Chapter 2: Information Asymmetry and the Architecture of Credit

A loan officer at a city commercial bank in Zhejiang receives two applications on the same morning. The first comes from a textile dyeing firm with twelve years of operating history, audited statements, a factory on pledged land, and a customer list that includes three listed garment makers. The second comes from a company that manufactures heat-pump control modules. It is four years old, rents its floor space, holds nine invention patents, and has one customer accounting for 61 percent of revenue. The first firm wants 8 million yuan to replace a boiler and install wastewater metering. The second wants 15 million yuan to double a production line that the loan officer has never seen and cannot fully evaluate.

Both applications are, on paper, green. The dyeing firm's project would cut chemical oxygen demand in its effluent by roughly 40 percent. The heat-pump modules would displace gas-fired heating in perhaps twenty thousand households over five years. The loan officer approves the first within a week and defers the second indefinitely. Nothing in the bank's credit policy tells her to do this. The policy manual mentions environmental compliance in two paragraphs and says nothing about carbon. She defers the second loan because she cannot price what she cannot see, and she cannot see a technology whose value depends on a market that does not yet exist at scale.

That deferral is the subject of this chapter. It is not a story about a banker's bias. It is a story about the architecture of credit—the institutions and instruments that stand between a saver's deposit and a borrower's project—and about how that architecture processes information. Environmental performance enters the architecture as a special kind of information: forward-looking, hard to verify, and correlated across borrowers in ways that ordinary credit risk is not. The architecture was not built to handle it. Understanding why requires going back to first principles.

2.1 The Nature of Information Asymmetry

2.1.1 What Credit Risk Actually Is

Start with a definition, because the vocabulary in this field is loose. Credit risk is the possibility of losing money because a counterparty cannot, will not, or does not timely honor a financial obligation. The counterparty is the obligor. The obligation is usually a contract, though oral contracts can bind. Three concepts get confused and should be separated.

Insolvency describes a financial state: liabilities exceed assets. Default describes an event: failure to meet a contractual obligation, typically nonpayment. Bankruptcy describes a legal proceeding: a court steps in, reviews the obligor's finances, and negotiates with management and creditors, aiming where possible to keep the entity operating through asset sales or renegotiated financing. A firm can be insolvent without defaulting, default without being insolvent (a liquidity squeeze), and be bankrupt without either term capturing the full situation.

For the loan officer in Zhejiang, these distinctions matter because each corresponds to a different information problem. Insolvency is a balance-sheet question, answerable in principle from statements. Default is a cash-flow question, answerable from projections and covenants. Bankruptcy is a legal question, answerable from jurisdiction and contract. Environmental performance cuts across all three. A firm can be pushed into insolvency by a remediation liability that never appears on its balance sheet until the regulator arrives. It can default because a pollution-control upgrade consumed the working capital it needed for payroll. It can enter bankruptcy because a single environmental penalty triggered a cross-default clause in a syndicated loan.

2.1.2 The Two Classic Failures

The economics of information asymmetry in credit markets rests on two mechanisms, both formalized in the 1970s and both still doing useful work.

Adverse selection, in the formulation George Akerlof gave it in 1970 and Michael Rothschild and Joseph Stiglitz applied to insurance in 1976, describes what happens before a contract is signed. The borrower knows more about the quality of the project than the lender does. If the lender cannot distinguish good projects from bad, it will price credit at the average quality of the pool. Owners of good projects, whose expected returns are high enough to justify a lower rate, withdraw. The pool worsens. The lender raises the rate. More good projects withdraw. In the limit the market may not clear at all, or it clears only at a rate that excludes the safest borrowers.

Stiglitz and Andrew Weiss applied this logic to bank credit in 1981 and got a result that surprised the profession: a competitive bank may rationally ration credit rather than raise rates to clear the market. Higher rates attract riskier borrowers (adverse selection) and induce existing borrowers to take riskier actions (moral hazard). The bank's expected return is not monotonic in the interest rate. Somewhere below the market-clearing rate there is a rate that maximizes expected return, and at that rate demand exceeds supply. The excess demand is not eliminated by price. It is eliminated by quantity, through non-price rationing devices: collateral requirements, covenants, loan size limits, relationship history.

Moral hazard describes what happens after signing. The borrower can take actions the lender cannot observe that affect the probability of repayment. In a green context the action space widens considerably. A firm that borrows to install emissions monitoring equipment can subsequently switch it off to save on operating costs. A firm that borrows against the promise of a technology upgrade can divert the funds to its existing, dirtier line. A firm can report emissions reductions that were achieved by outsourcing the polluting step to an unregulated supplier. None of these actions is visible to the lender without monitoring, and monitoring is costly.

The two mechanisms interact. The expectation of moral hazard raises the cost of monitoring, which raises the effective interest rate, which worsens adverse selection. The expectation of adverse selection makes the lender demand collateral, which screens out borrowers without assets, which is precisely the population of technology firms whose assets are intangible.

2.1.3 Why Environmental Information Is Different

The standard apparatus assumes that the lender's information disadvantage is about a project's cash flows. Environmental information adds three complications.

Horizon comes first. A conventional five-year loan is priced against a five-year cash-flow projection. Many environmental risks materialize over a longer span: groundwater contamination that migrates over a decade, a carbon price that ratchets after 2030, a supply chain that reconfigures when a major customer adopts scope-3 accounting. The lender's information disadvantage concerns the trajectory of a regulatory and physical environment that neither party controls, as much as the borrower's present condition.

The second is correlation. Conventional credit portfolios diversify idiosyncratic default risk. Environmental risk is systematic in a specific way: a tightening of effluent standards affects every dyeing firm in the province at once, and a carbon border adjustment affects every exporter in a sector at once. The loan officer's portfolio of twelve dyeing firms is not twelve independent bets; it is one bet on the province's regulatory trajectory, levered twelve times. This is the subject of Chapter 5, but it begins here, as an information problem: the lender cannot observe the correlation because it cannot observe the environmental exposure of each borrower on a comparable basis.

The third is verifiability. Cash flows can be audited. A firm's bank statements either reconcile with its sales ledger or they do not. Emissions reductions are harder. A reduction reported by the firm may reflect a change in measurement methodology, a change in the boundary of what is counted, a change in the grid emission factor used to convert electricity into carbon, or an actual reduction. The lender cannot distinguish these without a measurement infrastructure that most banks do not have.

None of this makes environmental information uniquely unknowable. It makes the cost of making it knowable high relative to the size of an individual loan, and it means the benefit accrues to the lender only if the information is comparable across borrowers. A single bank measuring a single borrower's emissions learns something of limited value. A rating system that measures all borrowers on the same scale converts a private cost into a public good. That is the argument of Chapter 3.

2.1.4 A Simple Model of Screening

Before moving to the institutional architecture, it is worth setting down the formal skeleton, because the rest of the book refers back to it.

