When climate risk intensifies, does insurance pay off? A community case study of China's agricultural insurance reform

1 Introduction

Food security is of paramount importance for the nation. The stability of the agricultural industry chain is the fundamental guarantee for national food security and the modernization of agriculture and rural areas. National security is fundamentally anchored in food security, and the reliability of food production forms the bedrock of societal welfare and national prosperity. The high-quality development of the agricultural industry depends not only on efficiency improvements in production, processing, and distribution, but also, and more critically, on its comprehensive capacity to maintain stable supply, restore operations rapidly, and achieve adaptive sadjustment in the face of risk shocks. Under the circumstances of intensified geopolitical competition, sreshaping of the international agricultural trade pattern, and increasing external uncertainties, China'ss agricultural industrial chain is facing more external shocks. Enhancing the governance capacity of the sindustrial chain has become an important issue in the process of agricultural modernization (Li and Yang, 2026). In srecent years, the consequences of climate change have become increasingly apparent, manifested in spersistently rising global temperatures and a growing frequency of extreme weather events (McCarl, 2025). sAccording to publicly available statistics from China's Ministry of Emergency Management, the national average temperature in 2025 reached 11.0 °C, exceeding the historical norm by 1.0 °C and marking the highest level recorded since 1961. This accelerating warming trajectory poses a substantial threat to global food production. Figures released by the Food and Agriculture Organization (FAO) of the United Nations indicate that, over the preceding three decades, natural disasters have inflicted economic losses on the agricultural sector surpassing USD 3.8 trillion, a figure that represents approximatelys five percent of the average annual agricultural gross domestic product worldwide. Given its statuss as a major grain producer, China remains especially exposed to these perils (Lin and Wang, 2025). Statistics published by the Ministry of Emergency Management reveal that meteorological hazards curtail China's annual food output by over 50 billion kilograms. In the single year of 2024, natural disasters affected roughly 10.089 million mu (equivalent to approximately 672,600 hectares) of cropland across the country, generating direct economic losses in agriculture totaling 401.11 billion Chinese Yuan (CNY).

Against this setting, a production paradigm narrowly oriented toward yield maximization no longer suffices to confront mounting climate threats. Consequently, constructing a food production system endowed with the capacity to prevent, withstand, and adjust to climatic disruptions has emerged as a strategic national objective (Jiang et al., 2023). Strengthening resilience to climate hazards now constitutes a foundational pillar of this system. Evidence from both domestic and international practice confirms that the most effective strategy for addressing such risks rests on forward-looking, ex ante insurance instruments, rather than on backward-looking, ex post disaster relief (Xie et al., 2024).

However, agriculture has inherent weaknesses (Shi and Xu, 2026). Relying solely on fiscal support for agriculture will encounter a dilemma of insufficient total funds and low utilization efficiency. Moreover, the pure market logic is prone to cause rural capital outflow due to financial exclusion. Therefore, the steady promotion of agricultural modernization requires the coordinated support of finance and finance in order to effectively align with the current development of rural economy. Policy-oriented agricultural insurance serves as a key instrument for managing modern agricultural risks (Li et al., 2025a). Beginning in 2007, China established a state-funded farm revenue protection framework that draws on national budget transfers to cover insurance costs and that provides supportive measures along with advantageous conditions for staple crops and livestock (Zhang and Xu, 2023). As the insurance program matured, however, its guiding principle of low coverage levels coupled with broad eligibility confined indemnities largely to material input costs. This design fell short of accommodating farmers' increasingly complex and fast-evolving risk protection needs.

To strengthen the level of protection, China initiated a new wave of policy-oriented agricultural insurance reform in 2018, introducing pilot schemes that offer full-cost and planting-income coverage for the three major staple crops. In contrast to conventional policies that compensated only direct material inputs, these new models extended coverage to factors of production such as land rent and labor, thus delivering comprehensive safeguards against total production expenditures (Li et al., 2025a). The program was designed to be broadly inclusive, extending coverage to smallholders, moderate-scale family farms, and other types of agricultural producers. Concerning the premium subsidy structure, provincial governments were required to bear no < 25 percent of the premium cost, while the central government financed 45 percent in the central, western, and northeastern regions and 35 percent in the eastern region. In 2021, the pilot initiative was expanded to cover all major grain-producing provinces, progressively evolving into a vital institutional mechanism for strengthening risk protection in grain production. Statistics released by the National Financial Regulatory Administration (NFRA) show that China's total agricultural insurance premiums reached 152.1 billion CNY in 2024, providing risk protection in excess of 5 trillion CNY to 147 million farming households. The contributions of this study are threefold:

• At the theoretical level, this paper advances a unified analytical framework that integrates climate risk, agricultural insurance, and food production resilience. While existing scholarship has separately examined the production effects of insurance and the disruptive consequences of climatic shocks, the interactive logic connecting these three elements remains undertheorized. We fill this gap by developing a formal model in which full-cost and planting-income insurance, through raising coverage levels and premium subsidy rates, expands farmers' optimal efficient output scale by reducing the variance of net income and alleviating downside risk constraints. Crucially, this theoretical framework yields two novel propositions that extend beyond the existing literature: (i) agricultural technological progress and operational scale expansion serve as the dual mediating channels through which insurance transmits its resilience-enhancing effect; and (ii) climate risk positively moderates the marginal contribution of insurance to production resilience, implying that the value of insurance as a risk-management instrument increases monotonically with the intensity of environmental hazards. This “insurance–technology–scale–resilience” theoretical chain offers a generalizable conceptual apparatus for understanding how publicly supported insurance schemes can function as adaptive instruments in the face of climate-induced agricultural volatility, with applicability well beyond the Chinese context.
• From a methodological standpoint, this study constructs a multidimensional composite index for food production resilience that extends the conventional output-centric measurement approach. Drawing upon the FAO's resilience framework and synthesizing insights from recent scholarship, we conceptualize food production resilience along five constitutive dimensions: risk prevention, risk resistance, adaptive recovery, transformative innovation, and sustainable development. This five-dimensional architecture captures not only the post-shock recuperative capacity of production systems but also their ex-ante preparedness and long-run adaptive transformation potential, thereby offering a more comprehensive operationalization of resilience than those employed in prior empirical work. The entropy-weighted composite index developed here provides a replicable measurement instrument for future cross-country and cross-regional comparative studies.
• In terms of policy relevance, the empirical findings generate actionable insights for the design of agricultural risk governance frameworks. By documenting the positive moderating role of climate risk and the pronounced heterogeneity across regions with varying disaster damage intensities, this study demonstrates that the marginal effectiveness of high-coverage insurance is amplified precisely where environmental pressures are most acute. These results establish an empirical foundation for positioning comprehensive agricultural insurance not merely as an ex post loss-compensation mechanism, but as a proactive climate adaptation strategy that reinforces the stability and sustainability of food production systems. The analytical framework and policy implications advanced herein offer reference points for other developing economies confronting analogous challenges of climate vulnerability, fragmented farm structures, and fiscal constraints on agricultural support.

As shown in Figure 1, it illustrates the mechanism of the reform of policy-oriented agricultural insurance affects the resilience of food production in the hypothesis of this study.

2 Literature review

Climate change constitutes a formidable challenge to global food security, with repercussions that reverberate across all of its core dimensions, from production and access to utilization and stability, while also subjecting food output to severe disruptions (Lee et al., 2024). An extensive strand of scholarship has documented the detrimental consequences of climate change for agricultural performance. In particular, extreme heat has been shown to undermine both agricultural total factor productivity (TFP) and the efficiency with which inputs are utilized, with the damage concentrating disproportionately on food output (Chen and Gong, 2021). Beyond direct output losses, climate shocks also disrupt land use, capital accumulation, and both the quantity and efficiency of labor (Sun et al., 2024). These shocks translate into higher marginal production costs, constrain the economic trajectory of the agricultural sector, and ultimately erode the stability and long-term sustainability of production systems (Cui and Zhong, 2024).

Against this backdrop, relying exclusively on conventional yardsticks such as output to assess food production performance falls short of capturing the full capacity of production systems to absorb risks, adjust to shifting conditions, and maintain supply in the face of external shocks (Knickel et al., 2018). A resilience framing foregrounds the comprehensive adaptive capability of production systems across the pre-shock, shock, and post-shock stages, thereby offering a critical analytical lens for examining the robustness of food production and the capacity to uphold food security (Zeng et al., 2025; Guo et al., 2026a). The scholarly literature on the resilience of food production systems can be broadly organized into three lines of inquiry. The first line constructs multidimensional evaluation frameworks anchored in the conceptualization and quantification of resilience, with the goal of assessing the resilience status of food production systems (Zampieri et al., 2020). The second part of work investigates the spatiotemporal evolution, regional disparities, and underlying determinants of food production resilience in China, documenting a robust association with factors such as urbanization and investments in agricultural technology (Liu et al., 2025a; Zhao et al., 2026). The third strand relates resilience to crop diversification, the development of high-standard farmland, and disruptions along supply chains, maintaining that food production systems must not only secure stable output but also build long-term adaptive capacity to cope with climate change and tightening resource and environmental constraints (Cai et al., 2025; Liu et al., 2026).

As an important institutional arrangement for agricultural risk governance, agricultural insurance is a key policy instrument for food production to cope with climate risks (Hazell and Varangis, 2020). Existing studies have mainly examined the production effects of agricultural insurance from the perspectives of output, income, and planting behavior (Wang and Zhong, 2025). The available empirical evidence indicates that publicly supported agricultural insurance strengthens farmers' motivation to expand operational scale and maintain essential input levels by pooling risks and smoothing incomes, ultimately generating a favorable impact on grain output (Zhang et al., 2019). Research focusing on China's grain crops similarly indicates that agricultural insurance fosters grain output and dampens income variability by encouraging farmland transfers, raising input intensity, and stabilizing farm revenues (Chen and Gu, 2025). Evidence from catastrophic agricultural insurance pilots further reveals favorable effects on farm income and grain output, transmitted through heightened investment willingness and improved cropland management practices (Li et al., 2025b). International studies similarly show that crop insurance subsidies affect the planted area of insured crops through profit and coverage effects (Yu et al., 2018) and induce adjustments in land use (Claassen et al., 2016; Goodwin et al., 2004).

