Chapter 1: The Limits of Single-Channel Green Lending

In the spring of 2021, a provincial commercial bank in eastern China announced a green credit portfolio of 4.2 billion yuan. The annual report put the figure on the cover. Eighteen months later an internal review found that 61% of the portfolio was loans to a single provincial power utility, refinancing coal-adjacent transmission assets that had been reclassified under a broader green taxonomy. The other 39% sat with a handful of solar developers and one wastewater operator. The bank hit its green lending target. It had not financed a single industrial decarbonization project.

That gap between the metric and the mission is what this chapter is about. Green credit has become the default instrument of sustainable finance policy almost everywhere. Central banks fold it into prudential frameworks, development banks anchor climate strategies to it, commercial banks report it in sustainability disclosures. The instrument works for a narrow class of assets: renewable generation with contracted offtake, certified green buildings, large infrastructure with a sovereign or utility standing behind it. It works badly for the harder and larger part of the transition—steel, cement, chemicals, shipping, heavy transport, and the small and medium firms that populate industrial supply chains. Those are the activities where the green premium is highest, the technology pathways unsettled, and the information a credit committee needs simply not available in the form it requires.

Green credit performs one function well: identification. The transition needs three. When only identification exists, capital flows to what is easiest to label, and the portfolio that results looks green in aggregate while being fragile in composition.

1.1 How Green Credit Became the Default

The modern green credit framework has three layers. At the top sit voluntary principles—the Equator Principles for project finance, launched in 2003 and now adopted by more than 80 financial institutions in 40 countries, and the Green Loan Principles issued in 2018 by the Loan Market Association, the Asia Pacific Loan Market Association, and the Loan Syndications and Trading Association. These define what counts as a green loan and how proceeds are tracked. Below them sits national regulation: statistical standards, directed lending quotas, preferential risk weights. The People's Bank of China introduced green credit statistics in 2013 and later accepted green bonds as collateral for medium-term lending facility operations. The European Central Bank moved more slowly, signaling that climate risk would enter supervisory review. At the bottom sits bank-level strategy: green product lines, sustainability-linked loan structures, internal targets cascaded down to branch managers.

The architecture rests on a simple bet. Banks hold the largest pool of intermediated capital in most economies. Redirect a slice of it toward low-carbon assets and the transition gets financed at scale, without new institutions or new fiscal outlays. The instrument is familiar, the origination process already exists, and the marginal cost of adding a green category to a loan application is close to zero.

By its own terms the strategy has worked. Global green bond issuance reached $487 billion in 2023, according to the Climate Bonds Initiative. China's green loan balance passed 30 trillion yuan by the end of 2023, the largest in the world. The European Investment Bank, which issued the first climate awareness bond in 2007, has committed over €100 billion to climate action since. These are flows, not pledges.

The composition of the flows tells a different story. The Climate Bonds Initiative's 2023 sector breakdown puts energy—mostly renewable generation—at 39% of green bond proceeds, buildings at 27%, transport at 19%. Industry, which produces roughly a quarter of global CO2 emissions from energy use and process reactions, took less than 5%. Agriculture and forestry, which together with land-use change account for about a fifth of emissions, took under 2%. I will return to this breakdown once, in section 1.4, where the Chinese loan data can be set beside it.

The pattern follows from how credit risk is assessed. A solar farm with a 20-year power purchase agreement has a predictable cash flow, a well-understood technology, and a counterparty that standard tools can evaluate. A steel plant replacing a blast furnace with hydrogen-based direct reduced iron has none of those. The technology is pre-commercial at scale. Input costs depend on electrolyzer prices and renewable tariffs that will move over a decade. The output competes with imports from jurisdictions without carbon pricing. A loan officer running a standard credit model cannot underwrite that project. The model rejects it, and the officer approves another solar farm.

1.1.1 The Green Premium as the Underlying Obstacle

The concept that organizes the problem comes from Bill Gates's 2021 book How to Avoid a Climate Disaster, though the economics is older. Gates defines the green premium as the cost difference between a zero-carbon technology and its fossil alternative:

where \(C\) is the levelized cost of the output—electricity, steel, cement, a transport service—including capital recovery, fuel, and operation. A positive premium means the zero-carbon option costs more and will not be chosen by a private actor responding to price signals alone. A negative premium means the transition pays for itself.

The size of the premium varies enormously by sector. A 2022 study by CICC (China International Capital Corporation) estimated green premiums for Chinese industries. Non-passenger transport and building materials (glass, cement) came in at 141% and 138%. Mature industries—paper, non-ferrous metals, steel, passenger vehicles, power—ranged from 3% to 17%. The spread matters for policy. A 5% premium can be closed with a modest carbon price or a small subsidy. A 140% premium cannot; it requires technology cost reduction, which requires deployment, which requires financing at a scale standard credit cannot supply.

