Intensity of crop and livestock insurance adoption: lessons from Mexico

Climate change poses escalating risks to agricultural communities worldwide, threatening crop yields, livestock health, and rural livelihoods (Rusmayandi et al. 2023; Mehraj et al. 2022). Over the past three decades, disasters have caused an estimated loss of USD 3.8 trillion in crop and livestock production, disproportionately affecting lower-middle-income countries where agricultural GDP losses range from 10% to 15% (Food and Agriculture Organization of the United Nations 2023). These impacts highlight the need to implement risk management strategies to protect vulnerable farmers from future natural disasters.

Among the various options for risk management, agricultural insurance plays an increasingly important role in managing production risk (Hill et al. 2019; Shahi Kiran and Umesh 2012; European Commission 2018), providing a financial buffer (Food and Agriculture Organization of the United Nations 2021; Chatterjee and Oza 2017), stabilizing farmers’ incomes (Cop et al. 2023; Bhuiyan et al. 2022), encouraging investment in modern techniques, smoothing negative shocks, improving resource allocation (Carballo and Dos Reis 2013), and providing socioeconomic stability and resilience (Epetimehin 2011).

Agricultural insurance is broadly classified into two groups: indemnity insurance, which compensates the policyholder for actual losses based on a loss verification exercise (Swiss Agency for Development and Cooperation (SDC) et al. 2022), and index insurance, which uses predetermined measurable parameters associated with weather or other relevant factors to calculate payouts (Miranda and Farrin 2012). Although indemnity insurance provides precise and tailored coverage that incorporates the risks and needs of individual farmers, its high administrative costs and susceptibility to moral hazard make it impractical for small-scale farmers (Hazell 1992; World Bank 2011). Index insurance has been promoted as an alternative solution for smallholders, offering rapid payouts in the event of adverse weather conditions (World Food Programme and International Fund for Agricultural Development 2011) and theoretically lowering costs by eliminating the need for loss assessment (Miranda and Farrin 2012).

In Mexico, these risks are compounded by socioeconomic vulnerabilities and the discontinuation of the CADENA program–a federal initiative that combined indemnity and index insurance to protect smallholder farmers. Mexico’s CADENA program–a government initiative combining indemnity and index insurance–illustrates this paradox. While the program insured around 8 million hectares of crops and 4.2 million head of livestock, covering 56% of the targeted smallholder population, it was ultimately discontinued (World Bank 2013). This outcome raises important questions about the drivers of insurance adoption at a macro scale.

Although research in developing countries has identified a range of adoption drivers, covering risk, behavioral, demographic, biophysical, and economic factors (Tang et al. 2021; Carrer et al. 2020; Takahashi et al. 2019; Nordmeyer and Mußhoff 2023; Cole et al. 2013; Marr et al. 2016)–these studies largely focus on individual-level decisions and fail to capture the institutional patterns that shape government-run insurance programs. Notably, no studies have examined the determinants of insurance adoption specific to Mexico’s CADENA program. For example, De Janvry et al. (2016) explored the post-impact of CADENA insurance payments on land cultivation and household expenditures, while Ritchie (2015) assessed the effects of weather-indexed insurance on small-scale farmers’ productivity. Fuchs and Wolff (2011) highlighted unintended consequences such as disincentives to invest in non-insured crops. However, these studies treat adoption as a discrete outcome and do not explore variation across regions and time.

This study addresses these gaps by examining the intensity of crop and livestock insurance adoption across Mexican states from 2003 to 2014. We distinguish between index-based and indemnity-based products and explore how product design affects uptake. Specifically, we ask:

1. 1.

   What are the factors that drive agricultural insurance adoption at the state level?

2. 2.

   How do these determinants differ between crop and livestock insurance?

3. 3.

   What adoption patterns exist between index and indemnity insurance?

To answer these questions, we conduct an empirical analysis using panel data from 2003 to 2014, focusing on the CADENA program’s implementation across Mexican states. Our findings contribute to both the academic literature and policy design in several important ways.

This study makes several important contributions to the literature on agricultural insurance and to the design of policy instruments aimed at enhancing climate resilience in agriculture. First, by analyzing the intensity of insurance adoption at the state level, rather than treating adoption as a binary outcome at the household level, the study provides a macro-level perspective that captures institutional and structural dynamics shaping insurance uptake. This approach offers a more comprehensive understanding of how public programs such as CADENA operate across diverse regional contexts. Second, the study distinguishes between crop and livestock insurance, and further disaggregates index-based and indemnity-based products. This differentiation reveals that the determinants of adoption are not uniform across product types, underscoring the need for tailored insurance schemes that reflect the distinct risk profiles and operational realities of different agricultural sectors. Third, the use of Poisson, Quasi-Poisson, and Negative Binomial count models allows for robust estimation of adoption intensity while addressing overdispersion in the data. This methodological rigor enhances the reliability of the findings and contributes to the growing body of empirical work on insurance uptake using count-based approaches. Fourth, the results yield policy-relevant insights. For instance, the positive association between poverty levels and crop insurance adoption suggests that targeted subsidies may be more effective than universal ones in expanding coverage among vulnerable populations. Conversely, the negative association between risk-sharing arrangements and insurance uptake highlights the need to better integrate formal insurance with existing informal safety nets. Finally, the findings have implications for the design of future insurance programs. Improving the benefit-to-premium ratio in livestock insurance could significantly increase adoption, while recognizing the substitutive role of direct support programs such as "Apoyos Directos" may help avoid crowding out formal insurance mechanisms. These insights can inform the development of more inclusive and responsive agricultural insurance strategies in Mexico and similar contexts.

The remaining paper is organized in the following manner: Section 2 presents the literature that motivates our examination of the intensity of agriculture insurance adoption. Section 3 describes the conceptual framework and variables, sources of data, and discusses the count models used to explain the intensity of adoption of agriculture insurance, and the products for which the intensity is measured. Section 4 analyzes the results of the estimated count models. Lastly, Section 5 summarizes the study’s findings.

Agricultural insurance adoption remains uneven globally, despite its potential to mitigate climate-related risks (Tang et al. 2021; Cole et al. 2013). Although over 100 countries have implemented agricultural insurance programs, and initiatives such as the World Food Programme’s index insurance protect millions, uptake remains low in many regions (Mahul and Stutley 2010; World Food Programme 2021; World Bank 2013). Globally, only 45% of economic losses from natural catastrophes were insured in 2022 (Banerjee et al. 2023). In addition, studies show low formal insurance uptake, for example, 16.30% in 16 developing countries surveyed during 2011-2018 (Panda et al. 2020) and 14% among smallholder food crop farmers in Ghana (Njue et al. 2018). This low adoption of agricultural insurance presents a puzzle for researchers and practitioners aiming to mitigate agricultural risk effectively.

