Chapter 6: Insurance as a Risk-Buffering Instrument: Weather Index and Forest Insurance

In the spring of 2019, I stood in a greenhouse in Wuliming, a township in Hebei Province, watching a farmer named Zhang Weiwei run his thumb along a cracked irrigation pipe. The cucumber seedlings behind him were three weeks behind schedule. A late frost in early April had killed the first planting; the replanting had cost him 18,000 yuan he had not budgeted for. He had a loan from the local rural credit cooperative, 120,000 yuan at 6.8 percent, and the repayment date was fixed. The frost did not care about the repayment date.

Zhang had no insurance. He had heard of agricultural insurance—the village committee had mentioned it once—but the premium seemed high, the payout process seemed complicated, and he had never met anyone who had actually collected. So the frost became a credit event. He missed one installment in June, caught up in September by borrowing from his brother-in-law, and spent the winter operating on a thinner margin than the year before. The loan was paid. The stress was not.

The stakes are concrete. A weather shock that hits a farmer like Zhang is, from the lender's perspective, a default risk that arrives without warning and without any change in the borrower's effort, competence, or willingness to repay. It is a risk that credit markets have historically handled badly, and the reason is not mysterious: the events that matter most to agricultural borrowers are the events that are hardest to verify at the level of the individual farm. Did the frost hit Zhang's plot, or did it hit the next valley and miss his? Did the drought reduce his yield, or did he manage the irrigation poorly? The information problem that Chapter 2 described in the context of credit rationing returns here in a different form, and it is the reason weather index insurance exists.

This chapter examines the two insurance instruments that matter most for the green transition in agriculture and forestry: weather index insurance, which pays out on a measured weather variable rather than on verified loss, and forest insurance, which covers fire, pest, and weather risks to a resource whose value is increasingly bound up with carbon sequestration. Then it examines how these instruments interact with credit—how a payout stream that is uncorrelated with the borrower's own effort can change the lender's assessment of default risk. The argument is not that insurance solves the problem of agricultural credit. It is that insurance changes the shape of the risk that credit carries, and that this change is large enough to matter for the financing of eco-agriculture and forestry.

6.1 Weather Index Insurance Design: Basis Risk, Trigger Levels, and Payout Structures

6.1.1 The Information Problem That Index Insurance Solves

Traditional crop insurance indemnifies verified loss. The insurer sends an adjuster to the field, measures the shortfall against an expected yield, and pays. This works in countries with dense networks of trained adjusters and reliable yield histories. It works badly almost everywhere else. The adjuster cannot be everywhere at harvest; the farmer has an incentive to under-report yield or over-report damage; and the cost of verification consumes a large share of the premium. In the language of Chapter 2, traditional crop insurance suffers from both moral hazard—the insured can affect the loss—and adverse selection—the highest-risk farmers are the most eager to buy.

Index insurance addresses both problems by severing the link between payout and verified individual loss. The contract specifies a weather variable—rainfall, temperature, wind speed, a vegetation index derived from satellite imagery—measured at a reference station or grid cell. If the variable crosses a threshold, the insurer pays a predetermined amount. The farmer's actual yield does not enter the calculation. Neither does the farmer's effort. Moral hazard is largely eliminated because the farmer cannot influence the reading at the reference station. Adverse selection is reduced because the contract is the same for everyone in the reference area.

The cost of this elegance is basis risk: the possibility that the index triggers when the farmer has suffered no loss, or fails to trigger when the farmer has suffered a large loss. Basis risk is the central design problem of index insurance, and it deserves to be treated as a first-class object rather than a footnote.

Consider the structure formally. Let \(Y_i\) be the yield of farm \(i\) and let \(I\) be the index, measured at a reference station. A simple linear index contract pays

where \(K\) is the trigger level of the index and \(\lambda\) converts index units into currency. The farmer's revenue net of the insurance payoff is \(p\,Y_i + P(I) - \pi\), where \(p\) is the output price and \(\pi\) is the premium. The farmer's residual risk is the variance of \(p\,Y_i + P(I)\), not the variance of \(Y_i\) alone. If \(Y_i\) and \(I\) are perfectly correlated within the reference area, the residual risk can be driven close to zero. If they are not, the farmer retains a component of risk that the contract cannot reach.

The basis risk of the contract for farm \(i\) can be written as the variance of the difference between the farmer's loss and the index payout:

where \(L_i = p\,(\bar Y_i - Y_i)\) is the realized loss relative to the expected yield \(\bar Y_i\). A contract that is well designed for the average farm in the reference area can have large \(\mathrm{BR}_i\) for farms at the edge of the area, on different soil, or at different elevation. This is not a technicality. It is the reason index insurance has struggled to achieve high uptake in many markets: farmers who have experienced a payout that did not match their loss are reluctant to renew.

The design problem, then, is to choose the reference variable, the trigger \(K\), the payout slope \(\lambda\), and the spatial unit so that \(\mathrm{BR}_i\) is small relative to the uninsured risk, while keeping the contract simple enough to be sold and cheap enough to be subsidized. These objectives trade off against each other. A finer spatial grid reduces basis risk but raises the cost of the weather data infrastructure. A lower trigger reduces the risk of no payout in a bad year but raises the premium. A steeper payout slope increases protection but increases the insurer's capital charge.

6.1.2 Trigger Levels and the Architecture of a Payout Schedule

The trigger level is the point at which the contract begins to pay. In a rainfall contract for a maize-growing region, the trigger might be 300 mm of cumulative rainfall over the growing season, with the payout increasing linearly as rainfall falls below 300 mm and reaching the full sum insured at 150 mm. The choice of trigger is an exercise in balancing false negatives against false positives. A trigger set too high—say 400 mm—pays out in years that are only mildly dry, which raises the premium and dilutes the protection value. A trigger set too low—say 200 mm—leaves the farmer exposed in moderately dry years that still cause substantial loss.

