Be Happy, Be Loyal? Exploring Drivers for Renewal of Mobile-Delivered Index Insurance

Abstract Agricultural microinsurance is a promising risk management tool for smallholder farmers. However, adoption rates remain low and only a small share of farmers renew their policy after the first period. With the increasing availability of cell phones, mobile-delivered insurance is gaining importance on the market. As for any agricultural microinsurance, it is essential for the longevity of a mobile-delivered insurance scheme to retain a solid customer base. To date, it is unknown what drives the decision to renew a mobile-delivered agricultural microinsurance policy. We address this question by performing mean comparisons and logistic regressions based on primary data collected from 479 smallholder farmers in Mali who purchased a mobile-delivered weather index-based insurance in 2020. Results show that the level of satisfaction with the insurance product was considerably higher among farmers who renewed. We found low levels of understanding of the product among all clients, but especially among those who did not renew. Both factors were confirmed as drivers for renewal. Consistent with previous findings, receiving a payout had the strongest effect on the decision to renew. We conclude that additional measures to foster client satisfaction as well as to promote understanding of agricultural insurance among smallholder farmers are highly recommended.


Introduction
Given the increasing frequency of extreme weather events such as droughts and floods, agricultural microinsurance gains importance as a risk management tool for smallholder farmers.Index-based insurance schemes, in particular, seemed to pave the way for microinsurance success through providing indemnities based on objectively measurable factors that are highly correlated with yield losses.Thereby, they largely overcome adverse selection and moral hazard, eliminate costly and time-intense on-farm loss assessments, and facilitate the sales process as the product is relatively easy to understand (Barnett & Mahul, 2007).
With digital technologies on the rise, insurance schemes increasingly rely on digitally-enabled marketing solutions.Cell phones are used to sell and distribute insurance products.Depending on the degree of digitization of an insurance scheme, interested farmers can subscribe to the insurance, pay premiums, receive indemnities and also renew the policy via their cell phone (Raithatha & Priebe, 2020).The use of cell phones, particularly of mobile money, is assumed to facilitate the customer management, for example by allowing timely indemnity payouts (Benami & Carter, 2021).If and how the change from conventionally distributed to mobiledelivered insurance also shifts drivers for demand of index insurance has-to the best of our knowledge-not been investigated yet.In the microcredit sector, the drivers of demand for mobile money loans were found to differ from drivers of demand for conventional microloans (Sarfo et al., 2021), thereby highlighting the need for similar assessments for the microinsurance industry.
Demand dynamics, particularly the adoption decision of insurance, have been intensively studied over the last years as the adoption rates of index-based insurance often lag behind expectations (King and Singh, 2020;Lichtenberg & Iglesias, 2022;Platteau et al., 2017).While knowledge on the adoption decision is crucial for insurance providers to successfully attract the target group, it is equally important to keep these early adopters and to build a loyal client base.The share of clients that renew their policy for the next period is coined as the renewal rate.Since insurance schemes do not only pool risk across their clients but also over time, the renewal rate is an important determinant of financial sustainability for an insurance scheme (Apostolakis et al., 2015).
The decision to renew differs from the adoption decision in that the policy holder already gained a first experience with the product.Findings regarding drivers of adoption cannot be transferred directly as some factors may lose importance while other additional influencing factors come into play.The most prominent additional factor is the potential experience of a payout.For conventionally distributed insurance, receiving a payout was found to have positive effects on future insurance purchases, directly as well as indirectly through, for example, increased levels of trust to the insurance (Cai et al., 2020;Cole et al., 2014;Hill et al., 2016;Karlan et al., 2014;Stein, 2016).
The objective of the present study is to shed light on the renewal decision of mobile-delivered insurance schemes in order to promote sustainable, mobile-delivered microinsurance programs.While a few studies addressed drivers for microinsurance renewal (Cai et al., 2020;Cole et al., 2014;Hill et al., 2016;Karlan et al., 2014;Stein, 2016), none of the studied insurance schemes was digitally distributed.To date, it is unclear what drives the renewal of mobile-delivered insurance policies and whether, compared to conventionally distributed insurance, there are additional factors to be considered.We specifically focus on the renewal decision as we assume that the adoption decision for a mobile-delivered insurance is similar to the adoption decision in a conventionally distributed insurance whereas the renewal decision may differ in both distribution channels.The main research question of this paper is therefore: what are drivers for renewal of a mobile-delivered index-based insurance policy?In the same context, we also address the question how self-stated motives for insurance adoption differ from self-stated motives for insurance renewal to see whether there is a difference between the adoption and the renewal decision.
In addition to that, this paper provides descriptive evidence on the question whether there are systematic differences between customers who renew and one-time buyers of weather indexbased insurance policy.So far, it is unclear if there are factors that predetermine whether a farmer is likely to renew the insurance or not.There could be systematic differences in socioeconomic characteristics between one-time adopters and farmers who intend to renew their insurance.Identifying such differences would be beneficial for improving targeting as well as client management throughout the product cycle.
To address these questions, we conducted a case study on the private insurance provider OKO Mali SaRL (OKO) which issues mobile-delivered weather index-based insurance policies in southern Mali.We took a quantitative research approach based on primary data of 479 farmers who were insured with OKO in 2020 and whereof 282 respondents renewed their policy for 2021.The results of our study are of special relevance to practitioners striving for long-term viability of the insurance programs they offer, especially when considering to offer mobile-delivered products.We also address policy makers when highlighting the need to further promote financial literacy and thereby the understanding of insurance products.
The present paper proceeds with a theoretical framework on insurance renewal (Section 2), followed by an overview of relevant literature (Section 3) and the description of the case study context (Section 4).In Section 5 we then outline our methodological approach before we present and discuss our findings in Section 6.Finally, we highlight the most important conclusions (Section 7).