Nor is the premium a single number anyone can look up. It is a function of location, scale, input prices, and time. For green hydrogen it depends on the electrolyzer's capacity factor, the price of renewable electricity, and the cost of capital. A project in Inner Mongolia with cheap wind and a 50% capacity factor might face a 40% premium. The same technology on a coast with higher power prices and a 30% capacity factor might face 120%. A bank evaluating both has to model variables whose values no one can verify ex ante.

Here the single channel breaks down. Green credit can finance a project with a small or negative premium because the cash flow covers debt service. It cannot finance a large positive premium unless a subsidy, a carbon contract, or a guaranteed offtake price closes the gap. The instruments that close it—carbon contracts for difference, advance market commitments, price guarantees—are fiscal or regulatory. A bank cannot create them. It can wait for them to exist and then lend against them.

1.1.2 What the Instrument Was Built to Serve

Be precise about the original use case. The first green bonds came from multilateral development banks—the European Investment Bank in 2007, the World Bank in 2008—to finance renewable energy and efficiency projects in developing countries. Those projects had three features: a discrete asset with a measurable output (megawatt-hours, tons of CO2 avoided), a contracted revenue stream, usually a power purchase agreement with a utility or government, and a sponsor whose balance sheet could absorb construction risk. Proceeds were earmarked, tracked, reported. The model was project finance with a green label.

The model changed when it reached corporate issuers. Vasakronan, a Swedish real estate company, issued the first corporate green bond in 2013. Proceeds were no longer tied to a single asset but to a portfolio of eligible projects. Reporting became aggregate. Credit risk became corporate rather than project-specific. The extension was necessary for scale, and it loosened the link between the label and the activity. A utility could issue a green bond to refinance a wind farm and use the freed capital to build a gas plant. The bond was green; the balance sheet was not.

Green loans followed the same path. The Green Loan Principles define eligible categories broadly: renewable energy, energy efficiency, pollution prevention, sustainable management of living natural resources, biodiversity conservation, clean transportation, sustainable water management, eco-efficient products and technologies. A loan to a company that manufactures equipment in any of these categories qualifies. The loan need not finance deployment of the equipment; it can finance the manufacturer's working capital. Environmental additionality is assumed, not measured.

The principles were designed to be workable, and a workable standard that captures most of the easy cases beats a strict standard that captures none. The trouble is that the easy cases are not where the emissions are. Technology transformation—as distinct from technology deployment—falls outside the framework by construction, because the credit risk cannot be assessed with the tools the framework provides.

1.2 Information Asymmetry in Green Lending

A bank evaluating a loan application faces three kinds of information problem. Each is worse for green projects than for conventional ones.

Start with the project itself. For a solar farm, project-level information is measurable: irradiance data, panel degradation curves, construction contracts. For a hydrogen-based steel plant it is not. Electrolyzer performance at scale is unproven. The DRI shaft furnace has been demonstrated at pilot scale, not at the 2-million-ton-per-year scale of a typical integrated mill. Operating cost depends on electricity prices set by a market that does not yet exist. The bank cannot verify the project's economics because the economics are not yet determined.

Then the borrower. Will the counterparty repay? For a green project this is complicated by the possibility that the transition disrupts the borrower's own business model. A cement company that invests in carbon capture may become uncompetitive if a rival finds a cheaper route. A steel company that bets on hydrogen may be stranded if carbon prices rise more slowly than expected. Credit risk here covers two questions—whether the project works, and whether it remains the right bet as the transition unfolds—and standard credit models carry no variable for technology pathway risk.

Third, environmental performance. How much carbon does the project actually avoid, and is the avoidance additional? The green label is supposed to certify this, but certification rests on ex ante estimates, not ex post measurement. A building certified LEED Platinum may consume more energy than modeled because of occupant behavior. A renewable project may displace less fossil generation than assumed because of grid constraints. The bank cannot verify the environmental outcome, and the borrower has an incentive to overstate it.

The three layers interact. Project-level uncertainty makes counterparty risk harder to assess, and environmental uncertainty makes it harder to know whether the loan achieves its stated purpose. A credit committee comfortable with the first layer may still balk at the second; a regulator comfortable with the first two may question the third. The result is a conservative bias. Banks lend where all three layers can be resolved—proven technology, creditworthy sponsor, measurable outcome—which is to say, renewable energy and building efficiency.

1.2.1 Why Better Data Does Not Fix This

The information problem is not a market failure that better data will cure. It is structural.

Take the due diligence a bank would need for a green hydrogen project. It would have to assess electrolyzer technology, which requires engineering expertise most banks do not employ. It would have to model electricity prices over 15 years, which requires a view on power market evolution, renewable penetration, and the policy regime. It would have to evaluate an offtake agreement that may sit with a steel company that is itself transitioning and may not be creditworthy in a decade. It would have to assess a carbon price trajectory that depends on political decisions in several jurisdictions. Each assessment is uncertain; the uncertainties compound. A credit model built for a manufacturer with stable input costs and a diversified customer base cannot absorb this.

