INTRODUCTION
Problem statement
Natural disasters result in significant loss of life and property, encompassing both direct and indirect damages The incidence of these disasters is rising in tandem with climate change According to the IPCC's 2007 report, climate change has led to an increase in global temperatures of approximately 0.76°C and a rise in sea levels of about 20 centimeters since 1900.
Climate change is contributing to a gradual increase in the damages caused by natural disasters, particularly in regions that are already vulnerable to such calamities (Pielke et al., 2005).
The Mekong River Delta, situated at the downstream end of the Mekong River Basin, faces seasonal flooding due to high flow rates exceeding 65,000 m³/s during the wet season and its low-lying terrain Research by the Vietnam Academy for Water Resources indicates that from 1991 to 2009, the area affected by annual floods rose from 1.6 million hectares to 2 million hectares (To & Tang, 2011).
Each year, local residents in the Mekong River Delta prepare diligently for the flood season to mitigate potential damage In 2011, the Vietnam Mekong River Delta Project, supported by the United Nations, focused on enhancing flood resistance for impoverished households This initiative involved reinforcing houses, thoroughly examining flood protection infrastructure, conducting aid drills, and ensuring the availability of essential medicines for prioritized groups Despite these extensive preparations, the floods of 2011 resulted in over 1,000 billion VND in property damage, affecting 27,000 hectares of rice and vegetable crops, with 10,000 hectares suffering total loss, along with nearly 12,000 hectares of fruit crops also impacted by flooding.
The statistical data of flood damage in the Mekong River Delta from 1990s to 2000s shows that flood damages in downstream Mekong River Delta become abnormal.
Figure 1.1 Flood damages in the Mekong River Delta from 1990s to 2000s
Source: Collect from Nguyen, 2006; Dao & Bui, 2009; MRC, 2011; MRC, 2012
Recovery in the Mekong River Delta following floods is a prolonged process, largely due to residents' reliance on aid from government and humanitarian organizations However, this aid is often inconsistent, highlighting the need for communities to seek self-sufficiency For instance, after the severe floods in 2011, local residents faced a wait of at least one year before receiving full assistance from the International Federation of Red Cross.
Figure 1.2 Disbursement process of IFRC funding contribution for Floods in MRD
Source: Report of International Federation Red Cross, 2013
In the Mekong Delta, post-flood relief efforts fall short of covering agricultural costs, as farmers typically invest around 15 million dong per hectare Despite the government's increase of post-disaster subsidies to 5 million dong per hectare for paddy following the severe floods of 2011, this amount only compensates for approximately 30% of the total agricultural investment (Ngoc Anh, 2011).
Natural disasters, such as floods, significantly heighten the financial burdens and default risks faced by farmers, often leading to the threat of double liabilities Due to financial constraints, many farmers resort to purchasing agricultural inputs on credit, incurring higher costs to pay after the harvest Typically, farmers and agents enter into agreements allowing for a four-month debit period; however, if farmers fail to repay the loan by maturity, they are subjected to ongoing debt at an interest rate of 3-4% per month (Ngo, 2013).
Farmers face the risk of accumulating debt despite having access to bank loans For instance, rice farmers can only secure loans of approximately 1 million VND per 1,000 square meters, while their actual cultivation and harvesting expenses range from 2.2 million VND to 2.5 million VND per 1,000 square meters.
1000� 2 , and transferring value is about 40-50 million VND (Ngo, 2013) Therefore, occurrence of flood might induce farmer to fall into debt piling up.
In the 2000s, the Mekong River Delta was vital to Vietnam, contributing over 48% of the nation's food production and 85% of its rice exports However, projections from the Flood Prevention Agency in HCMC indicate that by 2030, around 45% of the Delta's area may face salinity issues and severe flooding, potentially resulting in losses of up to $17 billion The construction and operation of multiple dams in the Mekong River Basin have left Vietnam vulnerable to unpredictable flooding and drought conditions For instance, heavy rainfall leading to dam discharges can trigger a domino effect across the entire system of 12 dams, causing significant damage This hydropower infrastructure poses a serious threat to the future sustainability of the Mekong River Delta and the broader region.
Adapting to floods, or "Living with floods," is essential for maximizing the benefits of flooding while preserving Vietnam's rice granary In addition to flood prevention infrastructure, disaster insurance plays a crucial role in this adaptation strategy For developing countries, investing in catastrophe insurance is a wise decision, as it aids in damage reduction and helps prevent disaster-induced poverty traps Well-designed insurance not only serves as an effective tool for managing risks and facilitating recovery after adverse events but also alleviates the financial burden on governments during recovery efforts, allowing society to return to normalcy more quickly Furthermore, in cases of severe natural disasters where households lack resilience, insurance companies can distribute risk by using premiums collected from other policyholders to cover losses.
Market principles incentivize private insurance companies to implement more effective risk-reduction strategies compared to public insurers (Priest, 1996) By rewarding climate-adaptable construction designs and offering premium discounts, insurers can significantly mitigate devastation Insurance policies can motivate households in flood-prone areas to adopt risk-reduction measures, such as using tile floors or flood-resistant materials, by excluding coverage for damages to wooden floors (Thieken et al., 2006) A post-2002 flood survey in Germany revealed that insurance buyers prioritize flood mitigation more than non-buyers (Thieken et al., 2006), ultimately leading to reduced recovery costs and less severe impacts from natural disasters.
Flood insurance and agricultural insurance are relatively new concepts for Vietnamese farmers, with agricultural insurance contributing only 0.015% to the total insurance revenue, as reported by the Ministry of Finance Since its pilot implementation in 2011, challenges such as undefined insurance targets, diverse local risks, and limited technical infrastructure have hindered its growth Additionally, issues like ambiguity, adverse selection, moral hazard, and correlated risks have made the private insurance market less attractive Therefore, researching the demand for flood and agricultural insurance in disaster-prone areas is essential for the development of effective insurance solutions.
The demand for flood insurance has been extensively studied in developed countries, utilizing insurance statistics to assess this need (Kunreuther et al., 2009; Michel-Kerjan & Kousky, 2010) However, this approach is not applicable in regions like Vietnam, where the disaster insurance market is still emerging Consequently, there is a growing reliance on primary data to better understand flood insurance demand (Aliagha et al., 2015; Brouwer & Akter, 2010; Brouwer et al., 2013; Reynaud et al., 2012) While some research has employed the Choice Experiment Model to explore potential flood insurance markets, many studies still contain significant errors despite valuable insights.
