Predicting entrepreneurship from brain structure and neural responses to risk and ambiguity

Entrepreneurs are recognized for their valuable contributions to society, including the creation of new ideas, job opportunities within their ventures, and overall economic growth¹. Understanding what makes these individuals unique has long been a subject of research, and neuroscience methods now offer promising tools to investigate the neural and behavioral traits that may distinguish entrepreneurs from others². In this study, we explore whether behavioral and neural responses in risk-based decision-making tasks can be used to classify individuals as entrepreneurs or non-entrepreneurs.

The concept of risk has been associated with entrepreneurship since the 17th century in the writings of economist Richard Cantillon, and the term “entrepreneur” remains strongly linked to a preference for risk-taking. In fact, Oxford Dictionaries define an entrepreneur as someone who sets up a business, taking on financial risks³. While various psychological and behavioral traits—such as need for achievement, internal locus of control, innovativeness, and Big Five personality traits like openness to experience, extroversion, agreeableness, emotional stability, and conscientiousness—have been associated with entrepreneurship^4,5,6,7,8, risk tolerance remains one of the most consistently emphasized traits distinguishing entrepreneurs from non-entrepreneurs. Although some studies do not find differences in risk-taking willingness between entrepreneurs and non-entrepreneurs^9,10, a substantial body of research indicates that entrepreneurs are generally more willing to take risks than others^{8,11,12,13,14}. Moreover, risk-taking propensity may vary across different phases of the entrepreneurial journey, highlighting its dynamic and central role in entrepreneurial behavior¹⁵. In our study, “risk” refers to situations in which individuals have objective knowledge of both the potential outcomes of a decision and the probability distribution associated with these outcomes. In contrast, “uncertainty”, or “ambiguity”, arises when the decision-maker does not possess objective knowledge of the probability distribution of outcomes. This distinction, crucial for understanding entrepreneurial behavior, is based on the foundational work of Knight (1921), Ellsberg (1961), and further explored, for example, by Becker and Brownson (1964) and Camerer & Weber (1992)^16,17,18,19.

Entrepreneurs have often been compared to managers in previous research, revealing that entrepreneurs exhibit a higher willingness to take risks and a greater tolerance for ambiguity due to fundamental differences in their work-related demands^20,21,22,23. This is evident as entrepreneurs, operating within the context of new ventures and typically introducing new products or services, face situations where the outcomes of their decisions are unknown^24,25. In contrast, managers work within established companies, utilizing historical information about products, services, and markets to inform their decisions. Additionally, entrepreneurs can act independently as they own the ventures they have founded, whereas managers tend to act cautiously due to their responsibilities to external owners, employees, and other stakeholders^26,27.

While previous research has used neuroscientific methods to identify unique neural activation patterns of entrepreneurs compared to managers in experimental tasks involving risk and ambiguity^28,29, there is limited work combining participants’ behavioral data with their functional and structural brain data. Acknowledging this gap, our study explores how entrepreneurs and a control group of managers/employees process risk and ambiguity and examines whether fMRI-derived brain activity and structural differences can predict entrepreneurial status. Consistent with Vantilborgh et al. (2015)³⁰, we define “entrepreneurial status” as the occupational designation of individuals who have recently founded a venture or company and derive their primary income from it—that is, entrepreneurial status refers to participants’ classification as entrepreneurs or non-entrepreneurs, based on their established professional roles. Using a machine learning approach, we predict participants’ likelihood of being classified as entrepreneurs, their entrepreneurial status, by comparing the predictive power of self-rated behavioral traits, brain responses during fMRI, and structural differences in gray matter volume (GMV).

This combination of methods allows us to interpret the timescale of brain mechanisms linked to risk-taking in entrepreneurs: fMRI responses reflect the immediacy and pattern of neural responses to risk, while GMV assessments provide insight into potential long-term structural brain differences fostered by prolonged engagement in entrepreneurial activities. By doing so, we build on prior research showing that occupational expertise can induce structural changes in the brain, such as increased GMV in specific regions among experts³¹. Our research also responds to calls in entrepreneurship research to use the brain as a predictor³².

The primary objectives of this study are twofold: First, to investigate whether entrepreneurs demonstrate distinct responses to risk and ambiguity compared to individuals without entrepreneurial backgrounds. To achieve this, we recruited venture founders actively involved in business operations with high growth aspirations. The control group, matched as closely as possible to the founders in demographic and occupational characteristics, consisted of male employees of the same age and educational level, primarily with managerial responsibilities but without entrepreneurial backgrounds (see Supplementary material, Table S1). We conducted a comprehensive analysis of behavioral responses, brain activation patterns, and gray matter volume (GMV) in both groups. Second, we employed a machine learning approach to evaluate the predictive power of these data in classifying participants as entrepreneurs. Specifically, we compared the contributions of self-rated behavioral traits, brain responses during fMRI, and structural brain features (GMV) to prediction accuracy. This dual approach allows us not only to identify differences in decision-making processes and neural structures but also to evaluate how well these differences can be harnessed for predicting entrepreneurial status.

To evaluate the decision-making processes of entrepreneurs and the control group under conditions of risk and ambiguity, we employed the Becker-DeGroot-Marschak (BDM) mechanism³³. In this auction-type task, participants assign a selling price to various risky options, or “lotteries”, which is then compared to a randomly generated buying price to determine whether the option is “purchased”. The BDM mechanism closely parallels real-world scenarios where individuals evaluate uncertain outcomes and decide how much to invest or risk—situations frequently encountered by entrepreneurs when allocating resources to projects with uncertain returns or weighing potential gains against risks. Its effectiveness and reliability in eliciting true valuations have established the BDM mechanism as a widely used tool in decision-making research, with applications ranging from assessing the value of food items to examining savings behaviors^34,35. By employing this well-established economic paradigm, we provide a robust framework for studying entrepreneurial decision-making.

We hypothesized that, consistent with their inclination towards variable rewards over the fixed income of traditional employment^11,36, entrepreneurs would propose higher selling prices in BDM tasks. This behavior would reflect a valuation pattern consistent with entrepreneurs’ proclivity for risk-taking and comfort with ambiguity–traits that are postulated to be more pronounced among entrepreneurs.

Participants

To recruit entrepreneurs, we reached out to the Finnish Federation of Entrepreneurs, which represents over 115,000 entrepreneurial firms of various sizes (https://www.yrittajat.fi/en). An email invitation was sent to its members, targeting founders actively involved in their ventures’ business operations and aspiring to achieve high growth, defined as an annual growth rate exceeding 20% over the next three years. Although our recruitment initially aimed to include both male and female entrepreneurs, an insufficient number of eligible female participants signed up during the data collection period. Since our aim was not to specifically study gender differences, including a small number of female participants would have resulted in an unbalanced sample, potentially complicating the interpretation of the results. As a result, the final sample consisted solely of male entrepreneurs.

For the control group, we recruited male employees without entrepreneurial backgrounds through advertisements posted in the City of Helsinki’s daycare centers. Interested individuals signed up online. This recruitment approach was chosen as our participants were also involved in experiments focused on parental attachment. Of the employees, 80% held managerial responsibilities in their work. On the recruitment webpage, employees were asked if they had subordinates or if they were responsible for key functions such as product development, sales, marketing, financial management, or human resource management. Our aim was to select employees with managerial responsibilities in at least one of these areas.

We collected functional and structural MR images and behavioral data from 43 healthy male participants, including 21 entrepreneurs and 22 managers/employees as controls. We had to exclude three participants from the analysis as they misunderstood the task, leaving data from 20 entrepreneurs and 20 managers/employees. In the entrepreneur group, 40% of participants had children. Statistical tests showed no significant differences in key characteristics such as risk attitude, confidence, optimism, or affect intensity between entrepreneurs with and without children (Table S3, two-sided t-tests, all p > 0.05). Similarly, in the full sample, no statistically significant differences were observed between participants with and without children in these variables. Since the study is not intended to examine how age or education affect participants’ decision-making, we homogenized the groups with respect to those two factors. The participants had an average age of 34 years (SD = 5.12, range 24−45) and an average of 15 years of education (SD = 2.48, range 9−17). The average age of entrepreneurs was 33.3 years, while that of managers/employees was 35.1 years. Statistical tests revealed no significant difference in age between the two groups (p = 0.27, two-sided t-test; D = 0.25, p = 0.56, two-sample Kolmogorov-Smirnov test), indicating comparable age distributions.

Experimental task

During functional brain imaging sessions, participants viewed graphical representations of lotteries, featuring a deck of 100 cards with a known (risky trials) or unknown (ambiguous trials) mix of red and blue cards, as illustrated in Fig. 1A. The playing cards visually represented the probability distributions involved in the experiment, and participants were asked to envision a random card draw from this deck. If the drawn card matched the indicated color, they would win the specified cash amount; if not, they received no reward. In the BDM mechanism, participants determined a “selling price”— the maximum amount they were willing to accept to forgo their chance to participate in the lottery. This price reflected their valuation of the potential win, shaped by the perceived probability of success and the uncertainty of each scenario.

Before the fMRI sessions, we explained to participants that after setting their selling price, a “buying price” would be randomly selected between zero and the potential win amount. If a participant’s selling price was lower than this buying price, they would sell their ticket at the declared price and forgo the lottery. If the selling price was higher, they would keep the ticket, and the lottery proceeds. This procedure ensures that participants have a financial incentive to declare a selling price that truly represents their valuation of the lottery.

Full size image

Experimental design. Panel (A) Participants evaluated the lotteries by moving a cursor on a bar at the bottom of the screen. The bar on the left side of the screen depicted the distribution of blue and red cards (100 in total), the playing card next to the bar represented the winning color, and the number next to the card represented the sum of money that could be won, if a card of that color was drawn. The reward amounts varied from 8 to 10. In the ambiguous trials, a gray block covered part of the blue-red bar. Panel (B) Among the risky trials, half (24 trials) featured a winning probability of 50%, while the other half consisted of varying probabilities such as 20%, 30%, 40%, 60%, 70%, or 80%, with four trials allocated to each probability. Two ambiguity levels were used in the ambiguous trials. In half of the trials, the ambiguity level was 25%, and in the other half, 75% (24 trials each).

We conducted 96 trials in total, evenly split between risky and ambiguous scenarios, and presented them in random order to each participant. In the risky trials, half featured a 50% probability of winning, while the remaining trials included varying probabilities, as shown in Fig. 1B. For the ambiguous trials, we used two levels of uncertainty: 25% and 75%, illustrated in Fig. 2B. Participants adjusted a cursor on a scale at the bottom of the screen to indicate their minimum selling price for each “lottery ticket”.

Experimental procedure

The experiment comprised two separate sessions for each participant. During the first session, held in a meeting room at the University of Helsinki, a member of the research team welcomed the participant and provided an overview of the study’s structure. Participants read general instructions and provided informed consent by signing a consent form. They then completed several self-rated questionnaires privately, assessing traits such as risk preferences^37,38, affect intensity^39,40,41, optimism⁴², and confidence, along with socio-economic background and entrepreneurial/work experience. Participants’ confidence was measured through a judgment task, where they evaluated their certainty in response to seven general knowledge statements with two mutually exclusive answers (see Supplementary Material). These traits were selected due to their relevance to both entrepreneurship and decision-making under risk and uncertainty.

At the end of the session, participants received a movie ticket valued at approximately €10, along with earnings from the incentivized risk aversion measure (€0.10–3.85). In their study, Holt and Laury (2002) introduced a risk attitude measure using incentivized choices between lotteries of varying risk³⁷. We omitted this measure from further analysis due to its significant negative correlation (r = − 0.32, p = 0.04) with the general risk attitude measure by Dohmen et al. (2011), which sufficiently captures the scope of our analysis³⁸. The negative correlation arises from the scales’ differing directions. The first session lasted approximately 30 min.

The second session took place at the Aalto University Advanced Magnetic Imaging Centre. Upon arrival, the research team welcomed participants and explained the study’s procedures in detail. Participants reviewed general introductions and safety instructions for the functional magnetic resonance imaging (fMRI) procedure and reaffirmed their informed consent by signing a consent form. Before entering the MRI scanner, participants answered two control questions and practiced the task on a computer (see Supplementary Material). The control questions ensured participants’ understanding of the task, and any mistakes were clarified. Participants were also informed that two lotteries would be randomly selected and played for real money at the end of the experiment. The in-scanner task lasted approximately 20 min.

After the brain imaging session, participants completed a brief questionnaire about their experience during the scan. They rated their general vigilance at the beginning and end of the scan on a scale of 1 to 4, with 1 representing “sleepy” and 4 representing “vigilant”. On average, participants’ vigilance decreased from 3.38 at the start to 1.95 at the end, with no significant difference between the groups (p = 0.19, two-sided t-test). Approximately 60% of participants reported no disturbances during the scan. Those who did report disturbances mainly mentioned minor physical sensations such as itching or slight numbness in the limbs. None of the participants reported anxiety- or stress-related factors. About 80% of participants in both groups confirmed their ability to remain still during imaging, and any movements reported were minor, such as toe, foot, or finger movements.

At the end of the second session, two lotteries were randomly selected, and participants’ selling prices were compared with randomly drawn buying prices. If the selling price was lower than the buying price, participants sold their “lottery ticket” and received the price they had set. Conversely, if the selling price exceeded the drawn price, the lottery was played for real. On average, the total payment was €10.30 (SD = €5.00). The study protocol was approved by the Ethics Committee of Aalto University and conducted in accordance with the principles of the Declaration of Helsinki.

GMV and fMRI data acquisition and analysis

We performed MR imaging using a Siemens MAGNETOM Skyra 3-Tesla MRI scanner at the Advanced Magnetic Imaging Centre (Aalto NeuroImaging, Aalto University, Finland). First, all participants underwent a high-resolution, T1-weighted three-dimensional magnetization-prepared rapid acquisition gradient echo (MPRAGE) anatomical scan (TI = 1100 ms, TR = 2530 ms, TE = 3.3 ms, flip angle = 7°, rectangular field of view = 100%, acquisitionmatrix = 256 × 256 mm, voxel size = 1 × 1 × 1 mm). Immediately after the structural scans, we acquired whole-brain fMRI data using T2*-weighted echo-planar imaging (EPI) sensitive to blood-oxygen-level dependent (BOLD) signal contrast. Acquisition parameters included 36 axial slices, 3 mm slice thickness, TR = 2170 ms, TE = 30 ms, flip angle = 70°, FOV = 192 mm, and voxel size = 3 × 3 × 3 mm³. A total of 550 volumes were acquired, preceded by four dummy volumes to allow for equilibration effects. All images were acquired using a 32-channel head coil.

We preprocessed the structural MR images using SPM8 (Wellcome Institute of Cognitive Neurology, http://www.fil.ion.ucl.ac.uk/spm) and applied voxel-based morphometry (VBM) routines⁴³. First, we segmented the T1-weighted images to identify grey and white matter. Next, we estimated the deformations for best alignment of all participants’ images using SPM DARTEL routines (Diffeomorphic Anatomical Registration using Exponentiated Lie Algebra)⁴⁴. Using these deformations, we generated grey matter images that were (a) normalized to MNI space, (b) Jacobian scaled (to preserve regional totals after normalization), and (c) smoothed using a Gaussian function with a Full Width at Half Maximum (FWHM) of 10 mm × 10 mm × 10 mm.

In the resulting preprocessed images with a voxel size of 1.5 mm × 1.5 mm × 1.5 mm, we examined grey matter (GM) volume in four predefined regions of interest (ROIs): the right posterior parietal cortex (PPC), the dorsomedial prefrontal cortex (DMPFC), the right anterior insula, and the left anterior insula. These ROIs were selected based on previous research demonstrating their relevance to risk attitudes and decision-making processes. The right PPC, in particular, has been implicated in the processing of risk and uncertainty. