Big Five Personality Traits, Poverty, and Environmental Shocks in Shaping Farmers’ Risk and Time Preferences: Experimental Evidence from Vietnam

2.1 Determinants of Risk Preferences

While socioeconomic determinants of risk preferences have been widely studied, psychological factors remain underexplored. Emerging research suggests that personality traits significantly influence risk preferences. Mishra and Lalumière (2011) found a strong correlation between personality and risk preferences, while Pinjisakikool (2018) identified all five Big Five traits – extraversion, agreeableness, conscientiousness, emotional stability, and intellect – as significant predictors. However, Nicholson et al. (2005) reported that only extraversion and openness were significant, and Weller and Tikir (2011) found that emotionality and conscientiousness negatively influenced risk-taking. Despite these mixed findings, the relationship between personality traits and risk preferences remains insufficiently examined. To address this gap, our study incorporates the Big Five personality traits to assess their influence, particularly among farmers in developing economies where risk-taking is crucial for financial and agricultural decision-making.

Social capital, especially social networks, is another key determinant. Strong networks enable rural households to manage risk through informal credit and mutual support (Karlan et al., 2009). Attanasio et al. (2012) found individuals form risk-sharing groups with friends and family, while Nielsen et al. (2013) reported a positive association between local organization membership and risk aversion, although results varied by context. However, the role of borrowing status in shaping risk preferences is underexplored. We propose that borrowing experience may better indicate risk tolerance, as it can directly influence future behavior. Carvalho et al. (2016b) found that access to financial services, such as savings accounts, increased willingness to take risks. Yet, most studies emphasize informal credit, overlooking the influence of formal borrowing. Our study fills this gap by examining how borrowing across different credit markets affects risk preferences.

The link between wealth and risk preferences remains ambiguous. Some studies, such as Bosch-Domenech and Silvestre (1999) and Hallahan et al. (2004), suggest that higher income and net assets increase risk tolerance, while Mosley and Verschoor (2005) found no significant relationship. Similarly, the impact of dependents is unclear – Hallahan et al. (2004) observed that more dependents reduced risk tolerance, whereas Picazo-Tadeo and Wall (2011) found no significant effect. These inconsistencies suggest that financial influences on risk preferences may depend on broader socioeconomic contexts. Traditional indicators like income and assets may not adequately capture household economic status. The multidimensional poverty (MP) approach (Alkire & Foster, 2007) offers a more holistic perspective by including education, health, and living standards. MP captures nonmonetary deprivation, which may influence decision-making in ways income-based metrics do not. Households experiencing MP may exhibit different risk behaviors than those assessed only by financial indicators. By integrating MP, our study offers a broader understanding of the wealth-risk preference relationship.

External shocks, such as natural disasters, also shape risk attitudes. Given agriculture’s vulnerability to climate shocks and evolving policies, understanding how disasters shape farmers’ risk attitudes is crucial for informing effective adaptation strategies (Iyer et al., 2020). Brown et al. (2018) found disaster-exposed individuals were more cautious in their decision-making, and Cassar et al. (2017) observed long-term increases in risk aversion among tsunami survivors. Brown et al. (2018) also noted that the impact of cyclones on risk aversion varied across ethnic groups, underscoring the role of contextual differences. In contrast, Kahsay and Osberghaus (2018) found a negative correlation between storm exposure and risk aversion, suggesting that repeated exposure might desensitize individuals to risk. Given agriculture’s vulnerability to climate shocks, we posit that disaster exposure critically influences farmers’ risk preferences in developing economies.

Personality traits, particularly within the Big Five framework, have been shown to influence risk preferences. For example, Zhang et al. (2023) and Tang and Ngoc (2021) found that extraversion and openness are positively associated with risk-taking, while conscientiousness and neuroticism are linked to risk aversion. These traits may shape how individuals respond to uncertainty, including whether they engage with programs like subsidized loans or adopt new farming methods. In rural Vietnam, where farmers regularly face income and environmental shocks, understanding these links is especially relevant for designing effective, behaviorally informed interventions.

In conclusion, while prior studies have explored the determinants of risk preferences, findings remain mixed and highly context specific. Our study contributes to this literature in several key ways. First, we incorporate Big Five personality traits, addressing the limited attention given to psychological determinants. Second, we use the MP indicator to capture broader aspects of household economic status beyond income. Third, we highlight borrowing status as a distinct factor shaping risk-taking behavior. Finally, by focusing on farmers in developing economies – an often-overlooked group highly exposed to economic and environmental shocks – we offer insights with direct relevance for policy. These contributions support a more nuanced and comprehensive framework for understanding risk behavior in underresearched and vulnerable populations.

2.2 Determinants of Time Preferences

The relationship between poverty and time discounting (impatience) has received limited attention. Most studies use income as a proxy for poverty, but findings are mixed. While Harrison et al. (2002) found a negative but insignificant relationship, Tanaka et al. (2010) reported a significant negative correlation. Some researchers argue that wealth, particularly physical assets, may better explain patience (Bauer & Chytilová, 2010; Becker & Mulligan, 1997), yet evidence remains inconsistent. Pender (1996) found a positive correlation between net wealth and patience in South India, whereas Kirby et al. (2002) found no link in Bolivia. These inconsistencies highlight the need for broader poverty measures beyond income or wealth.