In Vietnam, flood insurance is a relatively new product for farmers, making the choice experiment method particularly relevant for assessing demand While two studies have explored flood insurance demand in Central Vietnam using the Choice Experiment Model, they failed to address previous research errors and overlooked the influence of local residents' perspectives on their demand for flood insurance Additionally, there is a lack of research focused on understanding flood insurance demand in the Mekong River Delta region.
Facilitating the disaster insurance sector in developing countries, such as Vietnam, is essential Understanding the willingness-to-pay (WTP) of individuals in disaster-prone areas, along with the effects of various obstacles on their WTP, is crucial for improving this sector.
Research objectives
This study utilized Choice Experiments to assess the willingness to pay (WTP) for flood insurance among farmers in the Mekong River Delta.
Firstly, we estimate the impacts of flood insurance attributes on the utility of farmers in Mekong River Delta.
The development of the flood insurance market and local irrigation services is significantly influenced by challenges such as ambiguity, adverse selection, charity hazard, and the accessibility of pumping stations Previous studies highlight that these challenges impact local farmers' perceptions of their vulnerability and fear of flooding, as well as their views on government responsibility and the effectiveness of local irrigation services.
We will assess the willingness-to-pay of local farmers for various attribute levels and specific flood insurance packages, considering different combinations of policy types and providers based on the estimated results.
Fourthly, we determine the variation of probability of WTP of Mekong River Delta farmers who are willing to pay for flood insurance with the changes in premium level.
Finally, we would like to present some appropriate suggestions to insurance companies and policymakers.
Scope of the study
This study was conducted in October 2015, focusing on three districts in the Mekong River Delta: Gao Giong, Phu Loc, and Tan Cong Chi These areas have experienced significant impacts from flooding and other natural disasters in recent years.
Structure of this thesis
This thesis is structured into five chapters, beginning with Chapter 2, which explores theories related to choice experiments and the challenges associated with flood insurance, alongside a review of empirical studies on the demand for flood and similar insurance types Chapter 3 outlines the data collection methods and methodologies employed in this research In Chapter 4, the focus shifts to the interpretation and discussion of statistical findings and empirical results Finally, Chapter 5 presents the conclusions drawn from the study and offers recommendations for flood insurance companies.
Previous studies without using RUM
2.1.1 Researches on agricultural insurance using secondary data or combine with primary data
Despite the potential benefits of agricultural insurance in alleviating government burdens and mitigating financial difficulties from crop failures, empirical research on its demand remains limited, particularly in developing countries Most studies have concentrated on the North American market and other developed regions, often relying on secondary data or a combination of secondary and raw data for their analyses.
A study analyzing data from 135 counties in Georgia between 1978 and 2010 revealed that both economic and demographic factors significantly influence flood insurance purchasing, while flood mitigation assistance does not impact demand (Atreya, 2015) Building on this, another research combined the effects of the representative heuristic on residents' risk assessment behavior with the previous findings to further explore flood insurance demand (Dumm et al., 2012; Volkman-Wise, 2012) The results indicated that recent natural disaster damages positively affect disaster insurance demand; however, this influence diminishes over time, and pre-disaster predictions tend to be undervalued.
Researches not applying RUM on agricultural insurance using primary data
In developing countries, agricultural insurance remains a nascent product, often lacking sufficient secondary data Conducting surveys to assess the potential for crop or disaster insurance development is essential In Malaysia, key factors influencing the demand for flood insurance include demographics, exposure levels, resilience capabilities, and residents' perceptions of vulnerability (Aliagha et al., 2015) However, this study is limited in scope, failing to identify significant issues within the insurance market or recommend specific flood insurance packages for implementation.
In the insurance market, information asymmetries can lead to market failures like adverse selection and moral hazard To address these issues, research has expanded the application of Expected Utility Maximization within a theoretical framework while accounting for heterogeneity Specifically, studies have examined how adverse selection impacts the ability to accurately reflect individual risk perceptions in premium pricing (Smith & Baquet, 1996) A survey conducted in Montana by Smith and Baquet highlights these findings.
In 1996, Heckman's two-stage estimation procedures were utilized to analyze agricultural insurance decisions, employing a probit model for participation and OLS for coverage levels Various factors influenced local farmers' decisions to participate in insurance The findings revealed that adverse selection led to differing coverage level decisions between farmers with positive expected returns and those with negative expected returns, indicating that premiums are ineffective in reducing loss ratios.
A study by Ye et al (2015) utilized the Expected Utility Maximization and Random Utility Theorem to examine how perceptions of insurance contracts influence farmers' participation decisions and the demand for multi-peril insurance The research indicates that farmers' participation decisions are affected by factors such as premium costs, insurance perceptions, and indirect utility The Random Utility Theorem framework reveals a simultaneous correlation between insurance participation and perception, represented through simultaneous equations However, econometric analysis of survey data from Hubei, China shows that local farmers possess a low perception of insurance contracts, despite long-standing government subsidies This lack of perception may contribute to the absence of strong evidence supporting learning-by-doing in the insurance market (Ye et al., 2015).
After experiencing severe floods, the demand for flood insurance among farmers is expected to rise, particularly when factors such as risk behaviors, mitigation efforts, financial limitations, psychological influences, and information gaps are accounted for (Turner et al., 2014) Given the challenges in capturing farmers' risk behaviors, a lottery experiment was conducted to assess respondents' risk attitudes In the context of Pakistan, the Expected Utility Maximization framework was applied using a Probit Model, with a binary insurance choice experiment as the dependent variable, to explore the determinants influencing flood insurance participation (Turner et al., 2014).
Although using Binary Probit Models in Expected Utility Maximization framework is quite effective in finding the determinant of flood or crop insurance participation, it does not consider
The perception of potential clients regarding insurance packages can significantly influence their decision to participate in insurance programs When faced with actual flood insurance options, clients may develop negative perceptions or feel disappointed due to unrealistic expectations To address this challenge, research has utilized the Contingent Valuation Method (CVM) to effectively gauge the demand for agricultural insurance products.