Theoretical framework
Regardless of the distribution channel of an insurance, the renewal decision for a voluntary insurance scheme differs from the initial adoption decision as the policy holder already gained experience with the insurance product.We illustrate the difference in Figure 1.At the point in time t 0 , a decision-maker faces the choice whether or not to take out a certain insurance policy for the first time.Personal factors were found to influence the initial adoption decision.Among others, these factors are personal risk aversion (Elabed & Carter, 2015;King & Singh, 2020;Lampe & W€ urtenberger, 2020) and risk perception due to past shocks (Cai & Song, 2017;Cole et al., 2014;Stein, 2016), trust in institutions and towards the insurer (Cai et al., 2015;Cole et al., 2013;Cole & Xiong, 2017;Platteau et al., 2017), insurance understanding (Cai et al., 2020;Kramer et al., 2022;Lampe & W€ urtenberger, 2020;Stoeffler & Opuz, 2022), the use of other risk management tools that may work as substitutes (Berg et al., 2022;King & Singh, 2020) as well as socioeconomic aspects like wealth and available liquidity (Belissa et al., 2019;Casaburi & Willis, 2022;Cole et al., 2013;King & Singh, 2020).These factors constitute personal influencing factors as illustrated on the left-hand side of Figure 1.
Decision-makers also consider whether the design of the insurance policy is adequate for their needs.Product details like the premium (Hill et al., 2016;Karlan et al., 2014;Matsuda & Kurosaki, 2019;Stoeffler & Opuz, 2022), contract design (Elabed et al., 2013;Norton et al., 2014) as well as the design of the insurance policy, namely basis risk for weather index-based Be Happy, Be Loyal?Exploring Drivers for Renewal of Mobile-Delivered Index Insurance 833 insurance schemes (Elabed et al., 2013;Kramer et al., 2022;Lichtenberg & Iglesias, 2022), come into play here as shown on the right-hand side of Figure 1.
If the decision maker took out insurance at time t 0 , the person must decide at time t 1 whether or not to renew the insurance for the next period.Being insured between t 0 and t 1 implies that the policy holder went through the whole product cycle from initial consultation, subscription, premium payment, and potentially also claim management.This experience may alter personal factors such as trust towards the insurer or insurance understanding.While the characteristics of the product in t 1 are the same as in t 0 , experiencing the product may have improved the understanding of the product.Hence, the renewal decision is influenced by the new factor 'product experience' as well as by a slightly modified set of personal influencing factors and the understanding of the product.
Following this explanation, we argue that the purchase decision at time t 0 is different from purchase decisions at t 1 or later times t x .Other studies used panel data and did not differentiate between initial adoption and insurance purchases in subsequent periods (Cole et al. 2014;Stein 2016).In this study we explicitly focus on the insurance renewal decision at time t 1 , thereby ensuring that all respondents have exactly 1 year of experience with the product.

Literature background on insurance renewal
Even though knowledge on the renewal decision of microinsurance is scarce (Platteau et al., 2017), there are a few studies that assessed demand dynamics and potential drivers for microinsurance.These studies focused on microinsurance schemes that are sold via in-person discussions, namely door-to-door visits and village meetings (Cole et al., 2017;Hill et al., 2016;Stein, 2016).In the present study, we look at a mobile-delivered insurance scheme which does not entail personal encounters for the renewal decision.Despite this detrimental difference in sales modes, we outline previous findings regarding microinsurance renewal in the following.
There is wide consensus that experiencing a payout increases the demand for insurance in the subsequent period (Cai et al., 2020;Cole et al., 2014;Hill et al., 2016;Karlan et al., 2014;Stein, 2016).Even though payouts were only triggered based on previously defined events, all studies confirmed that it is the payout itself and not eventually incurred losses that increase the likelihood for renewal.The identified magnitude and duration of payout effects, however, vary.Analyzing an area-yield index insurance for rice farmers in China, Cai et al. (2020) observed a short-term effect for all indemnified clients but long-term effects only persisted among financially literate farmers.Cole et al. (2014), working on a rainfall index insurance in India, found that in the short-run renewal increased by 25-50% if another household in the same village received a payout regardless of one's own payout experience.The latter only gained importance in the long-run.Evidence on village level effects is mixed.While Karlan et al. (2014) also found spillover effects in their study among rural Ghanaians, other studies could not identify social effects of payouts on the renewal decision (Hill et al., 2016;Stein, 2016).Cai et al. (2020) showed that experiencing a payout in the village has a positive effect on demand among households that were uninsured, but not on the already insured farmers.
Payouts are part of the product experience when first taking out an insurance.They are sought to have indirect effects on other general factors that influence the demand for insurance products (see Figure 1).One explanation for the strong payout effect is that receiving a payout may improve product understanding through learning by using 1 (Hill et al., 2016;Stein, 2016).Cai et al. (2020) tested this hypothesis and showed that receiving a payout served as a one-time learning experience for households that received financial literacy training prior to the initial take up decision.In contrast, for households with low levels of financial literacy, the renewal decision was based on the most recent experience with the product.Hill et al. (2016) identified strong positive effects of intensive training on insurance understanding for the initial uptake of a rainfall index insurance in India, but these effects did not persist in the subsequent year with regard to the renewal decision.They did not explicitly link the effect of the training with payout experiences.Cole et al. (2014) also argued that the observed village level effects of payouts could potentially be explained by knowledge spillovers.
Building up trust towards the insurer could be another explanation for the strong payout effect on renewals (Hill et al., 2016).Again, trust towards the insurer is a personal factor influencing the insurance demand (see Figure 1).Trust is an important aspect in insurance business since clients need to trust the insurer to payout as agreed and, thus, actually receiving an indemnity may increase trust (Platteau et al., 2017).While Cole et al. (2017) reported that levels of trust increased strongly among Indian farmers who received a large payout, Cai et al. (2020) did not find a trust enhancing effect of payouts.They also tested for two additional indirect effects of payouts, namely a change in risk aversion and relaxed liquidity constraints due to the receipt of a payout, but neither was confirmed.
Besides payout related effects, the influence of prices, subsidies and habit formation on the renewal decision were assessed in greater detail.As to be expected, insurance policy price increases had negative effects on the purchase rates (Cai et al., 2020;Hill et al., 2016;Karlan et al., 2014).Full subsidies were found to reduce the payout effects as customers are less attentive to insurance outcomes when not paying for it (Cai et al., 2020).Price anchoring effects in the period following the receipt of a subsidy were not observed (Cai et al., 2020;Hill et al., 2016).Habit formation was ruled out as well which implies that insured farmers also actively take a new purchase decision (Cai et al., 2020).

Case study context
The present study is based on a cross-sectional primary dataset on maize farmers from Mali.Mali is a landlocked country located in West Africa that heavily depends on the agricultural sector in terms of GDP contribution (36.0% in 2021, World Bank, 2022), labor force employment (68.1% in 2020, ILOSTAT, 2021) and general livelihoods through subsistence farming (FAO, 2017).At the same time, high weather risks put agricultural success at stake.Due to climate change, more extreme dry and rainy seasons are expected.Given that crops are predominantly rain fed, drought hazards in particular are predicted to increase (Tomalka et al., 2020).Mali is considered a least developed country (LDC) and it is among the countries with the lowest Human Development Index (HDI) score (UNDP, 2022).In this context, a stable insurance scheme for smallholder farmers may be key to reduce the vulnerability of households to climate-related stressors.
The studied insurance scheme was designed by OKO and is the first commercial agricultural insurance scheme in Mali.It insures against droughts and floods based on satellite derived precipitation data.Strike levels differ depending on the location, time in the season, and the insured crop.Insurance premiums are determined on an individual basis using time series data on weather and site-specific characteristics such as elevation, slope, and proximity to water bodies.The insurance is only sold in southern parts of Mali, namely in the regions Kayes, Koulikoro, S egou, Sikasso and Bamako region.
OKO offers a mobile-delivered insurance and as such, the insurance has to be contracted via a mobile phone.Interested farmers can get informed by dialing certain combinations on their mobile phone, thereby navigating through a USSD 2 menu.Through the USSD menu farmers can request a callback to get in touch with an agent in the call center who will then provide further information and an offer.OKO's insurance product does not require the use of a smartphone.For initial subscription, OKO agents often travel to rural areas, advertise and inform about the insurance and facilitate the registration process.When it comes to renewing the insurance for the following season, most farmers decide whether or not to renew without meeting an insurance representative again.Premiums have to be paid via mobile money and, given the Be Happy, Be Loyal?Exploring Drivers for Renewal of Mobile-Delivered Index Insurance 835 client is eligible for a payout, it will be automatically disbursed to the same mobile money account.
The strong reliance on cell phones is facilitated by the high availability of cellular infrastructure in Mali which is clearly above average for an LDC.According to ITU (2023), the whole Malian population (100%) is now covered by the mobile cellular network.In 2022, there were 114 cell phone contract subscriptions per 100 inhabitants in Mali (ITU, 2023).Even though data on actual mobile phone ownership is not available, the number of mobile phone subscriptions suggests that most likely the majority of the Malian population has access to a cell phone.Hence, we assume that the sample selection bias through the distribution via mobile phones is only minimal.
OKO issued insurance policies for the first time in 2020.While they first focused only on maize insurance, they broadened their product range to four other crops in 2021.Out of 1815 clients who contracted OKO's maize insurance in 2020, 1316 clients renewed their insurance policy for 2021. 3Insurance renewal required active subscription to the scheme including premium transfer for the new period.With a renewal rate of 72.5% in the first year OKO is an ideal research subject for the analysis of drivers for renewal of weather index-based insurance.
The external validity of our findings in other country contexts is likely to be conditional on the availability of a sufficiently well-developed cellular infrastructure allowing for wide-spread access to the mobile-delivered insurance product.Other countries with similar socio-economic conditions will also reach full cellular network coverage in the near future, thereby making Mali a suitable reference point.In addition, the term "mobile-delivered" insurance has not yet been standardized which complicates the transferability of findings between insurance programs with a varying degree of digitization.It is important to consider the level of digitization of OKO's insurance product when projecting our findings on other insurance programs.

Data
The primary data were collected via an in-person survey conducted in October and November 2021.It focused on the customer experience with the OKO product in the main season of 2020.The time of the survey allowed to capture information on whether respondents renewed their policy for 2021.Respondents were sampled via a stratified random sampling approach using OKO's entire customer base of 4913 registered clients.The first sampling frame contained all clients who registered for information about the product, but finally did not take out the insurance (n ¼ 3098). 4For the renewal analysis, the second sampling frame that includes all clients who bought the maize insurance policy in 2020 (n ¼ 1815) constitutes our study population as only those who were insured also faced the decision whether or not to renew.The insured farmers were divided into four stratas based on the receipt of an indemnity (yes/no) and the decision to renew (yes/no).The sampling did not impose any regional focus.A total of 842 farmers were surveyed, 664 of whom were insured in 2020.
The collected survey data was complemented by data on compensation payments in 2020, and on insurance premiums for 2021 provided by OKO.The renewal analysis is based solely on those respondents for whom information on payouts for 2020 and premium payments for 2021 was available.As not all respondents could be unequivocally matched to OKO's payment data, 5 the final sample reduced to 479 respondents who were insured in 2020 and out of which 282 clients renewed their insurance policy for 2021.
Summary statistics of the sample used in this study can be found in Table 1.The sample mainly contains male respondents (95%) which is assumed to be in line with the target population.For agricultural insurance a predominantly male clientele is not uncommon (e.g.Ghosh et al. 2021;Temesgen Keno Belissa et al. 2020).In order to set sample household characteristics in a greater context, data collected for the Malian Agricultural Survey in 2017 served as a basis for comparison.The household characteristics in the sample largely correspond to typical agricultural households in Mali in terms of housing characteristics (walls, roof, and sanitary facilities) and characteristics of the household head (gender, age, education).The household size, however, is substantially larger than the average farming household in Mali (21.8 persons compared to 11.3 persons).Similarly, the cultivated area at ca. 15.2 ha is higher than the average cultivated area in southern parts of Mali which varies between 7 and 10 ha depending on the region.Yet, in the agricultural survey, farmers overestimated their plot size by 28% on average 6 (CPS 2018).In case a similar overestimation applies to our sample, the cultivated area per household still remains on average slightly higher within our sample as opposed to regional averages.
Regarding mobile phone usage, nearly all respondents in the sample owned a cell phone and more than half of the respondents used it frequently.Given the earlier mentioned mobile phone subscription ratio in Mali of 114 subscriptions per 100 inhabitants, this level of mobile phone ownership may correspond to national averages (ITU, 2023).
Despite our social group of interest being only insured farmers, our initial dataset also allows us to compare our sample to uninsured farmers. 7As expected, uninsured farmers differ from the insured farmers in key aspects.On average, farmers who did not buy the insurance in 2020 faced higher insurance premia, were less risk averse, and used mobile money less frequently.This indicates that we are looking at a select group of farmers. 8The insured farmers can be considered early adopters of the insurance scheme as we are looking at the first years of operation of the analysed insurance scheme.Possibly, our results would be less pronounced if we were looking at an established insurance program.While our cross-sectional data does not allow us to account for this potential bias towards early adopters, it is an important aspect to consider when discussing the magnitude of the results.

Approach to data analysis
Depending on the research question we took different approaches.We first performed mean comparisons to identify differences between clients who renewed and clients who did not renew their policy.For continuous variables, t-tests for two independent samples, here renewals (n ¼ 282) and non-renewals (n ¼ 179), were applied.We used Levene's robust test statistic to check for equality of variances and accounted for inequality in the t-tests accordingly (Levene, 1960).For binary variables, we performed proportion tests.
To identify drivers for renewal, we ran logistic regressions on the binary outcome whether or not a client renewed the insurance policy using robust standard errors.As the renewal rate in Be Happy, Be Loyal?Exploring Drivers for Renewal of Mobile-Delivered Index Insurance 837 our sample is lower than in the study population, we used sampling weights to account for the differing distributions. 9The model specification followed a specific-to-general approach (Brooks, 2008) in order to capture the sensitivity of results when adding additional variables to the right-hand side.The basic model includes the insurance premium for 2021, a binary variable on the receipt of an indemnity in 2020 and the perceived harvest success in 2020.This model was then gradually expanded, first to include satisfaction and product understanding, then variables on future expectations, and finally agricultural activity.In all models we controlled for socioeconomic characteristics of the respondent.Reported marginal effects are average marginal effects.
In insurance research, a common problem is to distinguish the effects of a payout from the effects of the triggering event leading to the payout since-ideally-the payout should compensate for incurred losses.In index insurance, however, basis risk, resulting from an imperfect correlation between the index used and actual losses incurred, can only be kept minimal but is unlikely to be ruled out.This also holds for the analyzed insurance product, especially since it was their first year of service provision.In addition to that, we used the perceived harvest success in 2020 as a simplified approximation for harvest loss.The clients were asked to rate how their maize harvest in 2020 compared to a typical year on a five-point Likert scale.This selfstated assessment was uncorrelated with both payout-related variables.We argue that we do not only solve the problem of endogeneity between payout and triggering event by using the perceived harvest success, but that it is the perception of the loss rather than the actual loss incurred that potentially drives the renewal.
The insurance premium as well as the indemnity are given in CFA-Franc BCEAO, which is pegged to the Euro with a fixed exchange rate of 655.957CFA ¼ 1 Euro.As more than half of the non-renewal clients did not request an offer for 2021, missing values for the insurance premium in 2021 were replaced by the village averages. 10For eight observations, district level averages had to be used.Similarly, nearly half of the respondents did not give an estimation about future harvest losses.Arguing that it is most likely that the expected frequency is close to the experienced frequency, we replaced missing values with the latter in order to be able to use the variables.Furthermore, we ran a principal component analysis to approximate wealth based on housing quality, sanitation facilities, and asset ownership.An estimation of the first component was then employed as a wealth index (see Supplementary Material C for estimation results).
Lastly, we looked only at the clients who renewed their insurance policy (n ¼ 282) and compared their motivations driving the decision to adopt the insurance with their motivations for renewing the policy.Therefore, we conducted two-sample proportion tests.The respondents were asked for their motives to subscribe and for their motives to renew.For both questions they could choose multiple answers out of a proposed list of answers.The respondents also had the option to answer freely.Transforming the answers into dummy variables allowed to conduct the two-sample proportion tests to identify changes in the importance of reasons for purchase between the adoption and the renewal decision.

Differences between clients based on renewal decision
Between smallholders who renewed their insurance policy and those who did not we found substantial differences in terms of experience and understanding of the insurance product, but close to no disparities with regards to socioeconomic characteristics.Results are presented in Table 2 and Supplementary Material D.
In terms of product experience, the largest difference was found in the share of clients who received a payout.It was 62% higher among clients who renewed their insurance than among non-renewals.The average size of the indemnity did not vary statistically significantly between the groups suggesting that it is not the size of the payout but rather the fact whether or not the Notes: a Dummy variables take the value 1 if variable statement is true and 0 otherwise.b Measured on a 5-point Likert scale where −2 indicated not satisfied at all and 2 indicated very satisfied.c Dependence on maize takes 1 if the majority or the total income is derived from maize cultivation.d A harvest loss was defined as a loss of at least 25 % of the harvest in a typical year.The frequency of the harvest loss refers to a 10-year time period.e Measured on a Likert scale where −2 was a lot lower and 2 a lot higher than in a typical year.f Respondents were asked to rate their level of concern for their cereal production in general where −2 was not worried at all and 2 was very much worried.
Be Happy, Be Loyal?Exploring Drivers for Renewal of Mobile-Delivered Index Insurance 839 client received a payout which mattered.Similarly, the level of satisfaction with the insurance product differed statistically significantly between the groups.Clients who renewed their policy were on average (very) satisfied with the service while non-renewals were rather neutral.Both groups also differed statistically significantly in terms of product understanding.Farmers who renewed showed substantially higher levels of understanding on eligibility criteria for an indemnity and on how to receive an indemnity.This difference was partially expected since the data collection took place after the indemnity of the season 2020 had been issued.Given the high share of farmers who received a compensation in the group of renewals as opposed to the low share in the other group, it is possible that renewal clients learned about the product through the receipt of the indemnity.Nevertheless, among farmers who renewed, the overall level of understanding also remained low as only 55% correctly answered the three questions regarding basic product characteristics.
Despite probable learning effects, statistically significant differences in the initial reasons for subscription suggest that the clients who renewed already showed higher levels of understanding from the beginning on.Among the renewals a substantially lower share (35% as opposed to 56%) stated that they had taken out the insurance merely out of interest in receiving an indemnity.Other less important reasons for subscription were peer behavior and cooperative recommendations.These reasons were more important for clients who did not renew as compared to those who renewed with both differences being statistically significant at the 10% level.Hence, farmers who renewed their insurance policy for 2021 already took the decision to buy the insurance for a first time in 2020 more independently than clients who did not renew their insurance.This allows us to assume that referral bonusses to attract new customers through peer referrals can only be considered a long-run investment if efforts are undertaken to ensure high levels of client satisfaction and a good product understanding.
In terms of socioeconomic variables, 11 there were only minor differences.We cannot derive direct implications for a specific target group.Both groups did not differ in terms of educational level, reading ability, household size, household composition, cultivated area and mobile phone use.Considering that there was a statistically significant difference in mobile money use between insured and uninsured farmers suggests that farmers who are not at ease with mobile phones may have already been excluded during the adoption decision.Regarding age, the groups differed statistically significantly but the difference was, at only three years, fairly small.It does not imply a necessary focus on an older or younger target group.The wealth indicator was statistically significantly higher for respondents who renewed.
Farmers who renewed seemed to perceive higher risks of harvest losses while farmers who did not renew engaged more in risk diversification.There were no large differences in terms of dependence on maize.In both groups, roughly one quarter of the respondents derived the majority of their income from maize cultivation.Similarly, there was also no variation in terms of level of worries for the cereal production.However, farmers who renewed their insurance policy reported on average a statistically significantly higher number of harvest losses due to droughts or floods during the last ten years, even if the difference was only small.They also expected more harvest losses to occur in the upcoming decade.The latter difference, however, has to be taken with caution as roughly half of the respondents did not give any estimate on future harvest losses.Still, the share of farmers who diversified risk by engaging in livestock farming was 19% higher among farmers who did not renew their insurance.This suggests that livestock farming and insurance products are perceived as substitutes rather than complements.

Drivers for renewal
The results of the logistic regression models on the binary variable whether or not a client renewed the insurance policy confirm payouts as the main driver for the renewal decision of the analysed mobile-delivered insurance scheme.Results are presented in Table 3 with the base model in column ( 1) and the full model in column (4).The corresponding marginals effects are presented in Table 4. Based on variance inflation factors of below 5 for all variables, we could rule out multicollinearity.
The direction of the effects and the statistical significance persisted across all models.While harvest success showed a slightly positive effect on renewal, the coefficient for the insurance premium in 2021 was slightly negative, but always close to zero.Yet, both effects were not statistically significant across all models.In contrast to that, receiving a payout, regardless of its amount, was found to strongly increase the likelihood for renewal in all models.Even though the effect size reduced when including more explanatory variables, its marginal effect in the full model still indicated a 27.9% increase in the likelihood for renewal if the client received a payout.
The strong effect of payouts is in line with previous findings for conventionally distributed insurance (Cole et al., 2014;Hill et al., 2016;Karlan et al., 2014;Stein, 2016).Given the negative, though not statistically significant effect of increasing premiums per ha, it is questionable whether implications for product design should be derived from the strong effect of insurance payouts.One could argue that insurance schemes that favor small but frequent payouts are probably more likely to achieve high renewal rates.However, this would come at higher costs and thereby probably shift the clientele towards better off farmers.Hence, this approach needs to be considered carefully.Be Happy, Be Loyal?Exploring Drivers for Renewal of Mobile-Delivered Index Insurance 841 Besides payouts, the level of satisfaction was found to be an important driver for renewal.A one unit increase on the 5-point rating scale for satisfaction was associated with, on average and all else being equal, a statistically significant 6.0% increase in the probability for contract renewal in model 2. It seems that farmers who are satisfied with the service are more likely to be loyal clients for the insurance scheme.This gives rise to the question of what, in turn, drives client satisfaction.One could assume that it is solely driven by payouts, but payouts and satisfaction were only slightly correlated (0.42, p < 0.01).There were 120 clients who were either not indemnified and still satisfied with the service (86 clients) or indemnified and yet not happy with the service (34 clients).In order to provide a first indication, we ran a simple ordinary least squares (OLS) regression on the satisfaction level including the receipt of a payout (yes/no), encountered problems (yes/no), and the contact to an agent as potential drivers.Results are presented in Table 5.While the receipt of a payout exhibited a strong positive effect, encountering problems had a negative effect on the satisfaction level.Contact to an agent, either to inform about the insurance or to help with the subscription process, was greatly valued.The statistically significant positive influence of being in touch with an agent for the subscription process is an important hint for insurers of mobile-delivered insurance products.As this is only a first exploratory analysis, exploring drivers for satisfaction constitutes an interesting venue for future research.
Getting back to drivers for renewal, the level of product understanding was also found to have a positive and statistically significant, though smaller, effect on the probability for insurance renewal.Increasing the level of product understanding by one point was on average associated with a 3.2% increase in the likelihood for renewal in model 2. Considering that the current measurement of product understanding only covers very basic aspects, ensuring this level of product understanding should be a comparatively easy step for insurance providers to achieve higher ethical standards while increasing long-term profitability.The association between product understanding and insurance demand is in line with previous findings on insurance knowledge (e.g.Cai et al., 2020;Lampe & W€ urtenberger, 2020).Here, we provide an indication on general product understanding as opposed to insurance understanding.
From model 3 and 4 we further see that neither future expectations nor agricultural activities influenced the renewal decision.The coefficients for variables on future expectation regarding the harvest were close to zero and not statistically significant.Similarly, indicators for income diversification (livestock farming and off-farm income) did not contribute to explaining the renewal decision.
In the full model, we also included the binary variable whether or not the household had received remittances in the last 12 months to control for potential external risk sharing mechanisms.The results show a statistically significant negative effect on the likelihood to renew.If a household received remittances, it was 7.2% less likely to renew its insurance policy for the next period.Remittances could either be seen as a substitute for insurance, or they could indicate a lack of liquidity that necessitates a remittance and consequently does not allow for the purchase of insurance.For health and funeral insurance schemes, there is already first evidence for the former explanation (Crayen et al., 2013;King & Singh, 2020).Yet, this relationship remains to be proven for agricultural insurance schemes.
In all models, we controlled for socioeconomic aspects (see Supplementary Material E for full models).Age was found to have a statistically significant positive, yet rather small, effect on the renewal decision while all other control variables were found to be not statistically significant in the full model.Estimating the same specifications using linear probability models confirmed the robustness of our results (see Supplementary Material F).

Differences between self-reported drivers for adoption and renewal
The paired sample proportion tests which were performed only on those clients who renewed their policy revealed that there were changes in the importance of the reasons for purchase (see Table 6).With an approval rate of 52%, 'feeling confident for the next season due to the insurance' was an important reason for renewal.Compared to the initial adoption decision, where only 45% named it as a reason for subscription, it gained in importance.Yet, the difference was not statistically significant.
Statistically significant changes were observed for peer influence and the interest in an indemnity payment.Initially, 9% of clients who renewed mentioned peer behavior as a reason for subscription while for the renewal decision only 4% named their peers' renewal as a motive.Similarly, 35% indicated that the desire to receive an indemnity was a motive for subscription, Be Happy, Be Loyal?Exploring Drivers for Renewal of Mobile-Delivered Index Insurance 843 while only 26% gave it again as a motive for The last reason that was comparable across the decisions was that the possibility of taking out a credit was conditional on subscribing to the insurance.Yet, this reason was neither very popular in the adoption (3%) nor for the renewal decision (2%).Even though the cross-sectional nature of our data does not allow to explicitly track knowledge gain, the observed changes suggest that insurance understanding evolved over the course of the insured period.Among those clients who renewed their insurance, the desire to receive an indemnity was mentioned less often whereas the feeling of confidence for the next season was given more often as a reason for renewal.When arguing that the sheer interest in an indemnity implies that the respondent did not fully grasp the concept of insurance, this would indicate that the levels of understanding increased.Similarly, the share of renewals that were done because peers renewed their insurance was statistically significantly lower than the share of initial subscriptions due to peer behavior.Again, this may be due to improved product understanding after a first experience with the product.
Besides the comparable reasons for purchase, the level of satisfaction with the service is an additional reason for purchase which mattered in the renewal decision but could not influence the initial adoption decision.It was the most important self-stated motive for renewal as 68% of those who renewed named satisfaction as a reason for renewal.Among the clients who did not renew 18% indicated that they did not renew because they were not satisfied with the service.Dissatisfaction with the service thus ranked second behind disappointment at not receiving an indemnity (28% of those who did not renew) when it comes to the most common reason for dropping out.

