Chapter 2: Information Asymmetry, Credit Rationing, and the Limits of Green Lending

The loan officer's name was Chen. She worked at a city commercial bank in Ningbo, and on a Tuesday morning in March she received an application from a company that made industrial heat pumps. The firm had 43 employees, three patents on high-temperature compressors, and a contract to retrofit the district heating system of a nearby industrial park. It wanted 12 million yuan for five years. Chen read the file twice. The technology was plausible. The contract was real. The balance sheet showed two years of losses and a third year of thin profit, which is normal for a firm that has spent everything on tooling. What Chen could not do was price the loan. She had no way to estimate the probability that the heat pump would perform to specification across a fifteen-year operating life, no way to value the carbon certificates the park expected to generate, and no way to know whether the compressor patents would survive a validity challenge. Her credit model took financial statements and collateral. It did not take physics.

She rejected the application. Not because she thought the project was bad. Because she could not write a number in the expected-loss field that she could defend to her risk committee. The rejection took four minutes. The firm spent nine months finding a different lender, a provincial bank with a green finance pilot quota, which priced the loan at 6.8 percent against a benchmark of 4.9 percent.

This chapter is about what happened in those four minutes. The loan officer's inability to price the loan is not a story about her competence. It is a story about a structural feature of credit markets: when the lender knows less about the borrower's project than the borrower does, the lender does not simply charge more. Often it declines. And when it declines, it declines in a pattern that is not random. The pattern falls hardest on exactly the firms that a green transition needs to finance.

The claim needs precision. It is common to say that green firms face a financing gap because banks are short-sighted or because green projects are risky. Both statements are too loose to be useful. The argument here is narrower and more mechanical. Information asymmetry between borrower and lender produces credit rationing. Credit rationing is not a uniform tax on all borrowers; it is a selection mechanism that sorts by the lender's ability to verify, not by the borrower's quality. Green innovators sit at the wrong end of that sort because their assets are intangible, their cash flows are back-loaded, and their environmental performance is measured in units that credit systems do not yet read.

The chapter proceeds in three movements. I set out the mechanics of information asymmetry in green finance: where the asymmetry sits, why it is worse for green projects than for conventional ones, and how it produces both adverse selection and moral hazard. I then connect that asymmetry to credit rationing, drawing on the Stiglitz-Weiss framework and testing it against the specific case of green innovators. The third movement presents evidence: case material from Chinese green credit practice, data on rejection rates and loan pricing, and the policy experiments that have tried to loosen the constraint. The conclusion is specific rather than optimistic: single-channel green credit fails because the information problem is upstream of the lending decision, and no amount of subsidized interest rates fixes an information problem.

2.1 Information Asymmetry in Green Finance

2.1.1 What the lender does not know

Start with the basics of a loan. A bank hands over money today in exchange for a promise of repayment plus interest. The promise is enforceable, but enforcement is expensive and slow, so the bank's real protection is its estimate of the borrower's ability and willingness to repay. That estimate is built from information. When the information is good, the loan is priced close to the risk-free rate plus a spread that reflects expected loss. When the information is bad, the spread widens, and beyond some point the loan is not made at all.

Information asymmetry means the borrower knows more about the project than the lender does. This is not a market failure in the sense of a missing market. It is a structural condition of any credit transaction. The relevant question is not whether asymmetry exists but where it sits and how much it costs to reduce.

For a conventional manufacturing loan, the asymmetry is manageable. The lender can inspect the factory, audit the receivables, check the order book against customs records, and value the collateral against a resale market. The borrower still knows more, but the gap is narrow enough that a loan officer can close it with a site visit and a financial statement analysis. The cost of reducing the asymmetry is a few thousand yuan and two weeks.

For a green technology loan, the asymmetry is wider and the cost of closing it is higher. Consider the heat pump firm. The lender would need to assess:

• whether the compressor design meets its claimed coefficient of performance at the operating temperatures of the specific industrial park;
• whether the patents are valid and whether they would survive a challenge from an incumbent manufacturer;
• whether the industrial park's carbon accounting methodology would be accepted by the verification body that issues the certificates;
• whether the park's commitment to buy heat over fifteen years is backed by anything more than a memorandum;
• whether the firm's management has the manufacturing experience to move from prototype to volume production.

Each of these questions requires a specialist. The bank employs credit analysts, not thermodynamic engineers. The cost of answering all five is not two weeks and a few thousand yuan. It is six months and a consultancy fee that may exceed the interest income on the loan. So the bank does not answer them. It declines, or it prices the loan at a spread that reflects its uncertainty rather than the project's actual risk.

Call this the verification cost gap. The asymmetry is a difference in the cost of closing the knowledge gap, and that cost scales badly against loan size. For conventional loans, it is low relative to the amount lent. For green technology loans, it is high relative to the amount lent, especially for small and medium enterprises where the loan is 10 or 20 million yuan and the technical due diligence is a fixed cost.

2.1.2 Why green projects are worse

It would be a mistake to treat "green" as a single category. A loan to a utility building a solar farm is not the same as a loan to a startup making heat pumps. The information problems differ. Three types of green project need separation, because the asymmetry operates differently in each.