2.1 The CADENA program in Mexico

Mexico, the 14th largest economy worldwide with a population of 128 million, heavily relies on crop and livestock farming, which contributed 59% and 41% respectively to its agricultural output in 2021 (Organisation for Economic Co-operation and Development 2023). Its economy and population are highly susceptible to extreme weather events. For example, the severe droughts in 2011-2012 and 2021 that significantly reduced food production and increased food poverty (United Nations Office for Disaster Risk Reduction 2012; NASA Earth Observatory 2021). In light of these extreme weather events, the CADENA program is one of the climate adaptation strategies that was established by the Mexican government (World Bank 2013).

Mexico’s CADENA program offers a unique lens into macro-level insurance adoption. Launched in 2003 and overseen by the Ministry of Agriculture, Livestock and Fisheries (SAGARPA), the CADENA program provided indemnity and index-based insurance to protect smallholder farmers from climate-related disasters. Designed as a social safety net, it covered losses from events such as droughts, floods, and frost, addressing crop failure, livestock mortality, and feed shortages. CADENA was especially critical given the limited viability of commercial insurance for smallholders, but was ultimately discontinued in 2018.

Throughout its operational lifespan, the CADENA program underwent several structural transformations. It began as the Fund for Assisting Rural Populations Affected by Weather-Related Disasters (FAPRACC) in 2003, was rebranded as the Programme for Adopting Climate Change Strategies (PACC) in 2008, and in 2011, it was integrated into the broader Programa de Prevención y Atención de Riesgos (PAR), assuming the name CADENA.

CADENA offered support through two main pathways: "Apoyos Directos" (direct financial aid to affected agricultural producers) and "Seguro Catastrófico" (catastrophic insurance coverage). For the "Apoyos Directos", the federal and state governments would share the risk associated with catastrophic disasters, each contributing towards assisting affected farmers. On the other hand, the "Seguro Catastrófico" encompassed four macro-level agricultural insurance products, including both indemnity and index insurance options tailored for crop and livestock farming. Notably, the federal or state government acted as the policyholder for these disaster risk transfer instruments and was responsible for paying premiums rather than individual farmers. The state government would also receive premium support from the federal government when it purchased insurance. In the event of a loss, the state government would receive the payout for further distribution to affected farmers. CADENA achieved widespread coverage in Mexico by having the state and federal government pay the insurance premium instead of individual farmers. Its termination provides a compelling case for analyzing and understanding the underlying program-specific issues associated with the adoption of agricultural insurance. Previous studies have focused on the impacts of agriculture insurance on productivity and cultivated land (Ritchie 2015; De Janvry et al. 2016) and the consequences of insurance on crop choices (Fuchs and Wolff 2011), ignoring the factors that influenced adoption or participation intensity at the state level.

2.2 Factors influencing agricultural insurance adoption

To understand the reasons behind the low adoption of agricultural insurance, previous research has examined multiple factors, which can be broadly categorized into risk, behavioral, demographic, biophysical, and economic factors^{Footnote 1} .

Risk and behavioral factors are important because they shape policyholder decisions. Loss aversion can influence insurance demand by affecting perceptions of benefits and willingness to pay (Lampe and Würtenberger 2020). Risk-averse policyholders are generally more likely to adopt index insurance (Nordmeyer and Mußhoff 2023), and this aversion can lead to technology uptake when bundled with insurance (Visser et al. 2020). The ability of insurance to provide timely indemnity payments also builds trust, which is essential for confidence and willingness to adopt (Biswal and Bahinipati 2022; Mahul et al. 2012; Cole et al. 2013; Sinha and Tripathi 2016; Madaki et al. 2023; Nordmeyer and Mußhoff 2023; Bao et al. 2021; Boyd et al. 2011; Fahad et al. 2018; Gassler and Rehermann 2022). Factors like perceived barriers, risk attitude, past adoption, policy changes, and self-coping strategies in complex and uncertain situations are linked to trust (Giampietri et al. 2020).

Socio-demographic and biophysical factors also play a significant role. Socio-demographic factors such as education is consistently associated with a better understanding of insurance and higher adoption likelihood (Nordmeyer and Mußhoff 2023; Carrer et al. 2020; Tang et al. 2021; Madaki et al. 2023; Amare et al. 2019). The influence of age is conflicting, with some studies suggesting younger policyholders are more likely to purchase (Doherty et al. 2021), while others indicate older, more experienced farmers show higher adoption (Sherrick et al. 2004; Giampietri et al. 2020; Yufei et al. 2022). Gender also plays a role, with women farmers sometimes less willing to pay due to lack of trust or financial literacy (Nordmeyer and Mußhoff 2023; Ankrah et al. 2021; Akter et al. 2016). Family or household size can influence adoption either negatively or positively: negatively due to low disposable incomes (Njue et al. 2018; Showers and Shotick 1994) and positively due to low risk tolerance towards agricultural losses (Olowa and Olowa 2020; Wassie et al. 2025). On the biophysical side, farm (herd) size, soil quality, and land tenure have been identified as contributors to insurance adoption (Nordmeyer and Mußhoff 2023; Carrer et al. 2020; Yufei et al. 2022; Singh and Chandel 2019; Dahal et al. 2022).

Economic factors, particularly wealth and liquidity constraints, are strongly correlated with insurance adoption. Policyholders facing liquidity constraints may be financially challenged and unwilling to buy insurance at prices that cover the cost (Tadesse et al. 2017). This is particularly relevant as the frequency of adverse events increases (Leblois et al. 2020). Thus, offering discounts or rebates on premiums has been suggested as a solution to increase uptake in smallholder agriculture (Hill et al. 2019). Farmers are sensitive to cost, with higher premiums reducing the demand for insurance Kong et al. (2011). In addition, farm income positively affects willingness to adopt, as it directly influences their ability to pay premiums (Amare et al. 2019; Abugri et al. 2017; Ntukamazina et al. 2017).

2.3 Intensity of adoption and state level dynamics

While existing literature on agricultural insurance adoption provides valuable insights into the factors, challenges, and strategies influencing policyholders’ decisions, a significant gap remains concerning the intensity of adoption. Prior studies have mainly focused on adoption as a discrete outcome (whether or not a policyholder adopts), whereas the intensity of adoption refers to the level or extent to which a product or service has been adopted (Kyire et al. 2023). Researchers emphasize that factors affecting simple adoption may differ from those affecting the intensity of use (Martey and Kuwornu 2021). Studies that do measure adoption intensity primarily focus on a diverse array of agricultural innovations, such as (Mishra et al. 2018; Thompson et al. 2022; Yang et al. 2022; Bamire et al. 2010; Bopp et al. 2019), precision agricultural technologies Kolady et al. (2021a); Mozambani et al. (2023); Palma-Molina et al. (2023), conservation agriculture practices Pedzisa et al. (2015); Ngaiwi et al. (2023); Arslan et al. (2014); Kunzekweguta et al. (2017); Akter et al. (2021), climate-smart agriculture technologies Mujeyi et al. (2022); Mthethwa et al. (2022); Zakaria et al. (2020); Aryal et al. (2018); Sardar et al. (2021); Teklu et al. (2023), and technology adoption Jara-Rojas et al. (2020); Mgendi et al. (2022); Miine et al. (2023).^{Footnote 2} These studies delve into the factors influencing the depth and extent of farmers’ uptake of these innovations but overlook the adoption intensity of agricultural insurance.