Most contracts use a piecewise-linear payout schedule with three regions: no payout above the trigger, linear payout between the trigger and the exit, and full payout below the exit. Let \(K\) be the trigger and \(K_{\mathrm{exit}}\) the exit, with \(K_{\mathrm{exit}} < K\) for a deficit index. The payout is

where \(S\) is the sum insured. This schedule is transparent, easy to explain to farmers, and easy to price once the distribution of \(I\) is known. Its weakness is that it concentrates the protection in the tail: a farmer who suffers a 20 percent yield loss in a year when the index barely crosses the trigger receives little or nothing.

An alternative is a payout schedule indexed to a vegetation condition index (VCI) derived from satellite data, which captures the joint effect of rainfall, temperature, and management over the season. The VCI is defined as

where NDVI is the normalized difference vegetation index and the min and max are computed over a historical reference period for the same pixel and the same week of the year. The VCI has the advantage of being measured at the pixel level, which reduces the spatial basis risk that arises when a farmer's plot is far from the reference rain gauge. It has the disadvantage of being affected by pests, disease, and management practices that have nothing to do with weather, which introduces a different kind of basis risk—the index can fall because of a pest outbreak, and the insurer will pay for what looks like a drought.

The choice between rainfall, temperature, and vegetation indices is ultimately a choice about which basis risk the contract can tolerate. For a region where the dominant risk is drought, a rainfall index with a dense gauge network is often the best compromise. For a region where the dominant risk is frost, a temperature index at a reference station is more appropriate. For a region where the risk is a compound of heat and moisture stress, a vegetation index may capture more of the variation in yield, at the cost of a noisier relationship to any single weather event.

6.1.3 Payout Structures and the Timing Problem

The timing of the payout matters as much as its size. A farmer who receives a payout in December, after the harvest has failed and the loan installment has come due in June, has received a payment that arrives too late to prevent a default. The design of the payout schedule must therefore consider the cash-flow calendar of the borrower, not only the agronomic calendar of the crop.

There are three broad approaches. The first is a post-harvest indemnity, in which the payout is made after the index is finalized at the end of the season. This is the simplest structure and the easiest to administer, but it does nothing for liquidity constraints during the season. The second is a staged payout, in which a portion of the sum insured is paid at a mid-season trigger and the remainder at harvest. This structure is more expensive to administer but aligns the payout with the moments when the farmer needs cash—for replanting, for fertilizer, for the loan installment. The third is a pre-season or early-season payout tied to a forecast index, which is the most ambitious and the most difficult to price because it depends on the skill of seasonal climate forecasts.

The timing problem connects directly to the credit interaction that Section 6.3 will examine. A lender who knows that the insurance payout arrives in December, after the loan installment is due in June, will not treat the insurance as a full mitigation of default risk. The lender will discount the insurance by the probability that the borrower cannot bridge the gap between the due date and the payout date. This is a design failure that can be corrected by structuring the payout around the loan schedule, and it is one of the reasons the integration of insurance and credit requires deliberate coordination rather than the mere coexistence of the two instruments.

6.1.4 Pricing and the Cost of Basis Risk

The premium for an index contract is determined by the expected payout plus a loading for administrative costs, capital, and risk. If the index \(I\) has a cumulative distribution function \(F_I\), the expected payout is

For a deficit index with a lower tail, this is the area under the payout schedule weighted by the density of the index in the payout region. The actuarially fair premium is \(\mathbb{E}[P]\), and the commercial premium is \((1 + \ell)\,\mathbb{E}[P]\) where \(\ell\) is the loading factor.

The loading factor for index insurance is typically higher than for conventional insurance because the insurer must hold capital against the possibility that the index triggers across a wide area simultaneously—a drought that hits one farm is a local event, but a drought that hits a province is a correlated event that can exhaust the insurer's reserves. This is the same concentration problem that Chapter 5 examined at the portfolio level, transposed to the insurer's book. The insurer's capital charge is driven by the tail of the aggregate payout distribution, not by the expected payout, and the tail is fat when the index is spatially correlated.

The cost of basis risk enters the farmer's decision differently. A farmer who is risk-averse will compare the certainty equivalent of the insured position to the certainty equivalent of the uninsured position. If the basis risk is large, the insurance reduces the variance of income by less than the farmer expects, and the willingness to pay falls below the commercial premium. This is the demand-side counterpart of the supply-side capital charge, and it explains why index insurance markets have often required heavy subsidies to reach scale. The subsidy is not a transfer to farmers so much as a payment for the public good of risk reduction—a point that Chapter 11 will take up in the context of performance evaluation.

6.1.5 A Numerical Illustration

It is worth working through a simple example to make the design trade-offs concrete. Suppose a region has a historical rainfall distribution over the growing season that is approximately normal with mean 450 mm and standard deviation 80 mm. A contract with a trigger at 350 mm and an exit at 200 mm, sum insured 30,000 yuan per hectare, has an expected payout of

With \(I \sim N(450, 80^2)\), the standardized trigger is \(z_K = (350 - 450)/80 = -1.25\) and the standardized exit is \(z_{\mathrm{exit}} = (200 - 450)/80 = -3.125\). The probability of a payout is \(\Phi(-1.25) \approx 0.106\), and the probability of a full payout is \(\Phi(-3.125) \approx 0.0009\). The expected payout works out to roughly 1,050 yuan per hectare, or about 3.5 percent of the sum insured. A commercial premium with a 40 percent loading would be about 1,470 yuan per hectare.

Now consider a farm at the edge of the region where the correlation between the farm's yield and the regional rainfall index is 0.6 rather than 0.9. The residual variance of the farmer's revenue after insurance is larger, and the certainty equivalent gain from the contract is smaller. If the farmer's coefficient of relative risk aversion is 2 and the uninsured income standard deviation is 8,000 yuan per hectare, the value of the insurance to the farmer may fall below the commercial premium even though the actuarial value is positive. This is the basis-risk trap: the contract is fairly priced on average but not fairly priced for the marginal farmer, and the marginal farmer is the one whose participation determines whether the market reaches scale.

The design response is to reduce basis risk through finer spatial resolution, better index construction, or hybrid contracts that combine an index trigger with a limited loss verification. Each of these responses raises cost. There is no free lunch in index design, only a choice about which cost to bear.

flowchart TD
    A["Weather shock hits region"] --> B["Reference station records index I"]
    B --> C{"I below trigger K?"}
    C -->|"No"| D["No payout; farmer bears full loss"]
    C -->|"Yes"| E{"I below exit K_exit?"}
    E -->|"No"| F["Partial payout: S x (K - I)/(K - K_exit)"]
    E -->|"Yes"| G["Full payout S"]
    F --> H["Payout timing vs loan schedule"]
    G --> H
    H --> I{"Payout before installment?"}
    I -->|"Yes"| J["Default risk reduced"]
    I -->|"No"| K["Liquidity gap; default risk persists"]
    D --> L["Basis risk realized if loss occurred"]

Figure 6-1: Payout logic of a weather index contract and its interaction with the loan calendar.

6.2 Forest Insurance for Carbon Sequestration: Covering Fire, Pest, and Weather Risks

6.2.1 Why Forests Are Different

A forest is not a field. A field is replanted every season, and a failed season is a loss of one year's output. A forest is planted once and harvested decades later, and a failed forest is a loss of the entire accumulated carbon stock, the land's productive capacity for a generation, and the collateral that underpinned the loan that financed the planting. The time horizon changes the insurance problem in three ways.

First, the risk is cumulative. A forest that survives a fire in year five but suffers a pest outbreak in year twelve has lost value twice, and the insurance must cover the joint distribution of these events over a long horizon. Second, the value at risk is not just the timber. It is the carbon sequestration service, which is increasingly the basis for revenue through carbon markets and results-based payment schemes. A forest that burns releases the carbon it had stored, which means the insurer's liability includes the reversal of a carbon credit that the forest owner may have already sold. Third, the asset is alive and its condition changes continuously, which makes loss verification more expensive and more contested than for a crop.

These differences explain why forest insurance has developed along a different path from crop insurance. In China, forest insurance is dominated by a policy-supported model in which the government subsidizes a large share of the premium and the insurer covers fire, pest, and meteorological risks under a standard policy. The central subsidy covers 50 percent of the premium for public welfare forests, with provincial and county governments contributing additional shares; the insured amount is set at the cost of reforestation rather than at the market value of the timber, which keeps the premium affordable but leaves the forest owner underinsured relative to the full value of the standing stock.

The underinsurance is not an accident. It reflects a deliberate choice about what the insurance is for. If the purpose is to protect the forest owner's income, the sum insured should reflect the timber value. If the purpose is to protect the public good of forest cover, the sum insured should reflect the cost of restoring the cover. The policy-supported model in China has chosen the second purpose, which is consistent with the public-good logic of the green transition but creates a gap between the insurance and the credit that the forest owner needs.

6.2.2 Fire Risk and the Structure of Forest Insurance

Fire is the dominant insured peril in most forest insurance schemes. The structure is typically a named-peril policy that covers fire, lightning, and sometimes windstorm and pest, with a deductible and a coinsurance clause. The deductible is usually expressed as a percentage of the sum insured, and the coinsurance clause requires the forest owner to bear a share of the loss above the deductible, which preserves the incentive to manage the forest.

The actuarial challenge is that fire risk is spatially correlated and heavy-tailed. A single fire can destroy thousands of hectares, and the insurer's exposure is concentrated in the regions where the forest is most valuable and most flammable. The standard response is to cap the sum insured per policy and to reinsure the tail. In China, the provincial forest insurance schemes are reinsured through the China Reinsurance Group, which spreads the catastrophe risk across the national balance sheet. This is a sensible arrangement, but it means that the price of forest insurance is ultimately determined by the cost of catastrophe reinsurance, which is volatile and rising as climate change increases fire risk in many regions.

The fire risk itself is changing. The 2023 Canadian wildfire season burned more than 18 million hectares, an area larger than Greece, and sent smoke across the Atlantic. The 2019–2020 Australian bushfires burned 17 million hectares. These events are not directly comparable to the Chinese forest estate, which is smaller and more fragmented, but they illustrate the tail that forest insurers are exposed to. A forest insurance scheme that is priced on historical fire frequencies will underprice the forward risk, and a scheme that is priced on forward risk will be expensive. The gap between the two is a public policy problem, not a technical one, and it is the reason many governments subsidize forest insurance.

6.2.3 Pest Risk and the Problem of Slow Onset

Pest risk is harder to insure than fire because the loss is slow and the boundary between a pest outbreak and poor management is contested. A pine forest attacked by the pine wood nematode (Bursaphelenchus xylophilus) can lose 30 percent of its stock over two years, but the loss is not visible in a single event and the forest owner may have contributed to it by delaying the removal of infected trees. The insurer faces a moral hazard problem that fire insurance does not: the forest owner can affect the loss by how quickly he responds, and the response is costly.

The standard solution is a parametric or index-based trigger. Instead of verifying the loss tree by tree, the policy pays when a pest population index crosses a threshold, or when a remote sensing index of forest health falls below a reference level. This is the same logic as weather index insurance, applied to a biological variable. The advantage is that it removes the moral hazard and reduces verification cost. The disadvantage is that it introduces basis risk: a forest owner whose trees are attacked by a pest that the index does not capture receives nothing.

A more ambitious approach is to insure the carbon stock directly, using a combination of remote sensing and periodic field sampling to estimate the biomass, and to pay out when the estimated biomass falls below a reference trajectory for reasons other than harvest. This approach is conceptually attractive because it aligns the insurance with the carbon value of the forest, which is the value that matters for the green transition. It is also expensive, because the measurement, reporting, and verification (MRV) system required to support it is costly and the insurer must bear the cost of the MRV in the premium.

