programs report the cost per ton of carbon avoided, and those that do rarely report the basis of the calculation. Is the cost the subsidy, the administrative cost, or the full cost of capital? Is the carbon avoided measured over the life of the project or annually? Is the baseline the counterfactual or the status quo?

Without a common reporting standard, comparison across programs is impossible, and ineffective programs survive easily. A program that reports a cost per ton without a stated basis cannot be evaluated.

The research agenda here is methodological before it is empirical. A standard cost-effectiveness metric for green finance programs—fiscal cost per ton of CO₂e avoided, with a specified baseline and time horizon—would allow comparison across the rating, insurance, and catalysis instruments. The metric is not hard to define. The hard part is getting programs to report it.

15.3.3 The Correlation Between the Three Functions

The chain framework assumes that identification, buffering, and catalysis are complementary—that a well-rated firm is more insurable, and an insured firm is more investable. The assumption is plausible and unmeasured.

Measurement is feasible. A dataset linking the rating scores of technology firms, their insurance uptake, and their access to venture capital would allow a test of the correlation. A strong correlation supports the chain logic and the policy recommendation to build the three functions together. A weak correlation suggests the functions are independent and the sequencing argument is weaker.

My prior is that the correlation is positive but modest. The rating score predicts credit performance; insurance uptake predicts risk management capacity; venture access predicts growth potential. These are related but distinct, and the chain's value may come from the complementarity of the three signals rather than from their redundancy. That is a testable proposition and it should be tested.

15.3.4 The Governance of Cooperative Finance

Chapter 10 discussed the governance problems of rural cooperative finance, and the Chinese material is blunt about the failure mode. The key members of a cooperative—founders, leaders, large shareholders—typically have more capital and more human capital than ordinary farmer members. They often do not engage in production; they enter at the sales end of the value chain. They run the cooperative like a firm. And they set the membership threshold, which in some cases excludes smallholders entirely: a poultry cooperative requiring annual output of 10,000 birds, an aquaculture cooperative requiring 30 mu of water surface and working capital, a vegetable cooperative requiring 5 mu of operating area.

Why do these entrepreneurs choose a cooperative rather than a firm? In the Chinese case, government support. The support extends the cooperative's survival space and, if large enough, can convert an investor-owned firm into a cooperative. That is a policy design problem. Support targeted at cooperatives that are functionally firms subsidizes the wrong organizational form and excludes the smallholders the policy is meant to help.

The research agenda is to measure the incidence of this conversion and its effect on smallholder access. The data exist in the agricultural ministry's cooperative registry. The analysis has not been done.

15.3.5 The Digital Infrastructure Question

Chapter 13 argued that digital infrastructure and carbon accounting are cross-cutting enablers. The claim is plausible; the evidence is thin.

The specific question is whether the marginal return on carbon accounting infrastructure exceeds the marginal return on the financial instruments it enables. A carbon accounting system that produces accurate firm-level emissions data makes the rating model more precise. But the system is expensive to build and maintain, and the precision gain may be small relative to the cost. The alternative—sector averages and proxy variables—is cheaper and less precise.

The answer will vary by sector. For cement and steel, where emissions are concentrated and measurable, the accounting infrastructure is worth building. For light manufacturing and services, where emissions are diffuse and small, proxies may be sufficient. A research agenda that maps the marginal return by sector would let policy allocate accounting investment where it pays.

15.3.6 The Behavioral Question

The final gap is behavioral. The chain framework is an institutional design, and institutional designs are implemented by people. The rating analysts who score technology firms, the insurance agents who sell weather index products, the venture partners who allocate catalytic capital—their incentives and biases determine whether the design works.

The material on credit risk management in the research base is candid about this. Bank managers come in two types: those who resist quantitative techniques and those who embrace them uncritically. Neither uses the models well. The first ignores the score; the second treats it as an oracle. The behavioral question is how to design the institutional context so the score informs judgment without replacing it.

This is a question for organizational economics and behavioral finance, and it has received almost no attention in the green finance literature. The chain framework's effectiveness may depend on it more than on any of the technical parameters.

15.3.7 A Research Agenda in Order of Priority

The gaps are not equally important. Table 15-2 ranks them. The ranking reflects one judgment: the additionality question determines whether the chain framework is worth the policy effort at all, and the other five questions refine a design whose basic value is already established.

Table 15-2: Research priorities

If green credit is not additional, the chain is a relabeling exercise. If it is additional, the chain is a transition mechanism. The difference is large and the evidence is not yet in.

15.4 Conclusion: What the Chain Requires

The identification–buffering–catalysis chain is not a single instrument or a single policy. It is an institutional architecture with three load-bearing functions, each with a public-good component that private markets underprovide. The policy implication is to fund the infrastructure—the rating platform, the index design, the reinsurance backstop, the fund governance—rather than the spread. The international implication is to transfer the infrastructure, not just the capital, and to build mutual recognition on a verification floor rather than waiting for full harmonization. The research implication is to measure additionality and cost-effectiveness before scaling, because the current evidence is too thin to support the scale of the policy ambition.

The county platform's 8 million yuan fixed cost is a small number against China's green credit balance. It is a large number for a county budget. The policy question is how to share it. A regional platform serving 40 counties at 200,000 yuan per county is affordable. County-by-county replication is not. The arithmetic is not glamorous. It is the kind of calculation that determines whether the chain is built or merely announced.

Designs fail in implementation. The rating model can be gamed by a firm that learns which environmental indicators the score weights and manages those indicators rather than its emissions. The insurance index can be mispriced when the meteorological record is short and the tail is fat, and the mispricing will surface as either an insurer loss or a farmer non-renewal, depending on which direction the error runs. The catalytic fund can crowd out the private capital it was meant to seed, by setting a price that private investors anchor to and then undercut. The evidence reviewed in this book is consistent with the framework's logic but does not yet demonstrate its effectiveness at scale.

The next step is not another framework. It is the additionality study: a difference-in-differences design on firm-level emissions and loan data, using the rating platform's scores as the assignment variable, run on the first province that operates a consolidated platform across its counties. That study will produce the ratio that matters—the share of green credit that would not have been extended without the chain. If the ratio is high, the architecture is justified and the policy should scale. If it is low, the policy should be redesigned before another yuan is spent. We do not know the ratio yet. That is the honest state of the field, and it is where the next round of work should begin.