Climate risk in banking is the possibility that physical climate hazards or the economic transition change a borrower’s cash flow, an asset’s value, an insurer’s obligations, or a bank’s ability to keep funding and payments working. Flood damage is one route. A factory’s rising energy costs, a change in customer demand, an uninsurable building, or a transition investment that runs out of cash before completion can be equally important. The practical task is to follow the financial consequences rather than attach a climate label and assume the analysis is finished.
Climate risk management and transition finance belong in the same conversation, but they answer different questions. Risk management asks what could impair repayment, liquidity, value and continuity. Transition finance asks how investment can support changes in production, infrastructure or business models. Financing a credible improvement can reduce future vulnerability; financing an implausible plan can add debt without creating repayment capacity. Physical risk, transition risk, climate scenario analysis, financed emissions and financial resilience become useful concepts only when their boundaries are clear.
The central proposition of this guide is simple: a climate-related financial loop is not closed when a risk score is produced or a green loan is signed. It closes when money changes something in the real economy, the resulting performance is measured, and that evidence changes the next financing decision. A flood barrier has to work. A new production line has to operate. An insurance recovery has to arrive. A transition milestone has to be verified. A lender must then distinguish genuine improvement from a better-looking report.
This is a mathematical learning guide from Bukit Timah Tutor, written for readers who want to connect banking, finance and real-world systems. It is not personalised financial advice or a compliance manual. Alicia, Tricia and Kai Kai are fictional learning characters. All company names, probabilities, financial amounts and scenarios in the worked cases are invented teaching assumptions unless a passage explicitly attributes a factual statement to an external source. Hypothetical carbon prices are not Singapore tax rates, and hypothetical lending conditions are not regulatory requirements.
Your 50-second route through the guide
Route one — understand the mechanism. Begin with the difference between a driver and a financial loss, then follow hazard, exposure, vulnerability and protection. These chapters explain why identical weather can produce different credit outcomes.
Route two — follow the money after a flood. Go to the four-month warehouse case. It separates a valuable insurance claim from cash available today, then compares an adaptation investment with emergency borrowing.
Route three — understand transition finance. Start with the manufacturing investment case. The calculations separate operating savings, debt service, commissioning delay and the difference between an attractive project and a borrower that can survive its installation.
Route four — challenge a climate model. Read scenario analysis, the NGFS model-version warning and how to avoid double counting. A model should make uncertainty visible, not hide it behind decimal places.
Route five — connect evidence to decisions. Use the monitoring design, the exercises and the questions readers ask. These routes turn the guide into a practical reasoning lesson rather than a list of definitions.
The wider map is Banking And Finance Closed Loop Systems: The Complete System. For the underlying mathematics, use the existing Finance and Banking Algorithms library. This article connects those mechanisms; it does not replace their specialist explanations.
Explore all chapters, worked examples and reference sections
- 1. Start with a financial question, not a colour
- 2. Draw the loop before building the spreadsheet
- 3. Separate hazard, exposure, vulnerability and protection
- 4. Acute, chronic and compound risk need different clocks
- 5. Transition risk is a change in business economics
- 6. Keep three scoreboards separate
- 7. Worked case: a warehouse survives on the right date, not on the right annual total
- 8. Worked case: adaptation can change the loss path, but its price matters
- 9. A flood probability is not a probability of loan default
- 10. Insurance transfers some loss; it does not move every payment date
- 11. Collateral is worth what can be recovered in the stressed state
- 12. Worked case: the same expected loss can hide very different portfolio danger
- 13. Worked case: transition investment must earn its return
- 14. Worked case: a positive NPV does not guarantee survival during commissioning
- 15. Green bonds, transition bonds and sustainability-linked terms ask different questions
- 16. Financed emissions are not a credit-loss measure
- 17. Good climate data begins with the question it must answer
- 18. A transition plan is a sequence of fundable commitments
- 19. Scenario analysis is a conditional question, not a prediction contest
- 20. The NGFS warning: model versions are part of the evidence
- 21. Avoid double counting before making the model more elaborate
- 22. Choose actions that remain useful when the forecast is wrong
- 23. Singapore: distinguish an issued guideline from an effective requirement
- 24. Disclosure, assurance and action are different layers
- 25. Monitoring should notice a broken promise before the payment fails
- 26. The borrower’s climate problem can become the bank’s liquidity problem
- 27. Worked case: different decision objectives can select different adaptations
- 28. Build a small model that can be explained from beginning to end
- 29. Exercises: test the mechanism, then read the worked answers
- 30. Questions readers ask about climate risk and transition finance
- Evidence laboratory: how a successful test should change confidence
- Working glossary
- Sources, dates and scope of authority
- The loop that matters
1. Start with a financial question, not a colour
Imagine a lender reviewing two warehouses. The first has a low-emissions electricity contract but stores expensive equipment at ground level beside a drainage bottleneck. The second has higher reported emissions but has elevated equipment, tested drainage, a diversified customer base and enough cash to withstand a temporary shutdown. Which is the safer loan?
There is no honest answer from the emissions figures alone. The first might have strong transition credentials and weak physical resilience. The second might withstand a flood but face future energy-cost or customer-demand pressure. A useful analysis needs the loan’s maturity, cash flows, collateral, operating dependencies and the borrower’s ability to adapt. “Green” and “safe” are not synonyms. Neither are “high emitting” and “certain to default.”
The Basel Committee’s report on climate-related risk drivers and transmission channels explains why these issues can enter familiar financial risk categories, including credit, market, liquidity and operational risk. That framing is helpful: a lender does not need to invent a separate kind of arithmetic for every environmental concern. It needs to understand how a driver changes the variables already used to assess a financial claim.
A borrower normally services debt with cash. A building normally protects a lender through recoverable value. A bank settles payments with usable liquidity. An insurer pays claims from resources available under its obligations. Those are the financial endpoints. Between a climate driver and those endpoints lies a chain that must be described rather than assumed.
A flood might damage stock, which reduces sales, which delays collections, which creates a cash shortfall, which causes a missed payment. But the chain can be interrupted. Stock may be stored elsewhere. Customers may accept delayed delivery. Insurance may cover part of the loss. Owners may inject capital. A bank may have already provided a usable committed facility. The result depends on the whole sequence.
Transition risk has the same need for a chain. A higher carbon cost might reduce margins, but only after considering the emissions covered, the company’s ability to pass costs to customers, the availability of substitutes, and the time required to change equipment. A business may respond by reducing output, improving efficiency, changing its product mix or investing in a new process. These responses alter the outcome the original scenario was supposed to estimate.
Alicia begins by asking, “What must still be paid?” Tricia asks, “Which asset or cash flow changes?” Kai Kai asks, “What can the borrower do before the deadline?” Their questions are more useful than asking which company has the more attractive sustainability presentation.
This approach also protects against the opposite mistake: treating uncertainty as a reason to do nothing. The exact loss may be unknown while the transmission channel is obvious. A business that relies on one exposed facility and has no backup needs closer examination even without a precise thirty-year loss estimate. Uncertainty should change the form of the decision, the size of the buffer and the intensity of monitoring. It should not be converted into either automatic rejection or automatic reassurance.
The educational discipline is to finish every broad statement with a financial consequence. “Heat exposure is high” becomes “the assumed cooling requirement increases operating costs and reduces available cash during peak demand.” “The transition plan is credible” becomes “the project has funding, an executable timetable and evidence that the new process can meet the customer specification.” The more precise sentence exposes what still needs to be checked.
2. Draw the loop before building the spreadsheet
A compact climate-finance loop has seven stages: a physical or transition driver; an exposed activity; an operational response; a change in financial performance; a lender or investor decision; a real-world adjustment; and new evidence. The stages are not a checklist to complete once. They are a cycle because financing decisions change the activity that will face the next shock.
Consider a cold-storage business. Higher temperatures increase cooling demand in the teaching scenario. Electricity costs rise. The business’s cash margin falls. A bank reviews the loan and offers financing for more efficient equipment, subject to ordinary credit assessment. The equipment is installed. Energy use per unit of stored product falls. Metered performance then helps determine whether the financing improved the borrower’s capacity to repay. The final observation is what closes the loop.
Now change one assumption. The equipment is purchased but the electrical connection is not ready. The borrower has new debt service without the expected savings. A project that appeared protective in the long term creates near-term liquidity pressure. The difference is not philosophical. It is a sequence of dated cash flows.
For teaching purposes, write the system as a state update: next state equals current state plus operating flows, financing flows, valuation changes and adjustments caused by the scenario. The state should include more than an environmental score. Useful components might be cash, debt, production capacity, energy intensity, insurance availability, collateral condition and the status of a transition project.
The driver also has a date. A storm is not the same as a gradual increase in heat-related operating cost. A policy announced today but effective later differs from a price change applied immediately. A ten-year technology transition can still affect a two-year loan if customers stop renewing contracts now. Time connects apparently distant risks to current financing.
A feedback arrow should name an actor. “The system adapts” is too vague. Does the borrower install new equipment? Does the bank extend maturity? Does an insurer alter coverage? Does a customer change procurement? Does a public authority improve drainage? Each actor has resources, incentives and delays. Without identifying the actor, an adaptation assumption can quietly become a promise that nobody has made.
The same action can improve one measure and worsen another. Selling a vulnerable factory may reduce the seller’s direct exposure but leave the factory operating under another owner. Raising lending rates may compensate a bank for higher expected risk while reducing the borrower’s cash available for repairs. Requiring insurance may strengthen recovery protection today but create a new problem if renewal becomes unaffordable. A loop diagram makes these side effects easier to see.
Students can test their diagram by removing one arrow. Suppose the bank’s annual review does not receive the equipment’s actual performance data. The loan continues, but the learning loop is broken. Suppose the data arrives but nobody can change terms, limits or monitoring. Measurement exists without an effective response. Suppose the bank changes its score but the borrower’s operating process remains unchanged. The financial record moved; the physical risk did not necessarily move with it.
A good spreadsheet follows a good diagram. It should not replace one. The diagram says which variables influence which other variables, and the spreadsheet tests the consequences under explicit assumptions. When the numbers surprise the reader, the diagram helps locate the cause. When the diagram has an unsupported arrow, no amount of numerical precision will repair it.
3. Separate hazard, exposure, vulnerability and protection
A hazard is a potentially damaging event or condition. Exposure describes what is in its path. Vulnerability describes how susceptible the exposed activity is to harm. Protection describes measures that can reduce damage or support recovery. These concepts are related, but combining them into one score too early destroys useful information.
In a fictional flood example, the hazard might be a specified water level at a site. Exposure might include machinery, stock, access roads and electrical systems. Vulnerability depends on where those items are located, how water enters, how long it remains and whether the business can operate elsewhere. Protection might include drainage, raised equipment, barriers, emergency procedures and insurance. Some protection reduces physical damage; some only helps pay for it afterward.
Two buildings on the same map can therefore have different outcomes. One houses a ground-floor server room; another uses the same floor as an easily cleaned loading area. One relies on a single access road; another has alternative routes. One can pause for a week without losing customers; another supplies a production line that immediately switches to a competitor. The geographic hazard is only the beginning of the financial story.
It is equally important not to overstate protection. A barrier designed for one event may fail under another. A backup generator needs fuel, maintenance and an operating connection. A second supplier may share the same port or electricity network as the first. A protection measure should be represented with a capacity, a reliability assumption and a failure condition, not as a permanent switch from risky to safe.
For a teaching model, one can write physical loss as exposed value multiplied by a damage fraction conditional on the event. That is not a complete engineering model. It is a transparent way to show the inputs. If exposed equipment is worth two million dollars and the assumed damage fraction is twenty per cent, direct equipment damage is four hundred thousand dollars. Neither the equipment value nor the damage fraction tells us how much business income is lost.
Business interruption requires another calculation. A firm could lose little equipment but suffer a long shutdown because a specialised component takes months to replace. Conversely, it might lose stock yet recover quickly using other facilities. Direct damage and interruption should therefore be estimated separately before they are connected to cash.
The distinction also changes data collection. A headquarters address may be irrelevant when the production facility is elsewhere. A property valuation may not identify the location of critical machinery. An insurance schedule may describe covered assets but omit uninsured revenue dependence. Asking for more data is not enough; the requested data has to correspond to an actual transmission channel.
The IPCC’s AR6 Synthesis Report headline statements emphasise compound and cascading risks. For financial teaching, the implication is to avoid assuming that every disturbance arrives alone. A heat event can coincide with high electricity demand. Flooded roads can prevent repair teams reaching an undamaged facility. A business may have strong on-site protection but weak access to the surrounding services it needs.
Alicia therefore marks the physical assets. Tricia marks the revenue dependent on them. Kai Kai marks the shared infrastructure outside the company boundary. Their combined map is more informative than colouring the company’s head office on a national risk map and treating the result as a loan decision.
4. Acute, chronic and compound risk need different clocks
An acute event produces a concentrated disruption: a flood, storm or other shock in a stated scenario. A chronic change alters the background in which a business operates. For a teaching example, repeated heat-related productivity losses or a persistent increase in cooling costs can gradually change margins. Compound risk arises when events or vulnerabilities interact rather than simply add up.
A one-month shock is often best understood through cash timing. Which receipts stop? Which bills continue? What must be repaired immediately? When can production restart? When does insurance pay? An annual average loss does not answer these questions. A business can survive an annual cost of one hundred thousand dollars but fail when that cost arrives in one week without access to liquidity.
A gradual cost change requires another model. Suppose operating cash before debt service is one million dollars a year and maintenance plus cooling expenses rise by fifty thousand dollars each year in an invented scenario. The first year may be manageable. By the fourth year, the cumulative effect can change the economics of the business, especially if revenue cannot rise. The important date may be the next equipment replacement or loan refinancing rather than the next storm.
Compound risk should not be modelled by mechanically adding independent losses. Suppose a site loses electricity and road access simultaneously. Electricity loss stops production; road loss prevents repair. The combined shutdown may be longer than the sum of two short independent incidents because one problem delays recovery from the other. The financial effect comes from interaction.
Another interaction appears in insurance. A property might remain physically repairable after an event, yet renewal terms become less favourable in the hypothetical next year. The borrower then faces repair costs today and higher recurring costs later. A model that ends when the insurance cheque arrives misses the changed operating state.