6.2.4 Weather Risk and the Compound Event

Forests are exposed to compound weather events: drought followed by fire, windstorm followed by pest outbreak, frost followed by disease. The insurance of compound events is difficult because the events are correlated and the loss is not additive. A forest that is weakened by drought and then burned has suffered a loss that is larger than the sum of the drought loss and the fire loss, because the drought made the fire more severe. Standard insurance policies that cover each peril separately will underinsure the compound event, and the forest owner will discover the gap only after the loss.

The design response is a multi-peril policy with a single deductible and a single limit, which covers the compound event as a single loss. This is the structure used in several European forest insurance schemes, and it is more expensive than single-peril insurance because the insurer must hold capital against the joint distribution. The alternative is a parametric policy that triggers on a compound index—for example, a drought index combined with a fire-weather index—which pays when the joint condition is met. This is cheaper to administer but harder to calibrate, because the historical record of compound events is short.

6.2.5 The Carbon Dimension

The carbon dimension changes the insurance problem in a way that is not yet fully reflected in commercial products. A forest that is enrolled in a carbon market has sold a credit for the carbon it is expected to sequester over a defined period. If the forest burns, the credit is reversed, and the forest owner may be liable to repay the credit or to purchase replacement credits. This is a liability that is not covered by a standard forest insurance policy, which covers the timber loss but not the carbon reversal.

The gap is beginning to be addressed by specialized carbon insurance products. In 2023, several insurers and carbon market participants began offering policies that cover the reversal risk for nature-based carbon projects, with payouts triggered by remote sensing of forest cover loss. The products are new, the actuarial basis is thin, and the premiums are high relative to the value of the credit. But the direction is clear: as carbon markets mature, the insurance of carbon reversal will become a standard part of the forest owner's risk management, and the insurers who develop the MRV capability early will have an advantage.

The connection to credit is direct. A forest owner who has sold a carbon credit and has an uninsured reversal liability is a riskier borrower than one who has insured the liability, because the reversal is a contingent claim that can materialize at exactly the moment when the borrower's cash flow is weakest. A lender who is financing a reforestation project should therefore require carbon reversal insurance as a condition of the loan, in the same way that a mortgage lender requires property insurance. This is a concrete example of the integration that this book argues for: the insurance is not a separate product but a component of the credit risk mitigation.

6.2.6 A Note on the Chinese Forest Insurance Experience

The Chinese forest insurance scheme is the largest in the world by area covered. By the end of 2022, more than 130 million hectares of forest were insured, with premiums of roughly 4 billion yuan per year and a government subsidy covering the majority of the premium. The scheme has been successful in extending coverage to collective forests, which were previously uninsured, and in providing a payout stream after major fires and pest outbreaks.

The scheme's weakness is the one identified above: the sum insured is set at the reforestation cost, which is typically 6,000 to 15,000 yuan per hectare, while the value of the standing timber and the carbon stock is several times higher. A forest owner who loses a mature stand receives enough to replant but not enough to compensate for the lost income or to repay the loan that financed the stand. This is a design choice, not a failure, but it limits the usefulness of the insurance as a credit enhancement. The next section examines how the credit interaction can be strengthened even when the insurance is partial.

6.3 Risk Buffering and Credit Access: How Insurance Enhances Creditworthiness

6.3.1 The Mechanism

The proposition that insurance enhances credit access is intuitive but needs to be stated precisely. Insurance does not increase the borrower's expected income. It reduces the variance of income, and in doing so it reduces the probability that income falls below the threshold at which the borrower defaults. If the lender's decision is based on the probability of default, and if the borrower is risk-averse, the insurance can increase the borrower's willingness to borrow and the lender's willingness to lend at the same time.

Consider a borrower with stochastic income \(\tilde Y\) and a fixed loan obligation \(D\). Default occurs if \(\tilde Y < D\). Without insurance, the default probability is \(\Pr(\tilde Y < D) = F_Y(D)\). With an index insurance contract that pays \(P(I)\), the default probability becomes \(\Pr(\tilde Y + P(I) < D)\). If the insurance payout is positively correlated with the shortfall—that is, if it pays when \(\tilde Y\) is low—then the default probability falls.

The size of the reduction depends on the joint distribution of \(\tilde Y\) and \(P(I)\), which is to say on the basis risk. If the insurance is a perfect hedge, the default probability can be reduced to zero for a sufficiently large sum insured. If the insurance is uncorrelated with the borrower's income, it does nothing for the default probability except through the premium, which reduces income. This is the formal statement of the point made in Section 6.1: basis risk is not an academic curiosity, it is the parameter that determines whether insurance is useful for credit.

A simple model makes the mechanism transparent. Suppose \(\tilde Y = \mu + \sigma \varepsilon\) where \(\varepsilon\) is a standard normal shock, and the insurance payout is \(P = \alpha \sigma \varepsilon + \eta\) where \(\alpha \in [0,1]\) is the hedge ratio and \(\eta\) is independent noise with variance \(\sigma_\eta^2\). The hedged income is \(\tilde Y + P = \mu + \sigma(1 + \alpha)\varepsilon + \eta\), which has variance \(\sigma^2(1+\alpha)^2 + \sigma_\eta^2\). If the insurance is a perfect hedge (\(\alpha = 1\), \(\sigma_\eta = 0\)), the variance is \(4\sigma^2\)—wait, that is wrong. The hedged income should have lower variance, not higher. The correct specification is \(P = -\alpha \sigma \varepsilon + \eta\), so that the payout is high when the shock is negative. Then the hedged income is \(\mu + \sigma(1 - \alpha)\varepsilon + \eta\), with variance \(\sigma^2(1-\alpha)^2 + \sigma_\eta^2\). A perfect hedge (\(\alpha = 1\), \(\sigma_\eta = 0\)) gives zero variance. A partial hedge (\(\alpha = 0.5\)) gives variance \(0.25\sigma^2\). The default probability is \(\Phi\bigl((D - \mu)/\sqrt{\sigma^2(1-\alpha)^2 + \sigma_\eta^2}\bigr)\), which falls as \(\alpha\) rises and \(\sigma_\eta\) falls.