Recognizing the limitations of income-based measures, some economists advocate for an expenditure-based approach, which is more stable over time and better correlated with well-being (Meyer & Sullivan, 2011). However, Deaton (2004) noted challenges such as outdated poverty lines and regional inconsistencies. Beyond financial indicators, poverty is increasingly viewed as multidimensional, encompassing education, healthcare, and living conditions (Alkire & Foster, 2007). The MP index, widely used by organizations like the UNDP (2020), provides a more comprehensive perspective. Building on this framework, our study integrates MP into time discounting research, marking a significant departure from conventional income-based measures.

Access to credit can influence time preferences by reducing uncertainty about the future (Nielsen et al., 2013). Carvalho et al. (2016b) found that access to formal financial services, such as savings accounts, increases patience. In rural areas, informal credit from friends and family also plays a crucial role (Sohns & Revilla Diez, 2018). By incorporating both formal and informal credit sources, our study provides a more comprehensive view of how financial access affects time discounting.

Beyond economic and demographic factors, personality traits may also shape time preferences. The Big Five personality traits – extraversion, agreeableness, conscientiousness, emotional stability, and openness – are associated with decision-making behaviors (Barrick et al., 2001; Chamorro-Premuzic & Furnham, 2003). Studies suggest that individuals with higher neuroticism tend to be more impatient, while those with greater conscientiousness exhibit higher patience (Manning et al., 2014). Hirsh et al. (2008) found that extraversion correlates with higher discounting, whereas emotional stability is linked to lower impatience. Greco and Rago (2023) further show that conscientiousness and agreeableness are positively related to delay of gratification. These patterns suggest that conscientious individuals – who are organized and future-oriented – may be more inclined toward patient decision-making and long-term investments, while those high in neuroticism may respond more impulsively to uncertainty. In rural Vietnam, where farmers face persistent poverty and environmental shocks, understanding how personality traits relate to intertemporal preferences can inform more behaviorally grounded development strategies. Our study expands this literature by incorporating personality traits into an economic model of time discounting, offering a more holistic perspective on decision-making.

Environmental factors, particularly exposure to natural disasters, may further influence time preferences, especially for farmers whose livelihoods depend on weather conditions. Disasters shape expectations about the future, altering economic behavior. For instance, Cassar et al. (2017) found that tsunami victims became more impatient, while Callen (2015) observed increased patience in some cases. Given agriculture’s vulnerability to climate risks, understanding how environmental shocks affect discounting behavior is crucial. Our study is among the first to systematically explore this relationship, contributing to a deeper understanding of how experiencing natural disasters influences time preferences.

Our study makes several contributions to the literature. First, we integrate the MP framework into time discounting research, moving beyond income-based measures to capture broader deprivations. Second, we examine the role of credit access (regardless of formal or informal sources) in shaping time preferences, an aspect often overlooked. Third, by incorporating personality traits, we bridge psychology and economics to better understand individual impatience. Finally, we explore how environmental shocks influence discounting behavior, offering insights into financial resilience amid climate risks. Together, these contributions provide a more comprehensive framework for understanding economic decision-making, with implications for policies aimed at improving financial behavior and resilience among vulnerable populations, particularly in rural and agricultural settings.

3.1 Research Design

We conducted interviews with a randomly selected sample of 575 farmers in the MD, Vietnam’s largest delta. The risk and time preference experiments were part of a broader study involving an extensive questionnaire comprising multiple tasks and questions. Given the study’s comprehensive scope and rural context, the experiments were designed to be concise and easy to understand.

The time preference experiment drew on the methodologies of Benhabib et al. (2010) and Hardisty et al. (2013). Participants were asked to specify an immediate monetary amount they considered equivalent to a fixed future amount. Hardisty et al. (2013) compared several elicitation techniques, including the matching method, choice titration, and the dynamic staircase method. They found no dominant approach and highlighted the difficulty of explaining complex designs, even in lab settings with educated participants. Similarly, Tucker (2017) reported that intricate designs were unsuitable in rural Madagascar due to participants’ limited literacy and memory for time delays and rewards.

Our experiment included delays of 3 days, 2 weeks, 1 month, 3 months, 6 months, and 1 year. The delayed amounts were Vietnam dong (VND) 300,000 (USD 13.04) and VND 1,000,000 (USD 43.48).^[1] Each participant answered 12 time preference questions. To capture the magnitude effect, we followed Hermann and Musshoff (2016), who used high stakes triple those of the low stakes. In our case, the high stake was approximately 3.5 times greater, allowing us to assess variations in discounting across reward sizes.

In response to the concern raised by Tanaka et al. (2010) about the potential confounding of time and risk preferences in real-reward designs, our study not only adopts hypothetical rewards but also enhances the experimental setup by clearly separating risk and time preference tasks and providing neutral, nonincentivized framing for both. Unlike Tanaka et al. (2010), who used real monetary stakes that may have triggered uncertainty over delayed payouts – especially in field contexts with limited trust in future delivery – we minimize such biases by ensuring that participants face no strategic incentive to misreport preferences for fear of nonpayment. This design allows for a cleaner isolation of time preferences, reducing both risk contamination and experimenter demand effects. In addition to the early support from Du et al. (2002), more recent evidence further strengthens the rationale for using hypothetical rewards in time preference elicitation. Notably, Brañas-Garza et al. (2023) conducted a comprehensive study across laboratory, field, and online settings, directly comparing intertemporal choices under real and hypothetical monetary incentives. Their findings demonstrate that hypothetical rewards yield time preferences that are statistically indistinguishable from those elicited using real payments. This convergence is particularly important in light of the logistical, ethical, and economic constraints often associated with implementing real delayed rewards. As such, carefully designed hypothetical scenarios remain a methodologically sound and practically advantageous approach for capturing structural preference parameters without compromising validity.