Type A: Green retrofits of existing assets. A cement plant installs waste-heat recovery. A textile mill switches to a closed-loop water system. The technology is mature, the engineering is standard, and the borrower has an operating history. The asymmetry here is modest. The lender can verify the technology against installed examples elsewhere. The main uncertainty is whether the energy savings will materialize as projected, which is a performance risk rather than an information problem. Type A projects are the easiest to finance and, not coincidentally, the ones that green credit guidelines were designed around.

Type B: Green infrastructure with long-lived assets. A solar farm, a wind project, a forest plantation. The asset is tangible and the cash flows are contracted, often with a state-owned offtaker. The asymmetry is about the long-run performance of the asset and the counterparty risk of the offtaker. These projects are financeable, but they require project finance structures that most Chinese commercial banks have only recently learned to use. The constraint is capability, not information per se.

The heat pump firm belongs to a third category. So does a company making solid-state batteries, or a firm developing a new membrane for industrial wastewater treatment. Here the asset is intangible, the cash flows are back-loaded, the technology is unproven at scale, and the environmental benefit depends on a measurement methodology that may not yet exist. This is where information asymmetry bites hardest, and this is where the green transition's long-run productivity gains will come from. Call it Type C: green technology innovation.

The distinction matters because policy tends to conflate the three. Green credit targets are usually set in aggregate yuan, which means banks meet them by lending to Type A and Type B projects, where the information problem is manageable. Type C firms are left out, not because they are excluded by rule but because they fail the pricing test. A green credit target that counts a loan to a solar farm and a loan to a battery startup as equivalent creates an incentive to do the easy thing.

Why is Type C worse? Four reasons, each of which maps onto a specific feature of the credit process.

Intangibility. A bank's credit model is built around collateral. Land, buildings, equipment, receivables—these can be valued and seized. A patent can be seized too, but its liquidation value is close to zero unless there is a buyer who can use it. The heat pump firm's most valuable assets were three patents and an engineering team. Neither appears on the balance sheet in a form a credit model can use. The firm's book value was 8 million yuan against a 12 million yuan request. On a collateral-based lending model, the loan was dead on arrival.

Back-loaded cash flows. Green technology firms typically spend years on R&D and tooling before revenue arrives. The credit model discounts future cash flows, but it discounts them heavily when they are uncertain and distant. A firm with two years of losses and a projected break-even in year four looks, in a standard credit scorecard, like a firm in distress. The scorecard cannot distinguish between a firm that is losing money because its product is failing and a firm that is losing money because it is investing in a product that will succeed. Both look the same in the financial statements.

Measurement uncertainty. The environmental benefit of a green project is often the justification for preferential treatment—a lower interest rate, a subsidy, a guarantee. But the benefit is measured in units that are not standardized. A tonne of CO2 avoided is a well-defined unit in principle, but the accounting methodology for a specific project may be contested. The heat pump's carbon certificates depended on a baseline that the park and the verification body had not agreed. If the lender cannot verify the environmental benefit, it cannot price the green discount, and the loan reverts to a conventional (higher) rate.

Technological risk that the lender cannot assess. A credit analyst can read a balance sheet. She cannot evaluate whether a solid-state battery design will survive the transition from lab to pilot line. Here the asymmetry concerns the lender's competence, not the borrower's honesty. No amount of disclosure fixes it, because the lender would not understand the disclosure even if it were complete. This is the hardest form of asymmetry, and it is the one that matters most for Type C.

2.1.3 Adverse selection

The standard treatment of information asymmetry in credit markets goes back to Akerlof's market for lemons and was applied to credit by Stiglitz and Weiss in 1981. The logic is simple and worth restating because it drives everything that follows.

Suppose a bank cannot distinguish between good and bad borrowers. It charges an interest rate that reflects the average quality of the pool. Good borrowers, who know they are good, find the rate too high and exit. Bad borrowers, who know they are bad, find the rate acceptable because they expect to default. The pool worsens. The bank raises the rate. More good borrowers exit. The process continues until the market clears at a rate that only bad borrowers accept, or the market collapses entirely.

This is adverse selection, and it operates in green finance with a specific twist. The twist is that the "good" green borrowers are not just the ones with low default risk. They are the ones whose environmental benefit is real and verifiable. A firm with a genuine green technology and a firm with a greenwashing story look identical to a lender that cannot verify the technology. The lender charges a rate that reflects the average, which drives out the genuine firm and leaves the greenwasher. The greenwasher is more likely to default because its technology does not work, which confirms the lender's belief that green loans are risky, which raises the rate further.

I have seen this cycle in the Chinese green bond market. In 2016 and 2017, a number of issuers labeled bonds as green without a credible verification framework. When some of those issuers ran into financial trouble, the label lost credibility. Investors began to demand a green premium rather than pay one. The market did not collapse, but it stratified: issuers with credible third-party verification could still borrow at a discount, while issuers with self-declared green labels paid a penalty. The stratification is the market's response to asymmetry. It works, but it excludes the firms that cannot afford verification, which are disproportionately the small innovators.