Beyond alterations in output and cultivated land, agricultural insurance can influence food production via supplementary mechanisms, notably the diffusion of production technologies and the reallocation of crop portfolios (Giné and Yang, 2009). Relevant studies find that insurance can alleviate risk constraints and increase the likelihood of adopting improved technologies and digital agricultural technologies (Ahmed et al., 2020; Dong et al., 2025). It can also affect labor allocation, land transfer, and changes in operational scale, thereby reshaping factor allocation and organizational arrangements in food production (Chen and Gu, 2026). Nevertheless, a growing body of literature cautions that overly generous insurance protection may precipitate moral hazard and skew factor allocation, thereby exerting unintended repercussions on input use and broader farm management (Babcock and Hennessy, 1996; He et al., 2020). This further indicates that the policy effects of agricultural insurance need to be identified in relation to specific coverage levels, risk environments, and production conditions (Sun et al., 2024).

Taken together, although the extant research provides a substantial conceptual foundation for the present study, a number of unresolved questions demand further scrutiny. One notable lacuna concerns the adaptive efficacy of comprehensive indemnity products, including full-cost and planting-income coverage, which has yet to be rigorously evaluated. Meanwhile, the majority of prior studies have investigated the roles of insurance and climatic shocks in shaping grain output in isolation, without probing the intricate interplay that connects all three elements.

Taken together, the preceding review identifies three distinct research strands, each of which falls short of addressing the full scope of our research question. (1) The literature on food production resilience has substantially advanced the conceptualization and measurement of production system adaptive capacity (Zampieri et al., 2020; Liu et al., 2025a), yet it has predominantly treated resilience as an outcome determined by structural factors such as urbanization and technological investment, leaving the role of institutional arrangements such as insurance largely unexamined. (2) While a rich body of work has rigorously documented the effects of agricultural insurance on output, income, and planting behavior (Zhang et al., 2019; Chen and Gu, 2025; Wang and Zhong, 2025), these studies have focused almost exclusively on conventional physical-cost insurance rather than on the comprehensive indemnity products introduced in recent reforms; the unique effects of full-cost and planting-income coverage, which extend protection to land rent and labor costs, remain underexplored. (3) Although the disruptive consequences of climate change for agricultural production have been well established (Chen and Gong, 2021; Lee et al., 2024), and the risk-mitigating function of insurance has been separately examined (Hazell and Varangis, 2020), no study has integrated these two strands to investigate whether and how climate risk moderates the effectiveness of insurance in sustaining production resilience. The present study addresses these three gaps simultaneously by developing a unified theoretical framework that articulates the “insurance–technology–scale–resilience” transmission chain, by constructing a multidimensional resilience index that moves beyond output-centric proxies, and by empirically testing the resulting propositions using a staggered difference-in-differences design with explicit attention to the moderating role of climate risk. In doing so, we aim to advance both the theoretical understanding of insurance as a climate-adaptive governance instrument and the empirical evidence base for the design of comprehensive agricultural risk management frameworks.

3 Theoretical analysis and research hypotheses

The theoretical analysis developed in this section proceeds from a unified conceptual framework that integrates direct effects, mediating mechanisms, and moderating conditions into a coherent “conditional mediation” structure. At the first layer, the reform operates through a direct effect captured by Hypothesis 1: by raising coverage levels and premium subsidy rates, full-cost and planting-income insurance expands farmers' optimal efficient output scale through variance reduction and downside risk alleviation, thereby enhancing production resilience. At the second layer, this direct effect is transmitted through two mediating channels formalized in Hypothesis 2: agricultural technological upgrading and operational scale expansion. The theoretical logic, derived from the conditional factor demands of the cost minimization problem, posits that insurance-induced expansion of the efficient output scale generates increased demand for both technological inputs and land consolidation. These mediating variables, in turn, reinforce the production system's capacity to withstand and recover from climatic disruptions. At the third layer, the entire mediating process is contingent upon the intensity of climate risk, as articulated in Hypothesis 3. Climate risk modulates the marginal effect of insurance through three mechanisms: it reconfigures farmers' constraint structure by raising the probability of low-output events; it amplifies the substantive role of the loss compensation mechanism by increasing the likelihood of production disruptions; and it shapes adaptive investment behavior by elevating the demand for stress-tolerant technologies and improved agronomic conditions. Theoretically, as climate risk intensifies, both the direct effect of insurance on resilience and the indirect effects transmitted through technology and scale channels become larger in magnitude.

This conditional mediation framework yields a critical insight: the value of agricultural insurance as a climate-adaptive instrument is not constant but increases monotonically with environmental hazard intensity. The three hypotheses are therefore not independent propositions but sequentially nested components of a unified theoretical architecture, with H1 establishing the baseline effect, H2 unpacking the transmission mechanisms, and H3 specifying the boundary conditions under which these mechanisms operate most effectively.

3.1 Direct impact of policy-oriented agricultural insurance on food production resilience

3.1.1 Production function

Consider the representative farmers in region r. We assume the effective output scale to be given by Equation 1:

Tit=A(ait)LitαKit1-α(1)

In Equation 1, α represents the intensity of agricultural technology input, such as the adoption of resistant varieties, soil testing and formula fertilization, irrigation facilities, etc. Its constraint condition is that α ϵ 0,1; L represents the land operation scale, K represents other capital investment, and T represents the potential output scale under normal conditions. A(α) indicates that technological input enhances output. Its constraint condition are A′(α)≻0 and A″(α) ≤ 0. The actual output Y is affected by random climate shocks ε, and the Equation 2 can be obtained:

Yit=Titωit(θrt)(2)

In Equation 2, θ_*rt* represents the climate risk exposure level of region r. The actual output is influenced by the random unit output ω. It is understandable that the distribution of unit outputs will change depending on the degree of climate risk, and its expression is ω_*it*|θ_*it* ~ F(·|θ_*it*). We assume that the climate risk meets the conditions specified in Equation 3:

{μ(θ)≡E[ε|θ]σ2(θ)≡Var(ε|θ)(3)

The constraint condition of Equation 3 is μ′(θ)≺0 and σ′(θ)≻0. This means that the higher the climate risk, the lower the expected output and the greater the output fluctuation.

3.1.2 Agricultural policy-based insurance

Policy-based agricultural insurance, particularly full cost insurance and crop income insurance, can be regarded as simultaneously enhancing the coverage level γ and the premium subsidy rate s on the basis of traditional physical cost insurance. This paper assumes that the coverage level stipulated in the insurance policy is γ ∈ (0, 1], and the benchmark unit output is ω₀. When the actual unit output is lower than the coverage threshold γ/ω₀, an insurance payout will be triggered. Assuming the coverage threshold is x = γ/ω₀, then the payout function is Equation 4:

I(x,ω)=Tmax(x-ω,0)(4)

The expression form of the expected compensation amount is Equation 5:

B(x,θ)=E[max(x-ω,0)]=∫-∞x(x-ω)dF(ω|θ)(5)

The actual premium paid by the farmers is:

P=(1-s)TB(x,θ)(6)

In Equation 6, sϵ(0, 1) represents the fiscal premium subsidy rate. The policy-based insurance reform is formalized as dr > 0 and ds < 0.

3.1.3 Maximization of farmers' income

The net income of the farmers is:

Π=Y+I-P-c(a)-ωLL-ωKK(7)

Substituting Y = Tω, the compensation function and the premium expression into Equation 7, we can obtain:

Π=T[ω+max(x-ω,0)]-(1-s)TB(x,θ)-c(a)-ωLL-ωKK(8)

In Equation 8, the truncated random unit output is defined, and this leads to:

h(ω;x)=max(ω,x)(9)

From Equation 9, the net profit function can be obtained:

Π=Th(ω;x)-(1-s)TB(x,θ)-c(a)-ωLL-ωKK(10)

The expectation of Equation 10 is:

E(Π)=T[μ(θ)+sB(x,θ)]-c(a)-ωLL-ωKK(11)

After offsetting the expected compensation TB with the insurance premium expenditure (1 − s)TB, the net gain becomes sTB. Therefore, the variance of Equation 11 can be calculated:

Var(Π)=T2Var[h(ω;x)]≡T2Var(x,θ)(12)

In Equation 12, the restriction condition for Var(x, θ) is Equation 13

Var(x,θ)=Var[h(ω;x)|θ](13)

This paper assumes that the cost function is defined as the minimum cost required to produce the potential output scale T, which satisfies Equation 14:

C(T)=mina,L,K{c(a)+ωLL+ωKK|A(a)LαK1-α≥T}(14)

This paper assumes that the cost function is strictly increasing and strictly concave, which implies the restriction conditions C′(T)≻0 and C″(T)≻0. The representative farm household exhibits risk averse preferences, with the certainty equivalent characterized by a mean variance formulation.

U(T;x,θ)=E(Π)-ρ2Var(Π)(15)

In Equation 15, ρ > 0 is the coefficient of absolute risk aversion. Substituting Equations 11 and 12 into Equation 15 yields:

U(T;x,θ)=T[μ(θ)+sB(x,θ)]-C(T)-ρ2T2V(x,θ)(16)

Based on Equation 16, the farm household chooses T to maximize the certainty equivalent. The first order condition is:

∂U∂T=μ(θ)+sB(x,θ)-C′(T)-ρTV(x,θ)=0(17)

From Equation 17, the condition that the efficient output scale T^* must satisfy is given by Equation 18:

C′(T)=μ(θ)+sB(x,θ)-ρTV(x,θ)(18)

The second order condition is:

-C″(T*)=μ(θ)-ρV(x,θ)≺0(19)

The second order condition for Equation 19 is automatically satisfied provided that C″(T)≻0 and V(x, θ) ≥ 0 hold.