Risk preferences were measured using a simplified version of the Eckel–Grossman (EG) gamble adapted from Dave et al. (2010), suitable for rural populations due to its clarity and accuracy. Participants made two risky decisions, each framed with a 50–50 probability. Before each choice, participants were informed whether real money was at stake. A coin toss following both choices determined whether the incentivized treatment was paid out.

Each EG task consisted of six binary lotteries with varying levels of risk and expected value. In the low-stakes version, expected payoffs ranged from VND 21,000 (USD 0.91) to VND 27,000 (USD 1.17), with the riskiest option yielding VND 52,500 (USD 2.28). The high-stakes version offered fivefold higher rewards, with expected values between VND 105,000 (USD 4.57) and VND 135,000 (USD 5.87) and a maximum of VND 262,500 (USD 11.41). These values equate to approximately 3.2 and 15.9 h of average farm labor wages in the MD, respectively.

The 575 farmers were assigned to six different treatment groups. The order of the two gambles and whether participants received real monetary rewards varied across treatments.

Further details on the risk and time preference experiments, as well as the MP indicator, are provided in the Supplementary material.

3.2 Study Area and Procedure

Our research was conducted in 16 communes in the MD region of Vietnam. To select the study communes, we analyzed variations in household income, educational levels, and the ethnic composition of household heads across 48 communes, using data from the Vietnam Household Living Standard Survey.^[2] On the basis of these variations, we selected 16 communes for our experiment. The results presented in this article are part of a broader survey. Each participant received a show-up fee of VND 30,000 (USD 1.304) for the entire session, with the possibility of additional payments based on their participation in the experiments. Interviewers informed participants about this before conducting the interviews. All participants received a consent form outlining the study’s purpose, data confidentiality, and the fact that no personal information was collected.

On the day of the experiment, the primary investigator contacted local commune authorities, who randomly selected 36 farmers from each commune. The list of households was provided by commune officers, from which we randomly selected 36 households per commune (except for one commune, where one household withdrew) using computer-generated index numbers corresponding to household entries on the list. The selection was conducted independently by the research team without substitutions or replacements, and the original lists were assumed to cover the full population of resident households without systematic omissions – thus ensuring a simple random sampling procedure. Once selected, participants were assigned to one of six treatment groups based on a prespecified allocation ratio: approximately 25% of participants in each commune were assigned to each of Treatments 1 and 2, and around 12.5% to each of the remaining four treatments (see Supplementary material for treatment details). This distribution was maintained across all communes to ensure balance and comparability. Each selected household was represented by one individual – typically the household head – who was guided through the experiment by a trained investigator. Interviewers were thoroughly trained in the protocol and followed standardized instructions to minimize variation and ensure that participants understood and followed their assigned experimental tasks. All sessions were conducted individually to avoid influence across participants or treatment groups.

One member of the research team remained with each household head to guide them through the task, while the officer continued introducing the team to the next selected household, repeating this process until all 36 households were visited. To protect voluntary participation, the enumerator read aloud an informed consent form at the beginning of each session, clearly stating that participation was entirely voluntary and that respondents could decline to take part or skip any task without facing any negative consequences or losing their show-up fee. The commune officer’s role was limited to logistical introductions and did not stay for the experiments – upon instruction by the research team, the officer left immediately after introducing the investigators. This approach was designed to avoid any pressure or undue influence from local authorities and ensured that all study tasks were conducted privately and individually.

We argue that the officer’s introduction was essential, as rural farmers tend to be risk-averse and passive in decision-making (Scott, 1976; Tran & Nguyen, 2016). Without an official introduction, farmers might have been hesitant to participate in experiments involving real rewards, such as risk preference assessments. To minimize any potential influence of the officer’s presence on participants’ task performance, the primary investigator instructed the officer to leave immediately after introducing the research team to a household. This approach also helped prevent time effects and treatment diffusion, as participants across different treatment groups in the same commune completed their tasks independently and simultaneously.

Since farmers were not informed in advance about the experiment nor were they aware that it was designed to measure risk preferences, we believe that differential attrition and demand effects did not occur (Bocquého et al., 2013). Notably, no participants withdrew from the experiments, suggesting the absence of differential attrition (Aaker et al., 2011) and selection bias (Holt & Laury, 2002).

1. Informed consent: Informed consent was obtained from all individuals included in this study.

2. Ethical approval: The research related to human use has been complied with all the relevant national regulations and institutional policies, and has been approved by the Ethical Approval Committee of the University of Otago, New Zealand (Approval No. 20/009).