Furthermore, most existing studies have concentrated on individual policyholder-level factors, with limited research investigating the macro-level or state-level factors that may also impact agricultural insurance adoption. Specific to the discontinued CADENA program in Mexico, although its impacts have been explored, there is a distinct gap in understanding the underlying drivers of the adoption of agricultural insurance at the macro level. This study fills this gap by examining the factors determining the intensity of crop and livestock insurance adoption within the context of the CADENA program’s index and indemnity insurance products, providing richer insights at the sub-national level than past studies.

3.1 Conceptual framework and hypothesis

This study explores how biophysical, socio-demographic, risk, and economic factors are associated with the intensity of crop and livestock insurance adoption across Mexican states. The framework is grounded in existing literature and contextualized within the CADENA program.

3.1.1 Biophysical

We examined the biophysical factor using farm (herd) size to serve as an indicator of the scale of operations. Larger farms or herds often imply greater exposure to potential losses from climate events or diseases, potentially increasing the perceived need for formal insurance to protect significant assets. However, researchers report contradicting results regarding the direction of the relationship between farm (herd) size and insurance adoption. Some studies suggest that larger farm sizes are associated with higher adoption rates (Wik et al. 2004). Conversely, other research indicates that smaller, subsistence farms might be more inclined to adopt certain types of insurance, especially credit-linked crop insurance (Duchoslav and Van Asseldonk 2018), or have lower adoption rates (Gassler and Rehermann 2022; Saliu et al. 2010; Aditya et al. 2018). For livestock, an increase in herd size has been consistently found to positively influence the adoption of livestock insurance (Singh and Chandel 2019; Dong et al. 2020; Madaki et al. 2023). We hypothesize that farm (herd) size is associated with the intensity of crop and livestock insurance adoption, respectively, reflecting variations in exposure and capacity to manage agricultural risks in different Mexican states.

3.1.2 Socio-demographic

The population variable is utilized as a macro-level proxy to reflect aggregated household-level dynamics and vulnerability, providing a rationale for its expected influence on agricultural insurance adoption in Mexico. Drawing from literature, a larger population within an area implies a greater number of livelihoods potentially vulnerable to disasters (Clarke and Dercon 2016). In addition, larger households are more likely to adopt innovative strategies under a risky environment (Abdulai et al. 2008). Given Mexico’s huge population size and its vulnerability to climate shocks, it would be expected that state governments with larger populations would necessitate greater adoption of agriculture insurance for the protection of livelihoods and food security. Therefore, we hypothesize that population is positively associated with the intensity of both crop and livestock insurance adoption.

We also employed the poverty or marginality index (used interchangeably), which measures the social and economic hardships in an area. They were utilized to assess how the economic vulnerability of a state government can influence adoption. Extreme climate events, such as droughts and floods, often lead to a loss of economic opportunities. This affects mainly the underprivileged population, significantly increasing the poverty levels (Duchoslav and Van Asseldonk 2018; Urama et al. 2019; Farhan and Hassan 2018; Ahmed et al. 2009). For these populations, agricultural insurance can serve as a vital safety net, helping them avoid falling into poverty traps (Noritomo and Takahashi 2020). Therefore, an increase in poverty levels can influence the adoption of agricultural insurance as a crucial mitigation tool. In 2002, before the introduction of the CADENA program, 51.7% of the Mexican population lived in poverty, and this was widespread (Garza-Rodriguez 2015). This creates the expectation that states with higher poverty or marginality index will adopt agricultural insurance to protect the poor. We hypothesize that poverty levels or marginality index is positively associated with the intensity of crop and livestock insurance adoption.

3.1.3 Risk and economic

Informal risk-sharing arrangements are a non-insurance alternative and a common coping strategy among agricultural communities, especially in developing economies. The relationship between formal insurance uptake and informal risk-sharing mechanisms is complex and often debated in the literature. Some studies suggest that informal risk-sharing can complement formal insurance by covering losses when an index insurance contract fails due to basis risk (Mobarak and Rosenzweig 2013; Berg et al. 2022). This leads to a higher adoption of formal insurance by mitigating its limitations. Conversely, other research indicates a negative impact, suggesting that informal risk-sharing arrangements might act as substitutes, reducing the perceived need for formal insurance (Boucher and Delpierre 2014). The complementary or substitution effect of the informal risk-sharing arrangement depends on effective risk aversion (Annan and Datta 2022). The CADENA program had a non-insurance alternative, "Apoyos Directos", where disaster losses would be met by federal and state governments instead of insurers. This creates the possibility that state governments may use "Apoyos Directos" either as an alternative risk management tool or as a tool to cover the gap in agriculture insurance. These diverse uses can either decrease or increase the adoption of agricultural insurance. Therefore, we hypothesize that risk-sharing arrangements is associated with the intensity of agriculture insurance adoption.

Lastly, we include damage and the sum insured per premium as explanatory variables to explain the intensity of insurance adoption. We hypothesize that farmers experiencing more significant crop or livestock damage are likelier to adopt agriculture insurance to protect their livelihood and vice versa. Similarly, an increase in the sum insured per premium means more compensation for the insurance cost. Research supporting our hypothesis includes Mußhoff et al. (2014), who found farmers are sensitive to cost. Thus, an increase in the cost of insurance reduces the farmer’s demand for index insurance, indicating a negative correlation of premium costs and a positive correlation between subsidy on premiums and the agriculture insurance uptake decision (Aditya et al. 2018; O’Donoghue 2014; Sahoo et al. 2018; Saliu et al. 2010). Similarly, Boucher and Delpierre (2014) argues that if the premium is set too high, it reduces risk-taking and welfare, thus suggesting the importance of appropriate pricing strategies.