This is the transmission mechanism from insurance to credit. It is not a metaphor; it is a variance reduction that changes the argument of the default probability function. The practical question is how large the effect is, and whether it is large enough to change the lender's decision.

6.3.2 Evidence from the Field

The empirical literature on index insurance and credit access is thinner than the policy interest would suggest, but the findings are consistent enough to support a cautious conclusion. In a randomized evaluation in northern Ghana, farmers who were offered rainfall index insurance were more likely to take up agricultural loans and borrowed larger amounts than the control group, with the effect concentrated among farmers who were initially credit-constrained. In a study of cotton farmers in Burkina Faso, insurance uptake was associated with a 20 to 30 percent increase in the area planted, which is consistent with a relaxation of the credit constraint rather than a pure risk-management effect.

The evidence is not uniformly positive. Several studies have found that index insurance uptake is low even when the product is subsidized, and that the farmers who do buy are not always the ones who would benefit most. The low uptake is partly a trust problem—farmers who have been offered insurance by a series of failed schemes are skeptical—and partly a design problem: the products on offer often have basis risk that is large relative to the premium, so the expected utility gain is small.

The credit channel can help overcome the trust problem. When insurance is bundled with a loan, the farmer does not have to decide separately whether to trust the insurer; the lender, which has a relationship with the farmer, intermediates. The bundling also solves a coordination problem: the lender can require insurance as a condition of the loan, which creates a captive demand that allows the insurer to reach scale and reduce the premium. This is the logic behind the "insurance-linked credit" products that have been piloted in several countries, and it is the logic that Section 6.3.3 examines in detail.

6.3.3 The Design of Insurance-Linked Credit

An insurance-linked credit product has three components: a loan, an insurance policy that covers the risk that the loan is exposed to, and a mechanism that links the payout to the loan repayment. The linkage can take several forms.

The simplest form is a requirement: the borrower must purchase the insurance as a condition of the loan, and the payout is paid to the borrower, who is then expected to use it to repay. This form relies on the borrower's willingness to repay and does not guarantee that the payout reaches the lender. It is common in mortgage lending, where property insurance is required but the payout goes to the borrower.

A stronger form is an assignment: the borrower assigns the insurance payout to the lender up to the amount of the outstanding loan, and the lender is the loss payee. This form is used in project finance and in some agricultural lending schemes. It guarantees that the payout reaches the lender first, which reduces the lender's loss given default, but it does not reduce the probability of default unless the payout arrives before the default.

The strongest form is an integrated product in which the insurance payout is triggered by the same event that triggers the loan's covenant and is paid directly into a reserve account that covers the next installment. This form requires close coordination between the lender and the insurer, and it is the form that is most likely to change the lender's credit decision. It is also the form that is hardest to implement, because it requires the lender and the insurer to share data and to agree on the trigger and the timing.

A concrete example illustrates the difference. A rural credit cooperative in Yunnan lends 200,000 yuan to a tea farmer for five years, with annual installments of 40,000 yuan plus interest. The farmer buys a frost index insurance policy with a sum insured of 100,000 yuan, triggered by a minimum temperature at the county weather station. Under the simple form, the farmer receives the payout and decides whether to repay. Under the assignment form, the lender receives the payout up to the outstanding balance. Under the integrated form, the payout is paid into a reserve account at the lender and is used to cover the installment that is due in the same quarter as the frost event. The integrated form is the one that allows the lender to reduce the interest rate, because it reduces the lender's expected loss given default and the probability of a liquidity-driven default.

6.3.4 The Limits of the Insurance-Credit Link

It is worth being clear about what the insurance-credit link does not do. It does not eliminate the borrower's risk. The basis risk remains with the borrower, and if the index does not trigger when the borrower suffers a loss, the borrower defaults anyway. It does not eliminate the lender's risk. The insurance payout is capped at the sum insured, and a large loss can exceed the cap. It does not solve the information problem that Chapter 2 described. The lender still needs to assess the borrower's creditworthiness, and the insurance is an input into that assessment, not a substitute for it.

There is also a moral hazard on the lender's side. A lender who knows that the insurance will cover the loss has less incentive to monitor the borrower and to intervene early when the borrower is in difficulty. This is the same problem that arises in credit insurance for trade receivables, where the insurer's coverage can weaken the policyholder's incentive to manage the credit book. The solution is a deductible or a coinsurance clause that keeps the lender's skin in the game—a point that the trade credit insurance literature has emphasized and that the agricultural insurance literature has largely ignored.

The empirical question is whether the net effect of these offsetting forces is positive. The evidence from the piloted schemes suggests that it is, but the schemes are small and the results may not generalize. The honest conclusion is that insurance-linked credit is a promising mechanism whose value depends on the design details, and that the design details are not yet well understood.

6.3.5 The Portfolio Effect

The credit interaction operates at two levels. At the individual level, insurance reduces the probability that a given borrower defaults. At the portfolio level, insurance reduces the correlation of defaults across borrowers, because the insurance payouts are triggered by a common weather event and are therefore negatively correlated with the borrowers' income at the moment when the correlation of defaults would otherwise be highest.

This portfolio effect is the same one that Chapter 5 examined in the context of copulas. The default correlation between two farmers in the same region is high in bad years and low in good years. Insurance does not eliminate the correlation—the farmers are still exposed to the same weather—but it changes the shape of the loss distribution. The insurer absorbs the correlated component of the loss, and the lender's portfolio becomes less concentrated in the tail. The lender's capital charge falls, which allows the lender to lend more at the same capital.

The effect can be quantified. Suppose a lender has a portfolio of \(N\) loans, each with default probability \(p\) and pairwise default correlation \(\rho\). The variance of the portfolio default rate is approximately \(p(1-p)/N + \rho\, p(1-p)\). The first term is the idiosyncratic risk, which diversifies away as \(N\) grows; the second term is the systematic risk, which does not. Insurance reduces \(\rho\) by transferring the common weather shock to the insurer, and it reduces \(p\) by reducing the probability that an individual borrower defaults. Both effects reduce the variance of the portfolio default rate, and the reduction in the systematic component is the one that matters for the lender's capital.