Loan tenor adds a further complication. A lender with a short contractual maturity might assume long-term climate risk is irrelevant. That conclusion can fail where repayment depends on refinancing, resale or an asset whose market value already reflects later conditions. A two-year loan against a thirty-year asset can carry long-horizon risk through its terminal value.
The opposite error is to treat a distant scenario outcome as a loss occurring immediately. A projected cost in year twenty cannot simply be charged against next year’s cash without an explanation of how expectations, repricing or contractual obligations bring the effect forward. A model must show the bridge between horizons.
For students, the useful exercise is to draw three timelines: physical change, contractual cash flow and decision opportunities. Mark when the hazard or cost changes, when debts and insurance renew, and when the borrower can invest, relocate or alter production. Vulnerability is often concentrated where an obligation arrives before an adjustment becomes possible.
This time-based reading also encourages proportionate decisions. A borrower does not need every possible adaptation installed tomorrow. It needs a credible sequence that matches material risks and financing deadlines. Some actions are urgent because they protect this year’s cash. Others can be incorporated into ordinary replacement cycles. The model should reveal that sequence rather than demand an expensive catalogue of improvements without considering timing or feasibility.
5. Transition risk is a change in business economics
Transition risk can arise when policy, technology, customer preferences, market pricing or legal expectations change the value of an activity. The simplest teaching example is an input cost that rises while selling prices do not. But the full question is broader: can the company remain competitive, finance the adjustment and produce something its customers will still buy?
Consider two invented manufacturers with the same emissions. The first sells a specialised component under contracts that allow some cost pass-through. The second competes on price in a market with abundant substitutes. A carbon-related cost increase of the same amount will not produce the same margin loss. Contract structure and demand elasticity matter.
Technology creates a similar distinction. A new production method may be technically available but unsuitable for a particular product specification. It may require additional electricity, water, skilled labour or maintenance. A financing plan that pays for equipment without funding those dependencies can leave the borrower with an asset that cannot operate as intended.
Transition should therefore be represented as a project with both investment and execution risk. There is an upfront cash requirement, a construction or installation period, a learning period, expected operating benefits and a range of failure outcomes. The financial model should not count the benefits from day one while postponing all costs to a footnote.
Customer behaviour also matters. A company can invest in a lower-emissions product but fail if buyers will not accept its price or performance. Alternatively, maintaining the old process can become riskier if major customers change procurement requirements. The credit question is not simply whether the company is changing. It is whether its chosen change makes commercial and financial sense under plausible conditions.
A bank faces two tempting shortcuts. One is to avoid all difficult sectors and declare the remaining portfolio improved. The other is to finance any transition-labelled project and assume improvement will follow. Both skip the central task: distinguishing a credible, financeable adjustment from a plan that cannot survive its own cash-flow demands.
The distinction between transition risk and physical risk is also porous. Electrifying a process can reduce one exposure while increasing dependence on a particular electricity network. Relocating a facility can improve hazard exposure but create new supplier or labour constraints. A useful analysis asks what new risk the proposed solution introduces.
There can be upside as well as downside. A company that improves efficiency may reduce costs, win customers or become less exposed to volatile inputs. The guide’s manufacturing case tests those possibilities explicitly. It does not assume that every transition cost is a loss or that every new technology produces a gain.
The essential teaching question is counterfactual: compared with what? The project should be assessed against maintaining the current process, investing in an alternative, scaling down, outsourcing or closing the activity. “Do nothing” is itself a scenario with costs and risks. So is “invest immediately.” Sound reasoning compares alternatives on the same time horizon with the same definitions of cash, output and risk.
6. Keep three scoreboards separate
The first scoreboard is financial resilience. Can the borrower service debt, maintain liquidity and preserve recoverable value? The second is the activity’s environmental performance. Are its physical emissions or other relevant impacts changing? The third is a financial institution’s attributed or portfolio-level environmental measure. These scoreboards can move together, but they can also diverge.
Suppose a lender sells a loan to another investor. Its exposure and attributed emissions may fall. The borrower’s factory may continue operating exactly as before. The lender has changed its portfolio, not necessarily the physical activity. That can still be a legitimate financial decision, but the environmental claim should match what happened.
Now suppose the bank finances equipment that cuts energy use and emissions but adds debt service during installation. Physical performance may improve after commissioning while near-term financial resilience initially weakens. Calling the loan successful requires more than pointing to the future emissions reduction. The borrower must reach that future state without failing along the way.
A third possibility is measurement change. A company improves the quality of its emissions data and reports a larger footprint than its previous estimate. The reported metric worsens even though nothing physical changed. Treating that increase as proof of deteriorating performance would discourage better measurement. The correct comparison distinguishes changes in the activity from changes in the method.
These examples suggest a simple reporting discipline: state the object being measured, the boundary, the period, the calculation and the reason for movement. Financial risk measures need loan and borrower definitions. Physical emissions measures need organisational and operational boundaries. Portfolio attribution measures need financing and denominator conventions. A number without those definitions invites the reader to tell the wrong story.
PCAF’s Global GHG Accounting and Reporting Standard for the Financial Industry separates financed, facilitated and insurance-associated emissions. Its standard page also treats avoided-emissions and forward-looking disclosures separately from the core financed-emissions inventory. Those distinctions are useful guardrails: one should not subtract a speculative future benefit from a current inventory and present the result as though the two measurements were interchangeable.
For credit assessment, emissions can be an input, not a complete probability of default. A high-emitting company can have strong cash flow and a credible transition investment. A low-emitting company can be highly leveraged, uninsured or exposed to a severe physical disruption. Financial strength does not excuse misleading environmental claims, and environmental merit does not automatically create repayment capacity.
Alicia keeps the operating scoreboard: what actually happened at the site? Tricia keeps the financing scoreboard: what happened to cash and debt? Kai Kai keeps the measurement scoreboard: did the metric change because of real performance, portfolio transactions or methodology? Their three explanations should be reconcilable without being forced to be identical.
The rest of this guide uses those separate scoreboards. The flood case focuses on cash timing and adaptation. The manufacturing case focuses on investment economics and financing. The portfolio case focuses on dependence. The reporting chapters focus on claims that can be supported by evidence. Together they show why a complete climate-finance system needs several measurements and a disciplined way to connect them.
7. Worked case: a warehouse survives on the right date, not on the right annual total
Harbour Store is a fictional distribution business. Its name does not identify a real company or a particular Singapore property. All amounts in this case are millions of Singapore dollars. The business begins the scenario with S$0.60 million of unrestricted cash. A flood interrupts deliveries and damages stock and equipment. Customers eventually return, and an assumed insurance recovery arrives in month four. The question is whether the company can reach month four.
We use a deliberately short cash model. Customer receipts are the cash actually collected, not invoiced sales. Operating outflows include the cash expenses shown in the assumptions, and debt service includes scheduled principal and interest. Repair payments are listed separately so they are not hidden inside an operating margin. The illustration omits new taxes, financing fees and incremental credit-line interest; a real implementation would need to include them.
In month one, Harbour Store collects S$0.20 million. It pays S$0.55 million of operating costs, S$0.35 million for repairs and S$0.10 million of scheduled debt service. The net outflow is S$0.80 million. Opening cash of S$0.60 million therefore leaves a S$0.20 million shortfall. A forecast ending the month at negative cash does not describe an actual cash account; it identifies financing that must be found or payments that cannot all be made as scheduled.
In month two, receipts recover to S$0.45 million, but operating costs are S$0.50 million, further repairs cost S$0.15 million and debt service remains S$0.10 million. The month consumes another S$0.30 million. The cumulative funding shortfall reaches S$0.50 million. Month three is cash-neutral: S$0.70 million of receipts meet S$0.60 million of operating costs and S$0.10 million of debt service.
Month four finally looks comfortable. Receipts are S$0.90 million, operating costs are S$0.65 million, debt service is S$0.10 million and the assumed insurance recovery is S$0.50 million. The month adds S$0.65 million. Across the four months, the model ends with S$0.15 million after using the original cash. Yet without additional liquidity, the company could not have paid everything in months one and two.
| Cash item, S$ million | Month 1 | Month 2 | Month 3 | Month 4 |
|---|---|---|---|---|
| Customer receipts | 0.20 | 0.45 | 0.70 | 0.90 |
| Operating outflows | 0.55 | 0.50 | 0.60 | 0.65 |
| Repair payments | 0.35 | 0.15 | 0.00 | 0.00 |
| Scheduled debt service | 0.10 | 0.10 | 0.10 | 0.10 |
| Insurance cash received | 0.00 | 0.00 | 0.00 | 0.50 |
| Net monthly flow | −0.80 | −0.30 | 0.00 | 0.65 |
| Cumulative cash before new financing | −0.20 | −0.50 | −0.50 | 0.15 |
This is the first important result: a positive end-of-scenario balance does not prove survival along the path. The minimum balance matters. Harbour Store needs at least S$0.50 million of additional funding to avoid a negative balance under the stated monthly assumptions. Exactly meeting the minimum leaves no room for timing errors, fees or another setback.
Suppose the business has a facility of S$0.60 million that is genuinely drawable at the start of month one. Drawing it produces closing cash balances of S$0.40 million, S$0.10 million, S$0.10 million and S$0.75 million before repayment. Repaying the S$0.60 million principal after the insurance recovery leaves S$0.15 million, before the incremental financing costs omitted from this illustration. The facility bridges a timing problem; it does not create the insurance recovery or remove the underlying damage.
A facility of S$0.40 million is not enough. It would leave a S$0.10 million cumulative deficit in months two and three. A facility approved in principle but unavailable until month three is also not enough. Amount, availability and timing are separate conditions. A student should never replace “committed and drawable under the scenario” with “the company probably has a good relationship with its bank.”
Now stress the recovery date. If the insurance payment arrives in month six instead, the four-month operating path adds only S$0.15 million in month four. The cumulative position before new financing remains negative S$0.35 million. The company may still have a valid claim, but the bridge must last longer. The model needs months five and six, the line’s maturity and the cost of carrying the extra debt.
Alicia marks the first missed-payment date. Tricia marks the maximum funding gap. Kai Kai marks the conditions under which the line remains drawable and the insurance recovery can actually be received. Their conclusion is more precise than “the borrower is insured” or “the business is profitable over the year.” It is: under these assumptions, a short-term facility of the right size and duration can bridge the disruption, but survival remains sensitive to recovery timing and access to funds.
This cash-path reasoning is developed more generally in Treasury, Collateral, Repo, Margin and Intraday Liquidity. The climate-specific task is to connect the physical interruption to that established timing model.
8. Worked case: adaptation can change the loss path, but its price matters
Return to Harbour Store before the hypothetical flood. Management can spend S$0.30 million on a package of raised equipment, water-resistant storage changes and tested recovery arrangements. We are not asserting that these measures would produce a particular engineering result at any real site. For teaching, assume the package reduces damage and shortens disruption as shown below. Its cost is paid from the original S$0.60 million cash balance, leaving S$0.30 million before the event.
With the adaptation package in place, receipts are assumed to be S$0.50 million in month one, S$0.75 million in month two and S$0.90 million in each of months three and four. Operating outflows are S$0.55 million, S$0.60 million, S$0.65 million and S$0.65 million. Repairs cost S$0.10 million in month one and S$0.05 million in month two. Scheduled debt service remains S$0.10 million each month. A smaller insurance recovery of S$0.20 million arrives in month four.
The monthly cash movements are negative S$0.25 million, zero, positive S$0.15 million and positive S$0.35 million. Starting from the post-investment cash balance of S$0.30 million, closing balances are S$0.05 million, S$0.05 million, S$0.20 million and S$0.55 million. Under these assumptions, the company does not require an emergency draw to avoid a negative monthly balance.
| Illustrative comparison | Without adaptation | With adaptation |
|---|---|---|
| Cash before the event | S$0.60m | S$0.30m after investment |
| Repair payments over four months | S$0.50m | S$0.15m |
| Insurance received in month four | S$0.50m | S$0.20m |
| Minimum cumulative cash before emergency financing | −S$0.50m | S$0.05m |
| Month-four cash before incremental financing costs | S$0.15m | S$0.55m |
The smaller insurance payment is not evidence that the adapted business is worse off. It has less damage to claim and more revenue preserved. Looking only at recoveries would reverse the economic story. The relevant outcome is the combined path of operating cash, repair cost, insurance and debt service after paying for the adaptation.
However, one favourable event comparison cannot establish that the investment is financially attractive. The package costs money whether or not a flood occurs. A complete investment test needs a horizon, a probability model, maintenance, replacement and the consequences of failure. It also needs the expected benefit in less severe and more severe events, not just the one event selected for illustration.
Consider a separate, simplified appraisal. Assume an annual event probability of ten per cent. Conditional economic loss is S$1.20 million without the package and S$0.40 million with it. Expected annual avoided loss is therefore 0.10 multiplied by S$0.80 million, or S$0.08 million. After assumed annual maintenance of S$0.015 million, the net expected benefit is S$0.065 million. These loss assumptions are a separate annual economic model, not a redefinition of the four-month cash table.
Over eight years at a seven per cent discount rate, the present value of those level annual net benefits is approximately S$0.388 million. Subtracting the S$0.30 million investment gives an illustrative net present value of about S$0.088 million. The calculation is positive under those assumptions. It is not a forecast, a site-specific engineering assessment or proof that the project should be funded at any price.
The result is sensitive to the assumptions. A lower event probability, smaller avoided loss, higher maintenance cost or shorter useful life reduces value. Better service continuity or lower interruption costs could increase it. A protection measure that fails in the most damaging events needs a more detailed loss distribution than the single conditional average used here.
The financing question can also differ from the investment question. Paying cash preserves debt capacity but reduces immediate liquidity. Borrowing preserves cash but adds fixed obligations. Installing the package in stages may reduce initial pressure but leave partial protection for a period. There is no universal answer without the borrower’s wider financial state.
The strongest conclusion is conditional and operational: adaptation is financially meaningful when its performance assumptions are credible, its cost is fundable and it changes the loss path enough to justify the resources consumed. The lender should be able to identify the physical evidence that would confirm the package works and the financial evidence that would show the borrower can maintain it.