Conclusion
For insurance schemes to operate efficiently and to be profitable in the long-run, it is important to maintain a solid client base.To do so, it is crucial to know which factors drive the decision to renew an insurance policy and whether there are predetermining factors for or against renewal.With digitally-enabled insurance products on the rise, this study is of primary importance since we provide first evidence on these aspects for a mobile-delivered insurance product.We analyzed real-word data on an index-insurance product provided by OKO in Southern Mali which has to be contracted via a mobile phone.
Our findings show that receiving a payout is also in mobile-delivered insurance the most important driver for renewal, just as it was previously shown for conventionally distributed insurance (Cole et al., 2014;Stein, 2016).We add to existing evidence on drivers for renewal by showing that product understanding and client satisfaction are also important for the long-term success of weather index-based insurance schemes.Consequently, we highlight the need for insurance products that fit the clients' needs thereby supporting high levels of client satisfaction.We provide first exploratory evidence that, at least for mobile-delivered insurance products, contact an agent is valuable to achieve high levels of client satisfaction.While explicit drivers for satisfaction constitute an attractive area for future research, it also implies that-until specific evidence is generated-generally accepted principles of client management should be applied.In addition, we see a need for further research on how new marketing opportunities enabled by mobile delivery of insurance, such as text message reminders for contract renewal, affect renewal rates.
Besides, we compared one-time adopters to farmers who showed ongoing interest in the insurance scheme.We did not find substantial differences between both groups which implies that marketing activities as well as the insurance product itself are suited to target a variety of farmers.Looking only at those respondents who renewed their insurance, we compared selfstated motives for initial adoption and for ongoing demand.The results illustrate that the adoption decision clearly differs from the renewal decision in that the latter is taken much more independently and satisfaction with the first product experience plays an important role.
To conclude, we emphasize that, especially in contexts where educational levels are low, thorough explanations of the insurance product and the concept of insurance more generally are key for the long-term success of commercial microinsurance products.While it is the insurance providers' responsibility to explain the respective product, it is the policy makers' duty to promote financial literacy as a whole.

Notes
1.While learning by doing implies learning in the course of a production process, learning by using refers to learning about product characteristics due to the use of the respective product.The concept was suggested by Rosenberg (1982).2. USSD (Unsupervised Supplementary Service Data) menus are a type of text-based menu system used by mobile network operators to provide interactive services to users via their mobile devices.3. Clients were counted as renewal clients if they purchased an insurance policy from OKO in 2021 regardless of the crop insured in 2021.In our final sample, 95% of the farmers who renewed their insurance took out maize insurance again (see section A of the Supplementary Material for more details).4. Using this number for the calculation of the uptake rate would suggest a high uptake rate of 36.9% (1-(3,098/ 4,913)).However, we consider this figure to be an inflated uptake rate since the client base does not include farmers that have been in contact with OKO but decided they were not interested in the insurance.OKO did not record how many farmers were contacted throughout the sales season.Hence, the actual uptake rate cannot be computed.5. Matching of payment records and survey data was done via telephone numbers.It was not always possible to unequivocally identify which payment record belongs to which survey observation because of two reasons: (1) some farmers may have changed their phone number in the time between the insurance purchase and the survey and (2) it is possible to register via someone else's phone so that there are multiple insurance policies taken out via the same telephone number.6.In the course of the survey, plot sizes were measured using GPS technology.At the same time the farmers were asked to state the respective plot size, thereby allowing a comparison between measured and self-stated surfaces (CPS 2019).7. The results of two-sample t-tests and proportion tests (for binary variables) are presented in section B of the Supplementary Material.8.While one could argue that excluding uninsured farmers introduces sample selection bias, we argue that this is not the case since farmers who were not insured in the first place do not have to take the decision whether or not they want to renew their insurance.This study particularly focuses on the renewal decision and hence, the social group of interest does not comprise all farmers (insured or uninsured) but only those who bought insurance.9.The renewal rate in our final sample equals 58.8%.It is lower than in the study population as we failed to merge more observations of renewal than of non-renewal clients during the matching of the payment and the survey data.We also estimated all models without sampling weights and come to the same conclusions.The unweighted estimation results are available upon request.10.The insurance premiums are not the same within a village because OKO's index also considers aspects like proximity to water bodies, elevation and slope for the calculation of the premiums.These aspects are likely to vary between the fields.
Be Happy, Be Loyal?Exploring Drivers for Renewal of Mobile-Delivered Index Insurance 845 11.Variables measured as set out in Table 1. reported in section D of the Supplementary Material.

Figure 1 .
Figure 1.Framework for insurance adoption and insurance renewal.

Table 2 .
Mean comparisons between clients who renewed and clients who did not renew their insurance policy for 2021

Table 3 .
Estimates of logistic regressions on outcome of renewal decision a Measured as set out in Table2.b Binary variable that takes 1 if the variable statement is true, 0 otherwise.In all models we controlled for gender, age, ability to read, mobile phone usage, and wealth approximation.Standard errors in brackets.Ã p < 0.1.ÃÃ p < 0.05.ÃÃÃ p < .01;n ¼ 479.

Table 6 .
Paired sample proportion tests on initial purchase and renewal motives Motivation for purchase/renewal: I bought the insurance because … Notes: a Dummy variable that takes the value 1 if variable statement is true and 0 otherwise.n ¼ 282.