Contingent Value Method was used to obtain the WTP for breeding-sow insurance (Wan, 2014).
In a survey, respondents selected their preferred payment card method, which included a table detailing premium, coverage, and the premium per coverage ratio If they struggled to find the most suitable option, open-ended questions were posed for further insights The Tobit Model was utilized as the willingness to pay (WTP) for premium and coverage is inherently positive Analysis revealed that the average WTP for premium and coverage levels was significantly higher than current figures (Wan, 2014) However, a key limitation of the Contingent Valuation Method (CVM) is the challenge of accurately determining maximum WTP from a single question, potentially leading to bias Additionally, restrictions in presenting scenarios and timeframes, along with the inability to simultaneously assess changes in insurance contracts, have prompted researchers to seek more effective methodologies.
Random utility model (RUM) and applications
Classical economic theory posits that consumers aim to maximize their self-interest, which is articulated through heterogeneous preferences theories In this context, consumer preferences are represented by a utility function U(x), where x denotes the quantity of goods consumed within a budget constraint defined by px ≤ a, with p as the price vector and a as income The demand function is expressed as x = d(a, p) + ε, where ε represents measurement errors in consumer behavior These disturbances may arise from random factors affecting the objectives or constraints of economic agents (Griliches, 1975).
In 1927, Thurstone's law of comparative judgment proposed that individuals respond to stimuli by selecting the option that offers the highest level of stimulation This decision-making process involves an objective level of stimulus and an associated random error term for each alternative.
In 1960, Thurstone's study on stimulus levels was adapted for economics, leading to the development of the Random Utility Maximization Model This model incorporates random factors to analyze the binary and multiple-choice probabilities associated with utility maximization (Marschak, 1960).
Marschak expressed as utility level will be � � = � � +� � = � � � +� � , where � � and � � also were considered as systematic component and random component respectively.
In 1966, Lancaster introduced the "new theory of consumer demand," which transformed standard microeconomic demand theory by emphasizing that consumers seek to acquire the characteristics of goods rather than the goods themselves This perspective enables researchers to analyze the desired characteristics of consumers and estimate demand curves for new products more effectively Building on Lancaster's framework, McFadden expanded the study of consumer behavior by incorporating not only the quantity of goods but also their specific characteristics into the systematic component of demand analysis.
In McFadden's 1978 study, the systematic component of utility for an alternative is expressed as a function of its attributes, represented by the equation U_j = β_0 + β_1X_1 + β_2X_2 + + β_kX_k Here, X_k denotes the level of attribute k for alternative j, β_k represents the marginal utility of that attribute, and β_0 is the alternative-specific constant reflecting preference independent of attributes The model incorporates an error term, which follows a Gumbel distribution, denoted as ε ~ G(μ, σ), where μ and σ are the location and scale parameters, respectively Consequently, the alternative with the highest utility is expected to be selected.
In the binary choice, the utility function U(x) of vector x is random utility indicator when:
In the multiple-choice, the utility function U(x) of vector x is random utility indicator when:
To sum up, the probability of choosing alternative j:
�=1� � � The coefficients of the utility functions are estimated by maximizing the log-likelihood function:
Where: � �� the choice of individual i on alternative j (1 = chosen)
The Random Utility Model incorporates a random factor influenced by unobserved heterogeneity, including individual experiences, preferences, and sources of information regarding various attributes This randomness leads to the development of choice probability models based on specific parameters Furthermore, the unobserved elements of consumer characteristics are presumed to correlate with the observed components, contributing to the formation of subjective perceptions Consequently, there exists a continuous random field index that bridges the gap between unobserved and observed characteristics.
2.2.2Researches applying RUM on agricultural insurance
The limitations and potential biases of Contingent Valuation Method (CVM) can be addressed by using Random Utility Models (RUM), which assess the maximum willingness to pay (WTP) by analyzing choices among various insurance contract options Additionally, offering well-structured insurance package drafts enhances the evaluation process.
∑ help potential clients to have a clear visualization Due to many advantages, Random Utility Model is very suitable to study about the demand of agricultural insurances.
Holistic insurance, similar to flood insurance, is a relatively new and complex product, making it essential to use Random Utility Models (RUM) to gauge market demand (Nganje et al., 2008) Utilizing a Choice Experiment methodology, a well-structured survey was conducted after testing attribute levels and employing D-optimal main effects to select alternatives, which is a commendable aspect of Nganje's research Neglecting this careful approach could lead to inaccurate results and create false expectations among potential flood insurance customers Many previous studies failed to assess the appropriate scale of attributes or consider the income levels of local farmers, relying instead on average data from past research (Mercadé, 2009) Additionally, some studies established inappropriate and unrealistic attribute levels, particularly concerning premium and coverage (Brouwer et al., 2013; Reynaud et al., 2012).
Surveys often utilize questionnaires that gather demographic data alongside choice experiment (CE) questions Respondents are presented with choice cards featuring three insurance packages and can select the one they find most suitable or opt for a "none of them" option (Nganje et al., 2008) To minimize confusion and encourage rational decision-making, some studies limit each choice card to just two insurance packages and the "none of them" option (Mercadé, 2009; Brouwer & Akter, 2010; Brouwer et al., 2013; Reynaud et al., 2012; Opiyo et al., 2014; Liesivaara & Myyra, 2014).
Holistic insurance integrates crop and health insurance, addressing various scenarios that farmers may encounter, such as experiencing farm disasters alongside differing health conditions According to Nganje et al (2008), these scenarios include combinations of good and bad conditions for both farm and health Typically, each insurance package consists of five distinct attributes tailored to these situations.
Table 2.1 Attributes in the insurance package in Nganje’s study
Type of insurance RAH: revenue assurance, farm selects target revenue health and dental.
AGRH: adjusted gross revenue guarantees farm historical revenue health, dental and vision.
MPCIH: multi-perial crop insurance coverage for yield-related losses and health.
Implementing agency Government, cooperative, private
Many attributes in Nganje’s study (2008) can be found in other studies, though some authors used the different names, the definition is the same.