The standard response is a guarantee or credit enhancement. A development bank or government agency provides a partial risk guarantee, shifting some uncertainty off the commercial bank's balance sheet. This works, and it does not scale. Guarantee capacity is bounded by the guarantor's fiscal space, and the guarantor must do the same due diligence the bank cannot do. The information problem is transferred, not solved.

A second response is relationship lending. A bank that has lent to a steel company for 20 years knows its management, cost structure, and competitive position. It can extend that relationship to a green project even if the project is novel, because it trusts the borrower. Much early green lending in Europe and Japan worked this way. It serves large incumbents with long banking relationships, not new entrants—often the ones holding the most innovative technologies.

A third response is the bond market, where investors may accept more technology risk than a bank credit committee. The bond market has its own information problems. Investors rely on issuer disclosure and external review, both built on the issuer's own estimates. The market has grown fast, and its sector composition shows the same bias toward easy projects. Investors are diversifying the information problem, not solving it.

1.2.2 The Rating Gap

The asymmetry is most visible in credit ratings. A rating agency evaluates a company's ability to repay: cash flow, leverage, industry position, management quality. It does not, as a rule, evaluate environmental performance except where environmental liabilities hit cash flow—a carbon tax, a pollution fine, a stranded asset. The rating measures financial credit risk, not environmental risk.

The gap that opens is straightforward. A company with a strong environmental profile and a weak balance sheet gets a low rating and pays a high rate, even if its environmental performance reduces long-term risk. A company with a weak environmental profile and a strong balance sheet gets a high rating and a low rate, even if its environmental performance raises long-term risk. The rating omits the transition risk that green finance exists to address.

Closing the gap is not simple. Environmental performance is multidimensional—emissions intensity, energy mix, water use, waste, supply chain, governance—and aggregating it into one score depends on sector and jurisdiction. A cement company's emissions matter more than a software company's, though the software company may have the larger supply chain footprint. A rating agency would need a sector-specific model validated against realized outcomes, and those outcomes are not yet available.

There is a methodological knot as well. Credit ratings are ordinal: AAA means very low credit risk, not a 0.1% default probability. Environmental scores are cardinal: 0.5 tons of CO2 per ton of product is a measurable quantity. Mapping one onto the other is not obvious. Does a 10% cut in emissions intensity offset a one-notch downgrade? Does a high environmental score with a low credit rating beat the reverse? The agencies have been slow to answer, and the questions remain unsettled.

So green credit is priced on financial risk alone. A green loan to a strong borrower earns a small greenium—typically 1 to 5 basis points—which does nothing for a project carrying a 50% green premium. The environmental performance is a label, not a risk factor. Capital allocation follows the rating, and the rating follows financial variables. The label does not change the decision.

1.3 Risk Concentration and Maturity Mismatch

Single-channel green lending concentrates risk in two dimensions.

The sectoral dimension is visible in the bond data cited above: renewable energy and buildings take roughly two-thirds of proceeds. Not because other sectors lack green projects, but because other sectors lack projects financeable with standard credit. The green finance system ends up heavily exposed to the renewable sector and to the policy regimes that support it. Cut a subsidy or let power prices fall, and the green loan book deteriorates. The concentration is not diversified across the transition; it is concentrated in the part of the transition already underway.

Geographic concentration is less visible and just as important. Green lending clusters in jurisdictions with supportive policy—the EU, China, Japan, a few others. A bank operating where there is no carbon price, no taxonomy, and no renewable portfolio standard has little reason to originate green loans: the premium is not closed and the environmental performance is not rewarded. Capital flows to the policy environment, not to the abatement opportunity. The countries with the largest industrial emissions—India, Russia, Brazil, Indonesia—attract the least green lending.

This concentration has a systemic edge. The European Central Bank's 2020 Guide to Climate and Environmental Risks and the Network for Greening the Financial System's 2020 report on macroeconomic and financial stability both treat climate risk as a source of systemic risk. Green lending concentrated in a few sectors and jurisdictions leaves the financial system's transition exposure undiversified. A shock to renewable energy—a subsidy change, a price fall, a supply chain disruption—hits many banks' green portfolios at once. The green label, meant to signal lower risk, may signal correlated risk.

1.3.1 The Maturity Mismatch

The second structural problem is maturity.

Bank funding is short—deposits, interbank borrowing, short bonds. Assets are longer: a 10-year corporate loan, a 20-year project loan, a 30-year mortgage. This transformation is the core of banking and is regulated through liquidity ratios and mismatch limits, on the assumption that assets can be sold if funding dries up or that the central bank will supply liquidity in a crisis.