9. A flood probability is not a probability of loan default
One of the easiest modelling errors is to take a hazard probability and treat it as the borrower’s probability of default. The events are different. A flood can occur without default. A borrower can default without a flood. Credit modelling has to connect the event to income, liquidity, leverage, protection and access to support.
Use a two-state teaching model. Suppose the probability of a flood during the year is ten per cent. Without a flood, the borrower’s assumed default probability is two per cent. Conditional on a flood, it is twenty per cent. The overall probability of default is the weighted sum: ninety per cent multiplied by two per cent, plus ten per cent multiplied by twenty per cent. The result is 3.8 per cent, not ten per cent and not twenty per cent.
This calculation is an application of the law of total probability. Its quality depends entirely on the conditional assumptions. The twenty per cent conditional default probability is not supplied by the hazard map. It has to come from an explicit credit model, evidence or an acknowledged teaching assumption. The map says something about the event; the credit model says something about the borrower’s response.
Loss severity may change with the same event. Let exposure at default be S$1 million. Assume loss given default is thirty per cent in the no-flood state and sixty per cent in the flood state. Expected loss is then the sum of the state-specific products: 0.90 × 0.02 × 0.30 × S$1 million, plus 0.10 × 0.20 × 0.60 × S$1 million. The result is S$17,400.
A tempting shortcut is to calculate the overall 3.8 per cent default probability and multiply it by an unconditional average severity of thirty-three per cent. That produces S$12,540, materially below the correctly conditioned result. The shortcut understates loss because the high-severity state is also the state in which default is more likely.
The familiar formula of probability of default multiplied by loss given default multiplied by exposure is not the problem. The problem is feeding it an unconditional severity average when it requires severity conditional on default, with dependencies handled consistently. In this toy model, the default-weighted severity is approximately 45.79 per cent. Multiplying 3.8 per cent by 45.79 per cent by S$1 million reproduces S$17,400.
Exposure can also change. A borrower might draw an unused line to pay for repairs just before default. If the flood-state exposure rises, expected loss rises again. The analyst must model the timing of the draw and whether the line is actually available. Assuming a larger exposure for loss calculation while assuming no draw for the bank’s liquidity test would make the two models inconsistent.
This example also explains why a single climate adjustment pasted onto every borrower’s probability of default is weak. Borrowers differ in cash buffers, insurance, operational flexibility and access to finance. Two companies exposed to the same hazard can have different conditional default probabilities. A meaningful model should preserve those differences rather than punish or reward an entire postcode mechanically.
The general credit concepts are explained in Credit Risk, Underwriting, Provisioning, Default, Recovery and the Learning Loop. Climate analysis contributes the state-dependent operational and financial inputs. It does not abolish the need for calibration, validation and careful definitions.
For readers checking an actual report, the useful questions are modest but powerful. Is the probability annual or cumulative? Is it the probability of a hazard, of damage, of interruption or of default? Is loss severity conditional on the same scenario? Are insurance and adaptation already included? Has the same loss been counted again elsewhere? These questions often reveal more than an impressive-looking final risk score.
10. Insurance transfers some loss; it does not move every payment date
In Harbour Store’s case, the insurance recovery is valuable because it replenishes cash after the event. But it arrives later than the repairs and debt service. That distinction is enough to change the financing result. A promise of future payment cannot settle a bill today unless the business can finance the interval.
For a simplified insurance analysis, separate five quantities: the gross loss, the covered portion, the deductible or retained amount, the eventual cash recovery and the recovery date. These are not necessarily equal. A model that writes “insured: yes” discards all five. It also obscures whether the policy covers physical assets, interruption, additional operating costs or only selected risks.
The contractual details of real insurance vary. This guide does not infer coverage from a product label. Instead, it asks readers to understand the model implications of whatever coverage is assumed. If a claim is excluded, the economic loss remains with the borrower. If it is covered but delayed, the borrower may need liquidity. If the insurer pays only after a specified process, that process belongs in the timeline.
Insurance also changes the allocation of loss, not necessarily the amount of physical harm. When an insurer reimburses repairs, someone still paid for the damaged equipment and interrupted activity. At the system level, the insurer’s asset portfolio, capital and liquidity support the recovery. The loss did not disappear; it moved to a different balance sheet.
A bank therefore needs to avoid inconsistent boundaries. At the borrower level, insurance can lower net loss. At the combined borrower-and-insurer level, subtracting the payout from the borrower without recognising the insurer’s obligation would understate the total resource cost. Which boundary is relevant depends on the question being asked.
Renewal introduces a second timeline. A policy covering this year’s event does not establish that identical terms will remain available throughout a long loan. In a teaching stress, the next premium may rise, the deductible may increase or some coverage may cease. These changes affect the borrower’s recurring cash costs and future recovery assumptions.
Consider an invented borrower with annual cash available before debt service of S$1 million and scheduled debt service of S$0.70 million. A premium increase of S$0.10 million reduces the remaining cash cushion from S$0.30 million to S$0.20 million, holding everything else constant. The borrower has not suffered a new physical event, but its financial resilience has changed. If the premium rise coincides with lower revenue, the effects need to be tested together.
There is also a danger in assuming perfect insurance recovery under a system-wide event. A real assessment would need evidence about the insurer, contractual protections and claims operations. For teaching, one can test a delay, a partial recovery and a full recovery as separate branches without pretending to know which will occur. The point is to see which branch changes survival.
Alicia follows the loss. Tricia follows the policy assumption. Kai Kai follows the cash receipt. The three are different objects. An insurance certificate can support a financing discussion, but it cannot replace a cash-flow model, a coverage assessment or a view of what happens at renewal. The fuller risk-pooling mechanism is explained in Insurance, Reinsurance, Asset-Liability Matching and Risk Transfer.
11. Collateral is worth what can be recovered in the stressed state
Collateral provides a lender with a potential recovery source. Its usefulness depends on legal rights, condition, market value, sale costs and the time required to realise cash. An appraisal before a disruption is not a guaranteed recovery after it. Climate-related stress makes this distinction especially important when the same event damages the borrower’s income and the collateral.
Suppose a loan of S$8 million is secured by a property valued at S$10 million before the event. The initial loan-to-value ratio is eighty per cent. In a teaching stress, physical repairs, higher insurance costs and weaker demand reduce market value to S$7 million. If realisation costs are S$0.50 million, the net collateral proceeds are S$6.50 million before considering claim priority or other recoveries. The shortfall against the loan is S$1.50 million.
It would be wrong to count the original S$2 million equity cushion as unchanged. The cushion is a residual. When the asset value falls, it is consumed before the lender’s recovery shortfall appears. Nor should a model automatically assume the collateral can be sold at its appraised value on the day default occurs.
Marketability can deteriorate even where the building remains usable. Buyers may need financing, insurance or planning approvals. If several owners sell simultaneously, market depth can weaken. These are assumptions to test, not claims that every exposed property will become unsaleable. The analysis should identify what would make the sale difficult and over what horizon.
Location matters at a finer level than the company’s registration address. Floor height, access, drainage, construction, maintenance and surrounding infrastructure can alter the damage path. Two properties on the same street may still have different recovery characteristics. A coarse map is useful for screening; it should not be presented as a complete property valuation.
Another complication is collateral concentration. A lender may have many small loans secured by properties that depend on the same protection system. The loans look diversified by borrower name but not by physical dependency. If that system fails, several recoveries weaken at once. The portfolio chapter next makes this statistical distinction explicit.
Adaptation can change recoverable value, but it must be assessed carefully. Spending S$1 million on protection does not automatically raise collateral value by S$1 million. A buyer may value the improvement differently, and the measure may require continuing maintenance. The credit model should not count the same investment once as a cash benefit, again as avoided damage and a third time as an equal increase in sale value without checking the relationships.
A lender can also reduce risk through structure, such as lower leverage, reserves or staged funding. These are not substitutes for sound physical assumptions. They alter who has capacity to absorb loss. A lower loan-to-value ratio can protect the lender while the borrower still suffers a large economic loss. Financial resilience and societal resilience remain different scoreboards.
The reader’s final question should be, “Recoverable by whom, when, and after which costs?” That sentence prevents collateral from becoming a comforting but undefined number. It connects property analysis to the established Real Estate Finance, Cap Rates and Refinancing framework without treating climate exposure as the only determinant of value.
12. Worked case: the same expected loss can hide very different portfolio danger
Consider two invented portfolios. Each contains one hundred loans of S$1 million. Every loan has a five per cent probability of default over the teaching horizon, and every default produces a fifty per cent loss of exposure. The expected loss on each loan is S$25,000, so each portfolio has expected loss of S$2.5 million. If the analysis stops there, the portfolios appear identical.
Portfolio A assumes independent defaults. One borrower’s default does not change the probability of another’s. This is a simplifying mathematical assumption, not a description of real banking. The number of defaults follows a binomial distribution with one hundred trials and default probability 0.05. Each default costs S$0.50 million.
Portfolio B uses an intentionally extreme common-shock assumption. With probability five per cent, one shared event causes all one hundred loans to default. Otherwise none defaults. Each loan still has the same five per cent marginal default probability, and the expected portfolio loss is still S$2.5 million. But the distribution is radically different: either zero loss or S$50 million.
Under the stated assumptions, Portfolio A’s loss standard deviation is approximately S$1.09 million. Portfolio B’s is approximately S$10.90 million, ten times larger. The ninety-ninth percentile loss is S$5.5 million for the independent portfolio and S$50 million for the common-shock portfolio. These are toy distribution results, not regulatory capital figures or calibrated predictions for a real bank.
| Teaching measure | Independent defaults | Single common shock |
|---|---|---|
| Number of S$1m loans | 100 | 100 |
| Marginal default probability | 5% | 5% |
| Loss per default | S$0.50m | S$0.50m |
| Expected total loss | S$2.50m | S$2.50m |
| Loss standard deviation | About S$1.09m | About S$10.90m |
| 99th-percentile loss | S$5.50m | S$50.00m |
The lesson is not that real portfolios have perfect common default. They usually sit between these extremes, with different dependencies across sectors, sites, suppliers and financing arrangements. The lesson is that the average does not describe the tail. Dependence can change the capacity required to survive even when every loan’s individual expected loss is unchanged.
Climate-related dependencies can appear through a shared catchment, transport route, grid, water supply, insurer, customer or technology. A bank might lend to a hotel, a food distributor and a manufacturer and call the sector mix diversified. If all three rely on the same vulnerable access road, one operational failure can connect their cash flows. Sector diversification is not the same as infrastructure diversification.
Transition risk can also be common. Several borrowers may depend on the same future technology becoming commercially available, the same supplier delivering equipment, or the same customer maintaining a purchase commitment. If the shared assumption fails, their transition plans can underperform together. A lender should not count three plans as three independent protections when they all depend on one untested component.
Portfolio mapping should therefore include dependency identifiers alongside borrower names. It need not begin with a perfect global network. Even a modest map of major sites, critical suppliers, common insurers and significant refinancing dates can reveal concentrations that an ordinary industry table misses. The map should record uncertainty where the dependency is not known.
A useful stress test then asks which dependencies fail together, how much exposure is affected and whether the bank’s own response adds another correlation. If the bank withdraws credit from all affected borrowers at once, some may lose the liquidity needed to repair or adapt. If every lender sells similar collateral, recovery prices may weaken. The portfolio is not merely receiving a common shock; its participants can amplify it.
This is where the climate guide joins the broader study of liquidity spirals and financial contagion. The specific driver changes, but the mathematical challenge remains: identify shared exposures, recognise nonlinear responses and test whether individually sensible actions remain feasible when many institutions take them simultaneously.
13. Worked case: transition investment must earn its return
Meridian Components is a fictional manufacturer. It produces industrial parts using a process that emits 100,000 tonnes of carbon dioxide equivalent a year under our teaching boundary. It is considering a new process costing S$12 million that would reduce the assumed annual emissions to 40,000 tonnes at the same output. The equipment would also save S$1.20 million in annual energy expenditure but add S$0.60 million in maintenance and other operating costs.
We first analyse the project without financing. This is important because borrowing does not make an uneconomic machine productive. The project’s operating benefit equals avoided carbon-related cost plus energy savings minus additional operating cost. Carbon prices in this example are invented scenario inputs applied to the stated emissions. They are not claims about any jurisdiction’s actual tax, coverage, allowances or exemptions.
At S$80 per tonne, the assumed emissions reduction of 60,000 tonnes avoids S$4.80 million of annual cost. Adding S$1.20 million of energy savings and subtracting S$0.60 million of additional operating cost gives S$5.40 million of annual net operating benefit. At S$20 per tonne, avoided carbon-related cost is only S$1.20 million, so the annual net benefit is S$1.80 million. The physical emissions reduction is unchanged; the financial value of that reduction differs by scenario.
Assume six years of level annual benefit, no terminal value and a ten per cent discount rate. In the higher-price case, present value of benefits is approximately S$23.52 million, giving net present value of S$11.52 million after the S$12 million investment. In the lower-price case, present value of benefits is approximately S$7.84 million, giving negative net present value of S$4.16 million. The same machine can therefore be attractive under one set of commercial assumptions and unattractive under another.
| Project-only teaching calculation | S$20 per tonne | S$80 per tonne |
|---|---|---|
| Annual avoided emissions | 60,000 tonnes | 60,000 tonnes |
| Avoided carbon-related cost | S$1.20m | S$4.80m |
| Energy savings less extra operating cost | S$0.60m | S$0.60m |
| Total annual net operating benefit | S$1.80m | S$5.40m |
| Six-year NPV at 10%, after S$12m investment | About −S$4.16m | About S$11.52m |
This is not an argument for choosing whichever scenario makes the project look good. It is an argument for identifying the variables on which the project depends. If a lender can only justify repayment under the highest assumed carbon cost, it should understand that dependence. If the project produces strong energy savings even at a low carbon cost, its financial case is less dependent on policy. The purpose of scenarios is to reveal those differences.