4.1 Risk Preferences Elicitation Methods

The calculation of risk preferences, derived from raw response data and used to generate the two dependent variables reported in this study, is as follows:

where i indexes individuals (i = 1 … 575); k indexes the two different measures of risk preferences (k [is a member of] {EGH, EGL}); x _ik represents the menu item number from the two sets of choices referenced earlier, reflecting increasing risk acceptance based on the selection of a gamble with higher risk and greater expected payoff, except in the specific case of comparing gambles 5 and 6, where the expected payoff is only weakly greater. Participant i is classified as minimally risk-accepting when x _ik = 1 for either EGH or EGL tasks, corresponding to A _ik = 0. Conversely, at the upper end of the bounded risk scale used in this study, a value of x _ik = 6 for EGH or EGL denotes complete risk acceptance, corresponding to A _ik = 100.

4.2 Discounting

We employed the general model of Benhabib et al. (2010) to measure the time preferences of farmers. Let x, y, and t denote immediate reward, future reward, and delay, respectively. The relationship between an immediate reward and future reward is represented by a four-parameter discounting model developed by Benhabib et al. (2010) as follows:

The four parameters α, θ, r, and b separately present bias (α), hyperbolicity (θ), time discount (r), and fixed cost component (b) of the discounting function, while x denotes an immediate amount that a farmer thinks it is equivalent to the given reward y in the future time (or delay) t. In estimation models, t is calculated by dividing the number of days of each question listed in Section 3 by 365. The monetary unit of x, y, and b is VND.

When θ = 1 and α = 1, the curvature of discounting function reduces to exponential form exp { − rt } − b y .

When θ = 1 and α = 1, the curvature of discounting function reduces to hyperbolic form 1 ( 1 + rt ) − b y .

The quasi-hyperbolic discounting appears if α < 0, while the fixed cost component appears if b > 0. According to Benhabib et al. (2010), the specifications (3) and (4) are consistent with present bias and magnitude effect phenomena. Present bias refers to time inconsistency – for example, a subject prefers $10 today over $20 tomorrow but opts for $20 in 366 days over $10 in 365 days (Frederick et al., 2002). The magnitude effect occurs when individuals are more patient with larger rewards. A subject might choose $10 today over $20 tomorrow, indicating a high discount rate, but prefer $200 tomorrow over $100 today, showing a lower discount rate for the same 1-day delay when the stakes are higher (Benhabib et al., 2010; Hermann & Musshoff, 2016).

To assess the reliability of responses, we conducted internal consistency checks on the time preference data. Specifically, we examined whether participants’ switching points followed a logical, monotonic pattern as delayed rewards increased. Most participants exhibited stable behavioral patterns consistent with established phenomena such as present bias and the magnitude effect (Benhabib et al., 2010; Frederick et al., 2002). A small number of inconsistent responses were automatically excluded during model estimation. While these checks are not reported in detail in the article, they provide reasonable confidence in the validity of the elicited time preferences.

4.3 Determinants of Risk and Time Preferences

To investigate the determinants of risk and time preferences, we employed linear regression models with separate specifications for each preference type. The general form of the regression model is as follows:

In models analyzing risk preferences, Y _i denotes the risk preference of farmer i, calculated using equation (1) for both EGL and EGH stakes. For time preference models, Y _i represents time preferences, estimated using equations (3) and (4). Across all four models, X _i includes explanatory variables such as the Big Five personality traits, MP status, borrowing status (credit access), social capital, dependency ratio, and demographic controls, as well as external environmental factors. Z represents instrumental variables (IVs) in the IV regression models.

To address potential endogeneity in credit access (borrowing status), we used the distance to the nearest bank as an IV, consistent with the studies by Koomson et al. (2020) and Bukari et al. (2021), who found it correlated with financial market participation but not directly with preferences. We also used the logarithm of unexpected expenses (e.g., funeral costs) as an IV, as such shocks often prompt borrowing.

For poverty, three IVs were employed: distance to the nearest industrial zone or company, arable land area, and water salinity levels. These are associated with household poverty but are unlikely to directly influence risk or time preferences, except through income effects. Industrial zones offer off-farm income opportunities; water salinity reduces crop yields, impacting income; and arable land determines farm income potential.

To address potential endogeneity in estimating the effects of credit access and poverty, we employ five IVs. For credit access, we use two IVs. First, the distance to the nearest formal credit institution, which captures variation in access while being plausibly exogenous to individual preferences, is widely used in the literature (e.g., Bukari et al., 2021; Churchill & Marisetty, 2020; Koomson et al., 2020). Second, the logarithm of unexpected nonliving expenses, primarily driven by culturally significant events such as funerals and weddings, reflects exogenous financial shocks. In rural Vietnam, such events often involve substantial upfront borrowing, regardless of poverty status, and hosts typically expect partial or full cost recovery through guest contributions. These characteristics make this variable a valid exclusion restriction.

For poverty, we use three IVs. The salinity level of nearby water sources, which affects crop yields and thus income, is based on site selection strategies from Tanaka et al. (2010). The other two IVs – distance to the nearest industrial zone and arable land area per household – are derived from arguments by Nguyen (2011) and Tran (2014), who emphasize the low geographic mobility of older rural Vietnamese and the role of inherited or assigned assets in determining livelihood options. These structural features imply that such geographic and environmental characteristics are largely exogenous and not subject to self-selection, making them suitable instruments for identifying poverty effects.