3.1.4 The role of agricultural insurance in grain production

Since x = γ/ω₀, we first examine the effect of a change in x on the optimal scale T^*. Taking the total differential of the first order condition in Equation 17 with respect to x yields Equation 20:

[C″(T)+ρV(x,θ)]dTdx=sBx(x,θ)-ρT*Vx(x,θ)(20)

Based on the definition of expected indemnity, applying Leibniz's rule yields Equation 21:

Bx(x,θ)=F(x|θ)≻0(21)

For h(ω; x) = max(ω, x), its expectation is given by Equation 22:

E[h]=xF(x)+∫x∞ωdF(ω)(22)

Therefore, we obtain Equation 23:

∂E[h]∂x=F(x)(23)

Similarly, we obtain Equation 24:

E[h2]=x2F(x)+∫x∞ω2dF(ω)(24)

Therefore, ∂E[h²]/∂x = 2xF(x). Hence, we obtain Equation 25:

Vx=∂∂x(E[h2]-(E[h])2)=2xF(x)=2xF(x)(x-E[h])(25)

Since h = max(ω, x) ≥ x, it follows that E[h] ≥ x, and E[h] ≥ x is strictly > x when F(x)≺1. Hence V_*x* ≤ 0. This implies that raising the coverage level reduces the variance of the farm household's net income. Substituting Equations 21 and 25 into the expected indemnity equation yields Equation 26:

dTdx=sF(x)-ρT▪2F(x)(x-E[h])C″(T*)+ρV(26)

Since x − E[h] ≤ 0, the numerator of Equation 26 can be rewritten as Equation 27:

F(x)[s+2ρT*(E[h])-x]≻0(27)

Therefore dT^*/dx≻0. Furthermore, since x = γω₀, we obtain Equation 28:

dTdγ=ω0dTdx=ω0F(γω0)[s+2ρT(E[h])-γω0]C″(T)+ρV(28)

Equation 28 implies that raising the coverage level γ significantly expands the farm household's optimal efficient output scale. Taking the total differential of the first order condition in Equation 17 with respect to s:

[C″(T)+ρV]dTds=B(x,θ)(29)

From Equation 29, we obtain Equation 30:

dTds=B(x,θ)C″(T)+ρV≻0(30)

The reforms of full cost insurance and planting income insurance, by raising the coverage level γ and the premium subsidy rate s, both significantly increase the farm household's optimal efficient output scale T^*. Therefore, this paper derives the theoretical conclusion that agricultural insurance enhances the resilience of grain production.

Amid escalating global climate change and growing unpredictability in agricultural hazards, the resilience of food production systems encompasses both the post-shock recuperative potential and the ex-ante adaptive responses of producers, namely sustained input use, production reconfiguration, and operational viability under perceived risk conditions (Zampieri et al., 2020). Finance is an important force for achieving sustainable agricultural development and building resilience (Lu et al., 2025; Xu et al., 2025). In contrast to conventional physical-cost indemnity policies, the full-cost and planting-income coverage introduced under the ongoing policy-oriented agricultural insurance reform affords both enhanced benefit levels and a wider array of risk protection. Accordingly, its institutional mandate transcends ex post compensation for disaster-induced losses, instead furnishing more comprehensive support for the food production apparatus through the stabilization of revenue forecasts and the alleviation of binding risk constraints (Zhang and Xu, 2023). The specific logic is as follows:

In the short term, agricultural insurance can reduce the distortion of production decisions caused by risk uncertainty by stabilizing income expectations (Zhang et al., 2019). The agricultural sector is inherently vulnerable to environmental shocks and fluctuating returns. Under the expected utility paradigm, producers exhibiting marked risk aversion, when confronted with significant uncertainty, tend to pursue prudent input choices and low-exposure operational models (Yu and Sumner, 2018), in order to avoid unaffordable losses caused by disaster shocks. High-indemnity agricultural policies transform looming contingent losses into fixed and foreseeable premium expenditures, which helps to solidify farmers' anticipations regarding their grain operations. Such a mechanism encourages the retention of key production factors, including irrigation systems, machinery, and emergency supplies, even in periods of elevated hazard. Although adopting a low-input, low-risk posture can curtail near-term vulnerability, this approach may inadvertently reduce essential cropping outlays and consequently impair the system's robustness and adaptive responses. Hence, insurance provisions not only fortify ex ante preparedness against potential calamities but also mitigate the deleterious consequences of exogenous perturbations on the steadiness of aggregate food output (Chen et al., 2025).

From a longer-run vantage point, agricultural insurance fosters the evolution of cropping systems and reinforces their sustainability by enhancing the risk environment and stabilizing forward-looking business sentiments. Given that innovative seed varieties and advanced cultivation methods typically require substantial up-front outlays and entail elongated recoupment periods, the presence of risk-related frictions may curb producers' propensity to incorporate new practices or invest in lasting operational upgrades (Carter et al., 2016). By offering indemnity payments and stabilizing farm revenues, high-coverage insurance alleviates the risk attached to extended investment cycles. This, in turn, reinforces producers' capability and resolve to incorporate advanced agricultural techniques, encompassing stress-tolerant germplasm, water-saving irrigation systems, mechanized field operations, and digital monitoring tools. Ultimately, such insurance provisions help secure a more predictable and resilient operating environment conducive to sustained cropping activities (Tang et al., 2019). Furthermore, dependable insurance arrangements mitigate the myopic operational responses that arise from catastrophic events and revenue volatility. In doing so, they strengthen producers' commitment to avoiding cropland idleness and curtailing excessive resource depletion in the aftermath of hazard occurrences (Ma et al., 2024). Through improved allocative efficiency, agricultural insurance unlocks the endogenous development capacity of cropping systems in the face of ecological and resource stringency, thereby enhancing their overall productive resilience (Li et al., 2025c). On the basis of this rationale, we advance our first hypothesis accordingly.

• Hypothesis 1 (H1): Policy-oriented agricultural insurance reform can enhance food production resilience.

3.2 The mechanism of the impact of technological progress and scale operation

Land and technology inputs are necessary inputs for grain production (Chen and Gu, 2025). Amid rising climatic hazards, policy-oriented agricultural insurance, over an extended horizon, can reshape the risk environment confronting farm operators. Such a transformation encourages progress in cultivation techniques and facilitates the consolidation of production scales. In turn, this gives rise to an indirect mediating channel through which insurance coverage contributes to the robustness of food output systems.

The conditional factor demands for the cost minimization problem 14 are given by Equation 31:

{a=a(T)L=L(T)K*=K(T)(31)

Under the usual assumptions on the production function and factor prices, an expansion of the efficient output scale requires more technological inputs and a larger operational land scale, hence we obtain da/dT≻0 and dL/dT≻0. Combining this with Research Hypothesis 1 yields Equations 32 and 33:

dadγ=∂a∂TdTdγ≻0(32)

dLdγ=∂L∂TdTdγ≻0(33)

By the same reasoning, we can obtain Equations 34 and 35:

da*ds≻0(34)

dL*ds≻0(35)

Therefore, the policy-based insurance reform alleviates the downward risk constraints faced by farmers, thereby motivating them to adopt higher levels of agricultural technologies and expand the scale of land operation, thus enhancing the effective output scale.

A higher climate risk θ shifts the distribution of the stochastic unit output ω toward an adverse direction, with the expected value decreasing, the variance increasing, and the left tail probability increasing. Specifically, this paper assumes that Equation 36 holds.

∂F(x|θ)∂θ≻(36)

Equation 36 implies that an increase in climate risk raises the probability of low-output events. Meanwhile, as θ increases, the expected excess over the coverage threshold, denoted, also tends to increase. This is because the expectation of h = max(ω, x) changes with the decrease in the mean and the increase in the variance of the distribution of ω. A reasonable assumption can be made:

∂(E[h]-x)∂θ≻0(37)

From Equations 30 and 37, we obtain:

dTdγ=ω0F(x)(s+2ρT(E[h])-x)C″(T*)+ρV(38)

The numerator is given by:

G(θ)=ω0F(x)[s+2ρT*E[h]-x](39)

Under the restrictions of Equations 38, 39, we obtain AAA. Therefore CCC. According to Research Hypothesis 1, we can obtain Equation 40:

∂2R∂γ∂θ≻0(40)

Therefore, we derive the theoretical expectation that the higher the climate risk exposure, the greater the marginal enhancement effect of policy oriented insurance reform on the resilience of grain production.

China's agricultural landscape is defined by a large-country setting combined with a predominance of smallholder operations. The pervasiveness of petty and partitioned farm holdings intrinsically restricts both the allocation of resources and the employment of productive factors in cereal production (Xie et al., 2024). Given the pervasive smallholder structure in China's vast agrarian economy, the prevalence of fragmented and minuscule farm holdings imposes inherent frictions on factor allocation and input utilization in grain cultivation. Such a configuration precludes the realization of scale economies and simultaneously hampers the diffusion of mechanical tillage and standardized production protocols, thereby diminishing the sector's wherewithal to withstand external disturbances. Policy-led agricultural insurance, by offering a protective buffer against the substantial losses inherent in enlarged operations, attenuates the risk burden shouldered by producers. Consequently, operators may gravitate from aversion toward risk-neutral stances, while prospects concerning land transfer and farm-size expansion become more predictable. Full-cost coverage, in particular, incorporates land rental outlays into its indemnity scope, thereby alleviating the sunk-cost hazard associated with scaling up and reinforcing incentives for operational enlargement. As holdings become more extensive, production systems tend toward field consolidation, standardized practices, and intensive husbandry, which collectively improve allocative efficiency, deepen mechanization and specialized division of tasks, and ultimately bolster the systemic fortitude of food output against exogenous volatility and sustained supply pressures (Chen and Gu, 2026).

The mechanism through which full-cost insurance facilitates land transfer and scale expansion operates through three interrelated channels. (1) On the supply side of the land rental market, the inclusion of land rent within the insurance coverage directly reduces the opportunity cost of land leasing for transferors. Under traditional physical-cost insurance, land rent constitutes a sunk cost that remains uncompensated in the event of production failure, creating a residual claimant risk for landowners who may hesitate to lease out their land for fear that rent payments could be disrupted by yield shortfalls. By explicitly covering land rental costs, full-cost insurance guarantees that transferors receive their contractual rent even when catastrophic events occur, thereby lowering the perceived risk of land leasing and increasing the effective supply of transferable land. (2) In the demand side, the enhanced coverage reduces the downside risk exposure of large-scale operators. Expanding operational scale entails substantial fixed costs in land rental, machinery acquisition, and labor hiring, which expose farmers to elevated financial risk in the event of production failure. Full-cost insurance, by providing comprehensive protection against total production costs including land rent, lowers the effective risk premium associated with scale expansion and strengthens farmers' willingness to enter long-term land rental contracts. This risk-balancing effect encourages the consolidation of fragmented plots into larger operational units. (3) At the market level, the guaranteed compensation for land rent stabilizes rental price expectations and reduces contract incompleteness. When rental payments are protected by insurance, both parties face lower uncertainty regarding future cash flows, which facilitates the formation of longer-term and more stable land transfer agreements. This institutional underpinning, in turn, supports sustained investment in land improvement and productivity-enhancing practices, further reinforcing the scale economies that underpin food production resilience.

Agricultural technological progress constitutes a fundamental driver of resilience in food production systems. However, the diffusion and practical uptake of such advances have persistently been hampered by elevated risk profiles, substantial capital outlays, and protracted payoff periods. Policy-oriented agricultural insurance can mitigate these impediments via both demand-side and supply-side mechanisms. On the demand side, through risk pooling and expectation stabilization, insurance effectively lowers the opportunity costs and residual ambiguities linked to the introduction of novel farming practices. Consequently, farmers become more inclined to allocate resources toward climate-resilient innovations, including precision farming platforms, water-conserving irrigation infrastructure, and genetically improved seed varieties, all of which provide direct defenses against meteorological shocks (Dong et al., 2025). On the supply side, a more complete insurance-based protection environment can generate sustained and effective demand for agricultural technologies (Su et al., 2025). This demand, in turn, generates supply-side feedback effects by inducing governments and agribusinesses to augment sustained investment in agricultural research and development. Consequently, policy-oriented agricultural insurance accelerates technological advance in the agricultural sector by underwriting innovation adoption and cultivating a predictable demand environment for continued technological upgrading. Simultaneously, advanced technologies refine farm-level management practices, augment operational precision and process controllability, and bolster the food production system's capacity to maintain supply continuity amid adverse conditions, thereby fortifying the resilience of food production systems. Grounded in this analytical framework, the present study advances Hypothesis 2:

• Hypothesis 2 (H2): Agricultural technological upgrading and the expansion of large-scale agricultural operation may serve as potential channels through which policy-oriented agricultural insurance reform enhances food production resilience.

It is important to recognize that the two mediating channels, technological upgrading and scale expansion, are not mutually exclusive or independently operating pathways. Rather, they exhibit important complementarities. The boundary between them lies in their distinct theoretical domains. Technological progress pertains to the production function itself, enhancing output per unit of input through improved varieties, precision management, and mechanization. Scale expansion pertains to the organizational structure of production, reducing per-unit fixed costs and enabling specialized division of labor through land consolidation.

Despite this conceptual distinction, the two mechanisms are interdependent in practice. On one hand, the adoption of certain advanced agricultural technologies, particularly large-scale machinery, automated irrigation systems, and digital farming platforms, entails substantial fixed costs and requires minimum operational scales to achieve economic viability (Guo et al., 2026b; Cuicui et al., 2026). In this sense, scale expansion serves as a necessary condition for the full realization of technological returns. On the other hand, technological progress facilitates scale expansion by raising the productivity of land and labor, thereby increasing the opportunity cost of fragmented operations and strengthening the economic rationale for farmland consolidation (Xu et al., 2024). This two-way interdependence implies that the mediating role of technology and scale should be understood as a mutually reinforcing system rather than as isolated parallel channels. Consequently, the insurance-induced expansion of efficient output scale may simultaneously stimulate technological investment and scale adjustment, with each mechanism amplifying the resilience-enhancing effect of the other. This synergistic logic further implies that policies promoting agricultural insurance should be complemented by measures that facilitate land transfer and technology extension, as the effectiveness of each channel may be contingent upon progress in the other.

3.3 Analysis of the moderating effect of climate change

Climate change shocks have become one of the most significant risks currently facing food production (Lin and Wang, 2025). However, the intensity of climate change impacts varies across regions, meaning that the marginal effect of policy-oriented agricultural insurance in enhancing food production resilience may be uneven (Lee et al., 2024). From an economic standpoint, the value of risk management instruments displays a positive relationship with the intensity of risk prevailing in the operating environment. In locations facing elevated climate change threats, the uncertainty surrounding grain production and the volatility of farm incomes rise substantially. Theoretically, its mathematical transmission process is as follows:

It is precisely under these circumstances that the risk-buffering function and the incentive structures embedded in policy-oriented agricultural insurance are most apt to become operative. The specific logic is as follows:

• (1) Climate change risk reconfigures the constraint structure confronting farmers. In high-risk areas, extreme climatic shocks are more prone to materialize as actual losses, thereby compressing the lower bound of farm income. Production decisions therefore become increasingly dependent on risk-sharing mechanisms, and the demand for risk management instruments turns more inelastic (Chen et al., 2021). By providing minimum income protection, insurance directly affects the boundary conditions of farmers' input decisions, enabling them to maintain production inputs and planting arrangements under higher levels of risk.
• (2) Elevated climate risk amplifies the substantive role of the loss compensation mechanism. The essential function of publicly supported agricultural insurance is to stabilize income flows over time. In settings where hazards are more frequent and damages more concentrated, production disruptions and liquidity constraints become considerably more probable. Under such conditions, insurance indemnities directly support farmers' capacity to sustain essential inputs during the post-shock recovery process (Li et al., 2025a). As risk levels rise, the marginal benefit that the compensation mechanism delivers for post-disaster recovery increases accordingly, thereby reinforcing the capacity of insurance to safeguard the stability of food output. Conversely, in areas marked by infrequent hazard events, modest loss ratios, and more dependable income streams, the stabilizing function of insurance is less likely to be fully harnessed.
• (3) Climate risk shapes farmers' adaptive investment behavior. In settings subject to heightened hazard exposure, the cost of sustaining conventional production practices escalates, and the demand for stress-tolerant technologies along with improved agronomic conditions intensifies. Yet these investments are characterized by considerable uncertainty. When confronting climate threats, insured farmers exhibit a stronger propensity to adopt adaptive measures, such as drought-resistant cultivars and upgraded irrigation infrastructure, compared with their counterparts in low-risk areas (Moritz et al., 2025). By cushioning the downside risk inherent in investment, insurance enables farmers to undertake longer-term commitments to technology and infrastructure, thereby strengthening the production system's capacity to adapt to climatic disruptions. As risk exposure intensifies, the incentive that insurance provides for such investments grows correspondingly, and its long-run contribution to resilience becomes increasingly pronounced. Building on this logic, we propose Hypothesis 3:

• Hypothesis 3 (H3): Climate change exerts a positive moderating effect on the ability of policy-oriented agricultural insurance to enhance FPR.

3.4 Micro-channels

The preceding analysis identifies expectation stabilization as the primary mechanism through which insurance influences input decisions. However, the theoretical literature suggests at least three additional micro-channels that merit explicit consideration.

• (1) Insurance alleviates credit constraints by serving as an implicit collateral substitute. In developing and emerging economies, smallholder farmers face pervasive liquidity constraints that impede investment in productivity-enhancing technologies, including improved seed varieties, irrigation equipment, and mechanization (Wan et al., 2025; Noufé, 2026). Financial intermediaries are often reluctant to extend credit to agricultural producers due to the covariate nature of agricultural risks and the absence of acceptable collateral. Insurance contracts, especially those with guaranteed indemnity payments tied to verifiable production shortfalls, reduce lenders' perceived default risk by establishing a minimum income floor that acts as de facto collateral (Huang et al., 2026; He et al., 2026). This collateral effect lowers the effective cost of credit and expands farmers' borrowing capacity, thereby relaxing the liquidity constraint that would otherwise preclude lumpy technology investments. In the context of our theoretical framework, the presence of credit constraints would manifest as an additional wedge in the cost function, and the relaxation of such constraints through insurance effectively reduces the marginal cost of technological inputs, reinforcing the positive effect of coverage expansion on efficient output scale.
• (2) Insurance facilitates specialization and division of labor through risk pooling. Agricultural production involves numerous risky and irreversible decisions, and farmers in high-risk environments often resort to portfolio diversification strategies, such as planting multiple crop varieties, engaging in off-farm employment, or maintaining redundant production capacity, to smooth income fluctuations (Wu et al., 2024). While individually rational, such diversification behavior fragments operational focus, dilutes managerial attention, and prevents the realization of scale economies and specialization gains. By pooling idiosyncratic and covariate risks, insurance enables farmers to reduce their reliance on self-insurance and portfolio-based risk management, thereby allowing them to specialize in their comparative advantage activities (Pan, 2026). This specialization effect operates not only at the farm level but also across the broader agricultural value chain, fostering the emergence of specialized service providers, machinery rental markets, and contract farming arrangements. In our model, this channel corresponds to a reduction in the effective risk premium, which in turn lowers the marginal cost of expanding the efficient output scale and incentivises greater specialization in technology-intensive production.
• (3) Insurance contracts embed dynamic incentive mechanisms that encourage technology adoption and best management practices. Standard crop insurance policies often incorporate features such as deductible amounts, coinsurance provisions, and experience-based premium adjustments that create incentives for policyholders to adopt loss-reduction measures and maintain prudent production practices (Tang aet al., 2019). Insurers may also provide technical assistance, early warning information, and conditional premium discounts for the adoption of recognized best practices, effectively serving as technology disseminators. Furthermore, the claims adjustment process itself generates valuable information on the agronomic and meteorological conditions underlying production losses, which can be fed back to farmers and extension services to inform future production decisions. This informational role of insurance, while often overlooked, constitutes an important but underappreciated mechanism through which insurance influences technology adoption at the farm level.

Collectively, these supplementary channels operate alongside the expectation-stabilization mechanism to reinforce the positive impact of comprehensive insurance on agricultural technology adoption, operational scale expansion, and ultimately, food production resilience. The empirical mediation analysis presented in Section 6 captures the aggregate effect of these multiple pathways, with agricultural research expenditure and land transfer rate serving as proxies for the combined influence of technological upgrading and scale expansion.

In sum, the theoretical analysis developed in this section yields three testable propositions that collectively constitute the core theoretical contribution of this study. Proposition 1 establishes that an increase in the coverage level and premium subsidy rate of policy-oriented agricultural insurance expands farmers' optimal efficient output scale by reducing the variance of net income and alleviating downside risk constraints, thereby enhancing food production resilience. Proposition 2 specifies the transmission mechanism: the resilience-enhancing effect operates through two mediating channels, namely agricultural technological upgrading and the expansion of operational scale, both of which represent conditional factor demands that increase with the efficient output scale. Proposition 3 articulates the moderating role of climate risk: as climatic hazard exposure intensifies, the probability of low-output events increases, and the marginal effect of insurance on production resilience becomes larger. These three propositions, derived from a unified microeconomic framework, move beyond the existing literature by providing a generalizable theoretical apparatus for understanding how insurance instruments interact with technological progress, structural transformation, and environmental risk to shape the resilience of food production systems.

4 Research design

4.1 Model construction

4.1.1 Baseline regression model

The staggered difference-in-differences (DID) design requires that policy assignment is exogenous conditional on covariates and fixed effects. The pilot provinces were selected by the central government based on observable criteria: grain production capacity, agricultural risk exposure, fiscal coordination capacity, and regional representativeness. The first cohort (2018) included six major grain-producing provinces (Hubei, Anhui, Henan, Shandong, Liaoning, Inner Mongolia); the second cohort (2021) expanded to seven additional provinces (Hebei, Jilin, Heilongjiang, Jiangsu, Jiangxi, Hunan, Sichuan). This selection mechanism implies non-random assignment correlated with observable agricultural and economic characteristics, potentially threatening exogeneity.

Nevertheless, the identifying variation in our staggered DID framework does not require random initial selection. Rather, it exploits the differential timing of policy adoption across provinces, conditional on province fixed effects (absorbing time-invariant unobserved heterogeneity), year fixed effects (absorbing common shocks), and time-varying covariates. The conditional parallel trends assumption is plausible because: (i) policy assignment was determined by central-level administrative rules rather than provincial self-selection; (ii) the expansion schedule was pre-announced based on time-invariant eligibility thresholds; and (iii) provinces could not strategically time their entry based on transitory resilience fluctuations.

Drawing on a panel of 31 Chinese provinces spanning the years 2013–2024, this study investigates the effect of the policy-oriented agricultural insurance reform on food production resilience. Because the reform was rolled out in a staggered fashion across provinces, we adopt a staggered DID identification strategy. This design isolates the causal impact by contrasting the evolution of the resilience outcome between provinces that had introduced the reform (the treated group) and those that had not yet adopted it (the control group), while conditioning on other relevant factors. The baseline regression model is specified as follows:

Yi,t=β0+β1DIDi,t+∑2nβncontroli,t+δi+μt+εi,t(41)

In Equation 41, Y_*i,t* denotes food production resilience in province i in year t. DID_*i,t* is the core explanatory variable and represents the treatment status of the full-cost and planting-income insurance pilot. It is constructed according to each province's first year of policy adoption. As shown in Table 1, for provinces in the first pilot cohort, DID_*i,t* equals 1 from 2018 onward and 0 before 2018. For provinces in the second pilot cohort, DID_*i,t* equals 1 from 2021 onward and 0 before 2021. For provinces that never entered the pilot program during the sample period, DID_*i,t* remains 0 in all years. Thus, DID_*i,t* is not a one-time post-policy dummy based on a common policy year, but a time-varying treatment indicator that reflects the staggered implementation of the policy across provinces.

Table 1. Evaluation index system for food production resilience.

Our primary coefficient of interest is β₁, which reveals the net effect of the policy on FPR. If β₁ is positive and significant, it indicates that the policy effectively enhanced resilience in the treatment provinces relative to the control provinces.control_*i,t* is a vector of control variables that may influence FPR,δ_*i* denotes province-fixed effects, μ_*t* denotes time-fixed effects, β₀ is the constant term, and ε_*i,t* is the stochastic error term.

4.1.2 Mediation effect model

To test Hypothesis 2, this study follows the recommendations of regarding mediation analysis in causal inference research (Jiang, 2022). Building upon the baseline regression, we construct the following models to test for mediation effects:

Mediatorit=α0+α1DIDit+θcontrolit+δi+μt+ϵit(42)

In Equation 42, Mediator_*it* denotes the mechanism variable, control_*it* is a vector of control variables. δ_*i* denotes province fixed effects, μ_*t* denotes year fixed effects, and ϵ_*it* is the random error term. The coefficient of interest is α₁. If α₁ is statistically significant, it indicates that the insurance pilot significantly affects the corresponding mechanism variable. Combined with the baseline estimation, this result provides suggestive evidence that the mechanism variable may be a potential channel through which full-cost and planting-income insurance enhances food production resilience. Since mechanism variables may also be affected by other unobserved factors, this study interprets the results as mechanism evidence rather than strict causal mediation effects.

4.1.3 Moderating effect model

To investigate the pathways through which policy-oriented agricultural insurance reform shapes food production resilience, we extend the baseline specification by interacting the policy variable with a continuous measure of climate change risk. All climate risk variables are mean-centered before constructing the interaction terms. The augmented model is specified as follows:

Yit=γ0+γ1DIDit+γ2ClimateitCit+γ3(DIDit×ClimateitCit)+θcontrolit+δi+μt+ϵit(43)

In Equation 43, Climate_*it*C_*it* denotes the mean-centered climate-risk variable. In the moderating effect model above, the primary focus is on the coefficient of the interaction term (DID_*it* × Climate_*it*C_*it*). The definitions of all other variables are the same as in the baseline regression model.

4.2 Variable selection

4.2.1 Dependent variable

The dependent variable of interest is food production resilience (FPR), a multidimensional and inherently dynamic construct. Prior scholarship has concentrated largely on the resistance and post-shock recovery capacities of grain systems, while giving comparatively limited consideration to their preventive functions, particularly the capacity to forestall climate-induced disruptions. Yet as climatic conditions exert an ever-larger influence on agricultural production, the ability to avert potential adverse impacts becomes an indispensable element of any thoroughgoing evaluation of production resilience. In the field of agricultural economics, the traditional model of ex post disaster relief concentrates on post-disaster loss compensation and emergency reconstruction. Such an approach not only incurs enormous economic losses and places heavy fiscal demands on relief efforts, but also fails to fundamentally resolve the pronounced vulnerability of agricultural systems. By contrast, agricultural economic resilience entails a governance approach characterized by a virtuous cycle of ex ante prevention and control, in-process response, ex post recovery, and transformation and upgrading (Zhou et al., 2026). It enables the agricultural industry to shift from passively reacting to risks toward proactively adapting to them, thereby providing systematic support for the stable development of the agricultural economy.

Drawing upon prior studies (Liu et al., 2026; Jiang et al., 2023) and following the conceptual framework of resilience outlined in the FAO's The State of Food and Agriculture 2021, we conceptualize food production resilience along five dimensions: risk prevention, risk resistance, adaptive recovery, transformative innovation, and sustainable development. We then apply the entropy method to construct a composite index. The full indicator scheme is reported in Table 1.

4.2.2 Core explanatory variable

The core explanatory variable in this study is DID, representing the policy-oriented agricultural insurance reform. The construction and assignment of this variable are based on official policy documents, including the Pilot Program for Full-Cost Insurance and Planting-Income Insurance for the Three Major Grain Crops and the Notice on Expanding the Implementation Scope of Full-Cost Insurance and Planting-Income Insurance for the Three Major Grain Crops.

• (2) Based on the relevant policy documents, six provinces (Hubei, Anhui, Henan, Shandong, Liaoning, and Inner Mongolia) were designated as the initial pilot regions in 2018; accordingly, their DiD indicator is coded as 1 from 2018 onward and 0 otherwise. Seven additional provinces (Hebei, Jilin, Heilongjiang, Jiangsu, Jiangxi, Hunan, and Sichuan) entered the pilot in 2021, with the DiD indicator set to 1 from 2021 onward and 0 before that. We adopt the full-cost and planting-income insurance pilots to capture the treatment of policy-oriented agricultural insurance. Relative to conventional crop insurance, this reform delivers more robust protection for grain production and permits a more accurate, in-depth assessment of the true impact and policy effectiveness of China's publicly supported agricultural insurance on food production resilience in a changing climate.

4.2.3 Control variables

Food production is influenced by a range of multidimensional factors, including the macroeconomic environment, production input structure, agricultural infrastructure, and grain market dynamics. Therefore, this study refers to relevant studies (Davis et al., 2020; Lee et al., 2024) and selects the level of urbanization development, the level of industrialization, the consumer price index (CPI) for food products, the degree of rural population aging, and freight volume as control variables.

4.2.4 Mediation variables

Following relevant studies (Carter et al., 2016), this paper identifies agricultural technology level and agricultural scale level as mediating variables. The agricultural technology level is measured by agricultural scientific research expenditure, representing technological advancement and innovation capacity. The agricultural scale level is measured by the land transfer rate, reflecting the degree of farmland consolidation and large-scale operations.

4.2.5 Moderating variable

The moderating variable in this study is climate change. In line with global warming trends, this study employs the climate warming risk index as a proxy for climate change (Montzka et al., 2011). Standard temperature measures, such as level, range, and extreme indicators, do not reliably separate the role of geographic location from that of climate warming itself. Accordingly, following Deschênes and Greenstone (2007), we capture climate risk using absolute temperature deviations. In particular, building on Wang et al. (2014), we operationalize climate change risk as the absolute standardized anomaly of the annual mean temperature relative to its long-term provincial average.

4.3 Data sources and descriptive statistics

Based on data availability, this study selects 31 provinces (municipalities and autonomous regions, excluding Hong Kong, Macao, and Taiwan) across China from 2013 to 2024 as the research sample. The main data are obtained from the China Statistical Yearbook, the China Rural Statistical Yearbook, the China Agriculture and Rural Affairs Database, and the China Meteorological Data Service Centre. During the indicator screening process, a small number of remaining missing observations were filled using linear interpolation.

5 Results analysis

5.1 Baseline regression

Table 2 presents the baseline estimation results. Column (1) reports the specification that omits control variables while including province and year fixed effects. The DID coefficient equals 0.0229 and is statistically significant at the 1% level, indicating that the full-cost and planting-income insurance pilot markedly strengthens food production resilience in the treated provinces. Column (2) augments the model with province-level covariates; the DID coefficient remains positive and significant at the 1% level, with an estimate of 0.0184. This finding confirms that, after accounting for provincial characteristics alongside province and year fixed effects, the policy effect stays positive and significant, lending further support to H1.

Table 2. Baseline estimation.

^***p < 0.01.

Robust standard errors in parentheses.

With respect to the covariates, the coefficient on industrialization level (sec) is significantly negative, likely reflecting the competitive pressures that industrial development exerts on agricultural production factors such as land and labor, which in turn constrains production resilience. In contrast, the coefficient on the grain consumer price index (cpi) is significantly positive, aligning with theoretical expectations: higher grain prices improve producers' revenue outlooks, incentivize greater input use and technology adoption, and thereby strengthen resilience.

Columns (3) through (5) further examine the pilot's effects on three selected dimensions of food production resilience. The dimension-specific estimates show that the insurance policy exerts significantly positive impacts on risk prevention, transformative and innovative development, and sustainable development. The positive effect on risk prevention suggests that higher-coverage insurance, by stabilizing income expectations and lowering risk exposure, enhances producers' capacity to maintain essential inputs and undertake disaster-prevention measures. The significant effect on transformative and innovative development indicates that insurance operates well beyond short-term loss compensation; by improving long-term business prospects and risk-bearing conditions, the policy encourages greater technological investment, adjustments to production practices, and adaptive transformation. The positive effect on sustainable development implies that high-coverage insurance helps sustain the continuity of grain production while reducing the adverse consequences of short-termist production behavior triggered by disaster shocks and market volatility. Taken together, full-cost and planting-income insurance systematically strengthens food production resilience through the combined mechanisms of ex ante risk prevention, adaptive transformation, and long-term sustainable development, providing further evidence in support of H1.

5.3 Robustness checks

5.3.1 Parallel trends test

To trace the temporal dynamics of the policy effect, we adopt an event-study framework and assess the parallel-trends assumption, using the year immediately before the reform as the reference period. The results are plotted in Figure 2. Prior to the reform, the point estimates are statistically indistinguishable from zero and hover around the null, confirming that no systematic divergence in food production resilience existed between pilot and non-pilot provinces before the intervention. From the implementation year onward, the coefficients become positive and statistically significant, and their magnitude grows progressively over time. This pattern indicates that the full-cost and planting-income insurance program raises food production resilience and that its effect endures over a longer horizon. These results validate the parallel-trends condition and strengthen the credibility of the baseline estimates.

5.3.2 Placebo test

To further evaluate the robustness of the findings, we conduct 1,000 placebo tests by randomly assigning pseudo treatment timing across provinces and re-estimating the model. This procedure yields a counterfactual distribution of the estimated treatment effects. Figure 3 displays the kernel density of the 1,000 placebo coefficients, and Table 3 reports the corresponding summary statistics. The placebo estimates are tightly centered near zero, with a mean of −0.0001, a median of −0.0001, and a standard deviation of 0.0035, implying that no systematic policy effect is detected under random assignment. By contrast, the actual DID estimate of 0.0184 lies far in the right tail of this distribution. The empirical p-value falls below 0.001, confirming that the true policy effect is statistically distinguishable from the placebo estimates.

Table 3. Placebo test results.

^*** denotes significance at the 1% level.

This table reports the results of 1,000 placebo randomizations. The empirical p-value is calculated as the proportion of placebo estimates whose absolute values are greater than or equal to the absolute value of the true DID estimate.

The logic underlying this exercise is straightforward. If the observed policy effect were driven by unobserved confounding factors rather than by the reform itself, then randomly reassigning the treatment timing to provinces should yield estimated effects that are as large as the true estimate, at least occasionally. Conversely, if the policy effect is genuine, the placebo coefficients should cluster around zero, and the true estimate should appear as an outlier in the placebo distribution. The results conform precisely to this latter pattern. The placebo estimates are centered tightly on zero, indicating that random policy assignment produces no systematic relationship with the outcome. That the true estimate of 0.0184 lies outside the range of nearly all placebo estimates provides strong evidence that the estimated policy effect is not a chance finding but is attributable to the insurance reform itself.

These results indicate that the positive effect of the policy on provincial food production resilience is unlikely to be driven by random factors, further supporting the robustness of the baseline findings.

5.3.3 Test based on PSM-DID

To address potential selection bias stemming from the non-random rollout of the pilot, we supplement the analysis with a propensity score matching difference-in-differences (PSM-DID) procedure. Using pretreatment provincial characteristics from 2013, we select industrialization, urbanization, the grain consumer price index, rural aging, and freight volume as matching covariates. A logit model is employed to estimate each province's propensity to be included in the full-cost and planting-income insurance pilot. Based on the estimated propensity scores, we construct matched samples using 1:4 nearest-neighbor matching and kernel matching, and then re-estimate the difference-in-differences specification.

Table 4 reports the covariate balance before and after matching. Prior to matching, the treatment and control groups differ markedly across the observed covariates, with a mean standardized bias of 61.66% and a median standardized bias of 86.84%. After matching, balance improves considerably: under 1:4 nearest-neighbor matching, the mean and median biases decline to 9.19 and 6.82%, respectively, while kernel matching yields corresponding values of 6.56 and 6.70%. Both methods retain a large majority of provincial observations, with sample retention rates of 87.10 and 83.87%, respectively, indicating that the matching procedure substantially enhances comparability between the two groups.

Table 4. PSM matching balance summary.

Balance statistics are based on the 2013 province-level matching sample.

Columns (1) and (2) of Table 5 present the PSM-DID estimates. The coefficient of interest equals 0.0147 and is statistically significant at the 5% level under 1:4 nearest-neighbor matching, and reaches 0.0193, significant at the 1% level, under kernel matching. These findings demonstrate that, even after controlling for observable differences between pilot and non-pilot provinces, agricultural insurance continues to significantly strengthen food production resilience. The baseline results are therefore unlikely to be driven by initial disparities in provincial characteristics.

Table 5. Replacement of the core explanatory and dependent variables.

^**p < 0.05.

^***p < 0.01.

Robust standard errors in parentheses.

5.3.4 Other robustness checks

To alleviate potential endogeneity concerns and further assess the reliability of the findings, we implement a series of supplementary robustness checks, including lagging the policy variable and winsorizing the continuous covariates.

We first replace the contemporaneous treatment indicator with its one-period lag (L.DID) to accommodate a possible delayed policy response. As shown in column (3) of Table 6, the coefficient on the lagged variable is 0.0171 and remains statistically significant at the 1% level, indicating that the reform's influence extends beyond the immediate implementation period and exerts a lasting, sustained effect on food production resilience. Second, we mitigate the influence of extreme observations by winsorizing all continuous variables at the 1st and 99th percentiles. The resulting estimate on DID, reported in column (4) of Table 6, is 0.0168 and retains significance at the 1% level. This stability confirms that the positive impact of the pilot on food production resilience is not driven by outliers.

Table 6. Robustness test.

^**p < 0.05.

^***p < 0.01.

Robust standard errors in parentheses.

5.3.5 Replacement of the core explanatory and dependent variables

We further assess the stability of the core findings by replacing both the key explanatory variable and the dependent variable. First, we move beyond the binary pilot indicator and instead capture the operational intensity of agricultural insurance using two alternative measures: total premium income and total claim payments. As reported in columns (1) and (2) of Table 5, the coefficients on premium income and claim payments are 0.0093 and 0.0132, respectively, both statistically significant at the 1% level. These results suggest that the development of agricultural insurance, whether reflected in expanded coverage or in higher indemnity levels, helps strengthen food production resilience, thereby corroborating the positive policy effect identified in the baseline estimation.

Second, we address the possibility that the measurement of food production resilience is sensitive to the chosen weighting procedure by constructing an alternative index using principal component analysis (PCA). Unlike the entropy method, PCA extracts the common variation shared across indicators to form a composite score, thus providing a different weighting strategy to test whether our conclusions hinge on a particular measurement technique. Column (3) of Table 5 shows that the coefficient on the policy variable remains positive and significant at the 5% level (coefficient = 0.012) when the PCA-based resilience score serves as the dependent variable. This indicates that the beneficial effect of the full-cost and planting-income insurance reform is not an artifact of the specific weighting method.

Overall, the estimates obtained with alternative explanatory and dependent variables are fully consistent with the baseline results. This pattern demonstrates that our main findings are not driven by a particular definition of the policy variable or by a specific technique for constructing the resilience index, further confirming the robust impact of agricultural insurance on food production resilience.

5.3.7 Robustness to treatment effect heterogeneity in staggered DID

Recent econometric literature has demonstrated that conventional two-way fixed effects estimators can produce biased estimates in staggered difference-in-differences settings when treatment effects are heterogeneous across cohorts and over time (Goodman-Bacon, 2021; Callaway and Sant'Anna, 2021; Sun and Abraham, 2021; Borusyak et al., 2024). The source of this bias lies in the fact that the TWFE estimator is a weighted average of all possible 2 × 2 comparisons, some of which use already-treated units as controls for later-treated units, generating “negative weights” that can distort the estimated treatment effect. Given that our policy roll-out occurred in two distinct cohorts (year 2018 and 2021), this concern is directly relevant to our empirical setting.

To diagnose the severity of this issue, we first implement the Goodman-Bacon (2021) decomposition. As reported in Table 7, the overall TWFE estimate of 0.0184 can be decomposed into three types of 2 × 2 comparisons: (i) early-treated vs. never-treated (weight = 0.421, estimate = 0.0197), (ii) late-treated vs. never-treated (weight = 0.385, estimate = 0.0189), and (iii) early-treated vs. late-treated (weight = 0.194, estimate = 0.0156). Critically, the “bad comparison” category, where early-treated units serve as controls for late-treated ones, which accounts for only 19.4% of the total weight, and its contribution to the overall estimate is modest. This diagnostic suggests that the conventional TWFE estimate is not severely contaminated by negative-weight bias.

Table 7. Goodman-Bacon decomposition and heterogeneity robust estimates.

^**p < 0.05.

^***p < 0.01.

Standard errors are clustered at the province level. The Callaway-Sant'Anna estimator uses “never-treated” as the control group with doubly robust estimation and 1,000 bootstrap replications.

To further ensure robustness, we re-estimate the treatment effect using the group-time average treatment effect estimator proposed by Callaway and Sant'Anna (2021), implemented via the csdid command in Stata. This approach addresses treatment effect heterogeneity by computing distinct average treatment effects for each cohort-time combination and then aggregating them into a summary parameter. We use the “never-treated” group as the control and employ the doubly robust estimation method with bootstrapped standard errors (1,000 replications). As shown in Table 7, the aggregated ATT from the Callaway-Sant' Anna estimator is 0.0168 (standard error = 0.0056), which is statistically significant at the 1% level and closely aligns with the baseline TWFE estimate of 0.0184. The magnitude and significance are nearly identical, confirming that the main findings are robust to the use of heterogeneity-consistent estimators.

We also estimate dynamic treatment effects using the event-study specification within the Callaway-Sant' Anna framework. The pre-treatment coefficients are statistically indistinguishable from zero, providing further support for the parallel trends assumption, while the post-treatment effects become positive and significant from the first year of treatment onward, with magnitudes increasing over time. Collectively, these diagnostic and robustness exercises indicate that treatment effect heterogeneity does not materially bias our baseline estimates. The positive and significant effect of the full-cost and planting-income insurance reform on food production resilience remains stable across alternative estimators that explicitly account for staggered adoption and heterogeneous treatment effects.

5.3.8 Eliminate the interference of concurrent policies

To address the concern that the estimated insurance effect may partially reflect the influence of other agricultural policies implemented during the same period, we augment the baseline specification with two additional control variables: the cumulative area of high-standard farmland and the per-unit-area agricultural support subsidy. As reported in Table 8, column (1) presents the baseline estimate for reference. Column (2) adds the high-standard farmland area as an additional control; the DID coefficient remains positive and statistically significant at the 1% level (0.0164, standard error = 0.0052). Column (3) adds the agricultural subsidy variable; the coefficient is 0.0169 (standard error = 0.0051), again significant at the 1% level. Column (4) includes both controls simultaneously; the DID estimate is 0.0158 (standard error = 0.0053), retaining significance at the 1% level. The marginal decline in the coefficient from 0.0184 to 0.0158, approximately 14.1%, suggests that a modest portion of the baseline estimate may be attributable to concurrent policies, but the insurance reform continues to exert a statistically and economically meaningful effect on food production resilience after partialling out these confounding influences.

Table 8. Robustness to concurrent agricultural policies.

^*p < 0.1.

^**p < 0.05.

^***p < 0.01.

Robust standard errors in parentheses.

6 Analysis of impact mechanisms

6.1 Mediation effect test

To verify H2, which states that policy-oriented agricultural insurance enhances FPR by increasing the agricultural technology level and the agricultural scale level, this study follows the mediation analysis framework proposed in prior literature (Jiang, 2022) and adopts a two-step mediation testing approach. Given that the impacts of agricultural technology advancement and agricultural scale expansion on production resilience are well-established in existing studies (Liu et al., 2025b; Tang et al., 2019; Chen and Gu, 2025; Liu et al., 2026), this study focuses on whether the core explanatory variable affects these mechanism variables, thereby providing empirical evidence on potential policy channels.

The estimates displayed in columns (1) and (3) of Table 9 indicate that the DID coefficients are positive and highly significant (p < 0.01), confirming that publicly supported agricultural insurance substantially lifts food production resilience, a finding fully consistent with the baseline specification. Columns (2) and (4) turn to the mechanism analysis; the DID coefficients for agricultural technological progress and large-scale farming operations are both positive and significant at the 1% level, suggesting that the reform drives agricultural R&D investment and the consolidation of farmland into larger operational units.

Table 9. Mediation effect test.

^***p < 0.01; ^**p < 0.05.

Robust standard errors in parentheses.

These results lend empirical support to the view that technological advancement and scale expansion serve as important transmission channels. On the one hand, insurance stabilizes income expectations and heightens the demand for adaptive technologies in grain production, with greater R&D investment potentially promoting advances in seed breeding, water-saving irrigation, and disaster-prevention measures. On the other hand, by strengthening income protection for scale operators and reducing the uncertainty surrounding land transfers and farm expansion, insurance bolsters expectations of production growth and facilitates the concentration of land resources among larger farms. Overall, the mechanism evidence aligns closely with theoretical predictions and the existing body of literature (Chen and Gu, 2026; Ma et al., 2024; Su et al., 2025). These results imply that policy-oriented agricultural insurance may strengthen food production resilience by fostering technological progress and the expansion of operational scale in agriculture, lending empirical support to H2.

To empirically examine the complementarity between the two mediating channels, we augment the baseline specification with an interaction term between agricultural technology level and agricultural scale level. As reported in Table 9 (column 5), the interaction coefficient is positive (0.0084) and statistically significant at the 5% level, indicating that the resilience-enhancing effect of technological progress is amplified in provinces with higher levels of operational scale, and vice versa. This result corroborates the theoretical discussion in Section 3.2 that technology and scale operate as mutually reinforcing rather than independent mediating channels. The finding also suggests that the insurance reform may generate compounded benefits when both mechanisms are simultaneously activated, highlighting the importance of coordinated policy efforts that address both technological diffusion and land consolidation.

6.2 Moderating effect test

Recognizing that climate change may alter the boundary conditions under which agricultural insurance operates, we further test whether climate risk moderates the resilience-enhancing impact of the insurance reform. The continuous climate risk variable is mean-centered before constructing the interaction term. As reported in Table 10, the DID coefficient is 0.0181 (p < 0.01), confirming that the pilot program significantly strengthens food production resilience at the average level of climate risk. The interaction term carries a coefficient of 0.0155 and is statistically significant at the 5% level, indicating that climate risk amplifies the policy's beneficial effect on resilience.

Table 10. Moderation effect test.

^**p < 0.05.

^***p < 0.01.

Robust standard errors in parentheses.

Table 11 reports the marginal effects evaluated at low, average, and high levels of climate risk. At low climate risk, the marginal effect is 0.0110 (p < 0.10), corresponding to an economic magnitude of 4.07%. At the mean risk level, the marginal effect rises to 0.0181 (p < 0.01) with an economic significance of 6.69%. At high climate risk, the marginal effect further increases to 0.0251, reaching an economic significance of 9.30%. This gradient indicates that the marginal contribution of the insurance policy to food production resilience grows monotonically with the intensity of climate risk.

Table 11. Marginal effects of the insurance pilot at different climate-change levels.

^*p < 0.1.

^***p < 0.01.

Robust standard errors in parentheses. Low, average, and high climate-risk levels are defined as mean −1 SD, and mean +1 SD, respectively. Economic significance equals marginal effect divided by the sample mean of FPR times 100.

This finding is consistent with studies showing that agricultural insurance can help mitigate climate-related risks (Lee et al., 2024; Madaki et al., 2023). A plausible interpretation is that climate risk intensifies the production constraints confronting farmers. In high-risk regions, climatic shocks are more apt to materialize as yield and income volatility, raising the likelihood of operational disruptions and constraining post-shock reinvestment capacity. Under these circumstances, insurance can sustain production continuity and foster recovery through risk pooling and income stabilization, thereby magnifying its impact on food production resilience. By contrast, where climate risk is low, output fluctuations remain relatively contained, and risk constraints exert a weaker influence on production choices; accordingly, the incremental benefit of insurance is smaller. These patterns imply that the value of agricultural insurance as a risk management instrument becomes more pronounced as climate risk escalates. Under conditions of elevated risk, insurance functions not only as a loss-compensation mechanism but also as a critical climate adaptation strategy for agricultural producers. This finding accords with theoretical expectations and provides empirical backing for Hypothesis 3.

6.3 Heterogeneity analysis

6.3.1 Heterogeneity test of disaster-damaged rate

The preceding analysis concentrated on the moderating role of policy-oriented agricultural insurance under chronic climatic pressure, captured by temperature deviations. This section investigates whether the policy effect displays an analogous pattern when actual disaster shocks are taken into account. Following related studies, we split the sample into high and low disaster-damage regions using the sample mean as the threshold. The disaster damage rate is defined as the ratio of the disaster-damaged area to the total disaster-affected area in a given year.

Table 12 reveals that the estimated coefficient on the treatment indicator reaches 0.0357 and is statistically significant at the 1% level for regions with high disaster damage rates, which implies that the insurance reform markedly strengthens food production resilience in these areas. By contrast, the corresponding coefficient for the low damage rate subsample fails to attain conventional significance, pointing to a substantially weaker policy influence. A Chow test yields a p-value of 0.000, confirming that the difference in insurance effects between the two groups is statistically distinguishable.

Table 12. Heterogeneity test of disaster-damaged rate.

^***p < 0.01.

Robust standard errors in parentheses.

These findings imply that where disaster shocks are more severe, the risk-sharing and post-disaster production-support functions of insurance are more readily mobilized, thereby amplifying its contribution to production resilience. Observational evidence from China corroborates this interpretation. Publicly available statistics from the National Financial Regulatory Administration indicate that in 2023, 14.479 million mu of grain crops in Henan Province sustained disaster damage, and agricultural insurance indemnities reached 3.87 billion CNY, substantially cushioning the impact on farmers' ability to resume production. Existing studies also show that agricultural insurance tends to have stronger policy effects in areas with higher agricultural production risks (Li et al., 2025a), which is consistent with the findings of this study.

Taken together, the moderation analysis and the heterogeneity assessment across disaster damage levels indicate collectively that the beneficial impact of agricultural insurance on food production resilience intensifies in settings characterized by elevated risk, regardless of whether that risk is proxied by sustained climatic pressure or by realized disaster losses. This evidence further implies that comprehensive insurance coverage can act as a climate adaptation instrument, operating through the stabilization of income expectations and the promotion of long-term production viability, thereby bolstering the robustness of the core conclusions.

6.3.2 Heterogeneity test based on different regions

Owing to China's expansive geography, its eastern, central, and western regions diverge markedly in economic development, agricultural production structures, natural resource endowments, and the intensity of policy support. As a result, the effectiveness of policy-oriented agricultural insurance is likely to exhibit pronounced regional heterogeneity.

To examine this issue in greater depth, this study, referencing relevant research (Chen and Gu, 2025), we split the sample into three regional subgroups, namely eastern, central, and western, and estimate separate regressions. Table 13 reveals that the insurance coefficient remains positive and statistically significant across all three regions. The effect attains significance at the 10% level in the eastern and central regions, whereas it reaches the 1% level in the western region, where the magnitude is largest. A Chow test yields a p value of 0.000, confirming that the policy effects differ statistically across regions. These findings indicate pronounced regional heterogeneity in the influence of the insurance pilot on food production resilience.

Table 13. Regional heterogeneity test.

^*p < 0.1.

^***p < 0.01.

Robust standard errors in parentheses.

The regional heterogeneity findings indicate that the efficacy of the insurance reform diverges notably across regions. The more pronounced effect observed in western China may partly reflect the combined influence of fiscal support and the existing base for large-scale farming. First, under the central government's administrative rules for agricultural insurance premium subsidies, the central fiscal subsidy covers 45% of crop insurance premiums in the central, western, and northeastern regions, whereas the subsidy rate stands at 35% in the eastern region. This higher subsidy share reduces participation costs and enhances both the accessibility and the depth of insurance coverage. Second, the western region benefits from a comparatively strong foundation for scale-based operations. Data from the Third National Agricultural Census show that the western region hosted 4.11 million large-scale agricultural operators, accounting for 31.8% of the national total and surpassing the shares recorded in the eastern and central regions. Large-scale operators typically exhibit stronger insurance demand and greater awareness of risk management, making it easier for insurance coverage to translate into stabilized input use and operational adjustment. Hence, through the joint operation of fiscal subsidies and large-scale farming, the insurance intervention in the western region is better placed to sustain production continuity and resilience, thereby producing a larger marginal impact.

The insurance effect in the eastern region remains positive and statistically significant at the 10% level, suggesting that the policy still contributes to food production resilience to a meaningful degree. This pattern may be associated with the region's more developed market environment, a more mature financial system, and greater insurance service capacity (Zhao and Cui, 2024). Such conditions can improve policy outreach, underwriting efficiency, and claims processing, thereby helping to translate risk coverage into production-stabilizing outcomes. At the same time, the grain sector accounts for a comparatively small share of the eastern regional economy, which narrows the scale of the policy's target population. This structural feature may explain why the marginal effect and statistical significance are less pronounced than those observed in western China.

In central China, the insurance effect remains positive and significant, yet the estimated coefficient is the smallest of the three regions. One plausible interpretation is that central China, as a core grain-producing area, already benefits from a relatively well-established agricultural insurance support system, which allows the reform to generate a positive impact. However, compared with eastern and western China, grain production in this region still relies more heavily on smallholder and fragmented farm structures, characterized by limited land transfer activity and weaker scale economies (Deininger et al., 2014; Jia and Petrick, 2014). In such an environment, insurance serves mainly as a household-level safety net, cushioning individual losses without triggering substantial reorganization of production structures or factor deployment. This restricted scope dampens its aggregate contribution to food production resilience.

7 Conclusions

Against the backdrop of intensifying climate change pressures, we exploit the staggered rollout of full-cost and planting-income insurance pilots as a quasi-natural experiment. Drawing on a provincial panel covering China from 2013 to 2024, we apply a staggered difference-in-differences design to systematically assess the effect of high-coverage agricultural insurance on food production resilience, and to explore its transmission mechanisms and heterogeneous responses across varying risk environments. Beyond the specific empirical findings, this study makes several broader theoretical contributions. First, by formalizing the “insurance–technology–scale–resilience” transmission chain within a unified analytical framework, we demonstrate that the resilience-enhancing effect of agricultural insurance operates not merely through ex post income stabilization, but more fundamentally through the reconfiguration of producers' ex ante investment incentives in technology and land consolidation. This theoretical insight challenges the conventional characterization of agricultural insurance as a passive loss-compensation device and reframes it as an active driver of structural transformation in food production systems. Second, the positive moderating effect of climate risk identified in this study advances the theoretical understanding of the contingent value of risk management instruments: the marginal return to insurance is not constant but increases with environmental hazard intensity, implying that insurance and climate adaptation are complements rather than substitutes in the resilience-building process. Third, the five-dimensional resilience measurement framework proposed herein offers a methodological foundation for future cross-country comparative research, enabling scholars to move beyond output-centric assessments toward a more comprehensive evaluation of production system robustness. These theoretical contributions, while grounded in China's institutional context, are generalizable to other developing economies confronting analogous challenges of climate vulnerability, agrarian structural transformation, and agricultural risk governance.

Regarding the aggregate impact, the reform substantially strengthens China's food production resilience. Descriptive analysis reveals that the overall resilience index experienced a marked acceleration following the 2018 pilot launch. The causal estimates corroborate this pattern, demonstrating that high-coverage, policy-oriented insurance effectively bolsters the resilience of the food production system. This finding withstands an extensive battery of robustness checks.

Mechanism analyses indicate that the insurance reform significantly spurs agricultural technological advancement and the expansion of large-scale farming, lending empirical credence to the view that these two factors function as possible transmission channels. Concretely, the policy appears to strengthen resilience by improving the enabling environment for technology adoption and by fostering operational consolidation, thereby delivering technological support and a scale-related foundation for enhanced resilience.

Moreover, climate risk positively moderates the resilience-enhancing impact of the insurance reform. Marginal effect estimates reveal that the policy's contribution to food production resilience intensifies monotonically as climate risk escalates, suggesting that agricultural insurance, as a risk-management tool, commands greater marginal value in settings exposed to elevated hazards.

Heterogeneity analysis further documents pronounced disparities in the policy effect across regions and across areas distinguished by disaster damage intensity. In particular, the effect is strongest in western China and is amplified in localities that suffer higher disaster damage rates. This pattern indicates that the policy's impact is more readily realized where fiscal support is more generous, where the groundwork for large-scale farming is more solid, and where the strain from disaster shocks is more acute.

Beyond their empirical relevance to policy design, the findings carry broader implications for agricultural risk governance. The positive moderating role of climate risk suggests that comprehensive insurance assumes increasing marginal value precisely where environmental pressures are most acute, repositioning it from a compensatory device to an adaptive strategy. The mediating roles of technological upgrading and scale expansion imply that insurance alone is insufficient; complementary investments in extension services and land consolidation are essential to unlock its full potential. The pronounced regional heterogeneity further indicates that uniform subsidy formulas are inefficient; fiscal resources should be directed disproportionately toward high-risk, high-impact regions. These insights, while grounded in China's institutional context, offer reference points for other developing economies confronting analogous challenges of climate vulnerability and agrarian structural transformation.

8 Limitation

Several limitations of this study warrant caution in the interpretation and practical application of our findings. (1) Constrained by data availability, we rely on macro-level provincial panel data, which makes it difficult to fully dispel concerns about omitted variable bias. The findings on agricultural technological progress and large-scale farming operations should therefore be interpreted as suggestive of possible transmission channels rather than as evidence of strict causal mediation. Future research could draw on micro-level data from farm households or agricultural enterprises to examine more precisely how agricultural insurance shapes technology adoption, land transfer decisions, and production input behavior, thereby pinpointing the underlying mechanisms with greater rigor. (2) The empirical analysis presented here captures only average treatment effects, thereby obscuring substantial heterogeneity across individuals, farms, and regional contexts. This aggregation necessarily masks the diverse pathways through which policy-based insurance may influence agricultural production, and our estimates should not be construed as universally applicable to all subpopulations. (3) The generalizability of our conclusions is constrained by the specific institutional and socio-economic context of China. While our findings may hold relevance for countries with comparable national conditions, particularly those characterized by similar agricultural governance structures, smallholder-dominated production systems, and evolving rural financial markets, they cannot be extrapolated to contexts with fundamentally different institutional arrangements or developmental trajectories. (4) And perhaps most critically, our results do not lend themselves to prescriptive guidance for practice. The policy implications we derive are necessarily general in nature, functioning more as broad appeals for attention than as actionable recommendations for specific interventions. This limitation stems from the inherent gap between population-level econometric estimates and the situated, context-dependent decisions that practitioners and policymakers must make. Given these constraints, we strongly encourage future research to adopt case study methodologies that can provide rich, contextualized explanations of how policy-based insurance actually operates to enhance agricultural production. Such qualitative and mixed-methods approaches would complement our quantitative findings by illuminating the mechanisms, enabling conditions, and lived experiences that aggregate analyses cannot capture, thereby offering more nuanced and practically relevant insights for both scholarship and policy development.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Mengying Feng, University of Roehampton London, United Kingdom

Reviewed by: Lidong Shi, Southwest University of Political Science & Law, China

Yong Li, University of Science and Technology of China, China