We have also assumed constant output and unchanged product quality. Those assumptions need attention. A process that emits less per unit but reduces output or requires a price premium may have a different commercial result. A process that improves quality and wins new contracts might generate additional revenue, but that revenue should not be included without an explicit demand assumption. Technical capability and commercial acceptance are distinct milestones.
Installation cost is another source of uncertainty. If the project costs S$15 million instead of S$12 million, net present value falls by S$3 million, holding the benefit path unchanged. If benefits start a year late, their present value falls and the borrower must fund the delay. If maintenance costs rise, the annual benefit changes. These are ordinary project-finance questions applied to a transition investment.
The comparison with doing nothing also needs discipline. Under the no-investment case, Meridian keeps the old process and remains exposed to the assumed carbon-related cost. But it might have other options: improve efficiency without replacing the entire line, change suppliers, modify the product, pass some costs to customers or reduce production. Those alternatives should be compared on consistent output and risk assumptions. Otherwise a model can make one preferred option appear superior simply by giving it an unrealistically weak comparator.
Alicia checks the operating quantities: tonnes, energy use and output. Tricia checks the cash conversion: which physical savings become actual money saved? Kai Kai checks the counterfactual: what would happen without this exact investment? Their approach prevents a projected emissions reduction from being treated as a financial return without the intermediate steps.
The core valuation method belongs to Corporate Finance, Cash Flow, Capital Structure and Investment. Transition finance adds the operational, technological and policy assumptions that make the cash-flow path different from an ordinary replacement project.
14. Worked case: a positive NPV does not guarantee survival during commissioning
Now finance Meridian’s S$12 million project with S$4 million of equity and S$8 million of new debt. Assume the debt is repaid through five equal annual payments at eight per cent interest. The standard annuity formula gives annual debt service of approximately S$2.004 million. Meridian already has S$6 million of annual debt service on its existing obligations.
To examine the borrower rather than the project alone, assume the old business generates S$12 million of annual cash available before the scenario’s carbon-related cost and before debt service. This is a simplified cash measure, not EBITDA and not a regulatory definition. At the higher assumed carbon price of S$80 per tonne, the old process incurs S$8 million of carbon-related cost, leaving S$4 million available for S$6 million of existing debt service. Without adjustment, the coverage ratio is about 0.67 times.
After the new process operates fully, carbon-related cost falls to S$3.20 million. Net energy and extra operating savings add S$0.60 million. Cash available for total debt service becomes S$9.40 million. Total debt service is S$6 million plus approximately S$2.004 million, or S$8.004 million. Full-operation coverage is therefore about 1.17 times. Under these assumptions, the investment improves the borrower’s high-carbon-cost state even after including its new debt.
But commissioning is not full operation. Suppose the first operating year produces only S$2 million of net incremental benefit after delays, learning costs and lost production. Starting from the S$4 million no-investment cash base in the higher-price scenario, Meridian has S$6 million available for approximately S$8.004 million of debt service. Coverage is only about 0.75 times. The project can have a positive six-year NPV and still create a first-year funding gap of roughly S$2 million.
| Higher-price teaching scenario | Cash available | Debt service | Coverage |
|---|---|---|---|
| Old process, no new investment | S$4.00m | S$6.00m | 0.67× |
| New process, commissioning year | S$6.00m | About S$8.004m | About 0.75× |
| New process, full operation | S$9.40m | About S$8.004m | About 1.17× |
The model has identified a problem, not selected a solution. More equity could reduce debt service. A reserve could cover the commissioning gap. A different amortisation schedule could move principal payments later. A phased project could reduce initial cash use. A smaller investment might preserve liquidity but deliver fewer benefits. Each option changes risk allocation and may have additional costs or feasibility limits.
It would be wrong to describe delayed principal as free funding. Interest still accrues under the chosen terms, later payments may be larger, and a longer final maturity can create refinancing exposure. Nor is sponsor support a fact merely because a spreadsheet includes it. A support assumption needs an amount, a timing condition and evidence that the sponsor can provide it when the project is under pressure.
The project should also be tested at the lower carbon-price assumption. The old process has S$10 million available for the existing S$6 million debt service. With the new process, cash available is S$11.80 million, but total debt service is approximately S$8.004 million. Coverage is about 1.47 times. The borrower can service the debt under these simplified full-operation assumptions even though the project-only NPV was negative at that carbon price. Affordability and value creation are different tests.
This distinction is important. A borrower may be able to repay a loan funding an investment that does not earn an adequate economic return. Conversely, a high-value project may require financing with enough patience to survive construction and commissioning. Credit analysis should not silently substitute one test for the other.
A robust transition-finance discussion therefore follows a sequence. First establish that the technical change can work. Then test its commercial value. Then build the financing path. Finally examine whether the borrower survives each important date, including adverse commissioning outcomes. Skipping to a sustainability label before completing that sequence risks confusing intention with bankability.
The related Project Finance, Infrastructure, Cash Waterfalls and DSCR guide develops the same timing discipline for longer-lived assets. A transition project may be smaller than a power station, but the need to fund the journey before enjoying the destination is the same.
15. Green bonds, transition bonds and sustainability-linked terms ask different questions
A financing label can help explain the purpose or terms of a transaction, but it cannot replace an analysis of the claim. The investor still needs to know who owes the money, how interest and principal are paid, what happens after a missed target and which risks remain. A label can improve information without guaranteeing credit quality or environmental success.
ICMA’s Green Bond Principles describe a use-of-proceeds approach: the relevant proceeds support eligible green projects, with associated process and reporting expectations. The borrowing entity and the financed project are not necessarily the same analytical object. A company can finance a qualifying project while still having other activities and ordinary credit risks.
A sustainability-linked instrument instead connects specified financial or structural terms to performance against defined measures and targets. ICMA’s Sustainability-Linked Bond Principles provide the reference framework. The reader should examine the exact target, baseline, observation date, verification process and consequence rather than assume all instruments share the same design.
ICMA also introduced Climate Transition Bond Guidelines in November 2025, alongside updated transition-finance guidance. The distinction matters because an issuer’s overall transition strategy and a particular financing transaction’s proceeds are related but different. ICMA guidance is a market framework; it should not be presented as a universally binding national law.
Consider a purely hypothetical S$100 million bond with a twenty-five-basis-point annual coupon increase after a missed target. The extra annual interest is S$0.25 million. If it applies for three years, the undiscounted extra cost is S$0.75 million. That may be economically relevant, but it is much smaller than a S$12 million equipment investment. The size of the step-up alone cannot establish that management will prefer completing the investment to missing the target.
Other incentives may matter: customer contracts, future financing access, operating savings, governance and reputation. The educational point is to compare magnitudes rather than assume a contractual penalty is automatically decisive. A target with a small financial consequence can still convey useful information, but it should not be described as a physical guarantee.
The observation date also matters. If performance is tested after most debt has matured, the financing consequence may arrive too late to influence the relevant decision. If the baseline can change without a clear adjustment method, reported achievement can become difficult to interpret. If the target only measures one facility while the issuer expands higher-emissions activity elsewhere, the boundary should be explained.
Verification should distinguish measurement from judgement. An external reviewer might confirm that a metric was calculated according to a stated method. That does not necessarily mean the transition plan is commercially executable, the borrower is low risk or the target represents the most ambitious feasible pathway. Readers should not give an assurance statement a wider meaning than it supports.
For a bank, the practical integration is straightforward in principle: use the financing framework to understand purpose and commitments, then use ordinary credit, liquidity and project analysis to test repayment and execution. For the reader, the strongest habit is to ask what changes when the target is met, when it is missed and when the data cannot be verified. Those three states reveal much more than the label on the cover.
16. Financed emissions are not a credit-loss measure
A financed-emissions calculation attributes a share of an activity’s emissions to a financing relationship under a defined method. It is useful for understanding a portfolio’s association with emissions. It is not an estimate of how much money a bank will lose. Tonnes and dollars answer different questions, even when the same transition driver affects both.
For a simple attribution illustration, suppose a business emits 100,000 tonnes a year and a bank’s financing share is twenty per cent under a stated toy denominator. Attributed emissions are 20,000 tonnes. The arithmetic is transparent, but it is not a complete PCAF calculation. Actual methods depend on asset class, denominator conventions, data and boundaries. The purpose here is to show why an attributed number can move for more than one reason.
First, physical emissions can fall. If the business reduces its annual emissions to 70,000 tonnes and the financing share remains twenty per cent, attributed emissions fall to 14,000 tonnes. That movement reflects a real change in the activity under the assumed boundary. A report can then explain what operational change produced the reduction.
Second, financing can change. If the bank’s share falls from twenty per cent to ten per cent while the business remains at 100,000 tonnes, attributed emissions fall to 10,000 tonnes. The bank’s portfolio measure improves, but the business’s physical emissions have not changed. The report should identify the financing change rather than imply operational decarbonisation.
Third, the denominator or methodology can change. In an illustrative calculation where the financing denominator doubles while the bank’s exposure and the business’s emissions remain constant, the attributed amount can halve. Whether that exact effect occurs in a real method depends on its rules. The broader point is that attribution is a relationship between quantities, so denominator movement can matter.
Fourth, data quality can improve. A previously estimated footprint may be replaced by measured information. If the measured number is higher, the reported portfolio footprint can rise without a corresponding increase in real activity. The correct explanation is a measurement update, not automatically a deterioration in operating performance.
Intensity metrics create a related issue. A business producing one hundred units with one thousand tonnes of emissions has intensity of ten tonnes per unit. If it later produces two hundred units with sixteen hundred tonnes, intensity improves to eight tonnes per unit while absolute emissions rise sixty per cent. Both statements are true. Reporting only one can mislead a reader who assumes it answers the other question.
Credit analysis adds still more variables. The growing business may have stronger cash flow, greater debt or both. Its emissions intensity does not establish its debt-service capacity. Similarly, a company reducing output may report lower absolute emissions while becoming less able to repay. The direction of the environmental metric does not determine the direction of financial risk.
The useful reporting approach is a bridge: opening attributed emissions, change from borrower activity, change from financing, change from valuation or denominator, change from methodology and closing attributed emissions. A parallel financial bridge can show changes in exposure, expected loss, concentration and liquidity. The two bridges can then be discussed honestly without pretending they measure the same object.
For students, this is an excellent lesson in ratios. When a ratio changes, identify the numerator and denominator before inventing a story. When an attribution changes, identify the underlying activity and the financing share. A clear explanation can be more valuable than a more elaborate dashboard because it tells the reader what actually moved.
17. Good climate data begins with the question it must answer
More data is not automatically better data. A bank can collect a large sustainability questionnaire and still lack the address of the facility that generates most of the borrower’s cash. A detailed emissions inventory can coexist with no information about debt maturity or insurance renewal. Data quality should be judged relative to the decision it supports.
For physical risk, the useful unit may be a site, asset, access route or supplier. For transition risk, it may be a production process, energy source, contract, product line or planned investment. For credit risk, it may be a borrower or obligor group. The analyst needs a way to connect these units without assuming they are identical.
A fictional borrower might have its registered office in one city, its main factory in another country and its critical supplier elsewhere. Screening only the headquarters can miss the source of revenue risk. Conversely, applying the worst site-level hazard score to the entire group can overstate exposure if that site is small and easily replaced. The data model should preserve the share of activity that depends on each asset.
Every important input should have a date, a source and a status. Is it measured, estimated, modelled or asserted by the borrower? Does it describe current operations or a future plan? Has it been independently checked? Can the figure be reconciled to invoices, meters, production records or financial statements? These questions make the data usable for challenge.
Missing data deserves an explicit state. It is not zero risk, zero emissions or proof of misconduct. A model might use a conservative proxy, require additional evidence, narrow the conclusion or assign a wider uncertainty range. The choice should be proportionate to materiality and explainable. Treating every missing field as either harmless or disqualifying is rarely a good analytical habit.
Data should also be available at the time of the decision. Suppose a model predicting which borrowers complete transition projects uses final verified energy savings as an input. It may appear highly accurate in a retrospective test because it has leaked the outcome into the prediction. At loan approval, those savings were not yet known. The model has answered an easier question than the one the lender actually faced.
A similar error occurs when a past flood model is tested using updated site protections that were installed after the event. The retrospective result may understate the vulnerability that existed at the time. Good validation uses the information set available at the relevant date, then separately studies how later improvements changed outcomes.
Ownership and permissions matter as well. Climate analysis does not justify indiscriminate collection of personal or commercially sensitive data. The institution should ask for information relevant to the decision and handle it under applicable legal and governance requirements. A small borrower should not be asked to produce an elaborate model merely because a large borrower can afford one.
The most useful data set is often a compact, well-maintained set of material facts: key sites, important dependencies, cash-flow exposure, insurance terms, debt deadlines, funded investment milestones and the evidence supporting each. That is enough to build a meaningful first loop. Additional detail should be added where it changes the decision, not because the database has room for another column.
18. A transition plan is a sequence of fundable commitments
A transition plan can contain targets, aspirations and a description of intended changes. To become useful in finance, it needs a sequence: what will be done, who will do it, what it costs, how it is funded, what it depends on and how completion is tested. The plan should connect an operating change to a financing path.
Return to Meridian Components. “Reduce emissions by sixty per cent” is a target. “Purchase a new production line” is an activity. Neither is enough to establish bankability. The bank needs to understand whether the line can meet product specifications, whether the site has sufficient power, whether installation interrupts existing revenue and whether staff can operate the new process. The target becomes credible only through those intermediate conditions.
A milestone should produce evidence that changes uncertainty. Signing an equipment order proves commitment but not installation. Delivery proves arrival but not performance. Commissioning tests can demonstrate output and energy use under stated conditions. Several months of operating data can show whether the improvement persists. Each milestone answers a different question.
Financing can be linked to milestones without pretending that documentation eliminates risk. A staged draw might fund design first, equipment second and final payment after performance testing. The borrower still needs enough cash between stages, and the lender needs to understand what happens if a milestone is delayed for a legitimate reason. A rigid process can accidentally stop a viable project at the moment it most needs liquidity.
The plan should also specify what management can control and what depends on others. The company may control equipment maintenance but not the timetable for a grid connection. It may negotiate customer contracts but not guarantee future market demand. It may plan to refinance but not control capital-market conditions. External dependencies need their own evidence and contingency branches.
A credible fallback is not necessarily “complete the original project at any cost.” If a technology fails to meet specification, another process or a smaller investment may preserve more value. If demand changes, a different product mix may be appropriate. Financing documents and governance need to distinguish justified adaptation of the plan from abandonment or misrepresentation.
This is also where financial and environmental objectives must remain visible. A cheaper fallback may improve debt service but deliver a smaller emissions reduction. A more ambitious alternative may deliver greater reduction but require more equity. Neither trade-off should be hidden by updating a single overall score. The decision-maker should see both consequences.
Alicia records the technical evidence. Tricia records the funds already committed and the next cash requirement. Kai Kai records the unresolved dependencies. At each review, they ask what new information has arrived and whether it changes the preferred action. The plan remains alive because it learns; it is not credible merely because its first version was attractively designed.
The practical test is whether a reader can trace a target backward to a funded action and forward to an observable result. A target without a funded action is an aspiration. A funded action without a measurable result is difficult to evaluate. A result without a decision rule does not close the learning loop. Transition planning becomes financially useful when all three connections are present.
19. Scenario analysis is a conditional question, not a prediction contest
A scenario asks what would happen if a specified set of conditions occurred. It is not automatically a forecast of what will happen, and it does not automatically carry a probability. This distinction is essential in climate finance because long horizons, changing policy, technology and physical conditions make the future difficult to compress into a single most-likely path.
A useful scenario has an internally coherent story. It identifies the physical conditions, policy assumptions, economic environment, technology availability and timing of change. The borrower model then translates that story into cash flow, debt service, collateral and recovery. The scenario and the translation model are separate objects; each can be challenged independently.
For Meridian, one scenario might assume a gradual cost increase with enough time to install the new process. Another might assume a delayed but sharp increase, creating a narrow commissioning window. A third might combine lower carbon-related costs with more demanding customer specifications. A fourth might combine a physical interruption with installation delay. These are invented pathways for comparing vulnerabilities, not official scenario probabilities.
Do not call a scenario “severe” merely because one input is large. Severity depends on where the borrower is vulnerable. A modest energy-price increase can be serious for a thin-margin business that cannot pass through costs. A larger increase might be manageable for a business with strong contractual pass-through and low energy intensity. The same external path can produce different financial outcomes.
The horizon should match the question. A one-year credit decision needs near-term cash and refinancing assumptions. A thirty-year asset decision needs longer-term operating and residual-value analysis. A bank liquidity test may need daily or weekly timing even when the underlying driver develops over decades. One horizon cannot substitute for all three.
A common mistake is to combine the most adverse value of every variable without checking whether the combination makes sense. Another is to assume that every beneficial response happens perfectly while every adverse variable follows the stress path. Both create misleading results. A scenario should state which responses are allowed and whether they are constrained by cash, time, technology or counterparties.
There are also two different uses of a static balance sheet. Holding the portfolio fixed can reveal the vulnerability of today’s exposures. Allowing the portfolio to evolve can explore future business choices. Neither is inherently wrong. The mistake is to present a fixed-portfolio result as though it incorporates future risk reduction, or to let a dynamic model remove all difficult exposures without explaining who buys them and at what price.
When probabilities are not supported, the model can compare outcomes without calculating a probability-weighted expected value. It might show that an investment remains cash-positive under three pathways but requires extra liquidity in a fourth. That is useful information. Assigning arbitrary weights solely to produce a single number can hide the very vulnerability the scenarios were intended to reveal.
The NGFS describes its first short-term climate scenarios, published in May 2025, as tools for examining near-term macro-financial consequences through 2030. Scenario libraries can provide structured inputs, but users still need borrower-specific translation, model governance and an honest account of limitations. The following chapter explains why the exact version of a scenario library matters.
20. The NGFS warning: model versions are part of the evidence
On 3 December 2025, the NGFS issued a statement about physical-risk estimates in Phase V of its long-term scenarios. It noted that the Kotz et al. paper underpinning a damage function had been retracted from Nature. Users were told to consider that retraction when interpreting affected Phase V results. The NGFS also stated that outputs not incorporating those physical-loss estimates, and its short-term scenarios, were not affected. Its statement described scenarios as illustrative pathways, not forecasts, and said an updated methodology was planned for the next long-term iteration at the end of 2026.
That is a specific model warning, not a reason to declare every climate scenario invalid. It is also not a reason to continue using an affected number as though nothing happened. As of this guide’s review date, 20 September 2026, a methodology described as planned for the end of 2026 should not be presented as already delivered on the strength of that statement alone.
The general lesson is model provenance. A result should identify the scenario version, underlying variables, transformations and date of use. “Based on NGFS” is too broad if the reader cannot determine which release or component supplied the number. A source name is not a substitute for a reproducible chain of evidence.
Imagine a fictional bank report that imported a long-term physical-loss variable into a borrower model and then used that output to adjust collateral. When the upstream model changes, the bank needs to know which borrower results depend on it. Without lineage, it may have to treat the whole report as uncertain. With lineage, it can isolate affected calculations, retain unaffected work and rerun the relevant branches.
This is why a useful model record contains both the external source and the institution’s own transformation. The source might provide a macroeconomic path. The bank then maps that path into sector revenue, borrower cash flow and collateral. A correction to the source may affect some of those mappings but not others. Conversely, the source can be sound while the bank’s translation is poor.
A sensible teaching response to model uncertainty is to compare alternatives. Use more than one plausible damage estimate, show a range, test which decision changes and identify where a result depends on a particular function. If the financing choice remains the same across a broad range, the decision may be robust even though the exact loss estimate is uncertain. If a small change reverses the conclusion, that sensitivity deserves attention.
One should also separate a model’s intended purpose from a use it was never designed to support. A macroeconomic scenario is not automatically a site-level engineering assessment. A long-run GDP effect is not automatically a one-year loan loss. A regional variable does not identify which floor of a building floods. Each translation needs an explanation and, where possible, validation against relevant observations.
Alicia records what the source actually says. Tricia records the calculation that used it. Kai Kai records which decisions would change if the source is revised. That is a stronger form of confidence than pretending models never change. A trustworthy system can incorporate corrections without losing track of what it previously believed or why it acted.
21. Avoid double counting before making the model more elaborate
Adding more risk channels can make an analysis look comprehensive while accidentally counting the same consequence several times. Climate-finance models are especially vulnerable because physical damage, business interruption, revenue decline, collateral loss and credit deterioration can all describe different views of one event.
Consider Harbour Store. The cash table includes repair payments and lower customer receipts. Suppose a separate annual operating-profit adjustment already includes those same repairs and lost sales. If the analyst subtracts both the cash-table effects and the full profit adjustment, the model overstates the loss. The calculation has become more conservative only in appearance; it is actually inconsistent.
The opposite error occurs with insurance. A model may reduce direct loss by an expected recovery, then add the same recovery again as a cash inflow without recognising that the first reduction already represented it. The financial statements and cash model need a reconciliation. One cannot simply add every favourable item found in different data sources.
Collateral creates another trap. A flood damages a property, repairs are paid and the restored property is later sold. If the model assumes full repair cost and also values the property as permanently unrepaired, it may overstate the final loss. That might be appropriate only if the repairs do not restore the relevant value, and the reason should be stated. Conversely, assuming full restored value while omitting necessary repair spending understates the cost.
Transition models can double count savings. If the new process’s lower electricity use is already incorporated in operating expenditure, the same energy savings should not appear again as an additional project benefit. If a carbon-cost reduction is included in the lower unit cost, it should not be added twice through a separate emissions benefit. Keep the physical calculation and financial translation connected.
A simple reconciliation helps. Begin with the no-action financial path. Add each incremental effect once, identify the account or cash-flow line it changes, and show the new path. The difference between the paths should equal the sum of those incremental effects after interactions are handled. If it does not, the model needs explanation before it needs more variables.
Discounting should also be consistent. A cash flow that already includes an expected loss adjustment should not automatically receive another arbitrary discount-rate penalty for the identical risk. There may be legitimate reasons to adjust both expected cash flow and the required return, but the components must represent different things. Otherwise risk can be penalised twice without a clear economic interpretation.
A related issue is nominal versus real amounts. A scenario may report inflation-adjusted operating costs while debt service is stated in nominal currency. Combining them without conversion creates a false cash gap or a false surplus. The same applies to currency: a US-dollar cost path cannot be added directly to Singapore-dollar revenue without an exchange-rate assumption.
The final check is units. Tonnes are not dollars. Annual probabilities are not cumulative probabilities. A percentage-point change is not a percentage change. A gross loss is not a net recovery shortfall. A commitment is not a drawn exposure. These distinctions may look elementary, but they are exactly where a complex model can become confidently wrong.
For a student, the best audit begins with a clean one-page bridge. What did the borrower own and owe at the start? Which event changed which line? What cash moved? What remained as a claim? What is the closing state? Once that bridge reconciles, the analysis can be expanded without losing its accounting foundation.
22. Choose actions that remain useful when the forecast is wrong
A decision does not need a perfect forecast to be well reasoned. It needs a clear objective, a set of plausible conditions and an action whose benefits and costs are understood across those conditions. In climate finance, this often means comparing robustness rather than selecting the option with the best result under one preferred pathway.
For Harbour Store, a tested backup procedure might have modest value in several disruptions, not only a flood. Raised critical equipment could reduce damage in one range of events but not protect against loss of road access. A larger cash buffer could help with many shocks but would not reduce physical damage. These measures serve different purposes. The portfolio of responses may be more useful than an attempt to find one perfect solution.
For Meridian, the S$12 million equipment project has high financial value in the higher carbon-cost scenario and negative project NPV in the lower one. Management could examine whether a smaller efficiency investment produces useful savings across both scenarios. It could also investigate whether design work now preserves the option to install later. Such choices trade some immediate ambition for flexibility; whether that trade is sensible depends on the cost of waiting and the risk of being too late.
The option to wait is not free. A delayed project might face higher equipment prices, a missed customer contract or a more compressed installation window. Conversely, committing too early can lock the company into a technology that later proves unsuitable. A robust analysis identifies what information is likely to arrive during the waiting period and whether it is valuable enough to justify delay.
One way to compare actions is to calculate regret. For each scenario, identify the best feasible action with hindsight. Then compare the loss from choosing another action. An option with the highest average payoff might have a disastrous outcome in one plausible state; another might give up some upside but avoid that failure. The decision-maker can then see the trade-off rather than hide it in a weighted average.
Liquidity constraints should remain explicit in this comparison. An action can have excellent expected NPV but require more cash than the borrower can safely commit. A robust action is not merely one that performs well economically; it must be executable from the current state. Financing access is part of feasibility, not an optional appendix.
Trigger-based decisions can help where uncertainty is material. A company might prepare a project now, then commit to full investment when a specified customer contract, price condition or technical test is achieved. The trigger must be observable and connected to the economics. A vague promise to “review later” is not a decision rule.
Triggers can also create delay risk. If management waits for certainty that only arrives after the project should have started, the rule is poorly designed. A useful trigger should reflect both evidence and lead time. A factory needing eighteen months to install equipment cannot wait until the day before a contractual requirement changes.
For financial institutions, the same principle applies to monitoring and exposure. A model can define conditions that prompt a deeper review, a request for evidence or a reassessment of financing. It should not automatically substitute a broad score for an authorised credit judgement. The goal is to make the next decision more informed, not to create an unchallengeable machine.
Kai Kai’s question is, “What would we do if this forecast is wrong?” That question is not pessimistic. It is the beginning of a resilient plan. A system that can update its actions when evidence changes is more useful than one that can only explain why its original prediction should have been correct.
23. Singapore: distinguish an issued guideline from an effective requirement
Singapore provides a useful example of why dates and legal status matter. In March 2026, MAS issued sector-specific transition-planning guidelines as addenda to its environmental risk management guidelines for banks, insurers and asset managers. The MAS-authored release states that they take effect in September 2027, following an eighteen-month transition period. It describes proportionate, forward-looking management of environmental physical and transition risks and emphasises constructive engagement rather than indiscriminate withdrawal from customers or investments. The release is reproduced at The Asian Banker; the MAS source page is the official reference.
As of this guide’s review date, 20 September 2026, “issued” and “effective” therefore describe different states. A reader should not convert an implementation period into a claim that every detailed expectation is already in force. Nor should future effectiveness be mistaken for irrelevance: preparation, data and governance can require time. The appropriate interpretation depends on the institution and the actual final documents.
This guide does not replace those documents or determine an institution’s legal obligations. Its purpose is to explain the financial reasoning that makes proportionate transition planning useful. The small-business borrower in our warehouse case does not become easier to assess simply because a bank asks it to complete a much larger questionnaire. The bank needs information connected to material cash-flow and recovery risks.
For a Singapore-based learner, the international dimension is especially important. A business can have local banking relationships while earning revenue from facilities, suppliers or customers elsewhere. A lender’s exposure may therefore depend on foreign physical conditions, policy and energy markets. The country of the bank account does not define the full risk boundary.
The same applies to educational examples associated with Bukit Timah. This article does not imply that a named local school, building or household faces a particular hazard or credit risk. Such a conclusion would require location-specific evidence and a proper assessment. The local value of the guide is intellectual: it connects mathematics learned in ratios, probability, cash flow and systems reasoning to a wider financial problem.
Proportionality can be understood as a decision about information value. A large loan to an asset-intensive company with several exposed sites may justify detailed site and investment data. A small exposure to a simple business may require a more focused assessment. The question is whether an additional data request changes the understanding of risk enough to justify its cost, not whether every borrower can produce the same report.
Engagement also needs a financial definition. It is not a promise to lend regardless of risk. It means understanding the borrower’s plans, financing constraints and feasible improvements before reaching a decision. A company may have a credible adaptation project but need a different financing schedule. Another may have an attractive target but no executable investment plan. The lender should be able to distinguish them.
For students, the key habit is to read the date, the status and the scope before repeating a regulatory claim. A consultation is not a final rule. A final guideline may have a later effective date. A framework for one sector is not automatically a requirement for every company. These distinctions are part of factual accuracy, not technical trivia.
