Chapter 3: Rating-Based Identification: Incorporating Environmental Performance into Credit Assessment

A credit committee at a provincial commercial bank sits down on a Tuesday morning to review a loan application from a mid-sized solar-cell manufacturer. Revenue has grown 40 percent over three years, debt-service coverage sits at 2.1, and collateral covers 130 percent of the requested facility. By every traditional measure the file is a straightforward approval. The committee hesitates anyway. The manufacturer's wastewater discharge permits were renewed only after a local environmental bureau inspection found elevated fluoride levels. Its largest customer is a state-owned utility whose payment behavior has deteriorated over two quarters. The production line depends on a polysilicon supply chain that trade policy has whipsawed twice in eighteen months. None of these facts appears in the rating score. The committee approves at a standard risk premium. Fourteen months later the facility is restructured after the utility's payments stall.

That vignette describes what this chapter calls the green blind spot of traditional credit rating: a systematic failure to encode environmental performance and transition exposure into the probability-of-default estimate. The failure is structural. Rating models were built to predict repayment, and the variables that predict repayment were selected from decades of default data in which environmental liabilities were rare, unpriced, and absorbed by public authorities. The U.S. Superfund legislation of the early 1980s changed the liability regime—for the first time, firms could be held responsible for pollution they caused, and lenders began to guard against credit risk arising from contaminated collateral—but the modeling conventions had already hardened. In 1992, more than 100 financial institutions signed the UNEP Statement by Financial Institutions on the Environment and Sustainable Development in New York, committing to integrate environmental considerations into lending. Three decades on, the integration remains partial, and in most emerging markets it has barely begun.

This chapter sets out how to close the gap. The first section diagnoses why traditional ratings neglect environmental performance and why that neglect is a credit-risk problem rather than only an ethical one. The second turns to indicator design: which environmental variables carry information about default, how they should be measured, and how they should be weighted. The third assembles the pieces into a modified rating framework that combines financial and environmental scores for green technology firms, then works through the calibration questions that determine whether the framework improves discrimination or merely adds noise.

The argument stays narrow. A rating is an instrument for sorting borrowers by default probability. If environmental performance carries information about default probability, excluding it degrades the rating. If it does not, including it is decoration. The empirical question is which environmental indicators carry that information, at what horizon, and for which sectors. The design question is how to add them without double-counting risks the financial score already captures.

3.1 Traditional Credit Rating and Its Green Blind Spot

3.1.1 What a Rating Is For

A credit rating is a forward-looking opinion about an obligor's willingness and ability to meet its financial obligations in full and on time. In the quantitative tradition that dominates bank internal ratings, the rating feeds a two-step procedure: assign the obligor to a grade, then map the grade to a historical default frequency. Expected loss on a facility is the product of three parameters:

where \(PD\) is the probability of default over the horizon, \(LGD\) is loss given default (one minus the recovery rate), and \(EAD\) is exposure at default. The rating's job is to order obligors by \(PD\). Pricing, provisioning, capital allocation, and portfolio limits are all downstream of that ordering.

The ordering is only as good as the variables that feed it. A typical corporate rating model draws on four families of inputs: profitability and cash flow (EBIT margin, funds from operations to debt), leverage and coverage (debt to EBITDA, interest coverage), liquidity (current ratio, cash to short-term debt), and qualitative factors (management quality, industry position, ownership support). Each variable earns its place because it has been shown, across large samples and long horizons, to separate defaulters from survivors. Fitch's global corporate finance transition tables—one-year and five-year average transition rates, and average annual default rates by rating—are the kind of evidence that underwrites variable selection. A variable that does not improve out-of-sample separation does not belong in the model, however intuitively appealing it may be.

Environmental performance has, until recently, failed that test. The reasons are partly about data and partly about economics. Environmental liabilities were historically rare and concentrated in a handful of heavy-industry sectors, so a pooled default model estimated on a broad corporate sample would find little signal. Where environmental costs were imposed, they were often imposed on the public balance sheet rather than the firm's—through state-funded remediation, tolerant regulators, unpriced water and air. A firm that polluted heavily but paid nothing for it had no worse a cash-flow profile than a firm that polluted lightly. The rating was right, given the prices.

That condition is ending, and three changes matter for the rating model. Liability regimes have tightened: Superfund in the United States, the EU Environmental Liability Directive, China's Environmental Protection Law and its successive amendments, and the pollution-discharge permitting systems that now attach to individual facilities. A firm can be made to pay for contamination, and the payment can be large relative to its cash flow. Transition policy has begun to reprice assets through carbon pricing, emissions-trading systems, renewable portfolio standards, and the withdrawal of permits for high-emission capacity. A firm whose asset base is misaligned with the transition faces a cash-flow shock that trailing financial ratios do not capture. Disclosure has improved enough to measure, through mandatory environmental reporting in the EU, China's enterprise environmental credit evaluation, and the spread of satellite and sensor data on emissions and discharges.

The blind spot is a lag between the world and the model: liability regimes, carbon pricing, and disclosure have moved faster than rating conventions. The rest of this section explains why the lag persists and what it costs.

3.1.2 The Four Channels of Environmental Credit Risk

Environmental performance affects default probability through four distinct channels. Distinguishing them matters because they call for different indicators and different weights.