This is the portfolio-level argument for insurance-linked credit, and it is stronger than the individual-level argument because it does not depend on the borrower's behavior. Even if the borrower's behavior is unchanged, the lender's portfolio is safer because the common shock has been transferred.

flowchart TD
    A["Weather shock"] --> B["Farmer income falls"]
    A --> C["Index triggers at station"]
    C --> D["Insurer pays claim"]
    B --> E["Farmer cannot repay loan"]
    D --> F["Payout assigned to lender"]
    F --> G["Loan installment covered"]
    E --> H["Default without insurance"]
    G --> I["Default avoided"]
    H --> J["Lender loss given default"]
    I --> K["Lender capital charge falls"]
    J --> K
    K --> L["More credit at same capital"]

Figure 6-2: Transmission from weather shock to credit supply through index insurance.

6.4 Basis Risk, Trust, and the Demand Side

The supply-side story—insurers design products, lenders accept them, capital charges fall—is only half the picture. The demand side determines whether the products are bought, and the demand side is shaped by trust, by experience, and by the cognitive difficulty of understanding a contract that pays on a variable the farmer cannot observe.

6.4.1 The Trust Problem

Index insurance is a hard product to sell. The farmer pays a premium every year and receives nothing in most years. When the index triggers, the payout maynotmatch the farmer's loss, and the farmer concludes that the insurance does not work. The experience of a single bad year can destroy a relationship that took years to build.

The trust problem is compounded by the history of agricultural insurance in many developing countries. State-run crop insurance schemes have often failed, paying claims late or not at all, and farmers who remember those failures are reluctant to try a new product from a new provider. Selling index insurance means showing the farmer, over several seasons, that the payout arrives when it is supposed to. That is a slow argument, and it cannot be won with a brochure.

The credit channel can help, but it can also hurt. If the insurance is bundled with a loan and the payout is assigned to the lender, the farmer may perceive the insurance as a fee that benefits the lender rather than a protection that benefits the farmer. The perception is not entirely wrong: the assignment does benefit the lender, and the farmer's net benefit is the reduction in the interest rate minus the premium. If the reduction is small, the farmer will see the insurance as a cost.

The design response is to make the farmer's benefit visible. A product that pays the farmer directly for the portion of the loss that exceeds the loan obligation, and assigns only the portion that covers the loan, is more transparent than a product that assigns the entire payout. A product that returns a portion of the premium in years with no payout—a no-claims bonus—is more attractive than one that does not, even if the expected value is the same. These design features are not cosmetic; they address the behavioral barrier that keeps uptake low.

6.4.2 The Cognitive Barrier

A farmer who has spent his life reading the sky and the soil finds it difficult to trust a contract that pays on a number recorded at a weather station twenty kilometers away. The number is not the farmer's experience. The farmer knows that his field was dry while the station recorded rain, or that the station recorded frost while his field was sheltered. The gap between the index and the experience is the basis risk made visible, and it is the reason many farmers who understand the product still decline to buy it.

The cognitive barrier is not irrational. It reflects a real limitation of the product. The response is not to persuade the farmer that the index is accurate but to design the index so that it is more closely aligned with the farmer's experience. A dense network of low-cost weather stations, a satellite-based index at the field scale, or a hybrid index that combines station data with farmer-reported observations can all reduce the gap. The cost of the denser network is higher, but the cost of low uptake is higher still.

There is also a role for farmer organizations. A cooperative that buys insurance on behalf of its members can aggregate the demand, negotiate a better premium, and provide a trusted intermediary between the farmer and the insurer. The cooperative can also monitor the index and advocate on behalf of members when a payout is disputed. The Chinese supply and marketing cooperatives, which have a long history of organizing farmers at the grassroots level, are a natural channel for this function. The same institutional capacity that allows a cooperative to help farmers form a cooperative and apply for registration—the kind of support that the Wanzhou District cooperative provided in 2007–2008—can be turned to the task of explaining and distributing insurance.

6.4.3 The Role of Subsidy

The demand-side barriers are large enough that index insurance has rarely reached scale without subsidy. The subsidy can take several forms: a premium subsidy that reduces the farmer's cost, a public reinsurance layer that reduces the insurer's capital charge, or a public investment in the weather station network that reduces the cost of the index.

The economic case for subsidy rests on the public-good component of the insurance. A farmer who buys insurance reduces the probability of default, which reduces the lender's loss, which reduces the cost of credit for the whole community. The farmer does not capture the full benefit of this externality, so the private demand is below the social optimum. A subsidy that closes the gap is a Pigouvian correction, not a transfer.

The case is strongest when the insurance is linked to a public objective—the preservation of forest cover, the adoption of a conservation practice, the reduction of agricultural emissions. Here the insurance serves environmental policy as much as farm risk management. The subsidy is justified by the environmental benefit, and the design of the insurance should be aligned with the environmental objective. A forest insurance policy that pays for replanting with native species, or a crop insurance policy that pays only when the farmer has followed a conservation protocol, is a green insurance product in a stronger sense than a policy that merely covers a weather risk.

6.5 Forest Insurance and the Carbon Economy

The forest insurance discussion in Section 6.2 focused on the traditional perils—fire, pest, weather. The carbon economy adds a new dimension that deserves separate treatment because it changes the economics of the insurance and the nature of the risk.

6.5.1 The Reversal Liability

A forest carbon project generates credits by sequestering carbon relative to a baseline. The credits are sold to buyers who use them to offset their emissions. If the forest is later destroyed, the carbon is released and the credit is invalidated. The project developer is typically liable for the reversal, either by repaying the credit or by replacing it with a credit from another project.

The reversal liability is a contingent liability that can be large relative to the project's cash flow. A project that has sold 100,000 credits at 50 yuan each has a liability of 5 million yuan if the forest burns. The liability is not covered by a standard forest insurance policy, which covers the timber loss but not the carbon reversal. The project developer who has not insured the reversal is exposed to a risk that can bankrupt the project.