Consider a lender facing a population of borrowers indexed by \(i\). Each borrower has a project with expected gross return \(R_i\) and a probability of success \(p_i\). If the project succeeds, the borrower repays principal plus interest, \(1 + r\). If it fails, the lender recovers a fraction \(\lambda\) of principal (the recovery rate) and the borrower loses everything. The borrower's project requires one unit of capital and the borrower has collateral \(C_i\) that the lender can seize on default.

The lender's expected return on loan \(i\) is:

The borrower's expected return is:

The borrower applies for a loan if \(E[u_i] \geq 0\). Rearranging, the borrower's participation condition is:

The right-hand side is the borrower's expected cost of borrowing, which rises with the interest rate and with the collateral requirement. The left-hand side is the expected gross return, which depends on project quality.

Now suppose the lender cannot observe \(p_i\) but knows the distribution of \(p\) across the population. If the lender sets a single rate \(r\) and a single collateral requirement \(C\), the borrowers who apply are those for whom the participation condition holds. Differentiate the participation condition with respect to \(p_i\): the left-hand side rises faster in \(p_i\) than the right-hand side when \(R_i\) is high, but the collateral term \((1 - p_i)C_i\) falls as \(p_i\) rises, so high-quality borrowers with low collateral benefit most from applying. The composition of the applicant pool depends on the terms offered. Raising \(r\) or \(C\) changes who applies, and not in a direction the lender can predict without knowing the joint distribution of \(p_i\), \(R_i\), and \(C_i\).

This is the formal content of adverse selection. The lender's problem is not that it lacks a price; it is that the price selects the pool.

Now add environmental performance. Let \(e_i \in [0,1]\) index the borrower's environmental quality, with higher values meaning cleaner. Suppose environmental quality affects the project in two ways. It raises the probability of regulatory interruption: a dirty firm faces a hazard rate of shutdown that reduces \(p_i\). And it raises the expected remediation liability in the event of default, which reduces the recovery rate \(\lambda\).

Write \(p_i = p(e_i, x_i)\) and \(\lambda_i = \lambda(e_i, z_i)\), where \(x_i\) and \(z_i\) are other borrower characteristics. The lender's expected return becomes:

If the lender observes \(e_i\), it can price the loan correctly or decline it. If the lender does not observe \(e_i\), it prices at the average \(\bar{e}\) and the clean borrowers subsidize the dirty ones. Clean borrowers with high \(R_i\) face a rate above their fair rate and may exit. The pool tilts dirty. The lender, observing rising default rates, raises the rate further. The spiral is the same as in the classical model, with environmental quality as the hidden characteristic.

The policy implication is direct. Anything that raises the lender's ability to observe \(e_i\) on a comparable scale improves the pool. A rating that maps environmental performance into a discrete grade, applied to all borrowers in a sector, does exactly this. It converts \(e_i\) from a continuous, unobservable, borrower-specific variable into a coarse but observable and comparable one. The information loss from coarsening is real; the gain from comparability is larger, at least at the current stage of measurement infrastructure.

flowchart TD
    A[Lender sets rate r and collateral C] --> B[Borrowers self-select]
    B --> C{Composition of applicant pool}
    C -->|High-quality borrowers exit| D[Average p falls]
    C -->|Low-quality borrowers remain| D
    D --> E[Default rate rises]
    E --> F[Lender raises r or C]
    F --> B
    D --> G[Observable environmental rating e]
    G --> H[Lender prices by grade]
    H --> I[Clean borrowers no longer subsidize dirty]
    I --> J[Pool composition stabilizes]

Figure 2.1: The adverse selection spiral and the stabilizing role of an observable rating.

2.1.5 Moral Hazard in the Green Context

The post-contract problem deserves separate treatment because it is where most of the practical damage occurs.

A lender that finances an environmental upgrade faces three monitoring problems. The first is whether the upgrade was installed. This is verifiable at modest cost through site inspection, though inspection frequency is limited by the lender's staffing. The second is whether the upgrade is operating. This is harder; a wastewater treatment unit can be installed and then bypassed at night. The third is whether the upgrade achieved its intended effect. This requires measurement of outcomes, not inputs, and it is the hardest of the three.

The standard contractual response is covenants. A loan agreement can require the borrower to maintain environmental permits, submit to periodic audits, and report specified metrics. Covenants work when violations are detectable and when the lender has a credible threat: acceleration of the loan, an increase in the interest rate, or a refusal to renew. The threat is credible only if the lender monitors, and monitoring is costly.

There is a further problem specific to green lending. The lender's objective is not the borrower's environmental performance per se; it is the lender's own exposure to environmental risk. These coincide when environmental failure raises default probability. They diverge when the lender can offload the environmental risk onto someone else. A bank that has sold its loan into a securitization pool has less incentive to monitor the borrower's emissions than a bank that holds the loan to maturity. This is the classic originate-to-distribute problem, and it applies to green lending with particular force because the environmental performance of the underlying asset is precisely what the buyer of the securitization cannot observe.

The empirical record here is thin. A 2019 study by the Climate Bonds Initiative of 1,500 green bonds found that the presence of an external reviewer reduced the issuance spread by roughly 5 basis points, which suggests that verification is priced. Work by the People's Bank of China on green loan pricing found that relationship length reduced the interest rate on green loans by 10 to 15 basis points relative to first-time borrowers, which suggests that repeated interaction substitutes for formal verification. Neither finding tells us what happens when the relationship is short and the reviewer is paid by the issuer. That question remains open, and it belongs on the research agenda in Chapter 15.

2.1.6 The Insurance Analogy

It is worth pausing on insurance because the book returns to it repeatedly. Insurance is the canonical market for information asymmetry. The insured knows more about the risk than the insurer. The insurer's response is a combination of deductibles, co-payments, exclusions, and experience rating, all of which shift some of the risk back onto the insured so that the insured has an incentive to exercise care.

Weather index insurance, discussed in Chapter 7, is a partial escape from this problem. It pays out on an objective trigger, such as rainfall measured at a specified weather station, rather than on the insured's reported loss. This eliminates the moral hazard of over-reporting losses and the adverse selection of only the highest-risk farmers buying coverage. It introduces a new problem, basis risk: the trigger may not match the actual loss, so the farmer can suffer a loss and receive no payout, or receive a payout without a loss. Basis risk is the price of objectivity.

The lesson for credit is that objective, third-party-measured triggers can substitute for costly monitoring. A green loan whose interest rate steps down when an independent auditor certifies an emissions reduction is a partial index contract. It does not eliminate the verification problem, but it moves verification from the lender's staff to a specialized third party, and it makes the result comparable across borrowers. This is the design logic behind the rating function in Chapter 3 and the buffering layer in Chapters 6 through 9.

2.2 Credit Rationing and Its Consequences

2.2.1 The Rationing Result

The Stiglitz-Weiss result is worth restating precisely because it is often misstated. The claim is not that credit markets never clear. The claim is that a profit-maximizing lender may choose to ration rather than raise the rate, because the expected return on the loan is not monotonic in the rate. At low rates, raising the rate increases revenue. At some point, further increases reduce expected return because the marginal borrowers attracted by the higher rate are worse risks, and because existing borrowers respond to higher rates by choosing riskier projects. The lender's optimal rate is below the market-clearing rate, and the excess demand is rationed by non-price means.

Collateral, covenants, loan size limits, maturity limits, relationship requirements, sectoral concentration limits: the familiar list. Each is a screening device. Collateral screens on wealth. Covenants screen on the borrower's willingness to accept constraints. Relationship requirements screen on history. Sectoral limits screen on the lender's own portfolio composition.

For green firms, each of these devices is biased against them, and the bias is structural rather than incidental.

Collateral screens on tangible assets. A technology firm's assets are patents, software, and human capital. A patent can be pledged in some jurisdictions, but its liquidation value is uncertain and the market for patent collateral is thin. The result is that a technology firm with a strong green project and weak collateral faces a higher effective cost of credit than a dirty firm with a factory.

Covenants screen on the borrower's ability to accept constraints. A young firm has less financial slack and less negotiating leverage; it may be unable to accept the covenants a lender demands. This is not because its project is worse but because its balance sheet is thinner.

Relationship requirements screen on history. A firm that has borrowed for ten years has a track record. A firm that is four years old has a shorter track record, and a firm that is two years old has almost none. The relationship premium is a rational response to information asymmetry, and it is precisely the firms with the least history that have the most to offer in environmental terms, because they are the ones building new capacity rather than retrofitting old.

Sectoral limits screen on the lender's existing portfolio. A bank that is already concentrated in textiles will limit further textile lending regardless of the environmental quality of individual projects. The limit is a portfolio risk control, and it is correct as portfolio risk control. It is also an obstacle to a green transition that requires reallocating capital within sectors.

2.2.2 Quantifying the Rationing

How large is the rationing effect? The empirical literature gives a range, and the range is wide enough that we should be cautious about point estimates.

A 2020 study by the China Banking and Insurance Regulatory Commission of 1,800 small and medium enterprises across five provinces found that the probability of a loan application being rejected was 23 percentage points higher for firms without property collateral, controlling for profitability and sales. The magnitude varied by bank type; rural commercial banks and city commercial banks showed collateral effects 8 to 12 points larger than the large state-owned banks, which have more diversified portfolios and lower funding costs. Green lending specifically has been less studied, but the pattern is consistent: green loan approval rates are higher for firms with existing banking relationships and lower for first-time borrowers, and the gap is larger in sectors where the environmental technology is newer.

The loan officer in Zhejiang is not an outlier. She is behaving as the architecture predicts. The question is what the architecture costs.

2.2.3 The Cost of Rationing: Three Channels

Rationing has costs that do not appear on any bank's income statement.

The first is the foregone project. A heat-pump control module company that cannot borrow at a reasonable rate grows more slowly than it would otherwise, which means the emissions reductions it would have delivered arrive later or not at all. This is a social cost, not a private one, and it is invisible to the lender. The lender sees a declined application and moves on. The cost is borne by the households that would have heated their homes with gas.

The second is the composition effect. When credit is rationed by collateral, capital flows toward asset-heavy firms. In a green transition, the asset-heavy firms are often the incumbents whose assets are the problem. A coal plant has collateral; a solar developer has a power purchase agreement. The rationing mechanism systematically favors the incumbent. This is not a market failure in the textbook sense; it is a predictable consequence of using collateral as a screening device when the transition requires reallocating capital away from collateralized assets.

The third is the selection effect on technology. If only firms with tangible assets can borrow, the technologies that get financed are the ones that come bundled with tangible assets. Retrofit technologies, which attach to existing assets, are easier to finance than replacement technologies, which require new assets. Process innovations, which live in organizational routines, are harder to finance than equipment purchases. The bias is toward the familiar. A green transition that requires unfamiliar technologies will be underfinanced relative to a transition that requires familiar ones, even if the unfamiliar technologies are cheaper per ton of carbon abated.

2.2.4 Why Green Firms Are Rationed More

The general rationing result applies to all borrowers with information problems. Green firms face three additional sources of asymmetry.

The first is technological opacity. A loan officer can evaluate a boiler replacement because the technology is mature and the cost savings are calculable from fuel prices. She cannot evaluate a novel control module because the technology is new and the market is uncertain. The asymmetry is not about the borrower's honesty; it is about the lender's competence. The lender does not have the engineering knowledge to assess the project, and hiring that knowledge is not economic for a single loan.

The second is policy dependence. Many green projects are viable only with a subsidy, a feed-in tariff, a tax credit, or a carbon price. The lender must assess not only the project but the durability of the policy that supports it. Policy can change with an election, a budget cycle, or a ministerial decision. The lender's information disadvantage extends to the political process, which is opaque by design.

The third is measurement uncertainty. The environmental benefit of a project is often calculated using methodologies that are themselves contested. The emissions factor for grid electricity, the global warming potential of a refrigerant, the baseline against which a reduction is measured: each is a parameter that can be set in more than one way. The lender cannot verify the borrower's calculation without replicating it, and replication requires expertise the lender does not have.

These three sources compound. A technology firm with a novel process, dependent on a subsidy, claiming a reduction measured against a contested baseline, is the hardest possible case for a credit officer. It is also, frequently, the case with the largest environmental payoff.

2.2.5 The Chinese Evidence

The Chinese literature on green credit rationing is instructive because the policy environment has been explicit about the goal for longer than in most economies.

The China Banking and Insurance Regulatory Commission (CBIRC, now the National Financial Regulatory Administration) has issued green credit statistics since 2013. The headline numbers show rapid growth: green credit balances reached 30 trillion yuan by the end of 2023, with annual growth above 30 percent in several years. The composition is skewed toward infrastructure and clean energy, which are asset-heavy and collateralizable. Lending to small technology firms in green sectors is a small share of the total.

The gap between the headline growth and the composition is the rationing effect. The banks are lending green, but they are lending to the green projects that fit the existing architecture. The projects that do not fit, which are often the ones with the highest innovation content, are rationed out. A 2021 survey by the People's Bank of China of 340 branch managers found that the main obstacle to green lending was not the absence of projects but the difficulty of assessing project risk (cited by 61 percent of respondents), followed by the absence of qualified collateral (48 percent) and the lack of environmental data (39 percent). The first and third are information problems. The second is an architecture problem.

There is a specific institutional detail worth noting. The People's Bank of China's Measures for the Administration of Loan Cards specifies application requirements for individuals, enterprise legal persons, and public-institution legal persons. It does not cover farmers' professional cooperatives. The requirement that cooperative legal persons hold a loan card therefore excludes them from borrowing, regardless of the environmental quality of their operations. This is a small example of a large point: the architecture of credit is built from administrative categories, and firms that do not fit the categories are excluded before any assessment of their project begins.

2.2.6 The Consequences for the Transition

The rationing of green credit has consequences that accumulate over time.

The first is delay. A project that is rationed for two years may be abandoned, or it may be financed at a higher cost through informal channels. Either way the emissions reduction arrives later. The social cost of delay is the cumulative emissions during the delay, which is not zero even if the project eventually proceeds.

The second is substitution toward cheaper, dirtier technology. A firm that cannot borrow for a clean process may borrow for a conventional process, because the conventional process is easier to evaluate and the collateral is more familiar. The rationing does not prevent investment; it redirects investment toward the technologies the architecture can process.

The third is concentration. If only large firms with established relationships can access green credit, the benefits of the transition accrue disproportionately to large firms. This is a distributional consequence that is rarely discussed in the green finance literature, which tends to focus on aggregate totals. The aggregate totals can grow while the composition becomes more concentrated.

None of this is an argument against credit rationing as such. Rationing is a rational response to information asymmetry, and eliminating it by fiat would produce worse outcomes. The argument is that the architecture can be improved so that the rationing is based on better information. That is what a rating system does.

flowchart LR
    A[Green firm applies] --> B{Tangible collateral?}
    B -->|Yes| C[Standard credit process]
    B -->|No| D{Established relationship?}
    D -->|Yes| E[Relationship-based pricing]
    D -->|No| F{Technology familiar to loan officer?}
    F -->|Yes| G[Project finance possible]
    F -->|No| H[Rationed out]
    C --> I[Loan approved]
    E --> I
    G --> I
    H --> J[Project delayed or abandoned]
    J --> K[Emissions reduction foregone]

Figure 2.2: The rationing path for a green technology firm.

2.3 The Role of Rating in Reducing Asymmetry

2.3.1 What a Rating Is

A credit rating is an ordinal assessment of an obligor's willingness and ability to meet its obligations, expressed on a scale that is comparable across obligors and stable over time. The scale is ordinal, not cardinal: a BBB rating does not mean a specific default probability, though rating agencies publish historical default rates by grade that allow users to map grades to probabilities.

The rating is produced by an agency that specializes in a sector, employs analysts who follow a small number of obligors, and has access to management that a single lender may not have. The agency's comparative advantage is specialization and repetition. It sees many obligors in the same sector, which allows it to benchmark. It rates the same obligor repeatedly, which allows it to detect deterioration. It publishes its methodology, which allows users to understand what the rating does and does not capture.

The rating is not a guarantee. It is a summary statistic, and like all summary statistics it discards information. The question for this book is whether the information it discards includes environmental performance, and if so, whether that information can be added without destroying the comparability that makes the rating useful.

2.3.2 Rating as Screening

In the language of information economics, a rating is a screening device. It is produced by an intermediary that invests in information acquisition and sells the result. The lender buys the rating, or uses a public rating, instead of acquiring the information itself. The economics work when the intermediary's cost of information acquisition is lower than the lender's, which it is when the intermediary has scale and specialization.

Screening devices are valuable when they separate types. A rating that is uncorrelated with default probability is worthless; a rating that is perfectly correlated with default probability is redundant with the lender's own assessment. The value lies in the correlation being high enough to be useful and low enough to add information.

For environmental performance, the screening problem is harder because the outcome of interest is not default but environmental impact, and the two are correlated only imperfectly. A firm can have a high environmental impact and a low default probability (a profitable coal plant). A firm can have a low environmental impact and a high default probability (a startup with a clean technology and no revenue). A rating that measures environmental performance must be explicit about which outcome it is predicting.

This is the central design choice of Chapter 3. There are at least three possible objects of a green rating: the environmental performance of the obligor (how clean is the firm), the environmental performance of the use of proceeds (how clean is the project), and the environmental risk to the lender (how much environmental liability does the loan carry). These are different, and a rating that conflates them will be misleading.

2.3.3 Rating as Signaling

The mirror image of screening is signaling. A borrower that expects to be treated well by the market has an incentive to reveal information that distinguishes it from worse borrowers. The signal is credible only if it is costly for worse borrowers to mimic.

A green rating can function as a signal if the cost of obtaining it is correlated with environmental quality. A firm that commissions an audit, pays for certification, and discloses its emissions bears a cost. A dirty firm can also bear that cost, but the audit will reveal its dirtiness, so the signal separates. The separation is imperfect because the audit methodology can be gamed, but it is better than no signal.

The signaling literature suggests that mandatory disclosure can crowd out voluntary signaling. If every firm must report emissions, the act of reporting no longer distinguishes. This is a caution for policy design: mandating disclosure is valuable for comparability, but it may reduce the informational content of voluntary disclosure. The two can coexist if mandatory disclosure is coarse and voluntary disclosure is fine, but the design has to be deliberate.

2.3.4 The Limits of Rating

A rating is not a substitute for monitoring. It is a snapshot, updated periodically, and the interval between updates is a window of vulnerability. A firm can deteriorate between ratings, and the lender that relies on the rating alone will be surprised.

A rating is also not a substitute for relationship. The rating agency sees the obligor through documents and interviews; the lender sees the obligor through transactions. The two sources of information are complementary. A lender that uses the rating to screen and the relationship to monitor gets more than either alone.

For environmental performance, the limits are sharper. The rating methodology for environmental performance is less mature than the methodology for credit performance. The data are less standardized. The verification is less institutionalized. A green rating today is closer to where credit ratings were in the 1970s, before the SEC designated the nationally recognized statistical rating organizations (NRSROs) and imposed disclosure requirements. The comparison is instructive: it took decades to build the institutional infrastructure that makes credit ratings reliable, and the green rating infrastructure is at an earlier stage.

2.3.5 The Rating Architecture in Practice

The rating process in a large agency follows a recognizable sequence. Analysts specializing in a sector review the obligor's financial statements, industry position, management, and legal structure. They meet with management. They apply a criteria framework that specifies how each factor maps to a score. A committee reviews the recommendation and assigns the grade. The grade is published with a rationale, and it is reviewed periodically or when a material event occurs.

The strength of this process is specialization. An analyst who covers twelve chemical companies knows the chemical industry's cost structure, its regulatory environment, and its cyclicality. The analyst can benchmark a new obligor against the twelve. A loan officer at a bank that lends to forty sectors cannot do this.

The weakness is coverage. The major agencies rate large issuers of public debt. They do not rate small and medium enterprises, because the fixed cost of rating exceeds the fee the issuer can bear. The firms that most need a screening device, the ones with the least access to capital markets, are the ones the rating agencies do not cover. This is the gap that a bank-internal rating, or a sector-specific rating scheme, has to fill.

2.3.6 Internal Ratings and the Basel Framework

Banks that use the internal ratings-based approach under the Basel framework maintain their own rating systems. The internal rating maps each obligor to a grade, and the grade maps to a probability of default, which feeds the capital calculation. The internal rating is not published, but it is audited by the supervisor and it must meet statistical standards: the bank must demonstrate that the grades are ordered by default risk and that the default probabilities are calibrated.

The internal rating is a natural host for environmental information. The bank already has the infrastructure: the data pipeline, the committee process, the validation methodology. Adding an environmental dimension to the internal rating requires extending the data collection and the criteria, not building a new system from scratch. This is the practical argument for integrating environmental performance into credit rating rather than creating a separate green rating system. The separate system would duplicate infrastructure and would not feed the capital calculation.

The counterargument is that the internal rating is designed to predict default, and environmental performance predicts default only weakly at short horizons. A firm's emissions do not affect its twelve-month default probability much. They affect its five-year default probability through regulatory and physical channels, and they affect the lender's reputational and legal exposure through channels that the internal rating does not capture. Integrating environmental performance into the internal rating may distort the default prediction without capturing the environmental risk.

The resolution, developed in Chapter 3, is a two-track rating: a credit grade that predicts default on the conventional horizon, and an environmental grade that measures performance and forward risk on a longer horizon. The two grades are reported together but not collapsed into one number. The lender uses the credit grade for pricing and the environmental grade for portfolio management and disclosure. This preserves the statistical properties of the credit grade while adding the environmental information the transition requires.

2.3.7 The Evidence on Rating and Asymmetry

Does rating actually reduce information asymmetry? The evidence is indirect but consistent.

Work by Efraim Benmelech and Nittai Bergman on bond markets found that ratings affect spreads beyond what is explained by accounting variables, which suggests that ratings convey information not already in the financial statements. Research by the Bank for International Settlements on bank lending found that firms with ratings pay 15 to 30 basis points less and face fewer non-price constraints, controlling for size and profitability. Studies of rating changes by the credit rating agencies found that downgrades are followed by reductions in investment and employment, which suggests that the rating affects the firm's access to capital, not just its cost.

For green ratings specifically, the evidence is newer and thinner. A 2019 study by the Climate Bonds Initiative of 1,500 green bonds found that the presence of a second-party opinion reduced the spread by 4 to 6 basis points, and that the effect was larger when the opinion was from a more reputable reviewer. A 2020 study by the European Commission's Joint Research Centre found that firms that disclose more environmental information have lower costs of capital, though the direction of causation is contested: it may be that firms with lower costs of capital can afford to disclose, rather than that disclosure lowers the cost.

The honest summary is that rating reduces asymmetry, that the effect is larger for firms with less access to alternative information sources, and that the effect for green ratings is plausible but not yet well quantified. The research agenda in Chapter 15 includes the measurement of this effect.

2.3.8 The Institutional Preconditions

A rating system works only if certain institutional preconditions hold.

The first is a common taxonomy. If one agency's "green" includes nuclear power and another's excludes it, the ratings are not comparable. The development of green taxonomies, such as the EU Taxonomy and China's Green Bond Endorsed Projects Catalogue, is an attempt to establish the common language. The taxonomies are still evolving and they differ across jurisdictions, which is a barrier to cross-border green finance.

The second is verification. A rating that is based on the obligor's self-reported data is only as good as the obligor's honesty. Third-party verification, whether by an auditor, a certification body, or a specialized reviewer, is necessary to make the rating credible. The verification market is developing, but it is fragmented and the standards are not uniform.

The third is liability. If a rating agency assigns a green grade to a firm that turns out to be dirty, who bears the loss? The credit rating agencies have faced litigation over their role in the 2008 financial crisis, and the legal framework for green ratings is less developed. The liability question affects the agency's incentive to invest in verification, and it is unresolved.

The fourth is data infrastructure. A rating needs data, and the data need to be collected, cleaned, and stored in a form that allows comparison. The digital infrastructure for environmental data is less developed than the infrastructure for financial data. This is the subject of Chapter 14.

2.3.9 Why Rating Is Necessary but Not Sufficient

A rating reduces information asymmetry at the point of origination. It does not eliminate the asymmetry that arises after origination, when the borrower's behavior can change. It does not address the correlation problem, which is a portfolio issue. It does not provide the risk transfer that insurance provides. And it does not by itself create the market for early-stage green capital, which requires a different set of instruments.

This is why the book's argument is layered. Rating is the first layer, the identification function. Insurance is the second, the buffering function. Sci-tech finance is the third, the catalysis function. Each addresses a different failure, and the failures interact. A rating without insurance leavesthe lender bearing a risk it cannot diversify. Insurance without a rating leaves the insurer unable to price. Sci-tech finance without either leaves the government-guided fund investing blind.

The sequence matters. A rating system that is not credible will not be used by insurers, and an insurance product that is not priced on a credible rating will not be bought by lenders. The layers have to be built in order, and the first layer is the one this chapter has described.

2.4 The Architecture in Detail

Before proposing a fix we need a clear account of what is broken and where. A credit system, viewed as an information-processing machine, has five components. Each performs a distinct function, and each can fail independently.

2.4.1 The Five Components

Origination is the process by which a loan application becomes a credit decision. It includes the collection of documents, the assessment of the borrower's financial position, the evaluation of the project, and the approval or rejection. The origination function is where the loan officer in Zhejiang works. It is the point at which the lender's information disadvantage is most acute, because the borrower controls the flow of information and the lender has not yet established a monitoring relationship.

Pricing is the mapping from assessed risk to interest rate and fees. It includes the risk premium, the funding cost, the operating cost, and the capital charge. Pricing is where the rating feeds in: the grade determines the expected loss, the expected loss determines the risk premium, and the risk premium determines the rate. If the grade does not capture environmental risk, the price does not either.

Contracting is the specification of the legal terms: covenants, collateral, maturity, amortization, cross-default clauses, and events of default. Contracting is where the lender converts its assessment into enforceable constraints. A covenant that requires environmental permits to remain in force is a contracting response to environmental risk. A covenant that is not monitored is a contracting failure.

Monitoring is the ongoing observation of the borrower after origination. It includes the review of financial statements, site visits, covenant compliance checks, and early-warning triggers. Monitoring is where the lender detects deterioration before default. It is the most labor-intensive component and the one most often under-resourced.

Workout is the process that begins when the borrower defaults or is about to default. It includes restructuring, collateral enforcement, and, in the limit, bankruptcy proceedings. Workout is where the lender recovers what it can. Environmental liabilities can complicate workout substantially: a contaminated site may be worth less than the collateral valuation assumed, and the lender may inherit remediation liability if it takes possession.

The five components form a sequence. Information gathered at origination feeds pricing and contracting. Contracting sets the terms that monitoring enforces. Monitoring generates the early warnings that trigger workout. A failure at any point propagates forward.

flowchart LR
    A[Origination] --> B[Pricing]
    B --> C[Contracting]
    C --> D[Monitoring]
    D --> E[Workout]
    A -.->|Information| B
    B -.->|Risk premium| C
    C -.->|Covenants| D
    D -.->|Early warning| E
    E -.->|Recovery| A

Figure 2.3: The five components of the credit architecture and the information flows between them.

2.4.2 Where the Architecture Fails on Environmental Risk

Each component fails in a characteristic way.

Origination fails because the loan officer cannot evaluate environmental technology. The failure is not a lack of diligence; it is a mismatch between the training of a credit analyst and the knowledge required to assess a novel green project. The analyst can read a balance sheet and a cash-flow projection. She cannot assess whether a heat-pump control module will perform as specified, whether the market for it will materialize, or whether the patents that protect it are defensible. The origination function has no slot for this kind of information, so the information is not collected, and the decision is made without it.

Pricing fails because the risk premium is calibrated on historical default data, and the historical data do not contain the environmental risk that is now materializing. A sector that has not defaulted in the past because environmental regulation was lax will be priced as if it will not default in the future when regulation tightens. The pricing error is not random; it is systematically optimistic for dirty firms and systematically pessimistic for clean ones, because the historical data reflect a regulatory regime that is changing.

Contracting fails because the standard covenants are financial. A loan agreement typically requires the borrower to maintain a debt-service coverage ratio, a leverage ratio, and a minimum net worth. It may require insurance on the collateral. It rarely requires an emissions limit, a remediation reserve, or a transition plan. The covenants that would address environmental risk are not in the standard template, and the loan officer who wants to add them faces a negotiation with a borrower who has no reason to accept.

Monitoring fails because environmental data are not in the reporting package. The borrower reports its financial statements quarterly and its environmental performance, if at all, annually and in a form that is not comparable across borrowers. The lender's monitoring system is built around financial ratios, and the environmental indicators are not integrated. A deterioration in environmental performance that precedes a financial deterioration is invisible until the financial deterioration arrives.

Workout fails because environmental liabilities are contingent and hard to value. A lender that forecloses on a factory may find that the site is contaminated and that the cost of remediation exceeds the value of the land. The lender may also find that it has inherited liability under environmental law, depending on the jurisdiction. The workout function was designed for financial distress, not for environmental distress, and the two require different expertise.

2.4.3 The Failure Is Structural, Not Personal

These failures are not the fault of individual bankers. The loan officer in Zhejiang is competent and diligent. She defers the heat-pump loan because the architecture gives her no basis for approving it. If she approved it and it defaulted, she would be blamed for a decision she could not justify. If she declines it and it would have succeeded, no one will ever know. The incentive structure rewards the conservative decision, and the conservative decision is to lend to the firm with collateral.

The structural nature of the failure has a policy implication. Telling bankers to lend more to green firms, without changing the information architecture, will produce either cosmetic compliance or increased losses. The first is worse than the second because it hides the problem. The CBIRC's green credit targets have been met largely by lending to the asset-heavy projects that fit the existing architecture, which is why the headline growth has not been matched by a comparable growth in lending to green technology firms.

2.4.4 The Interaction Between Components

The components do not fail independently. A failure at origination propagates to pricing, because the price is based on the origination assessment. A failure at pricing propagates to contracting, because the covenant package is calibrated to the risk premium. A failure at contracting propagates to monitoring, because the monitoring system checks the covenants that were written. A failure at monitoring propagates to workout, because the early warning that would have allowed a restructuring arrives too late.

The propagation means that a partial fix is less effective than it appears. Adding environmental information at origination, without changing pricing, produces a better-informed decision that is then priced as if it were uninformed. Adding environmental covenants, without a monitoring system that checks them, produces paper compliance. The fix has to be systemic, which is why the book's argument is about a mechanism rather than a single instrument.

2.5 The Special Case of Technology Firms

The rationing problem is general, but it is acute for technology firms, and the book's prior research on credit rating for technology enterprises is directly relevant. This section sets out why technology firms are the hardest case and what that implies for the rating design in Chapter 4.

2.5.1 The Asset Structure Problem

A technology firm's balance sheet is weighted toward intangible assets: patents, software, trade secrets, and human capital. Intangible assets are poor collateral for three reasons.

The first is valuation uncertainty. A patent's value depends on its scope, its validity, and the market for the product it protects. These are contested in litigation and in licensing negotiations. A lender that forecloses on a patent may find that the patent is worth less than the valuation assumed, or that it is worth nothing because the key inventor has left.

The second is liquidity. The market for patent collateral is thin. A lender that seizes a patent may be unable to sell it at any price near its going-concern value. The liquidation value is a fraction of the book value, and the fraction is uncertain.

The third is control. A patent is only valuable if the firm that owns it can exploit it. A lender that takes possession of a patent may not have the capability to exploit it, and the value may be destroyed by the transfer. This is different from a factory, which can be sold to another operator without losing its productive capacity.

The consequence is that a technology firm's borrowing capacity is limited by its tangible assets, which are small, rather than by its cash flows, which may be large and growing. The firm is rationed not because its project is bad but because its balance sheet does not fit the collateral template.

2.5.2 The Cash-Flow Problem

The alternative to collateral-based lending is cash-flow-based lending. A firm with stable, predictable cash flows can borrow against them without pledging assets. The lender assesses the firm's ability to service debt from operating cash flow and monitors the cash flow over time.

Technology firms are hard to lend to on a cash-flow basis for two reasons. The first is volatility. A technology firm's revenue can grow rapidly or collapse rapidly, depending on whether its product finds a market. The volatility is not necessarily a sign of poor quality; it is a feature of the innovation process. But it makes the cash-flow projection unreliable, and the lender's risk premium rises accordingly.

The second is the investment phase. A technology firm in its growth phase is reinvesting its cash flow in research and development, which means that the cash flow available for debt service is small even when the firm is healthy. The firm may be profitable on an accrual basis and cash-poor on a cash basis. A lender that focuses on cash flow will decline the loan, even though the firm's long-run prospects are good.

The combination of weak collateral and volatile cash flow is what makes technology firms the hardest case. The two standard screening devices both fail. The lender is left with relationship, which takes time to build, and with the borrower's own track record, which is short.

2.5.3 The Environmental Dimension

For green technology firms, the problem is compounded by the environmental dimension.

A green technology firm's environmental benefit is its selling point, but the benefit is often not priced in the market. A firm that produces a control module that reduces gas consumption by 30 percent creates value for the household that buys it, but the value depends on the price of gas, which depends on policy, which depends on politics. The firm's revenue is exposed to a policy risk that the lender cannot assess.

The environmental benefit is also hard to verify. A firm can claim that its product reduces emissions by a certain amount, but the claim depends on a baseline that is contested. The lender cannot verify the claim without expertise, and the claim is the basis for the firm's green credentials, which are the basis for any preferential treatment the firm might receive.

The result is that green technology firms face the general technology firm problem plus an environmental verification problem. They are the hardest case in the hardest segment of the market. If the architecture can be made to work for them, it can be made to work for everyone.

2.5.4 What the Prior Research Found

The book's prior research on credit rating for technology enterprises offers some guidance. The research developed a rating framework that weights four factors: the quality of the technology (assessed through patent analysis and expert review), the market position (assessed through customer concentration and contract backlog), the management team (assessed through track record and domain expertise), and the financial position (assessed through cash flow and funding runway).

The framework was tested on a sample of technology firms in a Chinese province. The main finding was that the technology and market factors had more predictive power for default than the financial factors, at least for firms in the early growth stage. This is the opposite of the conventional wisdom, which weights financial factors most heavily. The explanation is that financial statements for early-stage technology firms are noisy and lagged, while the technology and market factors are more stable and forward-looking.

The implication for green rating is that the environmental factor should be treated like the technology factor: assessed through expert review and structured criteria, not through a single financial ratio. This is the design principle behind the environmental performance rating in Chapter 3.

2.6 The Comparative Institutional Picture

The architecture described so far is general, but its specific form varies across jurisdictions. A brief comparison is useful because it shows which features are essential and which are contingent.

2.6.1 The Bank-Centered Model

In China and much of continental Europe, credit is intermediated primarily through banks. The bank originates the loan, holds it on its balance sheet, and monitors the borrower. The bank's information advantage comes from the deposit relationship, which gives it a view of the borrower's cash flows, and from the lending relationship, which gives it a history of repayment.

The bank-centered model is good at relationship lending. It is bad at arm's-length lending to firms without a deposit relationship. The loan officer in Zhejiang works in a bank-centered model, and her deferral of the heat-pump loan is a predictable consequence.

The bank-centered model has an advantage for green finance: the bank can internalize the environmental risk over the life of the loan. A bank that holds a loan for ten years has an incentive to monitor the borrower's environmental performance over ten years. A bank that sells the loan after origination does not.

2.6.2 The Market-Centered Model

In the United States, a larger share of credit is intermediated through capital markets. Firms issue bonds, and investors buy them. The information is produced by rating agencies, auditors, and disclosure requirements. The investor relies on the rating and the disclosure rather than on a relationship with the borrower.

The market-centered model is good at arm's-length lending to large firms. It is bad at lending to small firms, which cannot bear the fixed cost of public issuance. The rating agencies do not cover them, and the disclosure requirements are designed for large issuers.

For green finance, the market-centered model has an advantage in comparability: the rating and disclosure are standardized, so investors can compare across issuers. It has a disadvantage in monitoring: the dispersed bondholders have less incentive to monitor than a single bank.

2.6.3 The Hybrid Model

Most economies are hybrids. Large firms issue bonds and borrow from banks. Small firms borrow from banks and from informal sources. The rating infrastructure covers the large firms and not the small ones.

The hybrid model is where the green rating problem is most acute. The large firms have access to the market-centered information infrastructure, which is beginning to incorporate environmental disclosure. The small firms do not. The gap between the two is the gap that a bank-internal environmental rating has to fill.

2.6.4 The Policy Implication

The comparison suggests that the design of a green rating system should depend on the segment of the market it serves. A rating for large issuers can rely on public disclosure and third-party verification. A rating for small borrowers has to rely on bank-collected data and simplified criteria. A single rating system that tries to serve both will be either too burdensome for small borrowers or too coarse for large ones.

This is why Chapter 3 proposes a tiered rating: a full rating for large borrowers and a simplified rating for small ones, with a common core of environmental indicators. The common core ensures comparability; the tiers ensure feasibility.

2.7 The Cost of Asymmetry: A Quantitative Sketch

It would be useful to put a number on the cost of information asymmetry, even if the number is rough. The exercise is not precise, but it clarifies the magnitude.

2.7.1 The Deadweight Loss

Consider a population of green projects, each requiring one unit of capital and each with an expected social return \(S_i\) and an expected private return \(R_i\). The social return exceeds the private return because the project produces an environmental benefit that is not captured by the project's revenue. Let the difference be \(b_i = S_i - R_i\), the external benefit.

A lender that observes only \(R_i\) will finance the project if \(R_i\) exceeds the cost of capital. A social planner would finance it if \(S_i\) exceeds the cost of capital. The projects with \(R_i\) below the cost of capital and \(S_i\) above it are the ones that are privately unprofitable but socially profitable. These are the projects that the architecture fails to finance.

The size of the gap depends on the distribution of \(b_i\). For a project with a large external benefit, the gap is large. For a project with a small external benefit, the gap is small. The projects with the largest external benefit are the ones most likely to be rationed, because their private return is furthest below their social return.

This is the deadweight loss of information asymmetry in green finance. It is not a transfer from one party to another; it is a loss of value that no one captures. The project is not built, the emissions reduction does not occur, and the social benefit is foregone.

2.7.2 Estimating the Magnitude

A precise estimate would require data on the distribution of \(b_i\) across projects, which we do not have. But we can bound the magnitude using the observed gap between the social cost of carbon and the private cost of emissions.

The social cost of carbon is estimated in the range of $50 to $200 per ton of CO2, depending on the discount rate and the damage function. The private cost of emissions, in the absence of a carbon price, is zero. The gap is the external benefit of abatement, which is the value that a green project creates but does not capture.

A green project that abates one ton of CO2 per year for twenty years creates an external benefit of $1,000 to $4,000 at a discount rate of 5 percent and a social cost of carbon of $50 to $200. If the project requires an investment of $10,000, the external benefit is 10 to 40 percent of the investment. This is a substantial fraction, and it is the fraction that the private market does not see.

The implication is that the rationing of green projects is not a marginal problem. The projects that are rationed are the ones with the largest external benefits, and the external benefits are large relative to the investment. The deadweight loss is correspondingly large.

2.7.3 The Counterfactual

The estimate above is a partial equilibrium calculation. It ignores the general equilibrium effects: the abatement reduces the social cost of carbon, which reduces the external benefit of further abatement, which reduces the deadweight loss. It also ignores the learning effects: the first projects in a new technology are more expensive than later ones, so the external benefit of the early projects includes the cost reduction they enable for later projects.

These effects work in opposite directions. The general equilibrium effect reduces the deadweight loss; the learning effect increases it. The net effect is uncertain, and the uncertainty is a reason for caution in using the estimate for policy. But the order of magnitude is clear enough: the cost of information asymmetry in green finance is not trivial, and it is large enough to justify investment in the information architecture.