One possible concern is that personality traits and economic preferences may not be causally linked, but instead develop in parallel or reflect common underlying factors. Personality traits, particularly those in the Big Five framework, are widely viewed as dispositional and temporally stable characteristics that develop early in life and remain consistent over time and across situations (Costa & McCrae, 1999; Soto & Tackett, 2015). This stability reduces concerns about reverse causality, in which measured preferences might influence personality. To assess potential multicollinearity, we computed Pearson correlations among all independent variables, including the Big Five traits and demographic controls. This result is not reported in the article to avoid distracting readers or overloading the text with technical details, especially given the presence of nonacademic audiences. All pairwise coefficients were below 0.3, indicating that multicollinearity is not a concern in our data (Gujarati, 1995). The correlations were generally weak and statistically insignificant for most traits, except for Openness, which showed a small and marginally significant association. These results suggest limited concern about endogeneity in the current context as well.

Poverty was measured using both income-based indicators and Vietnam’s MP indicator (introduced in 2016), which captures nonfinancial deprivation across five dimensions: education, healthcare, housing quality, water/sanitation, and information access. Each dimension has two binary indicators, totaling ten. Households with three or more deprivations are classified as MP-poor, two as MP-near-poor, and fewer than two as MP-nonpoor (details in the Supplementary material).

The Big Five personality traits were measured using the Ten-Item Personality Inventory developed by Gosling et al. (2003). Respondents rated their agreement with ten statements on a 7-point Likert scale (1 = Strongly Disagree to 7 = Strongly Agree). The ten items are as follows:

1. Extraverted, enthusiastic

2. Critical, quarrelsome

3. Dependable, self-disciplined

4. Anxious, easily upset

5. Open to new experiences, complex

6. Reserved, quiet

7. Sympathetic, warm

8. Disorganized, careless

9. Calm, emotionally stable

10. Conventional, uncreative

Based on respondents’ answers, the Big Five personality traits were calculated as follows (Gosling et al., 2003):

Extraversion = (score of item 1 + reverse-score of item 6)/2

Agreeableness = (score of item 7 + reverse-score of item 2)/2

Conscientiousness = (score of item 3 + reverse-score of item 8)/2

Emotional Stability = (score of item 9 + reverse-score of item 4)/2

Openness = (score of item 5 + reverse-score of item 10)/2

Across all models, we control for demographic factors to refine the analysis of risk and time preferences’ determinants.

Descriptive statistics of variables in our models are in Table 1.

5.1 Validity of Risk and Time Preferences Measured from Our Experiments

5.1.2 Magnitude Effect and Time Inconsistency in Time Preferences

Classical economic models assumed exponential time discounting, where discount rates increase monotonically with delay length (Frederick et al., 2002). However, many studies have rejected this, noting deviations such as present bias and the magnitude effect (Benhabib et al., 2010; Hermann & Musshoff, 2016). These behavioral patterns challenge the standard exponential model and support alternative frameworks like hyperbolic discounting.

Comparing the 95% confidence intervals vertically in Figure 1 at each of the six time delays, the mean height of the raw response ratios in the large-reward condition was significantly higher in both the economic and statistical sense. This figure illustrates how indifference ratios (x/y) change with increasing delay durations for small (pink) and large (green) rewards. The decline in ratios over time reflects temporal discounting, with steeper declines for small rewards, indicating greater impatience. The fitted curves and confidence bands show that participants discounted delayed rewards less steeply when the reward was larger, consistent with magnitude effects in intertemporal choice. These vertical differences are prima facie evidence of the so-called magnitude effect.

Figure 1

Delay discounting patterns by reward magnitude.

The upper half of Table 6 is the uneven rate at which r its appears to decrease in time and the two (highlighted) nonmonotonicities in the sample median of r its with respect to time moving from the first to the fourth time delay. We note that none of the seven discounting models we will estimate econometrically in the next section can closely fit or explain such patterns. All the discounting functions considered respect the time value of money in the sense that they are strictly decreasing in t . Nor can any of the models produce nonconvexities in which the rate of decrease D( t , y s ) increases. Because we already demonstrated the existence of a significant magnitude effect exhibited by the mean participant in Figure 1, we focus on the following three nonoverlapping restrictions within each reward-size condition and test hypotheses corresponding to each of these restrictions implied by time consistency:

ln ( D ( 1 y , y s ) ) − ln ( D ( 6 m , y s ) ) ] / 1 2 ,

(6) = ln ( D ( 3 m , y s ) ) − ln ( D ( 1 m , y s ) ) ] / 2 12 ,

(7) = ln ( D ( 2 w , y s ) ) − ln ( D ( 3 d , y s ) ) ] / 11 365 .

Table 6

Evidence of time inconsistency

t	mean r	Std. err of mean	Median	Change in med r ∆r ijt /∆t	Mean r	Std. err of mean	Median	Change in med r ∆r ijt /∆t
(A) Mean r its by t and reward size ( s ) with tests of the time-consistency restriction: r its = ∆ for all t and s
	s = small-reward condition (6 tasks)				s = large-reward condition (6 tasks)
3d	7.34	0.74	2.45		4.80	0.52	1.22
14d	3.48	0.16	2.45	0.03	2.24	0.14	1.44	7.43
1m	3.14	0.09	2.77	6.78	1.85	0.09	1.24	−4.43
3m	1.74	0.04	1.62	−6.97	0.98	0.04	0.75	−2.99
6m	1.23	0.02	1.19	−1.72	0.68	0.02	0.57	−0.69
1y	0.82	0.02	0.74	−0.90	0.48	0.01	0.42	−0.31
	N = 575 [for each of 12 (t, s)-specific empirical distributions of r its ]
Pooled r (over six t-specific estimates within-y)	2.96	0.13	1.47		1.84	0.09	0.75
(B) A second set of estimates of r and corresponding test: [ln( D ( t , y )) − ln( D ( t + ∆ t , y ))]/∆ t = r
∆t
14d−3d	2.43	0.13	2.00		1.54	0.10	1.14
3m−1m	1.01	0.03	0.97	−7.74	0.53	0.03	0.38	−5.66
1y−6m	0.40	0.01	0.35	−1.84	0.26	0.01	0.22	−0.47
Pooled over ∆t	1.28	0.05	0.79		0.78	0.04	0.38