Green projects run longer than conventional ones. A solar farm lives 25 to 30 years, a hydrogen electrolyzer 20, a carbon capture retrofit 25. Loans against them typically run 15 to 20 years, longer than the average corporate loan, so the mismatch is larger. The green premium adds a second layer: a conventional loan's cash flow is fairly predictable because the borrower sells a product at an observable market price, while a green loan's cash flow depends on policy, technology costs, and carbon prices. A project profitable at €50 per ton of carbon may be unprofitable at €30, and the loan's maturity outlasts the policy horizon.

The usual fix is to match loan maturity to the offtake agreement. A solar farm with a 20-year power purchase agreement supports a 15-year loan. A hydrogen project with a 10-year offtake supports a 7-year loan, but the electrolyzer lasts 20, so the loan covers less than half the asset's life. The bank carries refinancing risk at maturity; the borrower carries the risk that the offtake will not renew at a sufficient price.

This is structural, not a pricing problem. Charge a higher rate and the green premium rises, making the project less viable. Shorten the maturity and refinancing risk rises, making it less bankable. Require a guarantee and you run into limited guarantee capacity. There is no price at which the bank willingly lends and the borrower willingly borrows, because the risk cannot be diversified within the single channel.

1.3.2 The Refinancing Cliff

The mismatch produces a specific vulnerability that has drawn little attention.

A green project is typically financed with a construction loan that converts to a term loan at commercial operation. The term loan runs 10 to 15 years against a 20- to 30-year asset. At maturity the project must be refinanced. If it is performing and policy is stable, refinancing is routine. If not, it may not happen at all.

Refinancing risk is higher for green projects because the pool of willing lenders is smaller. A conventional project can be refinanced by any bank with appetite for the sector. A green project can be refinanced only by a bank with a green mandate, and there are few of those. Let the mandate change—through regulation, strategy, or politics—and the refinancing disappears.

The data on green bond maturities shows the cliff. Green bonds carry shorter average maturities than conventional bonds because investors will not commit for the full project life. The issuer refinances earlier, which raises refinancing risk. The short maturity is a symptom of the information problem: investors do not trust long-run environmental performance, so they limit exposure to the period they can monitor.

The systemic version: if many green projects reach their refinancing dates together—likely, since they were financed in a wave—refinancing demand can exceed the supply of green capital. The result is a credit crunch in the green sector at the moment the transition can least afford one. The single channel has no mechanism to smooth the cycle; it depends on the bond market and the banks, and both are procyclical.

1.4 The Empirical Record

The evidence on green credit's effectiveness is mixed, and the mix is informative.

On the positive side, green lending lowers the cost of capital for green projects. A 2021 study by the Bank for International Settlements found green bonds trading at a greenium of 1 to 3 basis points, a modest reduction in the issuer's cost of capital. A 2022 study by the European Central Bank found that banks with a higher share of green loans had lower non-performing loan ratios, which suggests green lending is not riskier than conventional lending. Both findings support the view that green credit works for the projects it serves.

The environmental evidence is weaker. A 2023 Climate Bonds Initiative review found that only 40% of green bond issuers report on the actual environmental impact of financed projects, and the reporting is not standardized. The European Commission's Joint Research Centre, in a 2022 analysis, found that the carbon avoidance attributed to green bonds is often overstated because the projects would have been financed anyway. Additionality is uncertain.

The most informative evidence is the sector composition. If green credit were an effective instrument for industrial decarbonization, green loans would flow to steel, cement, chemicals, and heavy transport. They do not. They flow to renewable energy and buildings, where the premium is small or negative and the technology mature. The instrument does what it was designed to do. It does not do what the transition requires.

1.4.1 China's Green Credit: Scale Without Transformation

China is the useful case because it has the largest green credit market and the most developed policy framework.

The People's Bank of China introduced green credit statistics in 2013 and updated the standard in 2021. The 2021 standard is broad, covering loans to projects that reduce pollution, conserve energy, and protect biodiversity. By the end of 2023 the green loan balance exceeded 30 trillion yuan, about 12% of total loans, having doubled between 2020 and 2023.

The composition reveals the limits. The PBoC's own sector breakdown shows the largest categories are infrastructure, including clean energy, and the manufacture of green equipment. Loans to heavy industry for process decarbonization are a small fraction. The Chinese Academy of Social Sciences put the share of green loans going to technology transformation in carbon-intensive sectors at under 5% in a 2022 report on green finance development; the rest went to projects already commercially viable. The figure is consistent with the bank-level anecdote that opens this chapter, and with the Climate Bonds data: the label is doing its work, and the work is not transformation.

China also illustrates the concentration problem. The green loan book is heavily weighted to the power sector, which is dominated by state-owned utilities. The credit risk is sovereign-adjacent rather than commercial. If utility finances deteriorate—rising coal prices, falling demand, a policy shift—the green loan portfolio feels it. The concentration is not diversified across the transition; it is concentrated in the part the state is financing.

None of this makes the Chinese policy a failure. Green credit mobilized capital for renewable energy at a scale no other instrument could have reached. It did not finance the hard part of the transition, and the framework has not solved the information and risk problems that keep the hard part from being financed.