3.2 Data sources and variable definitions

The datasets covering the period 2003-2014 were accessed from the World Bank. Table 10 provides a summarises the elements constituting the datasets. Our analysis benefits from four comprehensive datasets: Concentrado SAC 2003-2014, Bases Apoyos Directos 2003 to 2014, Municipio population1990-2030 and Indemnizacionesproc. The Concentrado SAC 2003-2014 dataset provides relevant information about Mexican municipalities and states in the context of the ’Seguro Catastrófico.’ It functions as a foundational repository of data for our study, providing an understanding of the geographic and administrative dimensions of insurance coverage for catastrophic events. The dataset includes annual policy purchases of index and indemnity crop and livestock insurance. The Bases Apoyos Directos 2003 to 2014 dataset offers extensive information regarding the direct financial assistance extended to farmers, constituting a pivotal component of our analysis and enabling an examination of the financial support rendered to agricultural producers following natural disasters during the specified timeframe. The Municipio population1990-2030 dataset encapsulates demographic statistics across Mexican municipalities. Finally, the Indemnizacionesproc dataset delineates disbursements to Mexican states and municipalities carried out within the framework of the ’Seguro Catastrófico’, emerging as an indispensable source of information for evaluating the financial implications and outcomes associated with insurance coverage for catastrophic events. The payouts relate to index and indemnity crop and livestock insurance.

Study variables in Table 8 were constructed from four administrative datasets. The population was obtained from Municipio population1990-2030 by summing the number of people in all municipalities within each state. Intensity of insurance adoption, drawn from Concentrado SAC 2003-2014, is measured as the number of crop or livestock insurance policies purchased by the state government; sum insured per premium is the total sum insured divided by the total premium paid. Poverty or marginality is proxied by the average poverty or marginality index within each state. The Indemnizacionesproc dataset provides four variables: the land size or livestock size, expressed as hectares or livestock units where payouts were effected, crop and livestock damage payouts is measured in pesos. Risk-sharing arrangements are calculated as the number of direct-support instances reported for each state in Bases Apoyos Directos 2003 to 2014. All monetary values are adjusted to constant pesos.

3.3 Model for count data

Consistent with count-based measures of adoption intensity in agricultural economics (Kolady et al. 2021b; Jara-Rojas et al. 2020), we operationalize intensity as the annual count of crop/livestock insurance policies purchased per Mexican state. This approach captures the extensiveness of adoption across regions, complementing individual-level studies that focus on binary uptake decisions. The demographic, insurance-derived, and non-insurance variables are used to estimate the intensity of insurance adoption. The variables include population, poverty index, payouts, sum insured per unit premium, land size, livestock size, and risk sharing arrangements. In our study, the dependent variable \(Y_{i,t}\), the number of crop and or livestock insurance policies purchased by the state, represents the count variable. We focus on index and indemnity crop and livestock insurance policies. We take advantage of the Poisson regression model’s ability to explicitly recognize the dependent variable’s non-negative integer character (Winkelmann 2008) following the rationale of Boucher and Guillén (2009); Baltagi (2021); Majo and van Soest (2011); Hausman et al. (1984); Green (2007).

The Poisson model requires that the mean equals the variance. However, there are situations where we find that the variance is greater than the mean, referred to as overdispersion. To account for overdispersion in observed count data, we employ three models to address bias in the estimation of parameters (Weisburd et al. 2022). Estimation of the Poisson model requires that the conditional mean, \(E(Y_{i,t} \big | x_{i,t},\alpha _i) = \mu _{i,t}\) and the standard errors in fixed effect estimators are adjusted for robustness and consistency even when distribution assumptions are violated in count and non-count dependent variables (Wooldridge 1999; Verdier 2018; Hoang and Wooldridge 2024). The Quasi-Poisson model scales the standard errors of the estimated parameters by a post-estimated overdispersion factor and this affects the probability value of for deciding significance of the coefficients (Ver Hoef and Boveng 2007; Weisburd et al. 2022). For the Negative Binomial model, the overdispersion parameter is incorporated in the estimation which changes the coefficients and standard errors. This is in contrast to the Quasi-Poisson model which adjusts the standard errors only (Weisburd et al. 2022; Green 2021). The fixed effect estimators for the Poisson, Quasi-Poisson, and Negative Binomial models are employed to consolidate and enhance the analysis, make comparisons, and draw insights from crop and livestock insurance intensity of adoption factors (Biswas et al. 2020; Weisburd et al. 2022; Lee et al. 2021).

3.4 Poisson regression model

To provide some contextual understanding of the modelling choice, the exponential family distributions can be formulated in a generalized linear model (GLM) through a link function to examine the relationship between dependent and independent variables (McCullagh and Nelder 1989). The Poisson regression model belongs to an exponential family and assumes that \(Y_{i,t}\) is independent over time, conditional on \(X_{i,t}, \alpha _i\). Thus, the conditional Poisson distribution of \(Y_{i,t}\) for i in time period t, given regressors \(X_{i,t}\) and individual effect \(\alpha _i\) with parameters \(\mu _i\) is expressed as:

where

Here \(\beta = \beta _1,....,\beta _p\) represents a vector of regression parameters for explanatory variables \(x_{i,t} = x_{i,t,1},...., x_{i,t,p}\) to be estimated using the regression model. The Poisson regression model can be further classified into fixed and random effects. With fixed effect model, \(\alpha _i\) is treated as a deterministic parameter. On the contrary, the random effect treats \(\alpha _i\) as stochastic and is independent of all the explanatory variables \(x_{i,t}\), following a distribution function with mean and variance. Thus, using Poisson regression as the benchmark, our model specification for the modelled crop and livestock insurance data follows Nishitateno (2023) and is represented as:

Here \(i=1,....32\) refers to the Mexican state, and \(t=2003,... 2014\) is the year. ICIA is the intensity of crop insurance adoption measured by the number of policies purchased and is modelled as an independent Poisson random variable. POP is the population, ASIPUP represents the sum insured per unit of premium paid for crop insurance, LSFPO indicates the land size for which a payout was made for crop insurance, PI depicts the average poverty index, CDP denotes the crop damage payout and NRS illustrates the number of risk-sharing arrangements that are non-insurance, with specific reference to the "Apoyos Directos" program under the CADENA. Lastly, \(\gamma _{i}\) and \(\omega _{t}\) are state individual and time-specific effects, respectively. Equations 3 and 4 both models intensity of crop insurance adoption but differ from each other because of the payout measures \(LSFPO_{i,t}\) and \(CDP_{i,t}\) respectively. We estimate the intensity of livestock insurance adoption with models 5 and 6. The livestock insurance modelling specifications are highlighted as:

where i is the state, \(i=1,,円\ldots ,32\) and y is the year, \(y=2003,,円\ldots ,2014\). ILIA is the intensity of livestock insurance adoption and PO shows the population. We use BPUP to represent the benefit in the sum insured for each unit of premium paid for livestock insurance. LSP is a variable for livestock units for which insurance payout were made. MI shows the marginality or poverty index that is associated with a state, measured as an average of the state. LDP indicates the livestock payouts made whilst RSA reveals the number of non-insurance arrangements in place based on the support received from the "Apoyos Directos". We denote the state and time specific effects by the variables \(\alpha _{i}\) and \(\varphi _{y}\) respectively. The estimated regression coefficients from the model are \(\nu _{1},,円\ldots ,\nu _{6}\). The error term associated with state and year is denoted by \(\varepsilon _{iy}\) and is independent and identically distributed with mean 0 and standard deviation, \({\sigma _{\varepsilon _{iy}}}\) . Models 5 and 6 have different livestock payout measures illustrated by LSP and LDP respectively. We use land (livestock) size where payouts were effected as physical exposure proxies and damage payouts as value at risk proxies. Estimating models with each provides a robustness check if associations persist across quantity and value-based exposure measures, which strengthens confidence in the patterns we report.