Contamination produces cleanup costs, fines, and third-party damages—contingent liabilities that can crystallize suddenly and, in the limit, exceed the firm's equity. The Superfund regime made this channel salient for U.S. lenders in the 1980s. China's central environmental inspection campaigns have made it salient for Chinese lenders since 2016, when inspections across provinces produced waves of fines, production suspensions, and forced relocations. What matters for credit is the size of the potential liability relative to cash flow and equity, adjusted for the probability of enforcement.

Operating permits can be revoked, tightened, or made conditional on capital expenditure. A cement plant whose emissions limit is lowered must either invest in abatement or lose the right to operate. The investment is a cash outflow; the loss of the permit is a terminal event for the facility. The credit-relevant variable is the gap between the firm's current environmental performance and the standard it will be held to over the loan's tenor, together with the capital cost of closing that gap.

A firm selling into a market that is being decarbonized faces demand erosion. An auto-parts supplier whose product mix is weighted to internal-combustion components faces a revenue cliff as electric-vehicle penetration rises. A coal-fired generator faces both a carbon price and a shrinking dispatch window. Here the credit-relevant variable is the share of revenue exposed to transition-sensitive end markets, and the firm's capacity to redeploy its assets.

Floods, droughts, heat stress, and water scarcity affect operations, input costs, and logistics. A manufacturer whose single plant sits in a floodplain carries a tail risk that a diversified competitor does not. A food processor whose agricultural inputs are exposed to drought carries commodity-price risk that shows up in margins. The credit-relevant variable is the geographic and supply-chain concentration of physical exposure.

The four channels are not equally important across sectors. For heavy industry, liability and permit risk dominate. For consumer-facing and technology firms, transition risk dominates. For agriculture and forestry, physical risk dominates, which is why the insurance instruments discussed in Chapters 5 and 6 are the natural complement to rating in those sectors. A rating framework that applies a single environmental score across all sectors will mis-weight the channels and degrade discrimination. The framework has to be sector-differentiated.

3.1.3 Why the Market Has Not Fixed the Blind Spot

If environmental performance carries credit information, why have rating agencies and banks not already incorporated it? Four explanations, each with a grain of truth.

The first is data scarcity at the obligor level. Aggregate ESG ratings exist, but they are firm-level composites built for equity investors, not facility-level measures built for lenders. ESG scores from the major providers correlate at roughly 0.3 to 0.5, and Berg, Kölbel and Rigobon (2022) trace the divergence to differences in scope, measurement, and weighting rather than to disagreement about the underlying facts. A lender cannot build a \(PD\) model on a score whose construction it does not understand and whose variance across providers approaches the variance across firms.

The second is horizon mismatch. Environmental liabilities often crystallize over five to fifteen years; corporate loan tenors are typically one to five. An analyst who discounts a remediation liability that may materialize in 2035 against a loan maturing in 2027 is behaving rationally, given the contract. The blind spot is therefore partly a horizon problem: it matters most for long-tenor project finance, infrastructure lending, and mortgage books, and least for short-tenor working-capital facilities.

The third is the pricing problem. A bank that incorporates environmental risk into its rating and prices the loan accordingly will lose the dirty borrower to a bank that does not. The first mover bears a cost; the free rider captures the margin. This is a collective-action problem that individual institutions cannot solve, which is why the policy response—mandatory disclosure, common taxonomies, supervisory expectations—matters. The EU's Action Plan on Financing Sustainable Growth (COM (2018) 97 final) and its successors, together with the SFDR classification of financial products, are attempts to change the equilibrium by making the information public and the comparison mandatory.

The fourth is model risk aversion. Adding variables to a rating model requires revalidation, backtesting, and supervisory approval. A model with a track record is an asset; a model under revision is a liability. Institutions therefore change models slowly, and the slow change is rational at the level of the individual institution even when it is costly at the level of the system.

None of these explanations suggests the blind spot is benign. Each explains why it persists and what would be needed to close it. The remainder of the chapter treats closure as the design problem.

3.1.4 The Cost of the Blind Spot: Two Vignettes

Make the cost concrete. Consider a steel processor in a province subject to episodic environmental inspections. The firm's financial ratios are stable; its rating is investment grade. An inspection finds that its wastewater treatment facility has been bypassed during peak production. The firm is fined, ordered to suspend production for three weeks, and required to install a new treatment line. The suspension removes a quarter's operating cash flow; the installation consumes the rest of the year's free cash flow. The firm misses an interest payment and is downgraded. In retrospect the downgrade looks predictable—the bypass was visible in the firm's discharge data and in the pattern of past inspections—but the rating model never saw it, because discharge data were not an input.

Now consider a solar-equipment manufacturer. Its financials are strong, its growth is rapid, and its rating is upgraded on trailing revenue and margin. What the model does not see is that 70 percent of revenue comes from a single province's subsidy program, that the program's budget has been cut in the current fiscal year, and that the firm's inventory of finished modules is rising while its receivables from the provincial utility are aging. The environmental performance is excellent—the firm produces a low-carbon product—but the environmental exposure is concentrated. A rating framework that scores environmental performance without scoring environmental exposure will upgrade exactly the firms most vulnerable to transition-policy reversal.

The two vignettes point in opposite directions. The first is a dirty firm whose environmental liability is invisible to the model. The second is a clean firm whose environmental revenue dependence is invisible to the model. A framework that only penalizes bad environmental performance will fix the first and miss the second. The design must score both performance and exposure.