Note: Evidence of time inconsistency using two estimates of a counterfactually constant exponential discount rate r and testing for equality of means across t and s: (A) Mean rits =[ln (ys)-ln (xit)]/t tested for equality across nonoverlapping subsets of the 12 time tasks; (B) log-differenced estimates of the annualized discount rate rits across three nonoverlapping changes in time delay (Dt): riDts = [ln(D(t, ys)) − ln(D(t + ∆t, ys))]/∆t.

The lower half of Table 6 reports means for each of the three log-differenced expressions in the aforementioned equations. If the exponential model were descriptively valid, then each of these log-differences should give an identical estimate of the annualized discount rate r . The data easily reject (6) and (7), both individually and jointly, for both reward-size conditions ( s ) as seen in the lower half of Table 6. Thus, our data replicate the frequently reported finding of time inconsistency. These findings reconfirm our decision to use time preferences measured from equations with fixed and variable cost components as dependent variables.

5.2 Determinants of Risk Preferences

Table 7 presents the regression results on the factors influencing risk preferences across the two experimental elicitation methods: EGL and EGH. The linear regression models include the Big Five personality traits, age, age squared, gender, years of education, ethnicity, dependency ratio (calculated as the number of household members without income divided by the total number of household members), MP (measured as the total score of ten MP indicators), logarithm of income per capita per month (calculated as the total annual income of household members divided by the total number of household members and then divided by 12), borrowing status (coded as 1 if a farmer has ever borrowed and is still repaying debt, and 0 otherwise), social capital (coded as 1 if at least one household member is involved in a local social/political association, bank, or local people’s committee, and 0 otherwise), distance to the nearest car-accessible road (log-transformed one-way distance), and experienced natural disaster shocks (e.g., droughts, typhoons, and flooding) that disrupted agricultural production within the past 12 months (coded as 1 if experienced, 0 otherwise).

The results in Table 7 show that the directional effects of most significant independent variables remain consistent across models measuring experimental risk preferences. Columns (1) and (4) present ordinary least squares (OLS) regression results, while columns 2, 3, 5, and 6^[3] show IV regression results for risk tolerance using the EGL and EGH, respectively. The Kleibergen–Paap rk Wald F statistic and Hansen J statistic confirm the IVs’ validity, rejecting concerns of underidentification or overidentification.

Borrowing status is positively and significantly associated with risk tolerance in the OLS models (coef. = 7.017, p < 0.05 and coef. = 5.762, p < 0.1, columns 1 and 4, respectively), suggesting that farmers with credit access are less risk-averse. However, this relationship disappears in the IV models (coef. = −10.944, p > 0.1 and coef. = −9.646, p > 0.1, columns 2 and 5, respectively), indicating that unobserved confounding variables may drive the observed effect in the OLS results. This suggests that credit access could be correlated with other factors influencing risk preferences, rather than serving as a direct determinant. Social capital, measured by membership in local organizations, does not significantly affect risk preferences in any model (coef. = −1.877, −1.998, 2.074, 3.492 with p > 0.1 in columns 1, 2, 4, and 5, respectively). This finding aligns with the study by Nielsen et al. (2013), who suggest that the formal organization membership may have limited or negligible influence on individuals’ risk-taking behavior, as our results show that social capital – measured by local organization membership – does not significantly affect risk preferences across all model specifications.

Among the Big Five personality traits, openness and emotional stability are consistently and significantly associated with higher risk tolerance across models (columns 1, 2, 4, and 5). Specifically, openness has positive and statistically significant effects, with coefficient values of 6.417 (p < 0.01) in column 1, 7.941 (p < 0.01) in column 2, 6.127 (p < 0.01) in column 4, and 5.407 (p < 0.01) in column 5. Likewise, emotional stability is positively associated with risk tolerance, with coefficients of 3.116 (p < 0.05) in column 1, 4.037 (p < 0.05) in column 2, 5.410 (p < 0.01) in column 4, and 4.664 (p < 0.01) in column 5. Openness, reflecting intellectual curiosity and receptiveness to new experiences (McCrae & Costa, 1997), significantly reduces risk aversion, consistent with Nicholson et al. (2005). Emotional stability, which aids in managing uncertainty, also positively influences risk tolerance, aligning with Simon et al. (1999) and Zhao and Seibert (2006). Conscientiousness negatively affects risk tolerance in the EGL IV model (column 2), with a coefficient of −3.759 (p < 0.1), but is insignificant in the EGL OLS model (−2.370 in column 1) and in all EGH specifications (−0.063 in column 4 and 0.228 in column 5), suggesting context-dependent effects or sensitivity to endogeneity correction. Agreeableness also shows a significant negative association with risk tolerance in EGL models, with coefficients of −3.600 (p < 0.1) in column 1 and −4.302 (p < 0.1) in column 2. This finding is consistent with the notion that agreeable individuals tend to avoid conflict and risky choices. However, its effect becomes statistically insignificant in the EGH models (−3.181 in column 4 and −2.346 in column 5).