There is a second layer of adverse selection specific to policy-driven green credit. When the central bank sets a green credit target and offers a subsidized funding rate to banks that meet it, the banks have an incentive to classify existing loans as green. The classification is based on a catalog of eligible projects, and the catalog is broad enough that a loan to a coal plant's efficiency upgrade can qualify. The result is that the subsidy flows to projects that would have been financed anyway, while the genuinely new green projects—the Type C firms—remain unfinanced. The policy selects for the borrowers who are easiest to classify, not the ones who need the subsidy most. This is adverse selection induced by the policy design, and it is one reason why green credit volumes have grown much faster than green technology investment.

2.1.4 Moral hazard

Adverse selection happens before the loan is made. Moral hazard happens after. Once the borrower has the money, she has an incentive to take actions that benefit herself at the lender's expense. In green finance, moral hazard takes three forms.

The first is the standard form: the borrower diverts the loan to a higher-risk, higher-return use. A firm borrows at a green rate to install a wastewater treatment system and uses the money to buy inventory for a speculative product line. The lender cannot monitor the use of funds continuously, and the green label gives the borrower a cover story. This is not unique to green finance, but the green label makes it easier because the lender's monitoring capacity is already stretched by the technology it does not understand.

The second form is greenwashing after disbursement. The borrower commits to a set of environmental performance targets—emissions reductions, water savings, certification—and then fails to meet them. The lender has no mechanism to enforce the targets because they are not written into the loan covenant in an enforceable way. Even if they were, the lender would need to monitor the borrower's environmental performance, which requires the same technical capacity it lacked at origination. The result is that the green commitments are, in most loan contracts, aspirational.

The third form is specific to the structure of green subsidies. When a project's revenue depends on a subsidy or a carbon credit price, the borrower has an incentive to overstate the environmental benefit to capture more subsidy. The lender, if it relies on the subsidy revenue to service the debt, is exposed to the risk that the subsidy is clawed back or the credit is invalidated. This is a moral hazard created by the measurement problem: the borrower controls the measurement, and the lender cannot verify it.

The mechanism deserves precision. Moral hazard does not require dishonesty. It requires a divergence between the borrower's incentives and the lender's. A firm that genuinely intends to meet its environmental targets may still cut corners when it faces a cash squeeze, because the penalty for missing the target is zero and the cost of meeting it is real. The lender, who priced the loan on the assumption that the target would be met, bears the cost. This is a standard principal-agent problem, and it is worse in green finance because the agent's action (environmental performance) is harder to observe than the standard action (financial performance).

2.1.5 The asymmetry is not symmetric

One more point before moving to rationing. The information asymmetry in green finance runs in both directions, but not equally. The borrower knows more about the technology and the project's environmental performance. The lender knows more about the credit market, the borrower's peer group, and the probability of default conditional on financial characteristics. The borrower may also know less about the lender's decision process than she thinks, which is why firms often misjudge which loans will be approved.

The relevant asymmetry for credit rationing is the one that affects the lender's ability to price. That is the borrower's private information about project quality and environmental performance. The lender's private information about default probabilities is real but it does not cause rationing in the same way; it causes mispricing that the market corrects over time. The borrower's private information causes rationing because the lender cannot correct it without incurring a cost that exceeds the loan's value.

This asymmetry has a temporal dimension. At origination, the borrower knows more about the technology. After origination, the borrower knows more about the project's actual performance. The lender's information disadvantage grows over time if it does not invest in monitoring. For a fifteen-year heat pump loan, the lender would need to monitor performance for fifteen years, which is a commitment most banks are not structured to make.

2.2 Credit Rationing and Green Firms

2.2.1 The Stiglitz-Weiss framework

The canonical model is Stiglitz and Weiss (1981), "Credit Rationing in Markets with Imperfect Information." The paper's central claim is that a bank may rationally choose to ration credit rather than raise the interest rate, because raising the rate worsens the pool of applicants through adverse selection and encourages riskier behavior through moral hazard. The bank's expected return is not monotonic in the interest rate. It rises up to a point and then falls. Beyond that point, the bank prefers to lend less at a lower rate than more at a higher rate.

The model is usually presented with a simple diagram. I will not reproduce the full derivation, but the key result is worth stating in a form that connects to the green case. Let \(r\) be the interest rate, \(\bar{R}\) the bank's expected return, and \(\theta\) the quality of the borrower pool. The bank's return is:

where \(\pi(r)\) is the probability of repayment, \(L(r)\) is the loan volume, and \(\rho\) is the recovery rate. The first term is the interest income on performing loans; the second is the loss on defaulting loans. As \(r\) rises, \(\pi(r)\) falls because good borrowers exit and bad borrowers stay. The product \(r \cdot \pi(r)\) may rise or fall. When it falls, the bank's return falls, and the bank prefers to ration.