Types of insurance mainly mention about the combination between crop insurance and health insurance with the difference in level of subjects insured Some author named this attribute
In Nganje’s study (2008), the suitability of various insurance types is linked to individual household circumstances and their specific needs, with families having medical histories tending to prefer AGRH, while risk-averse households may opt for MPCIH Importantly, these insurance types adhere to the rule of mutually exclusive alternatives in Random Utility Models (RUM) However, numerous studies on flood insurance demand have breached this principle, employing mutually inclusive policies such as property damage, crop damage, health damage, and unemployment income insurance (Brouwer & Akter, 2010; Reynaud et al., 2012), as well as yield and rainfall insurance (Jorgensen & Termansen, 2015).
Some studies refer to implementing agencies as "providers," and it is crucial that the attributes of these agencies remain consistent to avoid confusing potential clients Authors should present uniform levels based on the proportion of business owners, such as government and private companies, or the products offered by insurance and micro-credit companies Inconsistencies in the attributes of providers can negatively impact study results.
The term "coverage level" typically refers to the percentage of damage insured (Nganje et al., 2008), but it is more accurately described as "indemnity insured level." Using the percentage of damage insured may not effectively encourage insurance buyers to engage in mitigation activities Instead, this attribute should be replaced with "deductible," which represents the portion of the loss that is not compensated and must be borne by the buyers (Liesivaara & Myyra, 2014) By sharing the burden of crop failure, farmers may be motivated to participate in mitigation efforts.
“deductibles” into insurance package is a good contribution, but this have yet to be applied in flood insurance studies.
In various studies on flood insurance demand, coverage is identified as the maximum financial support farmers can receive when their productivity falls below a specified threshold due to a disaster.
Research indicates significant issues with the scale of insurance coverage for flood damage, as highlighted by Brouwer et al (2013) and Akter (2010) Specifically, Brouwer (2010) found that insurance coverage is inadequate, with flood damages ranging from 224 USD to 738 USD per household, leaving medium-level damages uninsured Additionally, some studies have employed vague terms like "low" and "high" to describe coverage levels, lacking clarity and failing to reference specific regulations regarding these cover levels (Opiyo et al., 2014).
Challenges of disaster insurance market
Despite the absence of mandatory flood insurance participation in the Mekong River Delta, companies face significant challenges in expanding their market in this region Key obstacles include ambiguity surrounding policy details, adverse selection issues, and the phenomenon known as charity hazard, which can hinder the growth of demand for flood insurance (Botzen, 2010).
Ambiguity i s the difficulty in assessing accurately the probabilities of flood and its potential damages, predicting these issues requires a lot investment and many technical methods (Botzen,
2010) Moreover, the combination of climate change phenomena and variation of socio- economic characteristics contributes to make prediction to be more complicated (IPCC, 2007).
Simulation scenarios of climate change and socio-economic trends are essential for assessing future risks, particularly regarding floods and other adverse events Historical data and statistical methods can provide insights into the magnitude and trends of these occurrences (Saunders and Lea, 2008; Schmidt et al., 2010) However, these methods are limited by their reliance on historical data Despite advancements in computer-based techniques, such as catastrophe models, predicting the consequences of natural disasters remains challenging and often imprecise (Grossi & Kunreuther, 2005) Consequently, ambiguity persists, complicating the establishment of appropriate pricing levels for flood insurance.
2.3.1.2The impact of ambiguity on flood insurance demand
22Empirical study found that the more ambiguous adverse events are, the more insurance premium would be charged (Kunreuther, 1996) The flood insurance companies have to set high premium
Natural disaster insurance premiums tend to be higher than those for more predictable events like fire insurance due to the uncertainty surrounding the frequency and severity of such disasters This elevated cost can lead to decreased demand for flood insurance, as potential policyholders may find the premiums unaffordable.
The ambiguity and low probability of flooding can lead to underinsurance, as individuals often underestimate the risks associated with such events Research indicates that people may disregard crucial information about potential disasters, resulting in a lack of insurance purchases (Sunstein, 2002; Kunreuther et al., 2009) After experiencing a severe flood, many assume that another similar event won't occur for a century, a phenomenon referred to as the "100-year return" concept This perception, coupled with the low probability of flooding, can cause some clients to view flood insurance as an inefficient investment, leading them to abandon their insurance contracts (Kunreuther et al., 1978).
Experiencing significant losses from low-probability events, such as severe floods, can lead individuals to develop heightened awareness and concern for safety This increased perception of risk often results in a greater willingness to pay (WTP) for flood insurance and risk reduction measures Consequently, establishing higher premiums for flood insurance may be an appropriate strategy to meet this growing demand for protection.
2.3.1.3Vulnerability perspective is a good method the capture the impact of ambiguity on flood insurance demand
Local farmers' awareness of flood threats plays a crucial role in the development of the flood insurance market, even when potential flood damages are ambiguous and unpredictable If farmers acknowledge the risk of flooding, this recognition can drive the growth of flood insurance options, despite the uncertainties surrounding flood impacts.
The increasing popularity of media amplifies people's awareness of their vulnerability, significantly influencing their insurance purchasing decisions As socio-economic developments like population growth and improved living standards heighten the perception of risk, households with more members and assets recognize their susceptibility to adverse events (Miller et al., 2008) Additionally, social advancements raise awareness of vulnerabilities linked to issues such as rising greenhouse gas levels and global warming Consequently, heightened concerns about local flood risks lead to an increased demand for flood insurance and a greater willingness to pay for such coverage.
A study by Botzen and Bergh (2008) reveals that adopting a vulnerability perspective significantly enhances the willingness to pay (WTP) for flood insurance, indicating its strong positive influence on insurance demand Therefore, flood insurance providers may find it more beneficial to utilize the vulnerability perspective to assess the impact of uncertain risks rather than relying solely on historical data.
In the Mekong River Delta, residents rely heavily on media such as TV and radio for updates on weather and natural disasters, with all surveyed individuals reporting regular use of these sources Despite thorough preparations each year to mitigate flood damage, significant losses remain unpredictable The 2011 flood, supported by the United Nations and local government initiatives aimed at enhancing flood resistance, resulted in over 1,000 billion VND in property damage and devastated approximately 27,000 hectares of agricultural land This highlights the urgent need for farmers in the region to understand their vulnerabilities, making it crucial to assess how these vulnerabilities influence the demand for flood insurance among local farmers.