3.5 Quasi-poisson model

Overdispersion is a concern in count data modeling. Thus, to address this problem, we utilize the Quasi-Poisson model. The Quasi-Poisson model circumvents the overdispersion problem in a Poisson setup by correcting standard errors of estimated regression parameters using an overdispersion measure (Ver Hoef and Boveng 2007; Asravor et al. 2022; Polkowska-Kramek et al. 2024; Yu et al. 2024). As a background to our modeling choice, early applications of the Quasi-Poisson model are found in renewal theory and counter problems (Smith 1956; Elizalde and Gaztanaga 1988). The use of Quasi-Poisson has expanded to other fields that include health, environment, ecology, and business. A Quasi-Poisson is a generalization of the Poisson, characterized by the mean and variance (Ver Hoef and Boveng 2007). Quasi-Poisson looks like the standard Poisson with the mean, \(\mu _{i,t}\) being the same, but the variance differs by an overdispersion parameter, \(\theta _{i}\) (Ver Hoef and Boveng 2007). For example if the variance of a standard Poisson is \(Var(Y_{i,t} \big | x_{i,t},\alpha _i) = \mu _{i,t}\) then the Quasi-Poisson’s variance is \(Var(Y_{i,t})=\theta _{i}\lambda _{i,t}\). An overdispersion parameter of 1 in the Quasi-Poisson equates to a standard Poisson model. Despite the non-existence of a probability formulation for a Quasi-Poisson, its semi-parametric form fits data better (Zeviani et al. 2014). Since a Quasi-Poisson model has no full likelihood, it can only be compared with other models within the quasi-framework (Ver Hoef and Boveng 2007; Zeviani et al. 2014). Studies applying the Quasi-Poisson model in cross-section data include Green (2021); Blevins et al. (2015), for time-series data they comprise of Xu et al. (2019); Vicuña et al. (2021); Imai et al. (2015); Rodrigues et al. (2020), and in panel data Armstrong et al. (2014) model daily stratified counts of death, Hossain et al. (2023) examine monthly cases of dengue fever across years, Chen et al. (2022) investigates daily seizure counts of patients, Hess et al. (2013) acknowledge the importance of a Quasi-Poisson in estimation of gravity models, and Shiau et al. (2023) assess regional road traffic accidents across months. We extend Quasi-Poisson modeling to crop and livestock insurance by estimating models presented in Section 3.4 and subsequently account for overdispersion in the standard errors.

3.6 Negative binomial regression model

Similar to the Quasi-Poisson model, the Negative Binomial model is another solution to ameliorate the overdispersion problem in count data as demonstrated in Biswas et al. (2020); Yee (2020). Therefore, we include the Negative Binomial regression model to provide an alternative remedy to overcome overdispersion by adding a discrete parameter \(\theta \ge 0\), which is used to explain the heterogeneity of the data. When \(\theta\) approaches infinity, it indicates that the count data is extremely overdispersed, whereas if it approaches 0, it means that the Negative Binomial model approximates to a Poisson distribution. Negative Binomial can be used to model cross-section, time-series, and panel count data. For instance, Negative Binomial regression analysis focusing on cross-section count data include Lee et al. (2021); Espinoza et al. (2021); Hayat and Özden (2023). Time-series applications of the Negative Binomial are found in Aleksandrov et al. (2023); Maruyama and Taguchi (2021); Stapper (2021), and panel data studies are illustrated in Sun et al. (2021); Diaz-Corro et al. (2021); Tran et al. (2020). Furthermore, the Negative Binomial for panel data is highlighted as:

Here \(\Gamma\) is the gamma function; dispersion parameter \(\theta\) is assumed to be constant over time for each state i and \({\mu _{i,t}}\) depends upon the covariates. The mean and variance of \(Y_{i,t}\) are given by \(E(Y_{i,t})= \mu _{i,t}\) and \(var(Y_{i,t})=(1+\theta \mu _{i,t}) \mu _{i,t}\). When \(\theta =0\) then the variance equals that of a Poisson distribution \(\mu _{i,t}\) depicting equal dispersion, for \(\theta < 0\) and \(\theta> 0\), this illustrates underdispersion and overdispersion respectively. We compare the Poisson regression model with the Negative Binomial model for the insurance data.

Our empirical strategy accounts for unobserved factors that remain constant over time and across states. We emphasize that results from estimating the specified models demonstrate associations rather than proven cause-and-effect relationships. Hidden factors (like other policy changes happening at the same time) or reverse causation (where the intensity of insurance adoption might affect poverty measures) could still influence our findings. Future studies could use stronger methods, such as natural experiments or instrumental variables (for example, using unexpected changes in subsidy rules), to better identify causal effects.

This section presents the analysis for investigating the intensity of crop insurance adoption using count data models. The analysis is performed in three phases. It begins with the descriptive statistics of the variable used, followed by the preliminary tests. Next, the count data model explicitly presents the Poisson fixed effect model results. Lastly, generalized Poisson and Negative Binomial models are used to highlight significant predictors that explain variations in the intensity of crop insurance adoption.

4.1 Descriptive statistics

Tables 7 and 8 describe the variables incorporated into the econometric model to assess the intensity of crop and livestock insurance adoption. In Table 7, Panel A, the dependent variable, the intensity of total crop insurance adoption, has an average of 170.21 policies per Mexican state, ranging from one to a maximum of 1,974. In contrast, in Panel D, the dependent variable, the intensity of total livestock insurance adoption, shows, on average, 51.34 policies per Mexican state, ranging from 2 to a maximum of 246. Thus, the adoption rate of crop index insurance is higher and more diverse across regions than livestock insurance contracts. Further, according to Panels B and C, crop index insurance uptake is lower than indemnity insurance contracts, as on average, the Mexican state purchased 161.05 indemnity insurance whereas 101.98 index insurance. Explanatory variables from previous research include risk-sharing arrangements, farm or livestock size, population size, poverty index, and sum insured per premium.