24. Disclosure, assurance and action are different layers
IFRS S2 focuses on climate-related information useful to users of general-purpose financial reports, including governance, strategy, risk management, and metrics and targets. The IFRS Foundation’s standard overview and scenario-analysis implementation material provide the authoritative starting points. Application through local reporting requirements must be checked separately.
An important boundary is that IFRS S2 does not itself require an entity to have a transition plan. The Foundation’s implementation resources distinguish disclosure about transition-related information from an obligation to create a plan. A reader should not transform a disclosure standard into a universal operating mandate.
Disclosure is valuable because it makes assumptions and decisions easier to examine. But publishing information does not automatically reduce risk. A bank can clearly disclose a concentrated exposure and still hold it. A company can transparently report weak progress. Good disclosure improves the evidence available for a decision; the decision and its real-world result remain separate stages.
Assurance adds another layer. Depending on the engagement and framework, an assurance provider may assess whether specified information is prepared according to stated criteria. The reader must understand the scope. Assurance over an emissions figure is not automatically assurance that a transition project will finish on time or that a loan will be repaid. The conclusion should not be stretched beyond the work performed.
A useful public explanation therefore separates four statements: what the organisation intends; what it has committed resources to; what has actually happened; and what has been independently checked. Mixing those statements makes progress look stronger than the evidence supports. A future target is not a current outcome. A signed contract is not a commissioned asset. A verified calculation is not a verified commercial forecast.
Consider Meridian’s new production line. An intention statement says management aims to reduce emissions. A commitment statement identifies the approved budget and financing. An implementation statement reports equipment installation and commissioning. An outcome statement reports measured energy use, output and emissions after operation. An assurance statement identifies which of those reported quantities was examined and under what criteria. Each statement has a different evidential weight.
The financial reader also needs a bridge from those statements to cash. If the equipment meets its emissions target but maintenance is more expensive than expected, debt-service capacity can still disappoint. If the project is delayed, the financial consequences should not wait until the annual sustainability report. Monitoring needs to operate at the speed of the risk.
Greenwashing can be understood, in part, as a failure of these boundaries. A claim about a limited project is presented as a claim about the whole company; a portfolio transfer is presented as a physical emissions reduction; an aspiration is presented as a funded plan; or a modelled future benefit is presented as a current result. The corrective habit is to ask for the boundary, the baseline, the date and the supporting evidence.
The opposite problem is excessive scepticism that ignores genuine but imperfect improvement. A company can make measurable progress without having solved every challenge. A good report can state the improvement, the remaining exposure and the uncertainty together. Precision does not require either celebration or condemnation. It requires saying exactly what the evidence supports.
For a closed-loop system, the final question is what changes after the disclosure. Does the board reallocate capital? Does the lender revise an assumption? Does the borrower alter the project? Does an investor reassess price? Information becomes part of risk management when it changes a decision that can be observed and evaluated later.
25. Monitoring should notice a broken promise before the payment fails
A climate-finance monitoring system should follow the promises on which the financing decision depends. If the loan assumes that equipment will be commissioned in September, commissioning evidence matters before the first full debt-service period. If the model assumes an insurance recovery in month four, the status of the claim matters before the liquidity bridge matures. An annual report may be too slow for either problem.
The starting point is a small set of decision-relevant observations. For Harbour Store, those could include unrestricted cash, delivery capacity, repair progress, the insurance claim’s status and the remaining drawable facility. For Meridian, they could include installation milestones, power availability, production quality, actual energy use, project spending and cash available for debt service. Each observation should connect to an assumption in the original model.
A useful trigger has four elements: a measured condition, a threshold, a responsible decision-maker and an authorised response. “Watch the project closely” lacks all four. “If commissioning is more than six weeks late, refresh the monthly cash forecast and escalate the expected funding gap” is more specific. The six-week threshold is merely an example; a real threshold would depend on the project’s cash and operational timetable.
Not every trigger should automatically reduce credit. A delay could be temporary and manageable with a revised sequence. It could also reveal a fundamental technical problem. The purpose of escalation is to obtain the right evidence and reassess the decision, not to substitute a rigid rule for judgement. A system that reacts identically to every delay can make good projects harder to finish.
Monitoring should include measures of whether its own process works. How late does information arrive? How often are milestones revised? Which forecasts are systematically optimistic? How many alerts produce a meaningful decision? If a bank collects data but cannot process it before the financing deadline, the system has a capacity problem, not merely a data problem.
One can model that capacity problem simply. Suppose a review team receives fifty material project alerts each month but can resolve only forty. Its unresolved queue grows by ten a month. After six months, sixty additional cases are waiting, assuming the rates remain unchanged. Calling the process “enhanced monitoring” does not make it effective. The workload, prioritisation and decision authority have to match the volume.
The quality of an alert also matters. A missed emissions target caused by a temporary production mix change may need a different review from a missed equipment-delivery milestone that threatens cash. If both are assigned the same urgency, the more immediate financial risk can be buried. Risk prioritisation should consider consequence and deadline, not only the size of a deviation.
The review should preserve the original forecast. Replacing it with the latest forecast without retaining earlier versions destroys the ability to learn. The institution should be able to compare what was expected at approval, what was known at each review and what ultimately happened. That comparison reveals whether errors came from bad data, weak assumptions, poor execution or an external change.
Alicia’s record asks what changed physically. Tricia’s asks what changed financially. Kai Kai’s asks whether the institution changed its decision. A complete monitoring cycle ends with an action, a reason and a future check. The action might be to maintain the original plan because the evidence still supports it. Doing nothing can be a decision, but it should not be confused with failing to review the evidence.
26. The borrower’s climate problem can become the bank’s liquidity problem
A bank may hold many loans to businesses that draw working-capital facilities when disruption occurs. Those facilities help customers survive, but their simultaneous use creates a bank-level funding requirement. The same event can also prompt deposit withdrawals, collateral calls or asset sales. A portfolio stress test that calculates credit loss alone can therefore miss the most immediate constraint.
Consider a fictional bank with S$80 million of settlement-ready liquidity at the start of a stressed day. Existing customers transfer S$60 million of deposits to other banks. Separately, affected businesses draw and externally spend S$25 million under committed lines. A further S$15 million is required for margin and other specified settlement obligations. Expected incoming payments of S$20 million arrive only on day three, while S$30 million of securities can be monetised on day two.
Across all three days, the arithmetic looks comfortable: 80 minus 60 minus 25 minus 15 plus 30 plus 20 leaves S$30 million. But on day one, the bank is S$20 million short before the later receipts. The timing of asset monetisation matters. A security that can produce cash tomorrow cannot settle today’s payment unless some other facility bridges the interval.
The S$25 million customer draw is defined here as an additional net external settlement outflow. It is not counted again inside the S$60 million of existing-deposit transfers. This boundary matters because new lending can create deposits before those deposits are spent. Counting both the loan draw and its external payment as separate cash outflows would overstate the liquidity need.
Credit loss is another state variable. Suppose the same event produces a hypothetical S$6 million economic credit loss and S$4 million recognised market loss against S$50 million of starting equity. The simplified equity state falls to S$40 million before other changes. Those losses are not automatically S$10 million of same-day cash outflow. Capital and liquidity need separate calculations even when the same shock causes both.
The bank’s response can feed back into customers. Drawing committed lines helps borrowers repair and continue operating, but it uses bank liquidity. Refusing new discretionary lending may preserve resources, but it can prevent a viable borrower from completing adaptation. Selling securities can raise cash, but a stressed sale may crystallise loss. Raising deposit rates can retain funding, but it increases future expense. No single action improves every state variable.
This is why assumptions about management action need discipline. A bank cannot simultaneously assume that every customer receives support, no additional liquidity is required, securities sell instantly at unchanged prices and funding costs stay constant. Those assumptions describe an unrealistically generous system. A useful model names the resources consumed by each action.
There is also a timing issue in customer engagement. A bank that understands material vulnerabilities before an event may be better able to distinguish temporary cash needs from permanent impairment. A bank that first gathers information after all facilities are drawn has fewer options. Early information can improve the allocation of limited liquidity without guaranteeing that every request will be met.
The example does not prescribe a bank treasury strategy. It shows how to connect the borrower-level climate analysis to ordinary banking mechanics. A climate event can be slow in origin and fast in settlement consequences. The institution must therefore preserve short-horizon cash models alongside long-horizon scenarios. The broader distinction is developed in Balance Sheets, Liquidity, Capital and Bank Survival.
27. Worked case: different decision objectives can select different adaptations
A small logistics company is choosing among three fictional actions for a single planning period. It can make no new investment, install a barrier, or relocate a vulnerable operation. The following table shows total economic cost under three invented scenarios. Total cost includes the stated investment and residual damage for that period. It does not include unpriced social effects, and it is not a site-specific recommendation.
| Action | No damaging event | Moderate event | Severe event |
|---|---|---|---|
| No new measure | S$0.00m | S$0.80m | S$3.00m |
| Barrier | S$0.30m | S$0.50m | S$2.00m |
| Relocation | S$1.20m | S$1.25m | S$1.40m |
The barrier costs S$0.30 million even if no damaging event occurs. Its moderate-event residual loss is S$0.20 million, producing total cost of S$0.50 million. Its severe-event residual loss is S$1.70 million, producing total cost of S$2 million. Relocation is more expensive initially but leaves smaller residual event losses. The table is deliberately constructed to show a trade-off rather than make one action best in every state.
A decision-maker focused only on the largest possible cost in these three scenarios selects relocation. Its worst cost is S$1.40 million, compared with S$2 million for the barrier and S$3 million for no new measure. This is a minimax-cost criterion: choose the action with the smallest worst-case cost within the scenario set.
A decision-maker focused on maximum regret gets another result. Regret is the difference between the cost of an action and the best cost available in that scenario with hindsight. In the no-event scenario, no investment is best. In the moderate-event scenario, the barrier is best. In the severe-event scenario, relocation is best.
The regrets for no new measure are S$0, S$0.30 million and S$1.60 million. The barrier’s regrets are S$0.30 million, S$0 and S$0.60 million. Relocation’s regrets are S$1.20 million, S$0.75 million and S$0. The barrier has the smallest maximum regret, S$0.60 million. Under this criterion, it is preferred even though relocation has the lower worst-case total cost.
Neither result is a universal answer. They reflect different objectives. Minimax cost puts particular weight on limiting the worst outcome. Minimax regret limits how far a decision could fall behind the best hindsight choice. A probability-weighted expected-cost approach would require probabilities or decision weights that have not been established here. The table should not be given arbitrary probabilities merely to produce a more familiar answer.
Feasibility can change the choice again. If the company can commit only S$0.50 million and cannot obtain additional financing, relocation may not be feasible. If the severe-event outcome would cause unacceptable service interruption or safety consequences, a purely monetary regret criterion may be insufficient. If relocation creates a new vulnerable dependency, the table itself needs revision.
The exercise also reveals the limits of a narrow boundary. A barrier might protect one property while affecting water movement elsewhere; a relocation might impose costs on workers or neighbouring communities. We are not asserting those effects in a real case. We are identifying questions that a financial-only table cannot answer. A decision should not silently treat an omitted consequence as zero.
Alicia checks whether the scenario outcomes are technically plausible. Tricia checks that investment and residual losses are not double counted. Kai Kai checks the objective and constraints. Their conclusion is not “the spreadsheet says build the barrier.” It is: the preferred action depends on what the organisation is trying to protect, which outcomes it can tolerate, and which actions it can actually finance.
This is a useful model of honest decision-making under deep uncertainty. It makes disagreement intelligible. Two people can read the same table and prefer different actions because they have different objectives, not because one has failed at arithmetic. The next conversation should clarify those objectives and constraints instead of searching for a spurious single correct forecast.
28. Build a small model that can be explained from beginning to end
For a learner, a compact and transparent model is often more valuable than a large model whose assumptions cannot be traced. The aim is not to reproduce a regulated institution’s systems. It is to understand one causal chain well enough to explain where each number comes from and how the result changes when an assumption changes.
Begin with a borrower and one material activity. Specify the operating cash flow, debt schedule, cash buffer and collateral. Then define one physical or transition scenario. The scenario must change an operational input: receipts, costs, production, asset condition or timing. Only after that step should it change the financial outputs. This order prevents the model from assigning a credit loss directly from an environmental label.
A basic cash table can use one row per month and separate columns for opening cash, customer receipts, operating payments, project spending, debt service, insurance, new financing and closing cash. Each row should reconcile. The next month’s opening cash should equal the previous month’s closing cash after any separately recorded adjustments. A negative balance identifies a funding need, not a magical overdraft.
A second table can show asset and liability changes. Repair spending, lost stock, new debt and recognised losses affect different parts of the balance sheet. The purpose is not to teach every accounting rule in one exercise. It is to stop the learner treating cash, profit, asset value and equity as interchangeable. A financially coherent story should be explainable in both the cash table and the balance-sheet bridge.
A third table can contain physical performance: output, energy consumption, emissions, downtime or repair progress under the chosen example. Those measures should not be mixed into the financial table until a conversion assumption is supplied. Tonnes avoided become money saved only through a stated price or cost relationship. Downtime becomes lost cash only through an operating model.
The model then needs a comparison. What happens without the project? What happens with it? What happens if implementation is late? Keep the same output and currency assumptions unless the scenario explicitly changes them. A fair comparison should not give the preferred option perfect execution and the alternative every possible disadvantage.
Sensitivity testing comes next. Change one parameter at a time to understand its effect, then combine parameters that could plausibly move together. For Harbour Store, test a delayed claim and slower customer receipts. For Meridian, test a higher installation cost and a lower annual benefit. For the portfolio, test common dependencies rather than only individual probabilities. Record which change first causes a cash, coverage or value boundary to be crossed.
Validation should include elementary tests. With no event and no investment, does the model return the base case? With zero debt, does debt service disappear? If the insurance payment is delayed, does earlier cash worsen rather than improve? If a project costs more with unchanged benefits, does NPV fall? If the exposure is doubled with all else fixed, does a simple expected-loss model scale appropriately? These checks can catch errors before any sophisticated statistical review.
The student should also label what the model cannot answer. It may not estimate actual flood probabilities. It may not determine insurance coverage. It may not model market-wide price effects. It may ignore tax, transaction costs or refinancing. Those limitations do not make it useless; they define where its conclusions stop. A small model with an honest boundary is more useful than a large one pretending to answer everything.