1.4.2 Europe: Taxonomy Without a Financing Mechanism

Europe took a different route, built on taxonomy and disclosure rather than directed lending.

The EU Taxonomy, adopted in 2020, defines which economic activities count as environmentally sustainable. The Sustainable Finance Disclosure Regulation, also 2020, requires financial institutions to disclose the sustainability of their products. The Corporate Sustainability Reporting Directive, adopted in 2022, requires large companies to report environmental and social performance. Together they create a common language and reduce information asymmetry.

The European model standardized the definition of green better than any other jurisdiction. It did not solve the financing problem. The taxonomy tells a bank which activities are green; it does not tell the bank whether a given project will succeed. Disclosure tells a bank what a company is doing; it does not tell the bank whether the strategy will work. The information is necessary and not sufficient.

Optionality is the second weakness. The SFDR lets institutions choose whether to consider principal adverse impacts. Most have chosen not to, or only partly. Disclosure is therefore not comparable across institutions, and the asymmetry persists. The regulation built a framework for disclosure without a mechanism for using disclosure in credit decisions.

European green bond issuance reached roughly €300 billion in 2023. The composition shows the same bias: energy and buildings dominate, industry is a small share. The European Investment Bank has been the largest issuer, and its projects are the renewable energy and efficiency projects that dominate the global market. The taxonomy formalized the composition; it did not change it.

1.5 Three Functions, and What Happens When One Is Missing

A green finance system has to perform three functions. Green credit performs one.

Identification is the first: determining which projects and companies genuinely contribute to the transition. Green credit does this through taxonomy, certification, and reporting. Identification tells you what is green. It does not tell you whether the green project is financially viable.

Risk buffering is the second: absorbing the uncertainty inherent in the transition—technology risk (will it work at scale?), policy risk (will the carbon price hold?), market risk (will the offtake price suffice?), environmental risk (will the promised reduction materialize?). A system that can absorb these risks finances projects a single bank cannot. A system that cannot absorb them finances only projects where the risks are already resolved.

Catalysis is the third: mobilizing capital for projects not yet bankable, by blending public and private resources, providing early-stage equity, creating markets for new technologies. Venture capital and development finance perform this function elsewhere in the economy. Green credit, designed for already-bankable projects, does not.

The single channel performs identification and assumes the other two are unnecessary. The assumption holds for renewable energy and building efficiency. It fails for industrial decarbonization, and the failure is one of kind: the credit model cannot underwrite technology risk at any volume of lending. A rating without insurance leaves the bank holding technology risk; insurance without a rating prices the wrong risk. Neither substitutes for the other, and neither substitutes for the catalytic capital that gets a first-of-a-kind plant built.

1.5.1 The Missing Markets

Each function corresponds to a market that is missing or underdeveloped.

The identification market is partly built. Taxonomies, certification bodies, and rating agencies assess environmental performance. The assessment is not integrated into credit decisions: a company's environmental score is not a variable in its credit rating, and a project's environmental performance is not a variable in its loan pricing. The market produces information the credit market does not use.

The risk buffering market is thin. There are insurance products for weather risk, guarantees for political risk, hedges for currency and interest rate risk. There is no instrument that buffers the technology risk of a pre-commercial green project. A bank cannot buy insurance against a hydrogen electrolyzer underperforming or a carbon capture system costing more than expected. The risk stays on the balance sheet, and the loan is declined.

The catalysis market is small relative to need. Venture capital for clean technology and development banks for infrastructure exist. BloombergNEF put global venture capital investment in climate technology at about $50 billion in 2023, against the $4.5 trillion of annual investment the International Energy Agency estimates the transition requires. The gap is not closing.

1.5.2 What Policy Has Tried, and Why It Has Not Worked

The policy response to the single channel's limits has been to expand the definition of green and increase the volume of green credit. Both are inadequate. A broader definition lets more projects carry the label without making them bankable. A larger volume sends more capital to the easy projects without creating the risk buffering or the catalysis the hard projects need.

The alternative is to build the missing markets: a rating system that folds environmental performance into credit risk, an insurance system that buffers risks banks cannot absorb, and an intermediary system that catalyzes capital for early-stage technologies. A rating system that identifies a green project is useful only if the insurance system can buffer its risk and the intermediary system can catalyze its financing. An insurance system that buffers risk is useful only if the rating system can identify the risk and the intermediary system can finance the project. Each link is a precondition for the next, and the links are where the current system fails.

1.6 Method and Evidence

1.6.1 The Evidence Base

The book draws on three bodies of evidence.

The academic literature on green finance, credit risk, and insurance identifies mechanisms and empirical regularities. It is less useful for institutional detail, which tends to be country-specific and poorly captured in cross-country studies.