In Panel A, the risk-sharing agreements, serving as proxies for risk aversion, illustrate a moderate level of readiness among states to collectively manage agricultural risks, with an average of 29.86 arrangements per state. Farm size exhibits variation across states, averaging 24.81 thousand hectares with a deviation of 28.07, showcasing diverse agricultural landscapes for which insurance payout was made. Population size and poverty index highlight heightened vulnerability to risks, with 2.26 million people living at a moderate poverty level. However, despite the moderate risk-sharing behavior and elevated vulnerability, the crop damage payout amounts to 22.19 million pesos. The average sum insured per premium unit is 8.68, suggesting potential barriers to adoption. In contrast, in Panel D, for the risk-sharing agreements, livestock index insurance exhibits a low level of readiness among states, with an average of 10.61 crop arrangements per state. Further, on average, states received a payout of 14.45 million pesos for the herd stock of 537.71, with a deviation of 35.81 million pesos and 1,130.01 herd stock. According to McPeak and Barrett (2001), herd size variability is mainly due to biological regulations such as births and deaths, thus implying that most of the livestock mortality risk remains uninsured. Furthermore, like Panel A, Panel B’s Population Size and Marginality Index showcases a heightened vulnerability to climate risks, with an average population of 2.79 million individuals in a state living at a moderate poverty level. Lastly, the sum of the average benefits insured per premium unit stands at 11.36, suggesting livestock index insurance provides higher compensation to its policyholders than crop index insurance.

Further, in Panels B and C, the average values of Population, Land Size, Poverty Index, and Crop Damage for Crop Indemnity Insurance are 2.09 million, 47.65 thousand hectares, 3.00, and 45.30 million pesos respectively. These values are comparatively higher than the values for crop index insurance, which are 1.9 million, 30.70 thousand hectares, 2.86, and 24.64 million pesos. On the contrary, the sum insured per premium and the number of risk-sharing agreements for crop indemnity insurance, i.e., 8.32 and 11.36, are lower than those for crop index insurance, which are 9.64 and 15.57. This shows that, on average, Mexican states with higher populations, Land size, Poverty, and Crop Damage are more likely to adopt crop indemnity insurance than crop index insurance due to lower premium cost.

Further, in Panels B and C, the average values of Population, Land Size, Poverty Index, and Crop Damage of Crop Indemnity Insurance are higher by 0.19, 16.95, 0.14, and 20.66 than the index insurance. On the contrary, the Sum Insured per Premium and the number of risk-sharing agreements of crop indemnity insurance are lower by 1.32 and 4.21. This shows that, on average, Mexican states with higher populations, Land size, Poverty, and Crop Damage are more likely to adopt crop indemnity insurance than crop index insurance due to lower premium cost.

4.2 Results of poisson fixed effect for crop insurance

The dependent variable is a count variable representing the number of crop insurance policies purchased by the states. Given a set of explanatory variables, we use the Poisson distribution model to explain the variation in the intensity of crop adoption. The model is specifically designed for count data and explicitly recognizes the non-negative integer character of the dependent variable (Winkelmann 2008).

Before proceeding with the model estimation, a preliminary dataset analysis was conducted. Table 9, which displays the correlation between explanatory variables, showed that the correlation between most explanatory variables is small and not greater than 0.5. However, there appears to be multicollinearity between crop damage (livestock damage) and land size (herd size). To mitigate this issue, two different models are estimated. In models 1 and 3, represented by Eqs. 3 and 5, the intensity of crop (livestock) index insurance adoption is regressed on population, average sum insured per unit of premium (Benefits in the sum insured for each unit of premium), land size (Herd size), poverty index (marginality index), and the risk-sharing agreements. Model 2 and 4 (represented by Eqs. 4 and 6) are like Model 1 and 3, with land (farm) size replaced by crop (livestock) damage as an explanatory variable.

The Poisson model is extended to include a fixed effect. In the fixed effect model, the heterogeneity parameter is treated as a parameter to be estimated for each individual state. In this case, no intercept enters the model. Further, it makes inferences conditional on the effect present in the sample (Verbeek 2021).

Table 1 presents the results of three separate Poisson fixed-effects models of crop insurance, specifically for equation (3) and (4). Each model regresses a vector of explanatory variables on the intensity of crop insurance adoption. The first model focuses on combined crop insurance adoption, the second on index-based crop insurance, and the third on indemnity-based crop insurance. Table 1 combined crop insurance adoption results for models 1 and 2 show that all variables, i.e., population, land size, poverty index, risk sharing agreement, crop damage risk, and the average sum insured per unit of premium, significantly explain variation in the adoption of index insurance. Firstly, the estimated coefficient population, poverty index, and land size significantly and positively explain the intensity of crop adoption at the 1% significance levels, respectively. This supports the findings of Skees et al. (2008) and Moritz et al. (2023), who observed that larger population, land size, and poverty dynamics are positively associated with index insurance adoption. Thus, assuming that Mexico is classified as an upper-middle-income country with vast social disparity and a population of more than 37.4% living below the poverty line as of 2020 (United Nations Population Fund 2023; OECD et al. 2022), index insurance targeting the poor provides a reliable way to manage weather risk and protect their households. Similarly, farm size is positively linked to intensity of index insurance adoption, since a one-thousand-hectare increase in farm size is associated with an additional 0.002 crop insurance policies per state, on average. This supports the idea that farmers in states with large pieces of land are more risk-averse than states with small farm sizes (Wik et al. 2004) and, therefore, are more likely to adopt index insurance. Conversely, the risk-sharing agreement is negatively associated with the intensity of crop adoption, with a unit increase in the risk-sharing agreement associated with a 0.006 decrease in intensity of adoption at 1% significance level. This finding is consistent with Annan and Datta (2022), who argue that a risk-sharing mechanism may reduce risk aversion, thus shaping uptake rates in complex ways. Furthermore, Model 1 yields similar results to Model 2, with the addition of crop damage significantly positively explaining variation in the intensity of crop insurance adoption, thus suggesting that increasing crop damage is associated with an increase in adoption of index insurance due to the need for protection against weather-related risk.

Furthermore, the results of index-based and indemnity-based crop insurance for models 1 and 2 in Table 1 support the combined findings with slight differences. For instance, all variables are found to significantly explain the intensity of index-based crop adoption except for land size and crop damage. Similarly, in the case of results for indemnity-based crop insurance, all variables are statistically significant except for land size. This highlights that the determinants of the intensity of insurance adoption differ based on the specific type of insurance product.

4.3 Results of poisson fixed effect for livestock insurance

Table 2 represents three separate Poisson fixed effect model results of livestock insurance for specifically Eqs. 5 and 6. Each model regresses a vector of explanatory variables on the intensity of livestock insurance adoption. The first model focuses on combined livestock insurance adoption, and the second on index-based livestock insurance. The results of the combined insurance show that only the sum insured per unit of premium, herd size, and livestock damage payout are found to be statistically significant in explaining the variation in the intensity of livestock insurance adoption. The benefits in the Sum insured per unit of premium in both Model 3 and 4 are significant at the 1% level of significance, implying that a premium that generates sufficient revenue to overcome the potential losses also provides financial security to pastoralists and is positively associated with intensity of adoption, suggesting value-for-money considerations are positively linked to participation. Further, Herd size and Livestock Damage are significant and negatively associated with intensity of adoption at 5% and 1% significance levels in Models 3 and 4, with coefficients of -0.00005 and -0.002, respectively. This contradicts the findings of Chand et al. (2016); Madaki et al. (2023) which suggests that larger herd size and damage motivate risk-averse pastoralists to adopt index insurance. Overall, the premium is found of utmost importance among the three significant factors when designing livestock insurance. Furthermore, the insignificance of population, marginality index, and risk-sharing agreement in explaining the variation in livestock insurance adoption contradict the researches such as Matsuda et al. (2019) who found that poverty levels influence the adoption of index insurance with pastoralists near the poverty threshold increases the livestock numbers after receiving insurance payouts. At the same time, the result support Takahashi et al. (2019) who argue that the presence of index insurance does not crowd out informal risk-sharing agreements among the pastoralists in southern Ethiopia.

Further, the index-based livestock insurance presents contradicting results for models 3 and 4 in Table 2 as only the population is reported to significantly explain the variation in the intensity of index-based livestock insurance adoption at 1% significance level. The contrary results signify the major differences in the design of index and indemnity livestock insurance of the CADENA program. For instance, in case of CADENA Program, the traditional livestock insurance, namely (Seguro Pecuario Catastrofico, SPC), provides coverage of decreased forage and extraordinary weight loss in animals based on perilous drought. Whereas the Livestock-Pasture NDVI, namely (Seguro Pecuario de indices de Vegetation, SPIV), uses a satellite-measured NDVI Index to insure against all perils that reduce pasture growth Bank (2013), thus leading to different factors affecting the adoption of combined and index livestock insurance. Therefore implies that factors affecting the intensity of adoption of index insurance vary with respect to index insurance products in focus.

4.4 Results of Quasi-Poisson and negative binomial model for crop insurance

To test the assumption that the variance is equal to the mean, we implement the Cameron and Trivedi (1990) test of overdispersion. As shown in Tables 13 and 14, the p-value is less than 0.05 for models 1,2,3 and 4, respectively. This presents strong evidence of over-dispersion. In this case, the statistical literature proposes Poisson Generalized Linear Model, Quasi-Poisson Generalized Linear Model, and Negative Binomial Models (Saputro et al. 2021; Rakhmawati et al. 2017; Ismail and Jemain 2007; Ver Hoef and Boveng 2007). In the Quasi-Poisson and Negative Binomial models, the model is fit using the quasi-likelihood and explicit likelihood model rather than the likelihood function, which allows for an estimate of the over-dispersion parameter. Further, the Negative Binomial model, in addition to the mean of the count variable, introduces another parameter that gives models more flexibility than a Quasi-Poisson model, thus, is considered a better approach to dealing with over-dispersion.

Tables 3 and 4 present the results of three separate Quasi-Poisson and the Negative Binomial model crop insurance models, specifically for Eqs. 3 and 4, respectively. Each model regresses a vector of explanatory variables on the intensity of crop adoption. The first model focuses on combined crop insurance adoption, the second on index-based crop insurance, and the third on indemnity-based crop insurance. The Quasi-Poisson and Negative Binomial models present slightly different results compared to the Poisson fixed effect model. For instance, in the case of combined crop insurance, according to both Quasi-Poisson and Negative Binomial models, only population, poverty index, and risk-sharing arrangements are significant at 1%, 10%, and 5%, respectively. This is mainly due to a sharp increase in the standard errors because of the over-dispersion effect, thus affecting the significance of the variables. For instance, in Tables 3 and 4, the Quasi-Poisson model standard error for average sum insured per premium and land size (Crop Damage) increases from 0.003 to 0.017 and 0.0003 to 0.002; similarly, in the Negative Binomial model’s average sum insured per premium and land size (Crop Damage), standard error increased from 0.003 to 0.01 and 0.0003 to 0.001, respectively, thus making both insignificant in explaining the intensity of crop adoption compared to the results of Table 1.

Compared to the combined results of crop insurance adoption, index-based and indemnity insurance adoption presents contrasting results in Tables 3 and 4 for Models 1 and 2. For instance, in the case of index-based insurance for QP models for Models 1 and 2, only population and poverty indices are significant at 1%, respectively. However, for the Negative Binomial model results for Models 1 and 2, only population significantly affects the intensity of index-based crop insurance adoption. Similarly, for indemnity insurance adoption, the Quasi-Poisson model results for Models 1 and 2 report significant results for the population. Meanwhile, QP model results for Models 2 report significant results for all variables except for the sum insured per premium due to increased standard error. Furthermore, since the standard error of the Quasi-Poisson and Negative Binomial model is the lowest for Models 1 and 2 for both index and indemnity-based insurance, it leads to the conclusion that only population significantly affects the intensity of crop index insurance adoption, whereas population, poverty index, crop damage, and number of risk-sharing agreements for intensity of indemnity based crop insurance (Tables 3 and 4).

4.5 Results of Quasi-Poisson and negative binomial model for livestock insurance

Tables 5 and 6 present the results of three separate Quasi-Poisson and the Negative Binomial models of livestock insurance, specifically for Eqs. 5 and 6 respectively. Each model regresses a vector of explanatory variables on the intensity of livestock insurance adoption. The first model focuses on combined livestock insurance adoption, and the second on index-based livestock insurance. According to the combined livestock insurance results in Tables 5 and 6, only Population and Benefits in the Sum insured per unit of premium, are found to significantly explain the intensity of livestock index insurance. Meanwhile, the herd size and livestock damage become insignificant due to changes in the standard errors in the case of the Quasi-Poisson and Negative Binomial models. Further, for the index insurance similar to the Poisson fixed effect model, also only the population is reported as statistically significant at a 1% level of significance.

The disparities in intensity of insurance adoption results between crop and livestock insurance have several explanations. First, crops face largely covariate weather–yield shocks, so larger land size, population, and poverty align with governments’ food security and social-protection mandates and show positive associations; crop payouts validate value and are likewise positive. In contrast, while livestock insurance products insure herds against mortality and pasture shortfalls, the availability of mitigation strategies, such as adjusting herd size, sales, or feed mechanisms, can result in livestock size having a negative association with insurance adoption. Second, differences in contract design: the crop products use yield and weather station-based indices, unlike the livestock products, which use satellite indices such as the normalized difference vegetation index (NDVI). Third, risk-sharing arrangements substitute for formal insurance in crops (negative association) but show no association for livestock, emphasizing that direct support has a substitution effect in crops but none in livestock. Fourth, because state governments, not farmers, are the policyholders, fiscal and political considerations can diverge by product: crop losses map more directly to food-price and income-stability goals than livestock losses. Finally, data coverage is thinner for livestock than for crops in our panel, which may affect the degree of statistical precision in the estimated models.

Although our intensity of insurance adoption models reveal statistically significant associations, they should not be interpreted as causal effects because the data are observational. Without experimental design or exogenous variation, the results could be affected by reverse causality and omitted-variable bias. Even so, the stability of coefficient signs and magnitudes across multiple specifications lends confidence that these patterns are systematic and merit testing in future experimental or hypothesis-driven research. Importantly, the unit of analysis is the insurance purchased by state governments under CADENA, not individual farmers’ purchases. The estimated associations therefore reflect how biophysical, socio-demographic, risk, and economic factors shape state level adoption, and may not translate directly to household level agriculture insurance demand. Caution should be made when making comparisons with micro level studies.

4.6 Mechanisms and channels of insurance adoption

The preceding models establish which state-level characteristics are significantly associated with the intensity of insurance adoption. To explain the underlying decision-making processes, we discuss potential mechanisms and channels through which population size, poverty, risk-sharing arrangements, and product value may influence state participation in CADENA.

4.6.1 Political and administrative channel

Higher state population exerts pressure on policymakers to secure agricultural risk protection. Larger populations increase the potential number of affected farmers during disasters, strengthening political incentives to purchase CADENA coverage to safeguard livelihoods and avoid social unrest (Clarke and Dercon 2016). This channel suggests that the size of a population acts not merely as a demographic factor but as a driver of state-level demand for agricultural insurance.

4.6.2 Social Safety Net Channel

Poverty levels (marginality index) are strongly and positively associated with adoption intensity, indicating that CADENA functioned as a redistributive tool. Poorer states may rely on insurance as an ex-ante safety net, complementing other social assistance programs to prevent households from falling into poverty traps after climate shocks (Noritomo and Takahashi 2020).

4.6.3 Substitution channel with informal mechanisms

The negative relationship between informal risk-sharing arrangements ("Apoyos Directos") and formal insurance adoption suggests that states may partially substitute fiscal transfers for insurance purchases when budgetary flexibility allows. This is consistent with findings by Boucher and Delpierre (2014), who argue that informal mechanisms can crowd out formal insurance in settings with overlapping risk-sharing instruments. Consequently, the substitution effect calls for a broader risk management ecosystem when designing and implementing insurance programs.

4.6.4 Value for Money Channel

For livestock insurance, the benefit-to-premium ratio is a key driver, implying that states respond to perceived cost-effectiveness when deciding whether to adopt insurance. This supports the notion that actuarially fair pricing and sufficient coverage limits are crucial for maintaining program participation (Mußhoff et al. 2014).

Climate change presents a persistent and escalating threat to agricultural systems, particularly in developing economies. The increasing frequency and severity of natural disasters have intensified the need for robust climate risk management strategies. Agricultural insurance has emerged as a key instrument in this regard, offering financial protection and promoting resilience (World Bank 2020). Nevertheless, a significant protection gap remains, with only 45% of economic losses from extreme weather events currently insured (Banerjee et al. 2023).

While numerous studies have examined the general determinants of index insurance adoption (Nordmeyer and Mußhoff 2023; Yufei et al. 2022; Tang et al. 2021; Takahashi et al. 2020; Skees 2008; Moritz et al. 2023; Annan and Datta 2022), the dynamics of insurance uptake vary across political, legal, cultural, and local contexts, as well as in response to global trends such as deregulation and globalization (Cummins and Venard 2008). This study contributes to the literature by analyzing the intensity of agricultural insurance adoption under Mexico’s CADENA program at the state level.

Our findings indicate that population size and poverty levels are positively associated with the intensity of crop insurance adoption, suggesting that insurance demand increases with socioeconomic vulnerability. Conversely, the number of risk-sharing arrangements is negatively associated with adoption intensity, implying that informal mechanisms may substitute for formal insurance in some contexts. These results align with existing literature that positions agricultural insurance as a safety net for economically disadvantaged and risk-averse populations.

In contrast, for livestock index insurance, population and the benefit-to-premium ratio are significant predictors of adoption intensity, underscoring the importance of perceived value in product design (Takahashi et al. 2020; Liu et al. 2021). The study further reveals that determinants of adoption vary by insurance type. For index-based products, population is the primary driver, whereas indemnity-based crop insurance adoption is influenced by a broader set of factors, including poverty index, crop damage, and risk-sharing arrangements.

These insights carry important implications for policy and program design. Differentiated strategies are needed for crop and livestock insurance, reflecting their distinct risk profiles, coverage needs, and payout mechanisms. Crop insurance typically addresses yield and weather risks, while livestock insurance targets disease and forage-related losses. Moreover, indemnity insurance relies on verified field assessments, whereas index insurance operates on predetermined triggers. As such, product-specific factors must be considered when designing insurance schemes.

The positive association between poverty and crop insurance adoption, alongside the negative impact of risk-sharing arrangements and the importance of premium value in livestock insurance, highlights the role of insurance design in shaping participation. These findings support the use of targeted subsidies and affordable premiums to improve accessibility for low-income groups (Boucher and Delpierre 2014; Vargas Hill et al. 2014).

Future catastrophic insurance programs should prioritize ex-ante coverage in high-risk and high-poverty regions, leveraging performance-based co-financing mechanisms that reward transparency and timely payouts. For livestock, enhancing product value through improved benefit structures may increase uptake. Additionally, aligning insurance with disaster relief budgets and clarifying the limits of ex-post aid can prevent the crowding out of formal insurance. Premium credits may be considered in regions with strong informal risk-sharing systems.

Finally, this study opens avenues for further research. While we identify the negative impact of risk-sharing arrangements on index insurance uptake, future work could explore strategies to enhance adoption in the presence of such arrangements, including the role of subsidies, awareness campaigns, and training programs (Annan and Datta 2022; Madaki et al. 2023). Moreover, incorporating stakeholder perspectives through local-level qualitative research could enrich understanding of adoption dynamics and inform the design of complementary programs such as PROCAMPO. Given the product-specific nature of insurance adoption, similar analyses could be extended to other insurance types, including livestock and drought insurance.