Finally, write a decision paragraph without looking at the formula. Explain what matters, what is uncertain, what action is being compared and what evidence would change the conclusion. If the paragraph cannot be written clearly, the model may contain more calculation than understanding. The strongest mathematical lesson is not a beautiful spreadsheet. It is the ability to connect assumptions, arithmetic, evidence and a bounded conclusion.
29. Exercises: test the mechanism, then read the worked answers
Exercise one: the delayed claim
A borrower starts with S$0.20 million in cash. During month one it receives S$0.30 million and pays S$0.60 million, including all stated operating and debt payments. An insurance payment of S$0.25 million is expected in month two. Does the claim remove the month-one funding need?
The answer is no. Month-one cash before additional financing is negative S$0.10 million. The claim may support a bridge, but it does not become month-one cash merely because it is expected. A complete answer also asks whether the claim is valid, when it will be paid, whether financing is available and what happens if the receipt is delayed. The arithmetic identifies the minimum gap; it does not establish access to the money needed to fill it.
Exercise two: intensity improves while absolute emissions do not
A factory produces one hundred units and emits one thousand tonnes. It then produces one hundred and twenty-five units at eight tonnes per unit. What happens to intensity and total emissions?
Intensity falls from ten to eight tonnes per unit, an improvement of twenty per cent. Total emissions remain one thousand tonnes because 125 multiplied by eight equals one thousand. Both statements should be reported. Claiming an absolute reduction from the intensity improvement alone would be incorrect. Credit analysis would still need revenue, costs, investment and debt information; neither emissions number gives a repayment answer.
Exercise three: a one-per-cent annual event is not a one-per-cent lifetime event
Assume, solely for teaching, a constant one per cent annual event probability and independent years. What is the probability of at least one event over thirty years?
The probability of no event in one year is 0.99. Over thirty independent years, no event has probability 0.99 to the power of thirty. Subtracting that from one gives approximately 26.0 per cent. The calculation is not a forecast for any real site. Changing probabilities or dependence across years requires a different model. The important lesson is to state the horizon and assumptions before interpreting a risk percentage.
Exercise four: conditional default and loss severity
Use the chapter-nine assumptions: ten per cent probability of flood; default probabilities of two per cent without flood and twenty per cent with flood; loss severities of thirty and sixty per cent respectively; exposure of S$1 million. Calculate overall default probability and expected loss.
Overall default probability is 3.8 per cent. Expected loss is S$17,400 from adding the two conditional state calculations. Multiplying 3.8 per cent by the unconditional thirty-three per cent severity average gives the wrong result because the more severe state also contains a larger share of defaults. A complete answer explains the conditioning rather than simply presenting the correct number.
Exercise five: the project’s break-even scenario price
In Meridian’s six-year, ten-per-cent-discount-rate model, annual benefit in S$ million is 0.06 multiplied by the hypothetical carbon price per tonne, plus 0.60. What constant scenario price makes the S$12 million project’s NPV approximately zero?
The six-year annuity factor at ten per cent is about 4.3553. Required annual benefit is therefore 12 divided by 4.3553, or about S$2.755 million. Subtract the S$0.60 million non-carbon benefit and divide by 0.06. The break-even scenario price is approximately S$35.92 per tonne. This is a sensitivity threshold inside an invented model, not a prediction of policy and not a real carbon-tax recommendation. Timing, costs and useful life would change the threshold.
Exercise six: a coupon step-up is not the same size as the project
A S$100 million instrument adds twenty-five basis points of annual interest for three years after a missed target. Compare the undiscounted extra cost with a S$12 million project.
Twenty-five basis points is 0.25 per cent. Annual additional interest is S$0.25 million; over three years it is S$0.75 million. That equals 6.25 per cent of the S$12 million project cost before discounting. This comparison does not determine the issuer’s preferred action because the project also has operating benefits and other incentives. It does show why one should calculate the consequence rather than assume a step-up automatically forces investment.
Exercise seven: two portfolios share an average, not a tail
Both chapter-twelve portfolios have expected loss of S$2.5 million. Why does that not establish equal resilience?
Expected loss is an average across states. The independent portfolio spreads losses across many combinations of borrowers. The common-shock portfolio concentrates loss in a single S$50 million outcome. Capital, liquidity and continuity questions depend on the distribution, not only its mean. A good answer identifies common physical and financial dependencies that could move a real portfolio away from an independence assumption, while recognising that perfect common default is an intentionally extreme teaching case.
Exercise eight: repair spending and collateral value
A model subtracts the full cost of restoring a damaged building, then values the final collateral as if no repairs had been completed. Is that necessarily correct?
No. It may count the same impairment twice unless the repairs fail to restore the relevant value or leave other permanent effects. The analyst needs an explicit post-repair valuation and a timeline. The opposite error is assuming restored value without paying for the repairs. The solution is a reconciled bridge from pre-event value through damage, spending, restoration and final recovery.
Exercise nine: effective date versus publication date
A guideline is issued in March 2026 with an eighteen-month transition period and effectiveness in September 2027. A report written in September 2026 says the full new requirements are already effective. What is wrong?
The report has confused issuance with effectiveness. It should identify the actual status and scope of the guideline. Preparation may already be appropriate, but that is different from asserting that the future effective date has arrived. The exercise also reminds readers to check whether a document is a consultation, a final guideline, a standard or a binding local requirement.
Exercise ten: define a useful monitoring trigger
A lender’s project assumption says full operations begin before the next annual debt-service payment. Write a trigger that can reveal a problem early enough to act.
One possible teaching answer is: if verified commissioning slips beyond the date needed to produce the assumed cash before debt service, refresh the monthly cash model and escalate the resulting gap to the authorised decision-maker. The trigger is linked to money and time, not merely a calendar label. A stronger answer also states who supplies evidence, who reviews it, what response options exist and when the revised forecast is checked again.
Exercise eleven: explain a falling financed-emissions number
A bank reports a twenty per cent decline in attributed emissions, but the borrower’s physical emissions are unchanged. Give at least two possible explanations that do not require operational decarbonisation.
The bank’s financing exposure or attribution share might have fallen. The relevant valuation denominator or methodology might have changed. The bank might have sold an exposure to another investor. Each explanation needs evidence. The correct conclusion is not that the reduction is meaningless, but that it should not be presented as a physical emissions reduction without showing the bridge.
Exercise twelve: write a bounded conclusion
A model shows that a warehouse adaptation package prevents a monthly cash shortfall in one specified event scenario. What can the analyst honestly conclude?
The analyst can conclude that the package improves the modelled cash path under the stated assumptions. It cannot conclude that the site is protected against every event, that insurance is guaranteed, or that the investment is optimal under every probability and cost assumption. A useful conclusion names the scenario, the investment cost, the material benefits, the remaining vulnerabilities and the evidence required to validate performance. Bounded conclusions are not weak; they are conclusions that can survive scrutiny.
30. Questions readers ask about climate risk and transition finance
Is climate risk a separate risk category from credit risk?
It is often more useful to treat climate-related factors as drivers that affect established financial risks. A flood can change a borrower’s cash flow and collateral, producing credit risk. A transition-related repricing can change market value. Simultaneous customer draws can create liquidity pressure. The point is to trace the channel rather than assume a new label replaces the existing financial model.
Does a low-emissions borrower always represent a safer loan?
No. The borrower may have high leverage, weak cash flow, a concentrated customer base or physical exposure. Emissions can be relevant to particular transition scenarios, but they do not determine every credit variable. The same caution applies in reverse: high emissions do not by themselves establish default. The analysis needs the business model, financing structure and feasible adjustment path.
Can adaptation be financed without a green label?
A financing label and a useful adaptation investment are different objects. Ordinary financing can support a project that improves resilience, subject to the lender’s assessment and transaction terms. A labelled instrument can improve clarity about purpose or commitments, but the project still needs credible performance and repayment assumptions. The financial analysis should not depend solely on the marketing category.
Is a climate scenario a forecast?
Not automatically. A scenario specifies conditions for examining consequences. Some models use probabilities; others compare pathways without assigning likelihoods. Readers should not invent probabilities for a scenario library or describe an illustrative pathway as the expected future. The model’s purpose, version and limitations should be visible.
Does the NGFS retraction warning invalidate all climate analysis?
No. The warning identifies an affected research input and associated outputs, with specific unaffected components described in the NGFS statement. The appropriate response is to trace dependence, review affected calculations and use alternatives or ranges where necessary. It is neither sound to ignore the warning nor to generalise it to every physical-risk model or every climate-related financial mechanism.
Does insurance remove a borrower’s physical risk?
Insurance can transfer defined financial losses under its terms. It does not necessarily prevent physical damage, customer loss or disruption, and recovery can arrive after cash is needed. The model should distinguish gross damage, covered loss, retained loss, receipt timing and renewal. A valid claim can be valuable while still leaving a liquidity problem.
Why can a profitable transition project fail financially?
The project’s benefits may arrive after construction costs and debt service. Delays, commissioning losses or weak working capital can exhaust cash before full operation. Positive NPV is a value test under assumptions; it is not proof of liquidity at every date. The financing schedule has to support the path to the expected benefits.
Are green bonds and sustainability-linked bonds the same?
No. A use-of-proceeds approach focuses on how relevant proceeds support eligible projects. A sustainability-linked approach connects specified terms to performance against targets. Transaction documents and applicable frameworks determine the details. Neither label automatically guarantees the issuer’s credit quality or the project’s success.
Should a bank immediately exit every exposed borrower?
That does not follow from the mathematics. Some borrowers may have credible, financeable ways to reduce risk; others may not. An assessment should distinguish temporary liquidity need, transition execution risk and permanent impairment. Continuing a relationship is not a promise to lend regardless of risk, and reducing an exposure is not proof that the underlying activity has changed.
Why can a financed-emissions metric fall without lower real-world emissions?
Attribution depends on financing relationships and method as well as the activity’s emissions. Exposure sales, changes in financing share, valuation denominators or data methods can move the reported number. A clear report decomposes those effects. Physical emissions and portfolio attribution should be explained together without being treated as identical.
What is the most useful first model for a student?
Start with one borrower, one material activity and a monthly cash table. Connect the scenario to receipts, costs, investment and recovery timing. Add debt service and available liquidity. Then compare a no-action path with an adaptation or transition path. This small model teaches the causal chain before more complex portfolio or macroeconomic models are introduced.
What is the final test of a closed-loop climate-finance system?
Evidence must return to a decision. If the barrier performs, the production line saves energy or the recovery arrives late, that information should change assumptions, monitoring or financing where appropriate. A score that is never challenged and a plan that is never updated remain open loops. The system becomes stronger when it can identify what it believed, what happened and what should change next.
Evidence laboratory: how a successful test should change confidence
The guide has repeatedly asked for evidence that an adaptation or transition measure works. But evidence is rarely perfect. A successful commissioning test should increase confidence without necessarily turning uncertainty into certainty. A failed test should prompt investigation without automatically proving that the whole project is worthless. Probability provides a useful way to understand that update.
Consider an invented protection system with two simplified states: it will perform adequately in the specified event, or it will not. Before a test, the analyst assigns an eighty per cent probability to adequate performance. That prior is a teaching assumption, not a statement about any real technology. A test passes ninety-five per cent of the time when the system is adequate. It also passes thirty per cent of the time when the system is inadequate. The test is informative but imperfect.
In a population of one thousand comparable teaching cases, the assumptions imply eight hundred adequate systems and two hundred inadequate systems. Of the adequate systems, seven hundred and sixty pass. Of the inadequate systems, sixty pass. There are therefore eight hundred and twenty passing results, of which seven hundred and sixty correspond to adequate systems. Conditional on a pass, the probability of adequate performance is 760 divided by 820, or approximately 92.68 per cent.
That is a substantial improvement from the eighty per cent prior. It is not one hundred per cent. The remaining uncertainty reflects false-positive tests. A report stating “the system passed, therefore failure is impossible” would overstate the evidence. The correct conclusion is that the passing result changes confidence under the assumed reliability of the test.
Now examine a failure. Forty adequate systems fail the test, and one hundred and forty inadequate systems fail it. Among one hundred and eighty failures, forty are still adequate. Conditional on a failed test, adequate-performance probability is therefore about 22.22 per cent. A failure is strong adverse evidence, but it is not perfect proof. The next action might be diagnosis, repair and retesting, depending on costs, urgency and the nature of the test.
To connect this evidence to a credit calculation, add another explicit teaching assumption. Suppose the borrower’s one-year default probability is two per cent when the protection system is adequate and fifteen per cent when it is inadequate, with all other conditions fixed. Before the test, the weighted default probability is 0.80 × 0.02 plus 0.20 × 0.15, or 4.6 per cent. After a pass, the same calculation using the updated probabilities gives about 2.95 per cent. After a failure, it gives about 12.11 per cent.
These are not calibrated lending probabilities. They demonstrate how evidence could change a financial estimate when the causal and statistical assumptions are explicit. The test does not directly measure default. It provides information about protection performance, which is then connected to default through another assumption. Confusing those two steps would make the model harder to challenge.
The example also exposes the importance of test design. If inadequate systems pass almost as often as adequate ones, a passing result carries little information. A certificate is not useful merely because it looks formal. Its value depends on what was tested, how the test relates to the relevant event and how often it can be wrong. A routine inspection may answer a different question from a full performance test.
Evidence from one operating condition may not transfer to another. A system tested under moderate conditions might not reveal performance in an extreme event. A production line tested at low utilisation might not reveal energy use at full output. A successful trial with expert staff might not establish performance under normal staffing. The analyst should identify the conditions under which the evidence was produced.
Selection also matters. If only the strongest projects volunteer for testing, the observed pass rate cannot be applied automatically to all projects. If failed tests are not recorded, the data will look more favourable than reality. If the testing method changes, old and new results may not be directly comparable. These are not reasons to abandon evidence. They are reasons to maintain an honest record of how it was generated.