The implication for green finance is direct. If green projects have a lower \(\pi(r)\) at any given \(r\)—because the lender cannot verify quality, so the pool is contaminated by greenwashers—then the bank's return curve peaks at a lower \(r\) and a lower \(L\). The bank rations at a lower volume. The rationing is not a sign that green projects are bad. It is a sign that the lender cannot tell good green projects from bad ones, so it treats them all as average and lends to fewer of them.

A refinement the original model does not emphasize: Stiglitz-Weiss assumes the bank can choose \(r\) and \(L\). In practice, Chinese banks also choose the collateral requirement and the loan tenor. These are additional screening devices. A bank that cannot price the technology can still lend if it takes enough collateral. The heat pump firm had 8 million yuan of book value against a 12 million yuan request, so it failed the collateral test. A firm with more tangible assets would have passed, even if its technology was weaker. The rationing therefore selects on collateral, not on quality. This is the mechanism by which green innovators are disproportionately excluded: they are asset-light by nature, so they fail the one screening device the bank can operate.

2.2.2 The rationing pattern

If the theory is right, we should observe a specific pattern in the data. Green firms should be rejected at higher rates than comparable non-green firms, and the rejection should be concentrated among firms with intangible assets and back-loaded cash flows. The rejection should not be correlated with the actual environmental performance of the project, because the lender cannot observe it. And the rejection should be partially offset by policy interventions that reduce the lender's information cost.

The available evidence is consistent with this pattern, though it is patchy. I will present what I have and be explicit about its limits.

Rejection rates. A 2021 survey of 312 small and medium enterprises in Zhejiang and Jiangsu, conducted by a research team at a provincial university, found that firms identifying as "green technology" had a loan rejection rate of 38 percent, against 22 percent for a matched sample of conventional manufacturers. The gap persisted after controlling for firm size, age, and profitability. The authors attributed the gap to "the absence of verifiable environmental performance data," which is the polite way of saying the lenders could not assess the projects. The survey has limits: it is a single study, the sample is not nationally representative, and "green technology" is self-reported. The 16-percentage-point gap is suggestive, not definitive. But it is consistent with the theoretical prediction, and it is consistent with what loan officers told me in interviews.

Pricing. When green firms do get loans, they pay more. The heat pump firm's 6.8 percent against a 4.9 percent benchmark is a 190-basis-point spread. A 2020 study of green bond pricing in China found that green bonds issued by firms with third-party verification carried a 12-basis-point discount to comparable conventional bonds, while green bonds without verification carried a 5-basis-point premium. The verification is the variable that matters. The market is not pricing "green"; it is pricing "verified green."

Tenor. Green technology loans are shorter. The median tenor for the green technology firms in the Zhejiang-Jiangsu survey was 2.1 years, against 3.8 years for conventional manufacturers. This is a problem because green technology projects often need five to seven years to reach cash-flow breakeven. A two-year loan forces the firm to refinance at exactly the point when its cash flow is weakest. The refinancing risk is a form of rationing: the firm gets credit, but not the credit it needs.

Collateral. The survey found that 71 percent of green technology loans required real estate collateral, against 54 percent for conventional loans. The higher collateral requirement is the bank's response to the information gap. It is also the mechanism that excludes asset-light firms.

These four patterns—higher rejection, higher pricing, shorter tenor, more collateral—are the observable signature of credit rationing under information asymmetry. They are not independent. They are four faces of the same decision: the bank cannot price the technology, so it protects itself through quantity, price, tenor, and collateral.

2.2.3 Why rationing falls on the transition, not just the green

There is a further twist that matters for the book's overall argument. The rationing does not just affect firms that are already green. It affects firms that are trying to become green. A steel mill that wants to install a hydrogen-based direct reduction line is a conventional firm with a long operating history, but the new technology is as opaque to the lender as the heat pump startup's compressor. The mill can get a loan for its existing operations. It cannot get a loan for the transition.

This is the distinction between green finance and transition finance that Chapter 3 will develop. For now, the point is that the information asymmetry is not a niche problem for green startups. It is a general problem for any borrower whose project involves a technology or a process that the lender cannot verify. The green transition requires a large volume of exactly this kind of investment. The International Energy Agency's 2021 net-zero scenario estimated that 35 percent of the emissions reductions to 2050 would come from technologies that were not yet commercially mature. Those technologies are, by definition, the ones that lenders cannot price.

The rationing therefore has a macro dimension. If the financial system systematically underfinances unproven green technologies, the transition will be slower than the engineering allows. The constraint is not the availability of capital in aggregate. Chinese banks have ample liquidity. The constraint is the conversion of that liquidity into loans for projects whose risk cannot be assessed.

2.2.4 The counter-argument: maybe the rationing is correct

I should address the objection that the rationing is efficient. If a project's risk cannot be assessed, maybe the project is too risky to finance. The bank's refusal is a market signal that the technology is not ready. On this view, the green technology firms that fail are the ones that should fail, and the ones that succeed will find financing because their technology will eventually become verifiable.

This argument has force, and I do not want to dismiss it. Some green technology firms are bad investments, and the lenders who reject them are doing their job. The problem is that the rejection is not correlated with quality. The lender rejects the good and the bad together, because it cannot tell them apart. The market test is not "is this technology good?" but "is this technology legible to a credit analyst?" Those are different questions, and the gap between them is where the financing gap lives.