In addition to utilizing a vulnerability perspective to assess the effects of ambiguity, we also account for irrigation improvements and access to pumping stations among respondents, as these elements significantly influence the demand for flood insurance.
2.3.2.1Definition of adverse selection and its impact on flood insurance demand
Adverse selection in insurance occurs when coverage is predominantly offered to high-risk individuals at elevated premiums This phenomenon arises from the positive correlation between insurance demand and the expected risks faced by individuals, with premiums ideally reflecting these risk expectations However, information asymmetries complicate this dynamic, as potential buyers often have a clearer understanding of their risks than insurance companies, making it challenging for insurers to accurately classify customers Consequently, if premiums are set based on average expected losses, insurers may encounter capital shortages, leading them to impose high premiums exclusively on high-risk or risk-averse individuals, thereby perpetuating adverse selection (Botzen & van den Bergh, 2008).
2.3.2.2Reason for using risk perception or fear of individuals to determine adverse selection issue
A study reveals that even if insurance companies assess flood risks for different customer groups, issues stemming from information asymmetry persist Individuals' risk perception, which is often based on intuition, does not align with the predictions made by insurance companies (Slovic, 1987; Slovic, 2000) According to Kunreuther (2001), emotional responses, such as fear, significantly influence insurance purchasing decisions over rational calculations Additionally, when government relief is involved, individuals who perceive higher risks are more likely to purchase insurance, while those who believe they are at low risk may depend on assistance from government or charitable organizations (Kim and Schlesinger).
2005) Therefore, risk perceptions of individual plays an important role in making decision about mitigation and choosing self-protective measures (Burn, 1999; Flynn et al, 1999).
Rothschild and Stiglitz (1976) propose that the insurance market can reach equilibrium by categorizing natural disaster insurance into high-risk and low-risk options, emphasizing the importance of understanding individual risk perceptions and fears to effectively address adverse selection However, there is a scarcity of studies that explore how fear influences willingness to pay (WTP) for flood insurance, and the topic of individual risk perceptions regarding floods remains underexplored, leading to mixed findings in existing research (Peacock et al., 2005).
Flood risk perception studies have explored various factors influencing how individuals assess their vulnerability, including socio-economic development, household characteristics, and experiences from past disasters Geographic elements, such as proximity to hazards, also play a significant role in these perceptions Overall, increased awareness and anxiety regarding flood risks can lead to positive outcomes in preparedness and response.
Flood risk assessments highlight a significant correlation between individual fear of flooding and expert evaluations, as noted by Seigrist and Gutscher (2006) Their research utilizing flood risk maps indicates that while this fear is generally aligned with expert assessments, variations still occur based on specific regional factors (Brilly & Polic, 2005).
Demand for flood insurance
Natural disasters frequently impact agricultural production in the Mekong River Delta, leaving local farmers dependent on subsidies and humanitarian aid to manage cultivation costs Despite being identified as a promising market for insurance companies since 2013, agricultural insurance, particularly flood insurance, remains largely in the pilot phase Given that flood insurance is a relatively new offering for farmers in this region, employing choice experiments to assess its viability is an appropriate approach.
In the insurance sector, flood insurance demand is often analyzed using statistical data (Kunreuther et al., 2009; Michel-Kerjan & Kousky, 2010) Recent studies have employed the Choice Experiment Model to explore potential markets for flood insurance; however, while these studies offer valuable insights, they also contain several notable errors.
The advantages of Choice Experiment compared to Contingent Value Method
The Contingent Valuation Method (CVM) is a widely used non-market valuation technique employed by economists to assess individual preferences for public goods, environmental quality, and other specific commodities (Carson et al., 2005).
The Choice Experiment (CE), also known as the Random Utility Model (RUM), offers a significant advantage by enabling the simultaneous evaluation of how changes in insurance contract terms, such as premium levels and attributes, or environmental conditions influence insurance decisions In contrast, the Contingent Value Method (CVM) limits the number of periods and circumstances that can be presented to respondents.
The Contingent Valuation Method (CVM) can introduce significant bias when assessing the maximum willingness to pay (WTP) due to its reliance on a single question In contrast, Choice Experiments (CE) address this limitation by enabling respondents to compare and select among various insurance contract terms, providing a more accurate evaluation of their maximum WTP.
Given the benefits of the Choice Experiment, it is preferred for assessing how attributes, barriers, and household characteristics influence the decision to purchase flood insurance, instead of utilizing the Contingent Value Method.
General model
From alternative j, respondent can obtains utility as: � � = � � + � � (Marschak, 1960) With:� � : systematic component, can be observed by researchers.
� � : random component, cannot be observed by researchers.
The systematic component does not only mention about quantity of good but also about the characteristics of goods (McFadden, 1987)
� �� is attribute k level of alternative j,
� � is the marginal utility of attribute k,
� 0� is the alternative specific constant.
� � : the random variable with Gumbel distribution
Estimation
The Conditional Logit Model is commonly used to analyze individual decision-making among various alternatives This model emphasizes the characteristics of each alternative and the specific set of options available to each individual Research utilizing this model often tests hypotheses regarding choice behavior, suggesting that individuals will select the alternative that offers the highest utility The systematic component of this utility is influenced by the attributes of the alternatives, as defined by a specified function.
� �� = Pr(� �� > � �� ) with all k not equal to j The probability that alternative j could be chosen by individual n (� �� ).
• � �� = Pr (j is chosen among C) = Pr (� ��� �� >� �� , ∀ � ≠ �, �, � ∈ �)
In the Conditional Logit Model, while the influence of a unit of Z remains consistent across alternatives, the values of independent variables must vary among the choices If Z does not show variation across alternatives, it will not impact the choice probabilities, as the differences in characteristic values are essential for determining these probabilities.
The coefficients of the utility functions can be estimated by maximizing this log likelihood function:
After derivative, this log likelihood function will be:
When the log likelihood function is maximized, its first derivative equal to zero So the values of β estimated by maximum log likelihood function obey first-order condition.