The final step returns to the decision. A passing result might justify releasing a later financing tranche, reducing uncertainty in a model or maintaining a planned timeline. A failure might justify remedial work, a revised cash forecast or a different project design. Those are possible responses, not automatic rules prescribed by this example. The authorised decision-maker must consider the financial state and the actual evidence.
Alicia asks whether the test represents the physical condition that matters. Tricia performs the probability update. Kai Kai asks what action changes and when the next observation will arrive. This is the closed loop in miniature: belief, observation, update, action and another observation. Confidence becomes stronger because it is revisable, not because the system refuses to admit uncertainty.
Applied extension: cost pass-through changes the transition calculation
A higher input cost does not always reduce a company’s cash flow by the entire amount of the increase. A business may raise selling prices, reduce production, change its process or accept a lower margin. But each response has consequences. Passing costs to customers can reduce demand. Keeping prices unchanged can protect volume while destroying cash generation. Financing an efficiency improvement can preserve operating margins while creating new debt service. This final worked extension connects those responses rather than assuming that customer behaviour stays fixed.
All quantities below are invented teaching assumptions, independent of Harbour Store and Meridian Components. The business sells 100,000 units a year at S$100 each. Variable cash cost is S$70 per unit and fixed annual cash overhead is S$2 million. Its simplified operating cash before debt service is therefore (100 − 70) × 100,000 − 2,000,000, or S$1 million. Existing annual debt service is S$400,000, producing coverage of 2.50 times. We omit taxes, working-capital changes and other investments to isolate the price-volume mechanism; this measure is not a complete cash-flow statement.
A transition-related input change now adds S$10 to variable cost per unit. This is a hypothetical cost shock, not an actual tax rate or a claim about a named industry. If the company keeps selling at S$100 and volume remains 100,000 units, operating cash falls to zero: the S$20 contribution per unit exactly covers S$2 million of overhead. It cannot meet the S$400,000 debt service from that year’s operating cash. This result does not prove immediate insolvency; opening liquidity, available financing and asset value are outside this small model.
To examine price increases, add an explicit local demand assumption. For the small price changes tested here, assume that each one per cent increase in the company’s own selling price reduces quantity demanded by one and a half per cent, with competitor prices and all other demand conditions held fixed. This is not an empirical estimate. It is a deliberately simple response rule that lets us see why recovering a cost through price does not necessarily preserve the original cash flow. The rule should not be extrapolated indefinitely: a local approximation can produce implausible quantities far from its starting point.
With partial pass-through, the company raises price to S$105. The five per cent price increase reduces assumed demand by 7.5 per cent, leaving 92,500 units. Contribution per unit is S$25 because the new variable cost is S$80. Total contribution is S$2,312,500, and subtracting overhead leaves S$312,500 of operating cash. Coverage is 312,500 divided by 400,000, or about 0.78 times. Pricing has repaired some of the problem but has not restored debt-service capacity under these assumptions.
With full pass-through of the S$10 unit-cost increase, price becomes S$110. Assumed demand falls fifteen per cent to 85,000 units. The original S$30 contribution per unit is restored, but it is earned on fewer units. Operating cash is therefore S$550,000 rather than the original S$1 million. Debt-service coverage is about 1.38 times. Full unit-cost pass-through does not mean full preservation of company earnings. The lost volume still has to be absorbed by a business with fixed overhead.
| Annual teaching case | Selling price | Units sold | Operating cash before debt service | Debt-service coverage |
|---|---|---|---|---|
| Original business | S$100 | 100,000 | S$1,000,000 | 2.50× |
| Shock, no pass-through | S$100 | 100,000 | S$0 | 0.00× |
| Shock, partial pass-through | S$105 | 92,500 | S$312,500 | 0.78× |
| Shock, full pass-through | S$110 | 85,000 | S$550,000 | 1.38× |
| Shock, successful process investment | S$100 | 100,000 | S$900,000 | 1.53×, including new debt |
The final row needs its own assumptions. Suppose the company can invest S$1 million in a process that avoids the S$10 variable-cost increase while preserving product quality and capacity. Annual maintenance adds S$100,000 to fixed cash overhead. The company finances the investment with S$250,000 of equity and S$750,000 of debt, amortised through five equal annual payments at eight per cent. Additional annual debt service is approximately S$187,842. Assume successful installation before the operating year begins; commissioning delays require a separate cash-path test.
At the original S$100 selling price, assumed demand remains 100,000 units and variable cost remains S$70. After the higher overhead of S$2.1 million, operating cash is S$900,000. Total annual debt service is approximately S$587,842. Coverage is therefore about 1.53 times. The operating improvement produces more cash than the price-only alternatives shown, but the investment also consumes equity and creates a five-year commitment. Comparing coverage alone does not establish the best investment decision.
Notice what the table does and does not rank. It compares a few specified responses, not every possible selling price or production plan. The full-pass-through case performs better than the partial case under this particular demand rule. That ordering might change if customers were more price-sensitive, if competitors raised prices too, or if contracts prevented immediate repricing. The process investment assumes that its technical performance and customer acceptance are achieved. Its apparent advantage could disappear after a cost overrun, production interruption or lower useful life.
A lender should therefore ask which demand relationship is being assumed. An individual company raising its own price while competitors hold theirs fixed faces a different experiment from an industry-wide cost change that moves many prices together. Historical sales during one kind of change may not identify customer behaviour during the other. Similarly, a contractual right to raise prices says what the business is allowed to do; it does not establish how many customers will continue buying afterward.
There is a balance-sheet consequence to the lost volume as well. If the company has already purchased materials for 100,000 units but sells only 85,000, unsold inventory may absorb cash. The simple table assumes production and purchases adjust immediately to the stated sales volume. Relax that assumption and the first year can be worse than the steady-state comparison. This is another reason to connect annual economics to a monthly financing schedule rather than treat a coverage ratio as a complete survival test.
The environmental interpretation also needs its own evidence. The process investment may reduce emissions or avoid a more expensive input, but that outcome must be measured under a defined boundary. The price-only response may reduce total production and therefore change emissions through lower activity rather than improved efficiency. Neither route should be described as the same physical transformation. The financial table identifies money consequences, not a complete account of environmental performance.
Alicia checks the quantity sold and the physical process. Tricia calculates contribution, overhead and debt service. Kai Kai challenges the assumption that customers, suppliers and competitors stand still. Their combined conclusion is conditional: transition risk reaches repayment through both costs and behaviour, and a useful financing plan must model the response as well as the original shock. That is the return path a single emissions score or a single assumed carbon price cannot provide.
Working glossary
Adaptation. A change intended to reduce vulnerability or improve the ability to cope with physical climate effects. In this guide, adaptation is evaluated through a stated performance assumption and a financial path. Spending money on adaptation does not by itself prove that the measure is effective or that its benefits exceed its costs.
Acute physical risk. Risk associated with a concentrated event in the chosen scenario. The financial model should identify direct damage, interruption and payment timing separately. A short event can have a long cash-flow effect if repair, replacement or customer recovery takes time.
Chronic physical risk. Risk associated with a persistent change in operating conditions rather than one isolated event. Its financial effect may accumulate through costs, productivity, maintenance or asset value. The relevant decision date can be a replacement or refinancing date rather than an emergency response.
Transition risk. The possibility that changes in policy, technology, demand or other transition-related conditions impair financial performance. The effect depends on the business model, contracts and available responses. It is not adequately measured by an emissions total alone.
Exposure. The asset, activity or financial claim affected by a risk. Physical exposure and financial exposure are different uses of the word. The location of a factory is a physical exposure; the bank’s outstanding loan is a financial exposure. A good model connects them explicitly.
Vulnerability. Susceptibility to damage or disruption given an event. Vulnerability depends on design, operation, dependencies and recovery capacity. It should not be inferred entirely from geography when important differences between sites or businesses are known.
Cash-flow gap. The amount by which available cash and receipts fall short of required payments during a defined period. A later recovery does not remove an earlier gap. The gap must be financed, reduced or reflected in a failure to make some payments as assumed.
Liquidity. The ability to meet obligations with usable resources when they fall due. Liquidity has a date, currency and legal-entity location. An asset can have substantial economic value while being unavailable for the immediate payment that matters.
Solvency. A value-based assessment of whether assets and loss-absorbing resources are sufficient relative to liabilities under the relevant framework. It is distinct from having enough cash today. A borrower or bank can face liquidity pressure before a final solvency conclusion is clear.
Probability of default. The probability of a defined default event over a stated horizon. A hazard probability is not the same thing. The model must connect the hazard to the borrower’s financial condition, and it must specify what counts as default.
Loss given default. The fraction or amount lost conditional on default under the chosen definition. It depends on recovery, collateral, timing and costs. Combining it with probability of default requires consistent conditioning, especially when the same shock increases both default and severity.
Exposure at default. The amount outstanding when default occurs. It can differ from the current balance if commitments are drawn or principal is repaid before default. Climate-related emergency funding can therefore affect both borrower liquidity and lender exposure.
Expected loss. The probability-weighted average loss over the modelled states. It is not the maximum loss and does not describe the whole tail. Two portfolios can have identical expected loss while requiring very different capacity to withstand a severe common event.
Dependence. The relationship between outcomes across borrowers, assets or time periods. Shared infrastructure, suppliers or financing can make losses occur together. Counting many borrower names does not establish that the underlying risks are independent.
Scenario. A specified set of conditions used to examine consequences. It is not necessarily a forecast or a probability estimate. Its usefulness depends on coherence, relevance, transparent assumptions and a suitable translation into the financial question being asked.
Model provenance. The record of where a model’s inputs and methods came from, which versions were used and how they were transformed. Provenance allows an institution to identify which conclusions need review when an upstream source is corrected or replaced.
Net present value. The discounted value of future incremental cash flows minus the initial investment under stated assumptions. NPV tests economic value; it does not guarantee that the borrower has sufficient cash at every intermediate date.
Debt-service coverage. A comparison of defined cash available for debt service with the payments due. The definition of available cash matters. A ratio based on accounting earnings should not be described as a cash test unless the necessary adjustments are made.
Commissioning risk. The possibility that a new asset is delayed, performs below specification or takes longer than expected to reach normal operation. Commissioning can create a period with new debt service but incomplete operating benefits, as the Meridian case illustrates.
Use-of-proceeds financing. Financing whose framework identifies how relevant proceeds will be applied to specified purposes or eligible projects. Understanding that purpose does not remove the need to assess the borrower, legal claim, implementation and repayment path.
Sustainability-linked terms. Terms connected to performance against defined sustainability measures or targets. Their effect depends on the metric, baseline, timing, verification and consequence. The label alone does not reveal whether the financial incentive is material.
Financed emissions. Emissions attributed to financing relationships under a defined accounting method. They are not credit losses and should not be used as though tonnes were a direct substitute for default probability, collateral recovery or liquidity.
Emissions intensity. Emissions per unit of an activity such as production or revenue under a stated boundary. Intensity can improve while absolute emissions rise. The numerator, denominator and change in activity should be explained together.
Counterfactual. The alternative path used to evaluate a decision. An investment’s incremental benefit depends on what would happen without it. A weak or inconsistent counterfactual can make the preferred action appear better than it is.
Double counting. Including the same loss or benefit more than once through overlapping measures. A reconciled bridge between operating events, cash flows and valuation helps prevent this error. Adding more variables is not automatically a more complete model.
Robustness. The degree to which an action or conclusion remains useful across different plausible assumptions. A robust decision need not maximise return in every scenario. It should have a clearly understood trade-off between upside, downside, flexibility and feasibility.
Regret. The difference between an action’s outcome and the best feasible outcome in the same scenario with hindsight. Minimising maximum regret is one decision criterion. It is not equivalent to minimising worst-case total cost or maximising expected return.
Closed-loop monitoring. A process in which new evidence updates assumptions and changes decisions when appropriate. Data collection alone is not enough. The system needs responsibility, authority, response timing and a later check on whether the action worked.
Sources, dates and scope of authority
This guide was reviewed on 20 September 2026. Its numerical cases are original teaching examples, not extracted bank data or forecasts. The source links support the institutional frameworks and dated developments; they do not endorse the fictional scenarios or calculations.
The Basel Committee’s transmission-channel report supplies the foundational financial-risk framing. Its principles for managing and supervising climate-related financial risks are a reference for the broader supervisory context. The IPCC AR6 Synthesis Report supplies scientific context, not borrower-specific default probabilities.
The NGFS short-term scenario release, December 2025 Phase V statement and scenario data portal should be read together when identifying versions and limitations. This article does not use an affected long-term physical-damage estimate to claim a numerical future loss.
For Singapore, consult the MAS transition-planning release and the reproduced MAS-authored release. For reporting, consult the IFRS S2 overview and related implementation guidance. Publication, effective date, sector coverage and local adoption should not be conflated.
For financing frameworks, consult ICMA’s Green Bond Principles, Sustainability-Linked Bond Principles and climate-transition guidance. For emissions accounting, consult PCAF’s standard. These resources have different purposes and should not be treated as interchangeable credit, environmental or legal certifications.
The loop that matters
Climate risk becomes a financial problem through a specific path: something changes in the physical or economic environment, an activity is affected, cash flow or asset value changes, and a financial obligation must still be met. Finance can then help fund a response, but that response consumes resources and can introduce new risks. The system becomes more resilient only when the response works in the real world and the evidence returns to the next decision.
Harbour Store taught us that a later insurance recovery does not settle an earlier bill. Meridian taught us that an attractive investment can still fail during commissioning. The portfolio exercise taught us that an average loss can hide a concentrated tail. The attribution exercise taught us that a lower reported number does not always mean a physical improvement. The evidence laboratory taught us to update confidence without pretending uncertainty has vanished.
These are not isolated warnings. They are parts of one method: define the object, preserve the timeline, keep the units consistent, identify the actor, test the alternative and ask what observation would change the conclusion. The method is useful far beyond climate finance because every financial system ultimately depends on real activities and decisions that occur through time.
For the broader human question of how communities share uncertain losses, readers can continue with eduKate’s How Societies Share Risk. For the financial mechanisms, return to the Banking And Finance Closed Loop Systems master guide.
A complete climate-finance loop ends neither with a slogan nor with a spreadsheet. It ends with a funded action, an observed result, an honest explanation of what changed, and a better-informed next decision.