The insurance solution is a carbon reversal policy that pays the replacement cost of the credits if the forest is destroyed. The policy is triggered by the same remote sensing that verifies the carbon stock, and the payout is used to purchase replacement credits or to repay the buyer. The policy is expensive because the risk is correlated and the actuarial data is thin, but the cost is small relative to the liability.

6.5.2 The Buffer Pool Alternative

An alternative to insurance is a buffer pool: a reserve of credits that is set aside at the time of issuance and used to cover reversals. The buffer pool is the approach used by several voluntary carbon standards, which withhold 10 to 20 percent of the credits issued as a buffer against reversal. The buffer pool is simple and does not require an insurer, but it is inefficient: it ties up capital in the form of credits that could be sold, and it does not distinguish between projects with different reversal risks.

The insurance and the buffer pool are substitutes, and the choice between them depends on the project's risk profile and the cost of capital. A project with a low reversal risk—a forest in a region with low fire risk and secure tenure—is better served by insurance, which prices the risk accurately. A project with a high reversal risk—a forest in a region with weak governance and high fire risk—may not be insurable at a reasonable premium and is better served by a buffer pool. The two mechanisms can also be combined: a buffer pool for the expected reversals and insurance for the tail.

6.5.3 The Credit Link for Forest Carbon Projects

A forest carbon project is typically financed with a mix of equity and debt, and the debt is secured by the project's cash flow from credit sales. The reversal liability is a risk to the debt, because a reversal can eliminate the cash flow that services the debt. A lender who is financing a forest carbon project should therefore require reversal insurance or a buffer pool as a condition of the loan, in the same way that a project finance lender requires completion insurance or a performance bond.

The requirement changes the project's capital structure. The cost of the insurance is a project cost that reduces the equity return, but it also reduces the cost of debt because the lender's risk is lower. The net effect on the project's weighted average cost of capital depends on the relative prices, but in general the insurance is worth it if the reduction in the cost of debt exceeds the premium. This is the same trade-off that Section 6.3 examined for agricultural credit, applied to a different asset.

6.6 Designing the Insurance Pillar: Principles and Trade-offs

The chapter has examined two insurance instruments and their interaction with credit. It is time to draw the design principles together. The argument is not that weather index insurance and forest insurance are the answer to the financing problem of the green transition. It is that they are the risk-buffering component of the three-pillar framework, and that their design must be coordinated with the rating and catalysis pillars if the framework is to work.

6.6.1 Principle One: The Index Must Be Aligned with the Loss

The first principle is that the index must be aligned with the loss that the borrower actually suffers. This is the basis-risk principle, and it is the one that is most often violated. An index that is measured at a station far from the farm, or that captures a weather variable that is not the binding constraint on yield, will have large basis risk and will not change the lender's assessment of default risk.

The alignment can be improved by investing in the measurement infrastructure—weather stations, satellite data, soil moisture sensors—and by calibrating the index to the local agronomy. The investment is costly, but it is a public good that benefits all lenders and insurers in the region, and it is a natural candidate for public funding. The Chinese meteorological authorities have invested heavily in automatic weather stations in recent years, and the density of the network is now sufficient to support index insurance in many agricultural regions. The remaining gap is in the calibration of the index to specific crops and specific loss functions, which is a task for agronomic research rather than for the weather service.

6.6.2 Principle Two: The Payout Must Be Timed to the Cash Flow

The second principle is that the payout must be timed to the borrower's cash flow. An insurance policy that pays after the loan is due does not prevent default. The payout schedule should be designed around the loan schedule, with a mid-season payment that covers the installment and a final payment that covers the remainder of the loss.

The timing principle has implications for the structure of the insurance. A staged payout is more expensive to administer than a single payout, but the additional cost is small relative to the benefit of preventing a default. The lender and the insurer should coordinate on the schedule, and the loan contract should specify the payout dates. This is a simple contractual innovation that is not yet standard practice.

6.6.3 Principle Three: The Insurance Must Be Priced to Reflect the Correlated Risk

The third principle is that the insurance must be priced to reflect the correlated risk. A weather index that triggers across a region is a catastrophe risk, and the premium must include a capital charge for the tail. An insurer that prices the expected payout without the capital charge will underprice the risk and will be unable to pay in a bad year.

The capital charge can be reduced by reinsurance, by diversification across regions, and by public catastrophe backstops. The Chinese agricultural insurance scheme uses all three: the insurers reinsure with China Re, the scheme is diversified across provinces with different weather patterns, and the government provides a catastrophe fund for the largest events. The combination allows the premium to be affordable while keeping the insurer solvent.

6.6.4 Principle Four: The Insurance Must Be Linked to the Credit Contract

The fourth principle is that the insurance must be linked to the credit contract. The link can be a requirement, an assignment, or an integrated product, and the choice depends on the strength of the lender's monitoring and the borrower's incentives. The integrated product is the strongest link and the one that most clearly changes the lender's credit decision.

The link also has a governance dimension. The lender and the insurer must share data on the borrower's exposure and the index, and they must agree on the trigger and the payout. The data sharing raises privacy and competition issues that are not trivial, but they are solvable with standard data protection agreements. The governance arrangement should be specified in the loan contract and in the insurance policy, so that the borrower knows what will happen when the index triggers.

6.6.5 Principle Five: The Insurance Must Be Part of a Broader Risk-Management Strategy

The fifth principle is that the insurance must be part of a broader risk-management strategy. Insurance is not a substitute for good agronomic practice, for irrigation, or for diversification. A farmer who has insurance but no irrigation will still suffer a loss in a drought, and the insurance will only partially compensate. The insurance should be bundled with the practices that reduce the risk, and the premium should be discounted for farmers who adopt those practices.