N is the size of sample, with some arrangements and mathematical operations:
In this equation, the left side represents the observed average value of the z attribute in the alternatives selected by respondents (�̅), while the right side denotes the predicted average value of the z attribute in the alternatives expected to be chosen (�̂) Therefore, at the point of maximum likelihood estimates, the observed average is equal to the predicted average (�̅ = �̂).
The coefficients (β) derived from maximum likelihood estimation ensure that the observed average (�̅) equals the estimated average (�̂) This allows for the replication of observed averages using the model based on maximum likelihood estimation This property remains consistent even when alternative specific constants are introduced, where a dummy variable for alternative j is employed to represent the utility of alternative i in capturing sample shares.
Another feature is reflected in the first-order condition:
As residual is differential between the choice of respondents (y nj ) and choice probability of alternative j (P nj ), ∑ n
The covariance of residuals and independent variables is represented by the expression (y ij − P nj )Z ij In maximum likelihood estimations, the coefficients (βs) of independent variables are adjusted to ensure that this covariance is zero, indicating no correlation between the independent variables and the residuals This principle is similarly applicable to the first-order condition of logit models, much like in linear regression models.
To solve the first-order condition equations, the method of moments estimator is utilized, which determines estimates based on moment conditions or correlations between variables and modeling errors (Train, 2003) This approach relies on a sample that closely resembles the population, ensuring that the assumptions regarding exogenous variables hold true, meaning there is no correlation between the independent variables and residuals Consequently, the chosen estimates are significant, and the absence of correlation between residuals and independent variables in the sample supports the model's validity In summary, these estimates facilitate regression on an analogous population sample where covariances are zero.
3.4.2Estimation on Subset of Alternatives
This study involves a significant number of combinations, totaling 63 alternatives The research team randomly selected combinations for each choice card, which includes two alternatives and an option for "none of them." Given that each alternative can be included in the subset, the utility function must be estimated for these subsets in addition to the overall sample It is assumed that all alternatives have an equal probability of being selected.
The probability that alternative j will be chosen in subset H given denotes as �(�|�) Subset or choice cards H given to respondent must contain alternative j, if �(�|�) > 0 If the subset
H does not contain j, q( H | j ) = 0 ∀ H In the full set, the probability that alternative j is selected is � ��
The primary objective of estimation is to calculate the conditional probability of selecting alternative j, given that the research group has chosen subset H for respondent n This probability is essential for understanding decision-making processes in research contexts.
The nominator represents the joint probability of selecting subset H and the respondent choosing alternative j, while the denominator reflects the probability of selecting subset H from all possible alternatives This relationship allows us to express conditional probability effectively.
The research group assumed that any alternative \( j \) within subset \( H \) has the same probability \( P(j|H) \) due to the random selection of alternatives This leads to the "uniform conditioning property," which indicates that the probability of choosing alternative \( j \) is consistent regardless of the specific choice made by the respondent As a result, \( P(j|H) \) can be eliminated from the expression of \( P(y|H) \), allowing \( P(y|H) \) to be formulated as a logit function based on the alternatives available in subset \( H \) for respondent \( n \) (McFadden, 1978).
) Under this property, the conditional log likelihood function (CLL):
The conditional log likelihood (CLL) function, when applied to a selected subset K of respondents, aligns with the traditional log likelihood (LL) function Consequently, the estimator derived from the maximum CLL is consistent with that from the maximum LL However, the CLL estimator lacks efficiency because it omits certain information that the LL function encompasses.
Description of variables
3.5.1Description of all attributes and levels
Table 3.1 Attributes of flood insurance packages
Variable Notation Insurance Attribute Level
Insurance Policy ������ ��� This attribute allows respondents to choose kinds of disaster causing damages that the buyers will be compensated.
Flood Flood and inundation Flood and windstorm Flood, windstorm and inundation
Provider �������� ��� Which types of business owner will provides the insurance packages.
Foreign company Private company Corporation
In an insurance policy, the coverage refers to the maximum amount of compensation that policyholders will receive for damages caused by specified types of disasters, after deducting the deductible amount.
(%) ���������� ��� This attribute is the percent of damage that insurance buyers have to bear if disasters occur.
� ����� ��� The amount of money that insurance buyers have to pay per farming season for insurance company.
15,000 VND30,000 VND40,000 VND50,000 VND65,000 VND
An insurance policy is essential for allowing individuals to select the types of disasters that will be covered for damages Previous studies have shown that providing a clear list of disaster-related damages can help potential clients better understand their insurance options (Nganje et al., 2008; Mercadé et al., 2009) The research team aimed to improve upon past studies by ensuring that they included a comprehensive range of insurance types in their analysis (Brouwer).
& Akter, 2010; Reynaud et al., 2012) Levels of insurance policy include flood, flood and thunderstorm, flood and inundation, flood, thunderstorm and inundation.
In this study on flood insurance, it is essential to identify the preferred provider types, as previous research indicates that provider attributes significantly influence decisions (Nganje et al., 2008; Brouwer & Akter, 2010; Reynaud et al., 2012; Brouwer et al., 2013; Opiyo et al., 2014) The research team categorized providers based on business ownership, initially identifying three levels: state-owned, foreign, and private companies However, early surveys revealed that some farmers exhibited a strong preference for state-owned companies, leading to a phenomenon termed "dominant choice," where respondents predominantly selected state-owned insurance plans To address this issue, the study replaced the state-owned company option with corporations that include state participation, effectively resolving the dominant choice problem Consequently, the final provider levels in this study are foreign companies, private companies, and corporations.
The cover represents the maximum payout an insurance buyer can receive after deducting the deductible, which helps insurance providers understand their indemnity limits For farmers, knowing the maximum indemnity amount aids in calculating cultivation costs and potential profits Previous studies have often confused respondents by expressing coverage levels as a percentage of damage, leading to misunderstandings (Reynaud et al., 2012; Nganje et al., 2008; Opiyo et al.).
In 2014, research indicated that the intelligibility of financial coverage for agricultural attributes is crucial, necessitating the establishment of specific monetary levels (Liesivaara & Myyra, 2014) To determine appropriate coverage levels, the research team analyzed statistics on flood damage and agricultural costs associated with paddy fields Farmers typically invest between 4 million to 5 million VND in their paddy fields each season, leading to the establishment of three coverage levels: 2,000,000 VND, 3,000,000 VND, and 4,000,000 VND per 1,000 square meters for each farming season.
Deductible attributes was used to capture the demand for crop insurance in the study of Liesivaara
According to Myyra (2014), there has been a lack of studies on disaster insurance that incorporate deductibles as a key attribute Utilizing deductibles is beneficial as it encourages farmers to take proactive measures in minimizing damage from floods and other disasters This study considers two deductible levels: 10 percent and 25 percent.
Many respondents are particularly focused on the premium attribute in agricultural insurance packages, as highlighted in previous studies However, the use of percentages to convey premium levels has led to confusion among respondents (Reynaud et al., 2012; Nganje et al.).
2008) The research team tried to use a specific amount of money as levels of premium
2009; Liesivaara & Myyra, 2014) Besides, the statistic results about farming profit per
1000� 2 from initial surveys was utilized to maintain the suitability of premium levels Profit per 1000� 2 are around 450,000 VND The levels of this attribute are 15 000, 30 000, 40 000, 50 000, 65 000
3.5.2Description of variables used to capture challenges for flood insurance market development
The development of the flood insurance market faces three key challenges: ambiguity, adverse selection, and charity hazard By examining the perspectives of farmers regarding their vulnerability, fear of flooding, and expectations of governmental support for post-disaster recovery, we can analyze how these factors influence the demand for flood insurance These perspectives are represented as dummy variables, allowing us to assess their impact on the utility experienced by potential flood insurance buyers.
This article explores the impact of ambiguity on local farmers' perceptions of flood vulnerability in the Mekong River Delta It examines whether farmers recognize the threat of floods despite the unpredictable nature of potential damages, suggesting that such recognition could facilitate the development of the flood insurance market Key variables analyzed include the interaction between flood vulnerability perception and insurance policy, as well as premium rates Additionally, the study investigates adverse selection by assessing how farmers' fear of floods influences the marginal utility of insurance attributes, with similar interaction variables considered for fear of flood and insurance policy and premium Furthermore, the concept of charity hazard is analyzed through farmers' perceptions of government responsibility and post-flood relief, employing interaction variables between charity hazard and policy, as well as premium.
The demand for flood insurance is influenced by various perspectives, which interact with the utility derived from different policy types and their associated premiums Understanding this interplay is crucial for insurance buyers to maximize the benefits of their coverage.
Table 3.2 Variables used to capture impacts of challenges for flood insurance development
Variable Notation Definition Perceptive Levels
This variable is used to capture the impacts of flood ambiguity.
It reflects the perceptive of flood vulnerability of respondent �.
High or extremely high flood vulnerability = 1 Low or very flood vulnerability = 0
This variable is used to capture the effects of adverse selection.
It reflects whether respondents have the anxiety or fear of flood damages or not
Fear of flood = 1 Not fear of flood = 0
This variable is used to capture the impacts of charity hazard.
This variable reflects whether respondents depend so much or hazard their recovery after flood on government
Hazard the post-flood recovery on government = 1, or not = 0
This variable is used to capture the impacts of irrigation improvement.
It reflects whether respondent recognizes that irrigation in local area are improved or not
This variable is used to capture the impacts of pumping station accessibility.
This variable reflects whether respondent’s fields can access to pumping to pumping stations or not
Having accessibility to pumping stations = 1
Table 3.3 The interactions between perceptive about flood insurance challenges and attributes
Interaction between flood vulnerability perceptive and utility of flood insurance policy types
This variable reflects the impact of ambiguity on utility of flood insurance policy types
Interaction between flood vulnerability perceptive and utility of premium levels
� � ��� This variable reflects the impact of ambiguity on utility of premium levels
Interaction between fear of flood and utility of flood insurance policy types
� ���� ��� This variable reflects the impact of adverse selection on utility of flood insurance policy types
Interaction between fear of flood and utility of premium levels
This variable reflects the impact of adverse selection on utility of premium levels
Interaction between charity hazard and utility of flood insurance policy types
� ���� ��� This variable reflects the impact of charity hazard on utility of flood insurance policy types
Interaction between charity hazard and utility of premium levels
This variable reflects the impact charity hazard on utility of premium levels
Interaction between irrigation improvement and utility of flood insurance policy types
This variable reflects the impact of irrigation improvement on utility of flood insurance policy types
Interaction between irrigation improvement and utility of premium levels
This variable reflects the impact of irrigation improvement on utility of premium levels
Interaction between accessibility to pumping stations and utility of flood insurance policy types
This variable reflects the impact of pumping station accessibility on utility of flood insurance policy types
Interaction between accessibility to pumping stations and utility of premium levels
� � ��� This variable reflects the impact of accessibility to pumping stations on utility of premium levels
Empirical models
3.6.1Empirical model with only attribute variables
� ��� denotes the indirect utility of individual n who has chosen alternative j.
Insurance policies can cover various types of disaster damage, including those caused by floods, thunderstorms, and inundations It is anticipated that the type of policy selected will have a positive relationship with the utility received by the insurance buyer Specifically, policies that cover damages from all three disasters are expected to offer the greatest positive impact, while those that cover only flood damages are likely to provide the least benefit.
The selected insurance provider for respondent n could be a foreign company, a private company, or a joint stock company Analyzing the coefficients of these providers will assist in identifying the most preferred options.
����� ��� is the maximum amount of compensation j for natural damage that insurance companies will pay for insured n, after minus the deductible (VND/1000� 2 / Farming season)
We expected this attribute will have positive impact on the utility of flood insurance buyers.
The percentage of damage that insurance buyer n must bear, denoted as j (%), indicates the level of shared responsibility between the buyer and the insurance provider This attribute is anticipated to negatively affect the demand for flood insurance, as potential buyers may be discouraged by the requirement to absorb some of the financial burden.
� � ��� is the insurance premium j that buyer n have to pay for insurance company (VND/1000� 2 / Farming season) We expected that premium will have the negative relationship with flood insurance demand.
� ��� is the unobserved random term.
3.6.2 Empirical model with attribute variables and their interaction with non- attribute variables
The Mekong River Delta lacks mandatory flood insurance participation, highlighting significant barriers to the development of the flood insurance market, including risk uncertainty, adverse selection, and charity hazard (Botzen & van de Bergh, 2008) To understand the effects of these challenges, local farmers' perceptions were analyzed in relation to flood insurance attributes (Botzen, 2010) The results from the model will reveal how the marginal utilities of these attributes fluctuate in response to the identified challenges.