Policy documents and data from central banks, regulators, and international organizations—the Bank for International Settlements, the European Central Bank, the People's Bank of China, the Network for Greening the Financial System, the Climate Bonds Initiative—supply the quantitative picture. Definitions are not always consistent across jurisdictions, but the broad patterns hold.

The author's prior research provides the micro-foundations: credit rating for technology enterprises, weather index insurance, and the impact of agricultural insurance on farm income. The rating work shows how environmental performance can enter a credit model. The insurance work shows how risk can be buffered. The agricultural work shows how buffering affects income stability and credit access.

1.6.2 What the Book Does Not Argue

The book does not argue that green credit should be curtailed. The renewable and efficiency projects it has financed are necessary, and the capital is real. Without the green bond market and the green loan framework those projects would have been financed more slowly and at higher cost. The critique concerns sufficiency.

It does not argue that banks behave irrationally. A loan officer who declines a hydrogen steel project and approves a solar farm makes a defensible decision with the tools available. The credit model was built for a different class of assets, and the officer applies it correctly. The tools are the problem.

It does not argue that one new instrument solves everything. No single rating methodology, insurance product, or fund structure finances the whole transition. The argument is that a system of complementary instruments is needed, designed as a system rather than as separately optimized parts.

Nor does it argue that the transition can be financed without public policy. The green premium is a market failure, and market failures require policy responses: carbon pricing, subsidies, standards, public investment. The argument is that policy should build the financial system's capacity to identify, buffer, and catalyze, not merely expand the volume of green credit.

1.6.3 The Audience

Three audiences. The policy community—central bankers, regulators, finance ministry officials, development bank staff—gets a framework for the division of labor between instruments and a set of design principles for the missing markets. The financial industry—risk managers, credit analysts, insurance underwriters, asset managers—gets methods for incorporating environmental performance into credit and insurance decisions, and an account of where those methods stop working. The research community gets a synthesis of the evidence and a set of testable hypotheses, stated in the concluding chapter.

1.7 Where the Single Channel Breaks: A Model

The argument so far has been verbal. A small model makes the mechanism precise and yields a comparative static the prose cannot state.

Consider a bank allocating a fixed portfolio \(L\) across projects indexed by \(i\). Each project has expected return \(r_i\), default probability \(p_i\), and loss given default \(\ell_i\). The bank maximizes expected profit subject to a capital constraint:

where \(x_i\) is the loan to project \(i\). Without a capital constraint the bank funds the highest risk-adjusted return; with one, it funds the highest return per unit of capital consumed.

Environmental performance enters only through \(p_i\) and \(\ell_i\). A green project with more stable cash flow has a lower \(p_i\) and gets more capital. A green project with unproven technology has a higher \(p_i\) and gets less. The tons of CO2 avoided do not enter the objective function. The bank maximizes risk-adjusted financial return, not environmental outcome. A project avoiding a ton of CO2 at $100 is treated identically to one avoiding a ton at $10, provided the financial risk and return match. The bank has no incentive to prefer cheaper abatement, and none to finance expensive abatement that is also riskier.

The green premium enters as a constraint on cash flow. Let \(C_i\) be cost per unit of output and \(P_i\) the market price. The project is viable if \(P_i \geq C_i\). For a green project \(C_i = C_i^{\text{fossil}} + G_i\), where \(G_i\) is the premium. If the premium is positive and the market price is set by the fossil marginal supplier, the green project is not viable without support. The bank will not finance it because cash flow does not cover debt service. Raising the price (carbon tax, standard) or lowering the cost (subsidy, technology improvement) is outside the bank's control. Hence green credit flows to projects with a small or negative premium—solar where levelized cost sits below the fossil marginal cost, efficiency retrofits where savings exceed upfront cost—and the large-premium projects go unfinanced however green they are.

Now add the information constraint. The bank does not know \(p_i\); it estimates \(\hat{p}_i\) with standard error \(\sigma_i\), and it is risk-averse. It treats the project as if the default probability were \(\hat{p}_i + \lambda \sigma_i\), where \(\lambda\) is its risk aversion. The penalty is larger for green projects because \(\sigma_i\) is larger. The interpretation is direct: an uncertain technology is more likely to have a default probability well above the estimate, and the bank's capital is exposed to that tail. The bank demands a higher expected return, the hurdle rate rises, and more projects are excluded—not because they are bad, but because the bank cannot distinguish good from bad.

The comparative static is the useful part. The hurdle rate for a project is the return \(r^*\) that satisfies

where \(\rho\) is the cost of capital. Differentiate with respect to \(\sigma\):

The hurdle rate rises linearly in the standard error of the default estimate, at a rate equal to the bank's risk aversion times loss given default. For a sector where \(\ell\) is high—steel, cement, chemicals, where a failed plant leaves little salvage value—a small increase in estimation uncertainty produces a large increase in the required return. This is why the single channel fails hardest exactly where the assets are most specific and the technology least proven. It is also why better measurement has leverage: reducing \(\sigma\) lowers the hurdle rate at rate \(\lambda \ell\), and the leverage is greatest in the sectors with the highest loss given default.