The evidence for this is the firms that do get financed through non-bank channels. The heat pump firm eventually got its loan from a provincial bank with a green pilot quota, which had a dedicated technical advisory team. The team spent three weeks evaluating the compressor design. The loan was priced at 6.8 percent, which is high, but it was made. The project is now in operation and the park has extended the contract. The technology was financeable. It just required a lender with the capacity to verify it.

The counter-argument also runs into a selection problem. If the market's rejection is efficient, then the firms that survive should be the high-quality ones. But the firms that survive are the ones with collateral, not the ones with the best technology. The selection is on the wrong variable. A firm with a mediocre technology and a factory can get a loan. A firm with a breakthrough technology and a rented office cannot. Over time, this selects for asset-heavy firms and against innovation, which is the opposite of what a green transition needs.

2.2.5 The rationing multiplier

There is a further effect that the standard model does not capture. Credit rationing in green technology has a multiplier. When a green technology firm cannot get a loan, it cannot scale. When it cannot scale, its technology remains unproven. When its technology remains unproven, the next lender faces the same information gap. The rationing reproduces the conditions that caused it.

This is a dynamic version of the adverse selection problem. The first cohort of green technology firms faces the highest information asymmetry because there are no installed examples to point to. If they are rationed, there are no installed examples for the second cohort either. The market can get stuck in a low-information equilibrium where no green technology is ever financeable because none has ever been financed.

The way out is through demonstration projects. A single installed heat pump at a reference site gives the next lender something to verify against. The cost of verification falls for the second borrower. This is a public-good argument for subsidizing the first deployment of a new technology, and it is the logic behind many of the policy experiments I will discuss in the next section. The subsidy is not primarily about the cost of capital. It is about buying down the information cost for the next lender.

This can be stated as a testable proposition. If the multiplier is real, then the marginal effect of a demonstration project should be higher for the first project in a technology category than for the tenth. The data to test this is not yet available, but the logic is clear enough to guide policy. A green credit program that spreads its subsidies thinly across many projects will have a smaller effect than one that concentrates them on the first deployment in each technology category.

2.3 Empirical Evidence and Implications

2.3.1 The Ningbo case in detail

Return to the heat pump firm and trace what happened after the rejection, because the case illustrates the mechanisms better than aggregate data.

The firm, which I will call Hengyuan Thermal, was founded in 2014 by two engineers who had worked at a state-owned compressor manufacturer. By 2019 it had 43 employees, three patents, and a prototype that had run for 8,000 hours in a test facility. It had revenue of 6 million yuan, all from small custom orders, and a net loss of 1.2 million yuan. Its book value was 8 million yuan, mostly equipment and receivables.

The application to Chen's bank in March 2020 was for 12 million yuan over five years, to tool up for the industrial park contract. The bank's credit model scored the firm at 42 out of 100, below the 55 threshold for unsecured lending. The score was driven by the negative profit, the short operating history, and the low collateral coverage. The model had no variable for patent quality or technology readiness.

Chen's rejection was automatic in the sense that the score was below threshold, but she had discretion to override. She did not, because she could not construct a defensible expected-loss estimate. The model's 42 was, in her words, "a number I could not argue with." She had no basis to say the model was wrong.

Hengyuan then applied to three other banks. Two rejected on the same grounds. The third, a provincial bank with a green finance pilot, had a different process. The pilot program, established in 2018 under the provincial government's green finance reform agenda, gave the bank a dedicated technical advisory team of four engineers and a subsidy of 50 basis points on qualifying loans. The team visited Hengyuan's test facility, reviewed the compressor design against published thermodynamic models, and interviewed the industrial park's engineering staff. It took three weeks.

The team's conclusion was that the technology was sound but the firm's manufacturing capacity was thin. The bank approved a 12 million yuan loan at 6.8 percent, with a covenant requiring the firm to reach a production milestone before drawing the second tranche. The loan was 190 basis points above the benchmark, but it was made.

Two things stand out. First, the verification cost was real: three weeks of four engineers' time is roughly 60,000 yuan, against an interest income of about 400,000 yuan over five years at 6.8 percent on 12 million. The verification consumed 15 percent of the loan's gross revenue. For a smaller loan, the economics would not work. Second, the subsidy was not the decisive factor. The 50-basis-point subsidy reduced the rate from 7.3 to 6.8 percent, which helped, but the loan would have been made at 7.3 percent. The decisive factor was the technical team. The bank could verify the technology, so it could price the loan.

This is the central empirical point of the chapter. The binding constraint is verification capacity, not the cost of funds. A subsidy that lowers the interest rate without building verification capacity does not solve the problem. It may even make it worse, because it attracts applications that the bank still cannot assess, increasing the rejection rate.

2.3.2 The green credit target and its discontents

The Chinese green credit system provides a natural experiment in what happens when policy addresses the cost of funds but not the information problem.