A particularly noteworthy finding is that farmers who experienced natural disasters in the past 12 months exhibit significantly higher risk tolerance across all models. The coefficients for the natural disaster variable are 8.018 (p < 0.05) in column 1, 12.275 (p < 0.01) in column 2, 7.750 (p < 0.05) in column 4, and 8.591 (p < 0.05) in column 5, indicating a consistently strong and positive effect. This contrasts with studies indicating disaster exposure increases risk aversion (Cameron & Shah, 2015), but aligns with Kahsay and Osberghaus (2018) and Eckel et al. (2009), who found increased risk-taking following disasters. Such behavioral shifts may reflect psychological adaptation, cognitive reframing, or expectations of postdisaster assistance. Demographic characteristics mostly do not significantly influence risk preferences, as variables such as age, gender, education, and dependency ratio show statistically insignificant effects across all models. An exception is ethnicity: being from a minority ethnic group is positively and significantly associated with risk tolerance, with coefficients of 7.002 (p < 0.1) in column 1, 12.431 (p < 0.01) in column 2, 5.258 (not significant) in column 4, and 8.322 (p < 0.05) in column 5. This aligns with studies (Henrich & McElreath, 2002; Picazo-Tadeo & Wall, 2011), which report weak associations between gender, age, or education and risk-taking. Ethnic minority farmers demonstrate significantly higher risk tolerance than Kinh counterparts, particularly in the EGL models, suggesting culturally influenced risk behavior.

In the IV models, the significance of emotional stability, openness, and disaster exposure is robust, while the earlier OLS association between borrowing status and risk tolerance disappears. MP is positively associated with risk tolerance in the OLS models, with coefficients of 0.250 (p < 0.1) in column 1 and 0.249 (p < 0.05) in column 4. However, this relationship becomes statistically insignificant in the IV models (columns 2 and 5), where the coefficients increase in magnitude (0.428 and −0.927, respectively) but are not significant. This pattern suggests that the observed effect of poverty on risk tolerance may be confounded by endogeneity or omitted variable bias, and the true causal relationship could be weaker or indirect. In the revised IV specifications (columns 3 and 6), MP is excluded and replaced with the logarithm of income per capita per month as a potentially more exogenous measure of economic status. This income variable is positively and significantly associated with risk tolerance in the EGH model (coefficient = 12.416, p < 0.1, column 6), but not in the EGL model (coefficient = 7.376, not significant, column 3). These results suggest that income may have a stronger influence on risk preferences in high-stakes settings (EGH), possibly because wealthier individuals are more willing or able to bear financial risk when larger outcomes are involved.

Overall, these results highlight the nuanced effects of psychological, socioeconomic, and contextual factors on risk preferences. The consistency of results for personality traits and disaster exposure across models underscores their substantive importance. The findings support the argument, consistent with the study by Nielsen et al. (2013), that experimental methods offer a more precise and reliable measure of individual risk preferences than self-reported or observational approaches.

5.3 Determinants of Time Preferences

Table 8 presents regression results examining determinants of time discounting rates across models: a general model with a fixed cost of present bias and an unrestricted model that includes both fixed and variable costs. Columns (1) and (4) display OLS regression results, while columns 2, 3, 5, and 6^[4] present IV regression results. Since higher dependent variable values indicate greater impatience, positive coefficients imply increased discounting rates, while negative coefficients suggest decreased rates. To address endogeneity, IV regression is applied. The Kleibergen–Paap test confirms the IVs are valid, rejecting underidentification at the 1 and 10% levels. The Hansen J test indicates appropriate IV specification. However, the Stock–Yogo test suggests that the instruments for MP may be relatively weak.

We begin with the impact of MP on time discounting. Coefficients for MP are small and statistically insignificant across most models (columns 1, 2, 4, and 5), indicating that MP does not meaningfully influence farmers’ time preferences. This suggests that MP alone does not make farmers more impatient. In the revised IV models (columns 3 and 6), MP is replaced with the logarithm of income per capita per month, which shows a significant negative association with impatience (e.g., column 6: −0.379, p < 0.1). This implies that higher income is linked to greater patience. The contrasting effects of MP and income-based poverty underscore that these measures are not interchangeable. The negative relationship between income and time discounting supports findings that poorer individuals are more impatient.

Borrowing status is not significant in OLS models, but the IV regression for the fixed and variable cost model (column 5) reveals a strong positive effect (0.988, p < 0.05), indicating that farmers with credit access are more impatient. This suggests that credit availability may encourage present consumption over long-term financial planning, possibly due to improved liquidity and eased borrowing constraints.

Among the Big Five traits, openness is consistently and significantly negative across all models (e.g., column 5: −0.248, p < 0.01), indicating that more open individuals are more patient. This aligns with the study by Mahalingam et al. (2014). Conversely, conscientiousness is significantly positive in IV models (e.g., column 5: 0.144, p < 0.1), suggesting that more conscientious individuals prefer immediate rewards. This finding contrasts with prior research (e.g., Lee et al., 2006; Manning et al., 2014), which often links conscientiousness to future-oriented behavior. Other traits – extraversion, agreeableness, and emotional stability – do not significantly influence discounting rates.