3.1.5 What "Green" Means for a Rating

Settle a definitional question that recurs throughout the book. In credit rating, "green" describes how an obligor's environmental characteristics bear on its default probability. A firm can be green in the sense of producing an environmental benefit (a solar manufacturer), green in the sense of low environmental harm (a software company), or green in the sense of being aligned with the transition (a firm whose asset base is compatible with a decarbonized economy). These are different properties, and they map onto the rating differently.

Environmental benefit affects the rating mainly through revenue risk. A firm producing a transition-compatible good has a demand tailwind, but it may also have policy dependence, subsidy dependence, and technology risk. The net effect on \(PD\) is ambiguous and must be assessed case by case.

Low harm affects the rating mainly through the liability and regulatory channels. A firm with low emissions and discharges carries less contingent liability and less permit risk. The effect on \(PD\) is usually favorable, but the magnitude is small for firms in sectors where environmental liabilities are already well priced.

Transition alignment affects the rating through the asset base. A firm whose assets are compatible with a decarbonized economy faces less transition risk than a firm whose assets are not. The effect on \(PD\) is favorable over long horizons and small over short horizons.

A rating framework must therefore distinguish performance (how much environmental harm the firm causes), exposure (how much transition and physical risk the firm faces), and alignment (whether the firm's products and assets are compatible with the transition). The three are correlated but not identical, and conflating them produces a score that is neither a risk measure nor a policy tool. The next section takes up indicator design with this distinction in hand.

3.2 Designing Environmental Indicators for Rating

3.2.1 Selection Principles

The environmental indicators that belong in a rating model are not the same as the indicators that belong in a sustainability report. A sustainability report aims to describe the firm's environmental footprint comprehensively. A rating model aims to predict default. Five principles follow from that difference.

Materiality to default comes first. An indicator earns a place only if it carries information about the probability of default over the relevant horizon. Carbon intensity matters for a cement producer because a carbon price raises its costs and threatens its permit; it matters much less for a software firm, whose carbon intensity is low and whose exposure to carbon pricing is indirect. Materiality is sector-specific and must be assessed sector by sector.

Measurability at the obligor level comes next. The indicator must be observable for the individual borrower, not only for the sector. Sector-average carbon intensity is a useful input to a sector outlook; it is not a rating variable. Facility-level discharge data, permit status, and remediation liabilities are rating variables because they attach to the obligor.

Timeliness matters because the indicator must be available at the frequency of the rating cycle. Annual sustainability reports are too slow for a quarterly review; continuous emissions monitoring, permit databases, and inspection records are fast enough. The Chinese enterprise environmental credit evaluation, which aggregates regulatory penalties, permit violations, and inspection findings into a firm-level score, is an example of a data source designed for this purpose.

Non-redundancy is the fourth test. The indicator must add information beyond the financial variables already in the model. A firm with high energy intensity has high operating costs, which show up in the margin; if the margin is already in the model, the energy-intensity indicator adds little unless it also captures a regulatory or transition exposure that the margin does not. The test is incremental discriminatory power, not conceptual appeal.

Robustness to gaming is the fifth. The indicator must be hard to manipulate. Self-reported sustainability metrics are easy to game; third-party monitored emissions, satellite data, and regulatory enforcement records are harder. Research on greenwashing detection—including work by Yu, Luu and Chen (2020) showing that activist short-sellers target firms with weak environmental disclosure—confirms that reported environmental performance diverges from actual performance when reporting is voluntary and unaudited.

These principles narrow the candidate set sharply. What survives is a small number of indicators, most of them regulatory or physical rather than self-reported. Table 3-1 organizes them by channel.

Table 3-1: Candidate environmental indicators by credit-risk channel

The table is deliberately short. A rating model that adds twenty environmental variables will overfit; a model that adds five well-chosen ones will generalize. The next subsections take the four channels in turn and specify indicator construction.

3.2.2 Liability Indicators

The liability channel is the oldest and best understood. Its credit logic is straightforward: a contingent liability is a claim on the firm's cash flow that ranks alongside debt. If the liability is large relative to the firm's capacity to pay, it raises \(PD\).

Construct the indicator as a ratio:

The numerator requires an estimate of expected cost, not the maximum statutory penalty. The estimate should combine the size of the contaminated site or the violation, the probability of enforcement, and the firm's track record of contesting or settling. In practice, lenders use a three-tier approach: known liabilities from regulatory records, probable liabilities from site assessments, and possible liabilities from sector exposure. Only the first two enter the quantitative score; the third enters the qualitative overlay.

China's environmental inspection campaigns provide a natural experiment. During the first round of central inspections in 2016–2017, provinces received thousands of enforcement actions, and firms in inspected sectors experienced measurable cash-flow shocks. Lenders with access to the inspection records could have anticipated those shocks; lenders without access could not. The data exist. The constraint is access, not availability.

3.2.3 Regulatory Indicators

The regulatory channel is the most predictive for permitted facilities. A firm that is out of compliance faces a spectrum of outcomes from a small fine to a production suspension to a permit revocation, and the expected cost is the probability-weighted sum. The indicator should capture both current compliance status and trajectory.

Construct a graded compliance score:

The first term captures the current state; the second captures the trajectory; the third captures the cost of remediation. The weights should be calibrated so that a current violation with a large compliance gap moves the score more than a historical violation that has been remediated.

For emissions-intensive sectors, the compliance gap is best expressed against the forward standard, not the current one. A cement plant that meets today's emissions limit but will not meet the limit scheduled for three years hence carries a known future capex requirement. The rating horizon should match the loan tenor; if the loan matures after the standard tightens, the gap belongs in the score.