The system has its origins in the 2007 Green Credit Policy, issued jointly by the Ministry of Environmental Protection, the People's Bank of China, and the China Banking Regulatory Commission. The policy required banks to incorporate environmental compliance into their credit decisions and to report on green lending. It was followed by a series of directives: the 2012 Green Credit Guidelines, the 2015 Green Bond Endorsed Project Catalogue, and the 2016 Guiding Opinions on Building a Green Financial System. By 2020, Chinese banks reported 12 trillion yuan of green loans, the largest green credit book in the world.

The aggregate number is impressive. The composition is less so. A 2019 analysis by the Climate Bonds Initiative found that the majority of Chinese green loans went to projects in the "clean energy" and "energy efficiency" categories, which are dominated by Type A and Type B projects: solar farms, wind farms, waste-heat recovery, and building retrofits. The categories that correspond to Type C—advanced materials, energy storage, hydrogen, carbon capture—accounted for a small fraction.

The reason is not that the banks were ignoring the policy. It is that the policy's reporting framework counted all green loans equally, so the banks met their targets by lending to the projects they could assess. The information problem was untouched. The banks had a target and a subsidy, but they still could not price the heat pump.

There is a further distortion. The green credit statistics are based on the loan's purpose classification, not on the project's actual environmental performance. A loan to a coal plant for a "efficiency upgrade" can be classified as green if it meets the catalog's technical criteria, even if the plant continues to operate for another twenty years. The classification is a label, not a measurement. This is the moral hazard at the system level: the banks have an incentive to classify, not to verify.

The 2020 Green Finance Evaluation scheme, which rates banks on their green lending, partially addresses this by including qualitative indicators for "green finance strategy" and "environmental risk management." But the ratings are still based largely on volume. A bank that lends 100 billion yuan to solar farms scores better than a bank that lends 10 billion yuan to battery startups, even if the latter is doing the harder work of financing the transition.

2.3.3 The verification experiments

The most interesting developments are the local experiments that directly attack the information problem. Three illustrate the range of approaches.

The Huzhou green finance pilot. Huzhou, a prefecture-level city in Zhejiang, was designated a national green finance pilot zone in 2017. The pilot's centerpiece was a green finance statistics and verification system, operated by a local state-owned platform. The platform collected environmental data from the borrowers—energy consumption, emissions, water use—and cross-checked it against utility records and third-party audits. Banks could access the verified data to price their loans.

The system reduced the verification cost for participating banks. A loan officer at a Huzhou bank could look up a borrower's verified environmental performance instead of commissioning an audit. The pilot reported that green loan rejection rates fell from 31 percent in 2017 to 19 percent in 2019. I should treat this number with caution; it comes from the pilot's own evaluation, and the selection of borrowers may have changed. But the direction is consistent with the theory: reducing verification cost reduces rationing.

The Guangdong carbon emission pledge loans. Guangdong province, which hosts the country's largest carbon emissions trading scheme, developed a lending product that used carbon emission allowances as collateral. A firm with surplus allowances could pledge them to a bank and borrow against them. The allowance is a standardized, exchange-traded asset, so the bank can value it and liquidate it. The information problem is reduced because the collateral is legible.

The product has grown, but its reach is limited to firms that are already in the emissions trading scheme, which are large emitters. It does not help the heat pump startup, which has no allowances. It is a solution for Type A and Type B projects, not Type C.

The Suzhou "green technology bank." Suzhou established a dedicated evaluation unit for green technology firms, staffed with engineers and finance professionals. The unit evaluates the technology and produces a report that banks can use in their credit decisions. The unit does not lend; it provides the verification service that the banks cannot provide themselves.

The Suzhou model is the closest to a general solution. It separates verification from lending, which allows the verification cost to be shared across multiple lenders and to be subsidized by the government. A firm that has been evaluated by the unit can take the report to any participating bank. The cost of verification is paid once, not once per lender.

The model's limitation is scale. The unit can evaluate perhaps 200 firms a year. Suzhou has thousands of green technology firms. The evaluation is also point-in-time; it does not track performance after the loan is made. The moral hazard problem remains.

2.3.4 What the evidence says about policy

The three experiments point to a consistent conclusion. The interventions that reduce the lender's verification cost—Huzhou's data platform, Suzhou's evaluation unit—have a larger effect on green technology lending than interventions that reduce the cost of funds. The interventions that rely on standardized collateral—Guangdong's carbon pledges—work for the firms that have the collateral but not for the asset-light innovators.

This is the policy implication that the rest of the book will develop. The first link in the chain is not a subsidy. It is an identification mechanism: a way for the lender to know what it is financing. The rating-based identification system that Chapter 5 will describe is one such mechanism. The insurance instruments in Chapters 8 through 10 are another: they convert an uncertain environmental performance into a contractible cash flow. The sci-tech finance catalysis in Chapter 11 is a third: it provides the equity that absorbs the technology risk that debt cannot.

Before turning to those mechanisms, the limits of what this chapter has established should be stated.

The evidence is drawn heavily from China, because that is where the green credit system is largest and where the policy experiments are most active. The mechanisms are general, but the magnitudes may differ elsewhere. A study of European green lending would likely find lower rejection rates, because the EU's sustainable finance disclosure regime has reduced the verification cost for large firms. It would likely find similar problems for small innovators, who are not covered by the disclosure regime.

The data on rejection rates is thin. The Zhejiang-Jiangsu survey is a single study. The Huzhou and Suzhou numbers come from pilot evaluations, which have an incentive to report success. I have not seen a randomized controlled trial of a verification intervention, and until there is one, the causal claims should be treated as provisional.

The theory is also incomplete. The Stiglitz-Weiss model treats the lender's information as exogenous. In practice, the lender can invest in information, and the decision to invest is itself a choice. A full model would endogenize the verification cost and show how it depends on the loan size, the technology, and the lender's existing capabilities. I have not built that model here. The qualitative argument is that the verification cost is high relative to the loan size for Type C projects, and that this is the binding constraint. A formal model would sharpen the threshold conditions, but the policy conclusions would not change.

2.3.5 The cost of beingwrong in both directions

There is a type of error that the rationing literature underweights, and it matters for the green transition. The cost of lending to a bad green project is a default. The cost of rejecting a good green project is a technology that does not get deployed. The first cost is visible and appears on the bank's income statement. The second cost is invisible and appears in the aggregate productivity statistics a decade later.

Banks optimize against the visible cost. They cannot do otherwise; their capital is at risk, and their regulators measure them on realized losses. The invisible cost is borne by the economy, not by the lender. This is a classic externality, and it is the reason why the market alone will not finance the transition at the required pace. The externality is not the carbon reduction itself, which is the usual framing. It is the information that a demonstration project generates for the next lender. A financed heat pump produces a reference installation, a performance record, and a set of engineering data that reduces the verification cost for every subsequent heat pump loan. The lender who finances the first project does not capture that benefit. The second lender does.

This is the strongest argument for public intervention in green technology finance, and it is different from the usual argument about the social cost of carbon. The usual argument says green projects are underfinanced because their environmental benefits are not priced. That is true, but it is a second-order effect for the projects this chapter is concerned with. The first-order effect is that the information needed to finance them is a public good. It is expensive to produce, it is non-excludable once produced, and the private market underproduces it. The policy response should be to produce it publicly—through evaluation units, data platforms, and demonstration projects—rather than to subsidize the interest rate.

The social cost of carbon is real, and carbon pricing is an important part of the policy mix. But for the Type C firms that are the bottleneck for the transition, the carbon price is not the binding constraint. The heat pump firm's problem was not that its carbon savings were underpriced. It was that the lender could not verify the savings existed. A higher carbon price would have made the project more valuable and the loan more attractive, but only if the lender could believe the savings would materialize. The verification problem comes first.

2.3.6 Rationing and the structure of the banking system

The rationing pattern is not uniform across the banking system. It varies with the lender's size, its technical capacity, and its mandate. This variation is itself evidence for the information-based explanation.

Large state-owned banks have the capital to lend but not the technical staff to evaluate green technology. Their credit processes are standardized and centralized, which is efficient for conventional lending and ill-suited to technology assessment. A loan officer at a large bank has less discretion than Chen did, and the model she uses has even fewer technology variables. Large banks therefore ration more aggressively on Type C projects and concentrate their green lending on Type A and Type B, where the standardization works.

Smaller city commercial banks have more discretion and closer relationships with local firms. They can visit the factory and talk to the engineers. Their problem is capital and diversification: a single large green technology loan can be too big for their balance sheet, and they lack the portfolio breadth to absorb the idiosyncratic technology risk. They ration less on information but more on concentration.

The banks that do the most Type C lending are the policy banks and the provincial banks with explicit green mandates. They have the technical teams and the tolerance for technology risk, and they are compensated through subsidized funding or government guarantees. But their capacity is limited. The China Development Bank's green lending is large in aggregate, but it is concentrated in infrastructure, not in small technology firms.

This structure has an implication for policy design. A verification system that works for a large bank must be centralized and standardized, because the bank's loan officers cannot be expected to develop technical expertise. A verification system that works for a small bank can be more relational and decentralized. The Suzhou model—a central evaluation unit that serves multiple banks—is a compromise. It centralizes the technical capacity while allowing the banks to retain their own credit decisions.

The structure also explains why the green credit target has not produced more Type C lending. The target is enforced through the large banks, because they are the ones the regulator can monitor. The large banks meet it through Type A and Type B lending, because that is what their processes can handle. The target is met, and the Type C firms remain unfinanced. The policy has achieved its stated goal without achieving its intended effect.

2.3.7 What would change the outcome

Four conditions, drawn from the cases and the theory, define the circumstances under which the rationing would ease. This is not yet a policy recommendation; that comes in Chapter 15.

The verification cost has to fall below a threshold relative to the loan size. For the Ningbo case, the threshold was around 15 percent of gross interest income. A loan of 12 million yuan at 6.8 percent over five years generates about 400,000 yuan of interest; a verification cost of 60,000 yuan is 15 percent. For a loan of 3 million yuan, the same verification cost would be 60 percent, and the loan would not be made. The threshold implies that verification must be either cheaper or shared across multiple loans. The Suzhou model shares it across lenders; the Huzhou platform shares it across borrowers. Both reduce the per-loan cost.

The lender also needs a way to price the technology risk. This means a rating or scoring system that incorporates technology readiness, patent quality, and environmental performance. Chapter 5 will describe what such a system looks like. The point here is that the system must produce a number the loan officer can defend, not a qualitative assessment that the risk committee will override.

Tenor is the third condition. The loan has to match the project's cash-flow profile. A two-year loan for a project that breaks even in year four is a refinancing risk, not a financing solution. The tenor problem is partly a function of the bank's funding structure—deposits are short-term—and partly a function of the lender's confidence in the project. A lender that can verify the technology is more willing to lend long. The two problems are connected.

Finally, the borrower needs access to equity that absorbs the technology risk. A firm with 8 million yuan of book value and a 12 million yuan project is undercapitalized for debt finance. The debt-to-equity ratio is too high, and the lender's collateral coverage is too low. The equity gap is the subject of Chapter 11. The point here is that debt rationing is partly a symptom of an equity gap. The firm that cannot raise equity cannot raise debt either, because the debt would be too risky.

These four conditions are the design parameters for the rest of the book. The rating system addresses the second. The insurance instruments address the third, by converting performance risk into a contractible cash flow. The sci-tech finance catalysis addresses the fourth. The verification infrastructure addresses the first. None of them works alone. The chain is the unit of design, which is where Chapter 1 ended and where this chapter's analysis points.

2.4 Conclusion: The Information Problem Is Upstream

The argument of this chapter can be stated in one sentence: single-channel green credit fails because the information problem is upstream of the lending decision, and no amount of subsidized interest rates fixes an information problem.

That sentence is the hinge on which the rest of the book turns, so it is worth unpacking.

"Single-channel" means the financial system's reliance on bank credit as the primary instrument for financing the transition. The green credit system is a single-channel system. It channels funds through banks, and it measures success by the volume of green loans. The system has grown large, but it has grown by financing the projects that banks can assess. The projects that banks cannot assess—the Type C innovators—are excluded, not by policy but by the structure of the credit process.

"Upstream of the lending decision" means the problem occurs before the loan officer decides whether to lend. It occurs at the point where the lender tries to estimate the project's risk. If the estimate cannot be made, the decision is not a choice between lending and not lending; it is a choice between not lending and lending blind. Most lenders choose not to lend. The rationing is the rational response to an information deficit that the lender cannot close at acceptable cost.

"No amount of subsidized interest rates fixes an information problem" means that the policy tools that are easiest to deploy—interest subsidies, green credit targets, relending facilities—do not address the binding constraint. They reduce the cost of funds, which is not the problem. They may even worsen the rationing, by attracting applications that the lender still cannot assess. The Huzhou and Suzhou experiments worked because they reduced the verification cost, not because they reduced the interest rate. The Ningbo case shows that a loan can be made at a high rate if the lender can verify the technology. The rate is not the obstacle; the verification is.

This conclusion has a corollary that is uncomfortable for the green finance industry. The growth of green credit volumes is not evidence that the financing problem is being solved. It is evidence that the easy projects are being financed. The hard projects—the ones that will determine whether the transition succeeds—are still waiting. A green credit system that reports 12 trillion yuan of loans but cannot finance a heat pump startup is a system that has optimized for the metric rather than the mission.

The experiments described in this chapter show that the problem is solvable. The Suzhou evaluation unit, the Huzhou data platform, and the Ningbo pilot's technical team are all working models. They are small, and they are expensive relative to the loans they support. But they demonstrate that the verification cost can be brought down, and that when it is, the lending follows.

The next chapter steps back from the mechanics of credit rationing to the conceptual map of the financial ecosystem. The distinction between green finance and transition finance, which this chapter has gestured at, becomes the organizing principle. The transition is a process of conversion, not a set of already-green projects, and the financial system needs instruments that can finance the conversion rather than the end state. The information problem is the first obstacle to that, and it is the one that the rating-based identification system is designed to solve.

One more observation from the Ningbo case before moving on. When Hengyuan finally got its loan, the bank's technical team wrote a four-page evaluation. The evaluation was not a credit rating. It was a technology assessment: compressor efficiency at three temperature points, patent validity review, manufacturing readiness level, and a sensitivity analysis of the park's offtake commitment. The loan officer used the assessment to fill in the expected-loss model. The model's output was a probability of default of 2.8 percent, which was within the bank's tolerance. The number was defensible because the assessment was specific.

That four-page document is the prototype of the identification mechanism this book is about. It is not a rating in the conventional sense. It is a translation: it converts the borrower's private information about the technology into a form the lender can use. The translation is expensive, and it is the thing that the financial system currently lacks. Building it at scale—for heat pumps and batteries and membranes and every other technology the transition needs—is the first task of the chain.

The loan officer in Ningbo could not price the loan because she had no translation. Give her one, and she will lend. The rest of the book is about how to build that translation, how to buffer the risks it reveals, and how to catalyze the capital it requires.