Natural disaster exposure is not statistically significant, though the negative coefficient implies a weak association with greater patience. This result contrasts with studies suggesting disasters increase impatience (e.g., Cassar et al., 2017) but supports Callen (2015), who found increased patience following disaster exposure. Social capital also lacks significance, indicating that informal networks may not substantially affect intertemporal decision-making.

Demographic effects are mixed. Age is significantly negatively associated with impatience in the IV model (column 5: −0.122, p < 0.05), while age squared is positively significant (0.001, p < 0.05), indicating a U-shaped relationship. This suggests that middle-aged farmers are more impatient than younger or older ones. Gender, education, and ethnicity are not significant, in line with the previous findings. However, the dependency ratio is significantly negative (column 5: −0.670, p < 0.1), implying that farmers with more dependents are more patient, possibly due to increased long-term responsibilities in larger households.

Overall, IV regression offers deeper insights than OLS. MP does not significantly affect discounting rates, but credit access increases impatience, highlighting the role of financial liquidity in short-term choices. Personality traits, particularly openness and conscientiousness, underscore the psychological factors influencing intertemporal decisions. The negative link between income and impatience further supports the notion that lower income individuals exhibit higher discount rates.

In summary, these results reveal the complex interplay between financial access, personality, demographics, and time preferences in rural Vietnam. Policies aiming to promote long-term financial decision-making should incorporate both economic and behavioral considerations.

6.1 Policy Implications

These findings have important implications for rural development and agricultural finance. The positive impact of credit access on both risk and time preferences suggests that policymakers should expand financial inclusion through microcredit programs, flexible loan repayment structures, and public credit guarantees. Reliable access to credit can encourage high-risk, high-reward agricultural investments, fostering innovation and economic resilience. The positive relationship between income and both risk-taking and patience implies that policies should support sustainable income growth and investment. Higher incomes enable individuals to take calculated risks and engage in long-term planning. Financial literacy programs and investment training can help ensure that risk-taking is strategic. Income support mechanisms, such as subsidies or savings incentives, can reinforce patient behavior and protect households from short-term financial shocks. To enhance practical relevance, differentiated financial tools could be developed for distinct behavioral profiles. For instance, impatient farmers might benefit from commitment savings products or mobile reminders that promote delayed gratification. Flexible loan products – such as income-contingent repayment or seasonal grace periods – may suit vulnerable or present-biased borrowers. Behaviorally informed training modules tailored to personality traits (e.g., planning strategies for less emotionally stable individuals) could improve financial decision-making. For disaster-prone communities, bundled tools that combine adaptive credit access with weather-indexed insurance and early warning systems could increase resilience and maintain productive risk-taking.

Given the role of personality traits such as openness and emotional stability in shaping individual preferences, rural development programs can benefit from more targeted behavioral interventions. Risk-management tools such as insurance and savings products could also be introduced using delivery formats and messaging strategies that align with different personality types. Strengthening local social networks and cooperatives can facilitate informal risk-sharing mechanisms and collective action, which may buffer the limitations of individual personality-driven behaviors. The observed increase in risk tolerance following disasters underscores the importance of adaptive and timely financial interventions, including weather-indexed insurance, emergency credit with flexible repayment terms, and community-managed disaster relief funds. These tools can help farmers recover from shocks while maintaining their investment capacity, especially when integrated with behavioral nudges such as mobile reminders or decision-planning apps tailored to time-inconsistent individuals.

These findings also align closely with the priorities of two major national policy frameworks currently guiding rural development in Vietnam. The National Target Program on New Rural Development (2021–2025) emphasizes inclusive finance, livelihood enhancement, and sustainable agricultural upgrading (Government of Vietnam, 2022a). Our evidence that credit access fosters both patience and risk-taking supports this agenda by highlighting the behavioral mechanisms through which credit empowers productive decision-making. Likewise, the Strategy for Sustainable Agriculture and Rural Development to 2030, with a Vision to 2050 includes a specific focus on developing smart and sustainable agricultural value chains, particularly in vulnerable regions like the MD (Government of Vietnam, 2022b). This strategy aims to promote climate-resilient and market-oriented farming systems. The observed effects of disaster exposure on risk preferences suggest that strengthening access to adaptive finance – such as weather-indexed insurance and post-shock liquidity support – can help maintain long-term investment capacity, even under growing climate uncertainty. The timeliness of this study lies in its potential to inform the design and implementation of these ongoing national efforts. As policymakers move to scale up financial inclusion and rural resilience initiatives, understanding how individual economic preferences respond to income, credit, and shocks can help maximize policy uptake and behavioral alignment.

Although this study is situated in the MD, its implications extend beyond the local context. Structural constraints common across rural areas in low- and middle-income countries – such as limited credit access, exposure to climate shocks, and income instability – often shape economic preferences in similar ways. Studies in regions such as sub-Saharan Africa and rural Southeast Asia have documented how poverty and environmental risk are linked to short-termism and risk aversion (Falk et al., 2018; Gloede et al., 2015). These parallels suggest that our findings can inform policy designs in other developing economies where similar behavioral patterns emerge. In such contexts, expanding access to credit and strengthening income security are likely to promote more forward-looking and risk-tolerant economic behavior.