3.2.4 Transition Indicators

The transition channel is the newest and the hardest to measure. A firm whose revenue or asset base is misaligned with the transition faces a cash-flow shock as policy, technology, and demand shift. The shock can be gradual (margin compression) or sudden (asset stranding).

Two indicators carry most of the information. The first is revenue exposure:

The definition of "transition-sensitive" is sector-specific and should be set by reference to a published taxonomy. The EU taxonomy and China's green bond endorsed project catalogue provide workable definitions; the lender's job is to map the borrower's segment disclosure onto the taxonomy.

The second is asset alignment:

For a coal-fired generator, the incompatible assets are the plants that cannot be retrofitted to meet the emissions standard or that will be retired before the loan matures. For an auto-parts supplier, the incompatible assets are the tooling dedicated to internal-combustion components. The indicator is a stranding measure, and it should be computed on a loan-tenor basis: an asset that will be retired after the loan matures is not stranded from the lender's perspective, though it is from the equity holder's.

The two indicators are correlated but not identical. A firm can have high revenue exposure and low asset exposure (a trading firm that sources transition-sensitive commodities but holds no long-lived assets) or the reverse (a utility with a long-lived asset base but a regulated revenue stream that passes through transition costs). The rating model should include both.

3.2.5 Physical and Supply-Chain Indicators

The physical channel matters most for agriculture, forestry, real estate, and utilities. A physical hazard that disrupts operations or raises input costs reduces cash flow and raises \(PD\). The indicator should capture exposure, not just historical loss experience, because the historical record understates forward risk when the climate is changing.

Construct a physical exposure index:

where \(s_{ij}\) is the share of the firm's operations (by revenue or asset value) in location \(j\), and \(h_j\) is the hazard intensity at location \(j\) for the relevant peril (flood, drought, heat, cyclone). The hazard intensity should come from geospatial data, not from the firm's self-report. The index is a weighted average of location hazards, and it can be computed for any firm with a facility register.

The supply-chain indicator is a variant:

where \(b_{ik}\) is the share of input \(k\) sourced from climate-sensitive regions and \(h_k\) is the hazard intensity in those regions. For a food processor, the input is an agricultural commodity; for a textile firm, cotton; for an electronics assembler, water for chip fabrication. The indicator captures the indirect exposure that the facility-level index misses.

3.2.6 Management and Governance Indicators

The management channel is the softest and the most prone to gaming. A firm with a functioning environmental management system is less likely to suffer an unanticipated environmental shock, because it monitors, detects, and corrects problems before they become liabilities. The indicator is best used as a qualitative overlay rather than a quantitative input, because the certification is binary and the underlying quality varies.

Two indicators are worth retaining. ISO 14001 certification establishes that an external auditor has reviewed the management system. Environmental incident history captures the firm's realized safety record. A firm with repeated incidents despite certification is a worse credit than a firm with a clean record and no certification; incident history should therefore carry more weight than the certificate. Neither indicator should be allowed to dominate the score. In the sector weightings below, management receives 10 to 40 percent of the total, with the higher end reserved for sectors where operational environmental risk is the main exposure and hard data are thin—software and services, where the environmental footprint is small enough that management attention is the best available proxy.

3.2.7 Weighting the Indicators

The weighting problem is the heart of the design. Three approaches are available, and they should be combined rather than chosen.

Statistical weighting estimates a default model with the environmental indicators as candidate variables and lets the coefficients set the weights. This is the gold standard, and it is rarely feasible because default samples with environmental data are small. The practical compromise is to estimate the model on a broader sample with a proxy for environmental risk (sector-level environmental intensity) and then calibrate the obligor-level indicators to the sector-level coefficients.

Expert weighting convenes a panel of credit analysts and environmental specialists, asks them to score the indicators' materiality by sector, and uses the average scores as weights. This is transparent and auditable, and it is the approach most banks actually use. Its weakness is that expert weights are not validated against default outcomes, so they can be systematically wrong.

Regulatory weighting adopts the weights embedded in a supervisory framework or a published taxonomy. The advantage is comparability across institutions; the disadvantage is that the regulatory weights are set for the average bank and may not fit a particular portfolio.

The recommendation is a hybrid: use statistical weighting where the data permit, expert weighting where they do not, and regulatory weighting as a floor. The weights should be documented, reviewed annually, and backtested against realized defaults. A weight that cannot be justified by one of the three approaches should be set to zero.

The weighting must also be sector-differentiated. The four channels have different relative importance across sectors, and a single weight vector will mis-weight them. Table 3-2 gives an illustrative sector weighting scheme.

Table 3-2: Illustrative channel weights by sector (sum to 1.0)

The weights are illustrative, not definitive. The point is that the weighting must be explicit and sector-specific; the specific numbers should be calibrated to the lender's portfolio and validated against default experience.

3.2.8 The Aggregation Problem

Once the indicators are constructed and weighted, they must be aggregated into a single environmental score. Two issues arise.

The first is compensability. A linear weighted sum allows strong performance on one indicator to compensate for weak performance on another. That is appropriate when the indicators are substitutes (a firm with low liability exposure can afford a higher transition exposure) and inappropriate when they are complements (a firm with a permit violation and a large compliance gap is worse than the sum of the parts). The practical fix is to use a linear sum for the base score and a penalty overlay for combinations that cross thresholds: a firm with both a current violation and a compliance gap above a cutoff receives an additional downgrade notch.