In this model, the expected relationships between attributes and flood insurance demand are similar to the basic model above.
There are expectation for interaction between flood insurance attribute and flood insurance market development challenges or irrigation improvement or accessibility to pumping stations in the circumstance t:
Flood vulnerability plays a significant role in influencing flood insurance demand, as it highlights the effects of uncertainty This relationship can be positively correlated with different types of insurance policies, indicating that higher flood vulnerability may lead to increased interest in flood insurance options The interactions between flood vulnerability and insurance policy types suggest a beneficial connection, emphasizing the importance of understanding these dynamics in the insurance market.
(������������� ��� ) and premium are expected to be positive.
Adverse selection is influenced by the fear of flooding, which we anticipate will positively correlate with the demand for flood insurance Consequently, the interactions between this fear and various insurance policy types are also expected to show a positive relationship Furthermore, we expect the coefficients of interaction between flood fear and insurance premiums to be positive as well.
Local farmers who rely on government assistance for post-flood recovery are less likely to purchase flood insurance This dependency on governmental support diminishes their motivation to invest in insurance, highlighting a significant relationship between perceived responsibility and insurance uptake.
�ℎ�����ℎ����� ��� , the coefficients of interactions of charity hazard (�ℎ�����ℎ����� ��� ) and flood insurance policy types (������ ��� ) might produce negative sign The coefficient of
�ℎ�����ℎ����� ��� ������ � ��� are expected to be negative.
Improvements in irrigation and access to pumping stations are likely to decrease the demand for flood insurance, similar to the concept of charity hazard Consequently, the interaction coefficients between these improvements and various types of insurance policies are anticipated to be negative.
� � ��� are expected to be negative.
3.7 Calculation of Willingness-to-Pay (WTP) for specific insurance packages and probability of buying specify insurance packages with the changes in premium
3.7.1 Calculation of Willingness to Pay (WTP) for each attribute and for specific insurance packages
The WTP for each attribute of choice is
A comprehensive flood insurance package should encompass five essential components: a flood insurance policy, a reliable provider, coverage amount (VND/1000 m² per farming season), deductibles (%), and premium costs (VND/1000 m² per farming season) It is crucial to eliminate any unrealistic insurance options to ensure effective protection against flood risks.
48 specific insurance packages in the survey.
Initial calculations reveal that the cover and deductible have minimal impact on the willingness to pay (WTP) of local farmers for flood insurance The accompanying graphs in the appendix illustrate the probability of WTP across various cover and premium levels Consequently, the analysis focuses solely on specific insurance packages with different combinations of policy types and providers, particularly those offering a cover of 2 million VND and a deductible of 25 percent This evaluation not only provides insights into farmers' WTP for flood insurance but also highlights how policy types and providers influence their willingness to pay.
Table 3.4 The insurance packages are used to calculate the WTP
The attributes of insurance packages are used to calculate the WTP
1 Flood Corporation 2 million VND 25 percent
2 Flood and Inundation Corporation 2 million VND 25 percent
3 Flood and Windstorm Corporation 2 million VND 25 percent
Windstorm Corporation 2 million VND 25 percent
5 Flood Foreign companies 2 million VND 25 percent
6 Flood and Inundation Foreign companies 2 million VND 25 percent
7 Flood and Windstorm Foreign companies 2 million VND 25 percent
Windstorm Foreign companies 2 million VND 25 percent
9 Flood Private companies 2 million VND 25 percent
10 Flood and Inundation Private companies 2 million VND 25 percent
11 Flood and Windstorm Private companies 2 million VND 25 percent
Windstorm Private companies 2 million VND 25 percent
In the basic model, calculation of Willingness to Pay (WTP) for specific insurance packages will be
� ∗ = � 0 + � 1 (Flood; Flood and thunderstorm; Flood and inundation; Flood, thunderstorm and inundation) + � 2 (Corporations; Foreign company; Private company) + � 3 (2 million VND; 3 million VND; 4 million VND) + � 4 (10%; 25%)
The second model examines how the development of the flood insurance market, along with challenges such as irrigation improvements and accessibility to pumping stations, affects individuals' willingness to pay (WTP) for flood insurance By incorporating individual characteristics (Z) of respondents, the model effectively captures the varying impacts on WTP for specific insurance packages.
∗ ∗ ∗ will equal the quotient of dividend (� � ) and (−� $ ).
The total utility of flood insurance packages, excluding the premium, represents the cumulative marginal utility derived from all their attributes This utility reflects the benefits a flood insurance buyer would enjoy if they were exempt from paying the premium Conversely, the marginal utility of the premium carries a negative value, indicating a decrease in overall satisfaction due to the cost involved.
The impact of natural disasters such as floods, windstorms, and inundations can significantly affect corporations, whether they are foreign or private companies Financial implications range from 2 million VND to 4 million VND, with potential losses varying between 10% to 25% Understanding these risks is crucial for businesses operating in areas prone to such events, as they must prepare for the possibility of severe weather-related disruptions.
� ���� + � 10 (Flood; Flood and windstorm; Flood and inundation; Flood, windstorm and inundation)* ��ℎ�����ℎ����� +
� 10 (Flood; Flood and windstorm; Flood and inundation; Flood, windstorm and inundation)* ����������� + � 12 (Flood; Flood and windstorm; Flood and inundation; Flood, windstorm and inundation)* ���������������
3.7.2Probability of buying specify insurance packages with the changes in premium levels
Based on the basic model, there is the calculation of probability of buying specify insurance packages with the changes in premium levels
Based on the calculation of Willingness to Pay (WTP) for specific flood insurance packages and their attributes, we will identify preferred flood insurance plans to assess the likelihood of purchasing these packages as premium levels fluctuate By comparing the probability of WTP across different premium levels, we can determine which flood insurance package is most viable for investment and launch in the Mekong River Delta.
To gather data for this study, we conducted a survey in three randomly selected districts in the Mekong River Delta—Gao Giong, Phu Loc, and Tan Cong Chi—previously impacted by flooding Households were chosen through a random selection process.