2.8 The Design Problem

The credit architecture fails on environmental risk because its five components were built for a world in which the relevant information was financial and the relevant horizon was short. The failure is structural, not personal. It produces rationing of green projects, especially green technology projects, and the rationing has a social cost that is large relative to the investment.

The design problem is to modify the architecture so that it can process environmental information without losing the properties that make it work: the comparability of the rating, the enforceability of the contract, the reliability of the monitoring, and the recoverability of the workout.

The solution has three layers, corresponding to the three failures that the architecture cannot address on its own.

The first layer is rating: a system that measures environmental performance on a comparable scale and feeds the measurement into origination, pricing, and contracting. The rating addresses the information asymmetry at origination. Chapters 3 through 5 develop it.

The second layer is buffering: a set of instruments that transfer or absorb the risk that remains after the rating. Insurance is the primary instrument. A weather index insurance contract transfers the rainfall risk that a farmer cannot diversify. A forest insurance contract transfers the fire risk that a forestry firm cannot self-insure. The buffering layer addresses the residual risk that the rating identifies but cannot eliminate. Chapters 6 through 9 develop it.

The third layer is catalysis: a set of institutions that channel capital toward early-stage green innovation, where the private return is furthest below the social return. Government-guided funds, sci-tech finance intermediaries, and blended finance structures are the instruments. The catalysis layer addresses the projects that the rating and the buffering layer cannot make commercially viable on their own. Chapters 10 through 12 develop it.

The three layers are not alternatives. They are complements, and they have to operate in sequence. A rating without buffering leaves the lender bearing a risk it cannot diversify. Buffering without rating leaves the insurer unable to price. Catalysis without either leaves the public fund investing blind.

2.9 Objections and Qualifications

Three objections deserve a response.

2.9.1 Is the Problem Really Information Asymmetry?

A skeptic might argue that the rationing of green projects is not an information problem but a profitability problem. If green projects were profitable, the market would finance them. The fact that they are not financed suggests that they are not profitable, and no amount of information will change that.

The objection has force for projects whose social return is genuinely below their private cost. For those projects, information is not the binding constraint; the economics are. But the objection conflates two cases. A project can be privately unprofitable because its social benefit is external (the case for catalysis), or it can be privately profitable but unfinanced because the lender cannot see the profit (the case for rating). The two cases require different remedies, and the book's argument is that both are present and both need to be addressed.

The evidence that the second case is real comes from the loan officer's behavior. She does not decline the heat-pump loan because she has calculated that it is unprofitable. She declines it because she cannot calculate. The distinction matters: if she had calculated and found it unprofitable, the remedy would be a subsidy; because she cannot calculate, the remedy is information.

2.9.2 Is Rating the Right Remedy?

A second skeptic might argue that rating is a weak remedy. The credit rating agencies failed to predict the 2008 financial crisis, and their reputation is damaged. Why should we expect a green rating to do better?

The objection is fair, and the answer is that the credit rating agencies failed for specific reasons that can be addressed in the design of a green rating. The failures were: a conflict of interest in the issuer-pays model, a reliance on historical data that did not cover the relevant scenarios, and a complexity that exceeded the agencies' understanding. A green rating can be designed to avoid these: a public or bank-internal funding model, a forward-looking methodology that incorporates scenario analysis, and a simplicity that keeps the rating within the range of the rater's competence.

None of this guarantees success. The green rating will have its own failures. But the choice is not between a perfect rating and no rating; it is between an imperfect rating and the current situation, in which the lender has no environmental information at all. An imperfect rating that is transparent about its limitations is better than no rating.

2.9.3 Does the Argument Apply Outside China?

The book's evidence is drawn heavily from China, and a reader might ask whether the argument generalizes.

The architecture described in this chapter is not specific to China. The five components are present in every credit system. The information asymmetry is present wherever the borrower knows more than the lender. The rationing of green projects is documented in the European Union, the United States, and emerging markets. The specific institutional forms differ, but the structural problem is the same.

What differs is the policy response. China has been more explicit about green credit targets, which has produced both the rapid growth in green lending and the composition problem described above. The European Union has relied more on disclosure requirements and taxonomy, which has produced better comparability but slower growth in lending to small green firms. The United States has relied more on tax incentives and market-based instruments, which has produced innovation but also volatility. The comparison is instructive, and Chapter 15 returns to it.

2.10 Summary of the Chapter

Credit risk is the possibility of loss from a counterparty's failure to honor an obligation, and it is analyzed through the lens of information asymmetry. Adverse selection operates before the contract, moral hazard after. Environmental information adds three complications: a longer horizon, correlated risk across borrowers, and verification difficulty.

Information asymmetry produces credit rationing. A profit-maximizing lender may ration rather than raise the rate, because higher rates attract worse borrowers and induce riskier behavior. The rationing devices—collateral, covenants, relationships, sectoral limits—are biased against green firms, and especially against green technology firms, whose assets are intangible and whose cash flows are volatile.

The rationing has a social cost. The projects that are rationed are the ones with the largest external benefits, and the external benefits are large relative to the investment. The cost is a deadweight loss that no one captures.

Rating is a screening and signaling device that reduces information asymmetry. It works when the rating is comparable, verified, and backed by institutional infrastructure. The credit rating model provides a template, but it has to be adapted: the object of the rating has to be specified (obligor, use of proceeds, or lender risk), the methodology has to be forward-looking, and the coverage has to extend to small borrowers.

The architecture has five components—origination, pricing, contracting, monitoring, workout—and each fails on environmental risk in a characteristic way. The failure is structural, not personal, and it propagates across components. A partial fix is less effective than it appears. The fix has to be systemic, which is why the book proposes a three-layer mechanism rather than a single instrument.

The next chapter takes up the first layer. It asks how a credit rating can be extended to incorporate environmental performance, what the design choices are, and how the resulting rating can be validated. The rating is a screening device, it operates at origination, and it feeds pricing and contracting. The question is how to construct it so that it is comparable, verifiable, and useful to the loan officer who deferred the heat-pump loan.

The answer, in brief, is that the rating has to be built from indicators that are observable at reasonable cost, weighted by their predictive power for the outcomes the lender cares about, and validated against realized defaults. The construction is the subject of Chapter 3. The application to technology firms is the subject of Chapter 4. The portfolio consequences are the subject of Chapter 5.

The loan officer in Zhejiang is not the villain of this story, and she is not the audience for a moral appeal. She is a professional making a rational decision within an architecture that gives her no basis for a different one. Change the architecture so that her rational decision becomes the one that finances the heat-pump modules, and the deferral stops. The first thing the new architecture has to give her is a number she can defend to her credit committee: a grade that tells her, on a scale her colleagues recognize, how the heat-pump manufacturer's environmental performance and forward risk compare with the twelve dyeing firms already on her book. Chapter 3 builds that grade.