The second is the treatment of missing data. Environmental data are patchier than financial data, especially for small and medium enterprises. A firm with no discharge data is not necessarily a clean firm; it may be an unmonitored firm. The aggregation should distinguish "no data" from "good data." The simplest approach is to assign missing indicators the sector-median value and to flag the score as low-confidence, which triggers a manual review. The alternative—dropping missing indicators and re-normalizing the weights—rewards firms for not reporting, which is the wrong incentive.

3.2.9 A Worked Example: The Tea Cooperative

Ground the design in a concrete case. A tea cooperative in a mountainous province applies for a working-capital facility. It has 240 member households, 180 hectares of tea gardens, and a processing facility. Its financial statements are thin: revenue is seasonal, receivables are concentrated in two buyers, and collateral is limited to the processing equipment and land-use rights.

Take the four channels in order. The processing facility uses a coal-fired boiler for drying. The boiler is subject to emissions standards, and the cooperative has no record of compliance testing. Replacing it with a gas or electric unit would cost roughly 1.2 times annual net income, which is material relative to cash flow. The water discharge permit is current, but the local environmental bureau has flagged the boiler; the compliance gap is the boiler replacement, and the ratio of replacement cost to annual capex is high because the cooperative undertakes little capex. Revenue is exposed to the price of tea, which is exposed to climate and to consumer preferences. Tea is not a transition-sensitive product in the way coal is, but the cooperative's buyers increasingly demand sustainability certification, which requires changes in fertilizer and pesticide use; the share of revenue going to buyers with sustainability requirements is rising. The tea gardens sit on a slope prone to erosion, and the region has experienced two drought years in the past five. The physical exposure index is high. The cooperative has no irrigation and no insurance.

The environmental score for the cooperative is therefore driven by physical exposure and by the boiler liability. The financial score is weak because the financials are thin. The combined rating falls below the threshold for the standard facility, but the loan officer can see that the binding constraints are the boiler and the drought exposure, both addressable: the boiler can be replaced with a subsidized green retrofit, and the drought exposure can be transferred through a weather index insurance policy (Chapter 5). The rating framework has identified the specific environmental risks that drive default probability and pointed to the instruments that can mitigate them.

The example also illustrates a limitation. The cooperative's environmental performance is good in the sense that tea is a low-carbon product and the gardens sequester carbon. A framework that scores only environmental harm would rate the cooperative well; a framework that scores environmental exposure rates it poorly. The correct rating depends on the channel, and the channel weights for agriculture put most of the weight on physical exposure. The cooperative is a good environmental actor with high environmental risk. The rating should reflect the risk, not the virtue.

3.3 A Modified Rating Framework for Green Technology Firms

3.3.1 Framework Architecture

The modified framework has three layers. A financial score carries the existing rating model, unchanged. An environmental score aggregates the indicators from Section 2. A combination rule determines how the two interact.

The architecture is deliberately modular. The financial score remains the anchor, because the default data that validate it are deep and the environmental data are shallow. The environmental score adjusts the financial score at the margin; it does not replace it. This is a statement about relative confidence, not relative importance. As environmental default data accumulate, the weight on the environmental score can rise.

The framework's output is a rating grade on the existing scale, not a new scale. This matters for adoption: a grade that maps onto the institution's existing PD master scale can be used immediately for pricing and capital, while a new scale would require a parallel reporting system. The mapping is the calibration problem taken up in Section 3.3.5.

3.3.2 The Financial Score

The financial score is the existing corporate rating model. For green technology firms, the model has three known weaknesses that the environmental score must not duplicate.

Green technology firms are often young and fast-growing, so their trailing financial ratios are unstable. A solar manufacturer that tripled revenue in two years has a debt-to-EBITDA ratio that reflects the investment phase, not the steady state. The financial model should use forward-looking ratios where available, or apply a growth-stage adjustment.

These firms also often depend on policy support—subsidies, feed-in tariffs, tax credits—that is embedded in revenue but not separately disclosed. The financial model should strip out policy-dependent revenue where possible, or flag it for the environmental score's transition channel.

Intangible-heavy balance sheets present the third difficulty: patents, licenses, and know-how that are not collateral in the traditional sense. The financial model should not treat these as worthless, but it should not over-rely on them either. The recovery rate for a failed solar manufacturer is driven by the resale value of its equipment, not its patents.

3.3.3 The Environmental Score

The environmental score is built in four passes. First, for each channel, compute the indicator as specified in Section 2 and normalize it to a 0–100 scale using the sector distribution, so that a score of 50 is the sector median and a score of 80 is the 80th percentile of environmental risk (higher score, higher risk). Second, combine the indicators within each channel using the sub-weights from Table 3-1; the channel score is a weighted average of its indicators, with a penalty overlay for threshold combinations. Third, combine the channel scores using the sector weights from Table 3-2 to produce the raw environmental score. Fourth, adjust for data quality: a score built on complete, third-party-verified data is more reliable than a score built on self-reported data with gaps, so the adjustment should widen the confidence interval rather than shift the point estimate, and the widened interval should trigger a manual review when it crosses a rating boundary.

The output is a score \(E_i \in [0, 100]\) with an associated confidence level.

3.3.4 The Combination Rule

Three rules are available for translating the environmental score into a rating adjustment.