This is the green insurance principle. A policy that rewards conservation, that pays for replanting with native species, or that covers the transition to a low-carbon practice is a green insurance product in a stronger sense than a policy that merely covers a weather risk. The discount for good practice is a price signal that aligns the farmer's incentive with the environmental objective, and it is the insurance-pillar counterpart of the rating-pillar's environmental score.

6.7 What the Insurance Pillar Does Not Do

It is worth being clear about what the framework does not claim. The insurance pillar does not solve the problem of agricultural credit. It reduces the variance of the borrower's income and the correlation of defaults across the portfolio, but it does not eliminate the information asymmetry that Chapter 2 described. The lender still needs to assess the borrower's creditworthiness, and the insurance is an input into that assessment.

The insurance pillar does not make the green transition profitable. It reduces the risk of the transition, but the return on a green investment still depends on the price of the green product, the cost of the technology, and the policy environment. Insurance can make a marginal project bankable, but it cannot make an unprofitable project profitable.

The insurance pillar does not replace public investment. The weather station network, the catastrophe backstop, and the premium subsidy are public goods that the private market will not provide at the optimal level. The insurance market can only function on top of a public infrastructure, and the infrastructure is a policy choice.

What the insurance pillar does do is change the shape of the risk that the financial system carries. It converts a correlated, catastrophic risk into a diversifiable, priced risk, and in doing so it allows credit to flow to activities that would otherwise be too risky to finance. This is the risk-buffering function, and it is the second pillar of the framework.

6.8 From Insurance to Integration

The chapter has examined weather index insurance and forest insurance as risk-buffering instruments, and it has examined how they interact with credit. The next chapter takes the argument one step further by examining the empirical evidence on agricultural insurance and farmers' income stability. The question there is whether the risk-buffering function translates into measurable income stabilization, and whether the stabilization is large enough to change the credit decision.

The answer, as we will see, is conditional. Insurance stabilizes income when the basis risk is low and the payout is timely, and it does not when the basis risk is high or the payout is late. The conditions are not always met, and the chapter will examine the cases where they are and the cases where they are not.

The broader point is that the three pillars are not independent. The rating pillar identifies the green projects that are worth financing; the insurance pillar buffers the risk that those projects carry; the catalysis pillar channels capital to the projects that the first two pillars have made bankable. The insurance pillar is the middle of the chain, and its design must be coordinated with the other two. A rating that does not account for the insurance will overstate the risk; a catalysis mechanism that does not use the insurance will underinvest. The integration is the subject of Chapter 14, and the empirical foundation for it is the subject of the next two chapters.

The frost that hit Zhang Weiwei's greenhouse in 2019 was not an unusual event. It was a normal risk of farming in Hebei, and it became a credit event because the financial system had no instrument to absorb it. The instrument exists. The design problem is to make it work at the scale and the price that the farmers and the lenders can accept. That is the task of the insurance pillar, and it is the reason this chapter has spent so much time on the details of trigger levels, payout schedules, and basis risk. The details are where the difference between a product that works and a product that does not is made.

6.9 Appendix 6A: A Minimal Model of Index Insurance and Default Probability

This appendix formalizes the transmission from index insurance to default probability for readers who want the algebra behind Section 6.3.

Let the borrower's income be \(\tilde Y = \mu + \sigma \varepsilon\), with \(\varepsilon \sim N(0,1)\). The loan obligation is \(D\). The default probability without insurance is

The borrower buys an index insurance contract with payout \(P = -\alpha \sigma \varepsilon + \eta\), where \(\alpha \in [0,1]\) is the hedge ratio and \(\eta \sim N(0, \sigma_\eta^2)\) is independent basis risk. The hedged income is \(\tilde Y + P = \mu + \sigma(1 - \alpha)\varepsilon + \eta\), and the default probability is

The reduction in default probability is \(p_0 - p_1\), which is increasing in \(\alpha\) and decreasing in \(\sigma_\eta\). For a perfect hedge (\(\alpha = 1\), \(\sigma_\eta = 0\)), \(p_1 = 0\) if \(D > \mu\) and \(p_1 = 1\) if \(D < \mu\)—the degenerate case where the income is certain. For a partial hedge (\(\alpha = 0.5\), \(\sigma_\eta = 0.3\sigma\)), the denominator is \(\sigma\sqrt{0.25 + 0.09} = 0.583\sigma\), and the default probability falls from \(\Phi((D-\mu)/\sigma)\) to \(\Phi(1.715\,(D-\mu)/\sigma)\).

The lender's expected loss is \(p_1 \cdot \mathrm{LGD} \cdot D\), where \(\mathrm{LGD}\) is the loss given default. If the insurance payout is assigned to the lender, the loss given default falls as well, because the payout covers a portion of the exposure. The combined effect on the lender's expected loss is the product of the reduction in \(p\) and the reduction in \(\mathrm{LGD}\), which is why the insurance-linked credit product can reduce the interest rate by more than the reduction in default probability alone would suggest.

The model is deliberately simple, and it omits several features that matter in practice: the borrower's risk aversion, the lender's monitoring, the insurer's capital cost, and the correlation of defaults across borrowers. These features are examined in the main text and in the empirical literature. The model's purpose is to show that the transmission from insurance to credit is a variance reduction, and that the size of the effect is governed by the hedge ratio and the basis risk—the two parameters that the design of the insurance contract controls.

6.10 Appendix 6B: Data Sources for the Chapter

The rainfall distribution used in the numerical illustration in Section 6.1.5 is a stylized calibration to the growing-season rainfall in the North China Plain, based on the China Meteorological Administration's station records for 1981–2020. The forest insurance figures in Section 6.2.6 are from the National Forestry and Grassland Administration's annual statistical yearbook and from the China Banking and Insurance Regulatory Commission's insurance statistics. The Canadian and Australian wildfire areas are from the respective national fire agencies' 2023 and 2019–2020 season reports. The Ghana and Burkina Faso studies cited in Section 6.3.2 are from the indexed insurance literature; the specific estimates should be read as indicative rather than precise, because the studies differ in their outcome measures and their identification strategies.