The maturity constraint closes the model. Let project life be \(T_i\) and loan maturity \(M_i < T_i\). The project must be refinanced at \(M_i\), and the probability that refinancing is available is \(q_i\). Expected cash flow falls by \((1 - q_i)\) times the cost of a forced sale or restructuring. The reduction is larger for green projects because \(q_i\) is smaller. Longer maturity would ease the borrower's problem and raise the bank's liquidity risk; a guarantee would help and is capacity-constrained; a higher rate would raise the premium and sink the project. The constraints interact, and no combination satisfies all of them.

The equilibrium is a portfolio concentrated in projects with small or negative premiums, low estimation uncertainty, and short maturities—renewable energy and building efficiency. Large-premium, high-uncertainty, long-maturity projects are excluded. The exclusion is the rational response of a financial system asked to perform a function it was not designed for, and it cannot be fixed by adjusting the channel's parameters. More volume does not change the composition, because composition is set by the risk-return profile. A broader definition does not change it either, because the definition is not the binding constraint. The binding constraint is the risk-return profile, and it moves only if the risk moves (insurance, guarantees) or the return moves (subsidies, carbon pricing).

1.8 Three Sectors, Three Failures

The model is abstract. The failure is concrete, and it looks different by sector.

1.8.1 Steel: The Technology Pathway Problem

Steel is the hardest of the hard-to-abate sectors. Global production is about 1.9 billion tons a year, roughly 7% of global CO2 emissions. The blast furnace–basic oxygen route uses coal as both fuel and reducing agent and emits about 1.8 tons of CO2 per ton of steel. Alternatives—electric arc furnace with scrap, hydrogen-based direct reduced iron, carbon capture—can cut emissions 50% to 90%, each with a different cost and risk profile.

A bank evaluating a green steel project faces a pathway problem. The sponsor may propose hydrogen DRI, demonstrated only at pilot scale. The largest operating plant, in Sweden, produces about 1.5 million tons a year and is not yet cost-competitive with the blast furnace. The bank cannot verify the economics because they depend on electrolyzer cost, renewable electricity price, and carbon price, all uncertain.

The premium is large. The CICC study estimated 3% to 17% for steel as a whole, but hydrogen-based steel probably runs 30% to 50% at current input prices—too large for a modest carbon price to close, too uncertain for a standard credit model to underwrite. The bank declines, and the project is financed, if at all, by a development bank or a government-backed fund. Public institutions absorb the technology risk because their mandate is to promote the transition, not to maximize risk-adjusted return. Commercial banks then free-ride on the public due diligence: once the technology is proven and the first plants operate, they finance the next wave. The single channel works for the second wave and not the first.

1.8.2 Agriculture: The Weather Risk Problem

Agriculture differs. The technology is mature, the green premium is often negative because regenerative practices cut input costs, and the projects are small. The barrier is weather.

A farmer adopting a green practice—cover cropping, reduced tillage,precision irrigation—takes on a risk that the practice fails in a bad year. Cover crops compete with the main crop for water; reduced tillage can raise pest pressure; precision irrigation may not pay back if rainfall is adequate. Farm income is already variable, and the green practice adds another source of variance. The bank lending to the farmer faces the same variance and prices the loan accordingly.

The result is that green agricultural practices are underfinanced even where they are profitable on average. The farmer cannot bear the risk and the bank cannot diversify it. The single channel fails because the risk is idiosyncratic and correlated with weather, the one variable neither party controls.

More credit does not fix this. Insurance does. A weather index policy that pays out when rainfall falls below a threshold protects the farmer against the downside of the green practice. It cuts the variance of farm income, which cuts the bank's risk, which makes the loan bankable. Insurance and credit are complements here, and the single channel has no insurance to offer, so the credit does not flow.

1.8.3 Forestry: The Long-Horizon Problem

Forestry has the longest horizon of the three. A forest matures over 20 to 50 years, and returns arrive at harvest. Fire, pest, disease, and weather compound across that horizon. The green premium is often negative—a forest sequesters carbon, and the carbon has value—but the value is realized only if the forest survives to maturity.

A bank evaluating a forestry project faces a maturity mismatch more severe than in any other sector. Loan maturity is typically 10 to 15 years against a project life of 30 to 50. The bank cannot lend for the full life, and the borrower cannot repay within the loan term. The project must be refinanced, and refinancing risk is high because the forest's value depends on carbon and timber prices at the refinancing date, both uncertain.

The single channel fails because the horizon exceeds what a bank balance sheet can carry. The fix is a different institution—a forestry fund with a 30-year horizon, or a government-backed carbon contract guaranteeing a price for sequestered carbon. With the long-term risk buffered by a contract or fund, the bank can finance the early stages. Without the buffer, the loan is declined.

1.9 What the Three Cases Have in Common

The steel case is a technology pathway problem: the solution is a rating method that can assess technology readiness alongside environmental performance, plus a public institution to absorb the risk until the technology is proven. The agriculture case is a weather risk problem: the solution is an insurance product that buffers idiosyncratic risk, plus a credit model that recognizes the buffering. The forestry case is a horizon problem: the solution is patient capital that matches the project's life.

Three sectors, three different binding constraints, one shared structure. In each case the project is green by any reasonable definition, the social return is positive, and the private credit channel declines it. In each case the missing piece is an institution that performs a function the bank cannot: assess technology readiness, buffer weather risk, supply patient capital. And in each case the missing institution is connected to the credit channel by a link that does not currently exist. A rating that no credit model reads is decoration. An insurance policy that no lender recognizes as risk mitigation is a standalone product. A fund that finances a first plant without a pathway to commercial refinancing is a one-off grant with extra steps.

1.10 The Institutional Landscape and Its Disconnections

Before the book turns to the ecosystem in Chapter 2, it helps to see which institutions already occupy the green finance space and where the connections fail. The space is crowded. The crowding is part of the problem: the institutions were built separately, optimized separately, and never wired together.

Three gaps stand out, each tied to a named institution and a transaction that failed because of it.

The first gap runs between the rating agencies and the credit committees. In 2022 a Chinese provincial bank's credit committee reviewed a loan to a mid-sized cement producer for a waste-heat recovery and alternative-fuel retrofit. The project had a credible engineering study, a 12-year payback, and a carbon reduction of roughly 180,000 tons a year. The borrower carried a rating two notches below the bank's threshold for unsecured term lending. The bank's internal environmental score for the borrower was in the top decile of its portfolio. The credit model read the rating and not the environmental score, because the environmental score had no mapping into the probability-of-default estimate. The loan was declined. The retrofit was later financed by a local government guidance fund at a rate the bank could not match, which is to say the project was financed by the state's balance sheet rather than the credit system's.

The second gap runs between the insurers and the lenders. Weather index insurance for agriculture has been sold in China since 2007, and the author's prior work on it shows measurable effects on income stability. The product has not been wired into agricultural lending. A cooperative in Heilongjiang that bought index cover for drought still could not use the policy as collateral or as a risk mitigant in its loan application, because the bank's credit model had no line for insurance coverage. The cooperative's loan was priced as if the drought risk were uninsured. The insurance reduced the variance of the cooperative's income and did nothing for its cost of credit, because the two institutions—insurer and lender—had no protocol for sharing the policy's effect.

The third gap runs between the catalytic institutions and the commercial banks. A first-of-a-kind hydrogen DRI pilot in northern China was financed in 2021 by a combination of a development bank loan and a government guidance fund. The plan assumed that once the plant operated for two years, commercial banks would refinance the construction debt. They did not. The commercial banks had no rating methodology for the technology, no insurance product for its performance risk, and no way to price the refinancing. The development bank carried the exposure past its intended exit. The catalysis happened; the handoff did not.

Each gap is a missing link between two institutions that both exist. The rating agencies produce environmental assessments that credit committees cannot read. The insurers produce risk mitigation that lenders cannot price. The catalytic institutions produce first-of-a-kind assets that commercial banks cannot refinance. The rest of the book works on these links.

1.11 Conclusion: What the First Channel Has Done, and What It Cannot Do

Green credit is the first channel of green finance, and it has done what a first channel can do. It mobilized capital for the projects that were already bankable, and it built the infrastructure—taxonomies, reporting standards, the bond market—that the rest of the system can use. The 30 trillion yuan of green loans in China, the €300 billion of green bonds in Europe, the $487 billion of global green bond issuance are real achievements.

The limits are structural. The channel can finance a small green premium and not a large one. It can finance proven technology and not unproven technology. It can finance a short horizon and not a long one. The credit model cannot underwrite technology risk at any volume of lending, and no expansion of the definition or the volume changes that.

The evidence is in the composition. Green credit flows to renewable energy and buildings and does not flow to steel, cement, chemicals, agriculture, or forestry. The sectors that need the most capital get the least. The pattern holds across jurisdictions and has not shifted as volume has grown.

The explanation is in the functions. Green credit identifies, and it identifies well. It does not buffer risk, because the banking system cannot absorb the technology, weather, and horizon risks the hard projects carry. It does not catalyze, because the banking system finances bankable projects rather than creating them. The missing functions are missing markets, and the missing markets are missing links between institutions that already exist.

The policy implication is a shift from expanding the first channel to building the other two and wiring them to the first. The shift is not a rejection of green credit. It is a recognition that green credit is one instrument among several and that the several must be designed together. The next chapter opens with a declined loan—a hydrogen steel project that a commercial bank turned down in 2023—and follows the file through the institutions that might have changed the decision. The structure of the book emerges from that file.