An additive adjustment converts the environmental score to a notch adjustment and adds it to the financial rating:

where \(\alpha\) is the maximum notch adjustment (typically 1 to 2 notches) and the term in parentheses maps the environmental score to a \([-2, +2]\) range. The rule is simple, transparent, and easy to backtest. Its weakness is that it treats environmental risk as a linear add-on, which understates the effect of tail risks.

A multiplicative adjustment works on the PD directly:

where \(\beta\) is the maximum proportional adjustment (typically 0.2 to 0.5). The rule captures the idea that environmental risk scales with the base default probability: a high-risk firm with a high base PD is more affected than a low-risk firm with a low base PD. Its weakness is that it can produce large PD adjustments for firms at the speculative-grade boundary, which may be appropriate but is hard to communicate.

A threshold overlay applies the environmental score only when it crosses a threshold:

The rule is conservative: it adjusts the rating only when environmental risk is clearly high or clearly low. Its weakness is the cliff effect. A firm at a score of 80.1 receives a two-notch downgrade while a firm at 79.9 receives one, which creates an incentive to game the score toward the threshold. The fix is to smooth the overlay with a linear interpolation between thresholds, or to use the additive rule for the base adjustment and the threshold overlay only as a review trigger.

The recommendation is to use the additive rule as the primary adjustment and the threshold overlay as a review trigger. The additive rule is transparent enough to explain to a borrower and calibrated against realized defaults, and it is stable enough to backtest. The threshold overlay catches the tail cases the additive rule understates. The multiplicative rule is best reserved for portfolio-level stress testing, where the proportional adjustment is applied to the whole book rather than to individual obligors.

The choice of \(\alpha\) is the calibration question. Set it too low and the environmental score is decorative; set it too high and the rating becomes a policy instrument rather than a risk measure. The empirical anchor is the observed default differential between high-environmental-risk and low-environmental-risk firms within the same financial rating grade. If the differential is 30 basis points, then \(\alpha\) should be set so that the notch adjustment reproduces it. In the absence of such data—the common case—a conservative starting value is \(\alpha = 1\), meaning the environmental score can move the rating by at most one notch in either direction, with a manual review triggered when the score is extreme.

3.3.5 Calibration and the Master Scale

The framework's output must map onto the institution's PD master scale, the table that converts rating grades to default probabilities and, through them, to pricing and capital. The mapping is not automatic: a one-notch downgrade corresponds to a different PD multiple at different points on the scale. At the investment-grade boundary, a one-notch downgrade might double the PD; at the speculative-grade boundary, it might increase it by 30 percent. Calibration must therefore be done at the grade level, not the notch level.

The procedure runs in four stages. Estimate the base PD by grade, using the institution's historical default data or the rating agency's transition tableswhere the institution's data are thin. Fitch's global corporate finance average annual default rates by rating are a usable benchmark for the upper grades. Then estimate the environmental PD multiplier by grade. This is the hard step. In the absence of obligor-level default data with environmental indicators, use a sector-level proxy: compare default rates for high-environmental-intensity sectors to low-intensity sectors within the same grade, and use the ratio as the multiplier. The proxy is crude, but it beats assuming a multiplier of one. Next, apply the multiplier to the environmental score for each obligor, computing the adjusted PD as the base PD times the multiplier implied by the environmental score and the grade. Finally, backtest. Compare the adjusted PDs to realized defaults over the trailing period. If the adjusted PDs are systematically higher than realized defaults, the multiplier is too large; if lower, too small. Repeat the backtest annually and revise the multiplier when the deviation exceeds a tolerance band.

Calibration is where most green rating frameworks fail. Indicator design is intellectually interesting; calibration is tedious and data-hungry. A framework that is not calibrated is a scoring exercise, not a rating model. The institution that adopts the framework must commit to the calibration work, or it should not adopt the framework.

3.3.6 A Worked Example: The Solar Manufacturer

Return to the solar manufacturer from the opening vignette and work the framework through.

The financial score reflects revenue growth of 40 percent, an EBITDA margin of 12 percent, debt-to-EBITDA of 3.5, interest coverage of 4.2, and a current ratio of 1.4. The financial model assigns a grade of BBB, with a base PD of 0.30 percent over one year.

The firm's own operations are clean: low emissions, no discharge violations, ISO 14001 certified. The liability channel score is 20 and the regulatory channel score is 25. Transition is where the risk sits. Seventy percent of revenue comes from a single province's subsidy program, and the program's budget has been cut by 30 percent in the current fiscal year. The revenue exposure indicator is 70, and the asset alignment indicator is 45 (the plant is modern and can be repurposed, but the tooling is dedicated to a single module design). The transition channel score is 60. The physical channel score is 30 (the plant is inland, low flood risk, but water availability is a constraint in dry years). The management channel score is 25.

For electronics, the weights are liability 0.10, regulatory 0.15, transition 0.30, physical 0.20, management 0.25. The raw environmental score is:

A raw score of 36 sits below the sector median of 50, which looks wrong given the transition exposure. The management and liability scores are very low, and they pull the average down. The transition risk is real but diluted by strong performance on the other channels.

This is the aggregation problem in practice. A linear weighted sum allows strong performance on low-risk channels to mask high risk on the channel that matters most. The fix is the penalty overlay: when the transition channel score exceeds 55 and the revenue exposure indicator exceeds 60, the raw score rises by 15 points. The adjusted score is 51, just above the median.

With \(\alpha = 1\) and an adjusted score of 51, the notch adjustment is:

which rounds to zero. The rating stays at BBB. The framework has not changed the rating, but it has flagged the transition exposure for monitoring. The loan officer now knows that the firm's rating depends on the provincial subsidy program, and the next review will check the program's status.

Whether that is the right outcome depends on the horizon. Over one year, the subsidy cut is a manageable margin shock; the firm can cut costs and redirect sales. Over three years, if the program is phased out, the firm faces a revenue cliff. The rating horizon should match the loan tenor: a one-year working-capital facility can stay at BBB, while a five-year project loan should be downgraded. The framework should therefore be applied at the tenor of the exposure, not at a single horizon.

The example also shows the limits of the score. The firm is a green technology firm in the sense that it produces a transition-compatible product. A framework that rewarded green products would upgrade it. The credit framework does not upgrade it, because the product's greenness does not reduce the firm's default probability—if anything, the policy dependence raises it. The rating should reflect the risk, not the benefit.

3.3.7 The Green Technology Firm Adjustment

Green technology firms have three characteristics that the standard framework handles poorly, and each needs a specific adjustment.

The first is policy dependence. A large share of revenue often comes from subsidies, feed-in tariffs, or procurement preferences. The transition channel captures this through the revenue exposure indicator, but the indicator should be refined to distinguish policy-dependent revenue from market revenue. A solar manufacturer selling into a competitive utility market is less exposed than one selling to a subsidized program, even if both have the same revenue concentration. Split the revenue exposure indicator into two components: policy-dependent share and market-concentrated share, with the former weighted more heavily. The distinction matters most for firms whose subsidy revenue is booked at fixed tariffs over multi-year contracts, because the contract length determines how long the cash flow survives a policy reversal. A firm with five years of contracted tariff revenue and a loan maturing in three is not exposed to the reversal; a firm whose tariff is reset annually is exposed immediately.

The second is technology risk. Green technology firms operate on the frontier, where the technology can be superseded. A manufacturer of a first-generation battery chemistry faces the risk that a second-generation chemistry makes its product obsolete before the loan matures. The standard framework has no technology variable. Add a technology-position indicator to the transition channel: the firm's patent portfolio, its R&D intensity, and its position relative to the technology frontier. The indicator is noisy, but it captures a risk that the financial ratios miss. Patent counts are a weak proxy on their own; what carries information is the citation-weighted age of the portfolio and the share of revenue from products introduced in the last three years. A firm whose revenue still comes from a five-year-old product line in a market where the cost curve is falling 15 percent a year is losing ground even if its current margins look healthy.

The third is intangible collateral. Green technology firms often have few tangible assets to pledge. The recovery rate on default is therefore lower than for a conventional manufacturer, which raises the expected loss for a given PD. Adjust the LGD rather than the PD for collateral quality: reduce the recovery rate assumption for firms whose collateral is predominantly intangible, and reflect the reduction in pricing. The size of the adjustment should track the composition of the collateral register. A firm pledging equipment and land-use rights recovers far more than a firm pledging patents and receivables from a single customer.

These three adjustments are refinements of the environmental score, not additions to it. The policy-dependence and technology-risk indicators belong in the transition channel; the collateral adjustment belongs in the LGD calculation. Keeping them separate avoids double-counting.

3.3.8 Validation: What Would Falsify the Framework?

A framework that cannot be falsified is not a model. State the conditions under which the modified framework would be rejected.

If the environmental score does not improve the separation between defaulters and survivors out of sample, the framework adds noise. The test is the area under the receiver operating characteristic curve (AUROC) or the accuracy ratio (Gini coefficient) of the modified model versus the financial-only model. An improvement of less than 2 percentage points in the Gini is within the noise of most credit samples and should not justify the added complexity.

If the environmental score's contribution to the rating varies sharply across sectors, time periods, or subsamples, the framework is not capturing a stable relationship. The test is the stability of the notch adjustment across subperiods. A framework that works in one year and fails in the next is a coincidence, not a model.

If the framework rewards firms for not reporting environmental data, or for reporting favorable data that is not verified, it will degrade over time. The test is the correlation between data completeness and environmental score. A positive correlation between missing data and favorable scores is a red flag.

If firms can restructure to improve their environmental score without reducing their actual risk—by outsourcing the dirty part of the production process to an unrated supplier, for instance—the framework is measuring the boundary of the firm, not the firm's risk. The test is the stability of the score after major corporate restructuring. A score that jumps after an outsourcing transaction is measuring the wrong thing.

These conditions are demanding. Most green rating frameworks in operation today would fail at least one of them. The point is not to discourage adoption; it is to set the bar for what counts as success. A framework that is adopted, calibrated, and backtested against these conditions is a genuine improvement. A framework that is adopted and never validated is a compliance exercise.

3.3.9 Implementation: The Operational Questions

The framework's design is only half the problem. The other half is operational, and it is where most implementations stall. Four questions must be answered before the framework goes live.

Who computes the environmental score? The credit analyst, the environmental specialist, or a centralized model? The answer should be the centralized model for the quantitative score, with the credit analyst responsible for the qualitative overlay and the data quality flag. Decentralizing the quantitative score to relationship managers invites gaming; centralizing the qualitative overlay removes local knowledge. The split should follow the data: quantitative where the data are objective, qualitative where they are not.

How often is the score updated? The financial score is updated at the rating cycle, typically annually or semi-annually. The environmental score has faster-moving components—permit status, inspection findings, fines—that should be updated more frequently. The practical solution is a two-tier update: the full score at the rating cycle, and an event-driven update when a trigger occurs (a permit violation, a fine above a threshold, a change in the applicable standard). The event-driven update should be automated where the data source is a regulatory database.

How is the score explained to the borrower? A rating downgrade based on environmental indicators is a commercial conversation, not only a credit decision. The borrower will ask which indicators drove the downgrade and what it can do to improve them. The framework should produce an explanation that names the indicators and the remediation options. The explanation is also a risk-mitigation tool: a borrower that invests in abatement improves its score and reduces its default probability, which is the outcome the framework is designed to produce.

How is the score audited? The framework's inputs and weights should be documented, versioned, and reviewed by internal audit annually. The audit should check the data lineage, the weight calibration, and the backtest results. An unaudited framework is a black box, and a black box cannot be defended to a supervisor or a board.

These operational questions are unglamorous, and they are the difference between a framework that works and one that sits on a shelf. The institutions that have made green rating work—among them ING, which built environmental risk into its sector policies and client assessments from 2015, and the Industrial Bank of China, whose green finance division has operated a separate environmental assessment track since 2008—have invested as much in the operations as in the model.

3.3.10 The Boundary Between Rating and Policy

A final design question, and the most important one. The rating framework is a risk instrument, not a policy instrument. Its purpose is to order borrowers by default probability, not to direct credit to favored sectors. The two purposes can conflict, and when they do, the rating should win.

The conflict arises because green technology firms are often the firms that policy wants to support, and the rating framework may downgrade them for policy dependence or technology risk. A bank that uses the rating to allocate credit will lend less to the firms the policy favors. A bank that overrides the rating to support the policy will accumulate unrecognized risk. The resolution is to separate the two decisions: the rating determines the risk-adjusted price, and the policy determines the subsidy. A green technology firm with a high environmental score can borrow at a higher spread, with the spread subsidized by a government guarantee or an interest-rate subsidy. The rating stays honest; the policy does its work through the price, not through the rating.

This separation is the theme of the book in miniature. The rating identifies; the insurance buffers; the sci-tech finance catalyzes. Each instrument does its own job, and the system works because the jobs are separated. A rating framework that tries to do the insurance's job—by treating policy support as a risk mitigant—will fail at both.

The next chapter takes up the scoring models and data challenges in more detail, including the construction of the environmental score for the specific case of green technology firms whose financial statements are thin and whose environmental data are sparse. The framework in this chapter is the architecture; the next chapter is the engineering.

3.4 Conclusion: What the Rating Can and Cannot Do

The chapter has argued three things. Traditional credit ratings neglect environmental performance for reasons that are historical and institutional, not analytical, and the neglect is now a credit-risk problem rather than only an ethical one. The environmental indicators that belong in a rating model are a small, sector-specific set organized by four channels—liability, regulatory, transition, and physical—and they should be selected by materiality to default, measurability at the obligor level, timeliness, non-redundancy, and robustness to gaming. The indicators should be combined with the financial score through a rule that is transparent enough to explain to a borrower and calibrated against realized defaults, adjusting the rating at the margin and triggering a manual review at the extremes.

The framework's limits are as important as its capabilities. It can identify the environmental risks that drive default probability; it cannot eliminate them. It can price them; it cannot transfer them. The transfer is the job of the insurance instruments in Chapters 5 and 6, and the catalysis of green investment is the job of the sci-tech finance mechanisms in Chapters 7 and 8. A rating that identifies a risk the system cannot buffer is a diagnosis without a treatment.

The tea cooperative and the solar manufacturer illustrate the two directions of the framework's work. For the cooperative, the rating identifies physical and liability risks that can be mitigated by insurance and a green retrofit. For the manufacturer, the rating identifies a transition exposure that the financial statements conceal. In both cases, the rating's contribution is to make the risk visible and specific, so that the instruments that follow can address it.

The measurement agenda is clear. The framework needs obligor-level default data with environmental indicators, which means the institutions that adopt it must commit to collecting and sharing the data. Sharing is the hard part: default data are commercially sensitive, and the institutions that hold them have little incentive to pool. The policy response—supervisory data collections, credit registries, and public-private partnerships—is the subject of Chapter 12. Until the data exist, the framework will be calibrated on proxies, and its coefficients will be uncertain. The uncertainty is a reason to be conservative in the weights, not a reason to postpone adoption.

A final observation on the framework's political economy. The rating is the point at which the financial system decides who is creditworthy. Incorporating environmental performance into that decision changes the terms of access for a large class of firms. The firms that benefit are those with clean operations and transition-aligned assets; the firms that lose are those with hidden liabilities and stranded assets. The losers will resist, and their resistance will take the form of arguments about model risk, data quality, and competitiveness. Some of those arguments will be correct. The framework's design—transparent weights, published calibration, backtested results—is the best defense against the arguments that are not.

The chapter closes where it began, with the credit committee. The committee that approves the solar manufacturer at a standard premium is behaving as its model instructs: it ignores the subsidy dependence and the transition exposure because those variables are not in the score. Changing the model changes the decision. The next review of this file will ask three questions the current model cannot answer—what share of revenue survives a subsidy cut, what the tooling is worth if the module design is superseded, and how much of the receivables balance is owed by a single utility—and the committee will need a score that has already answered them.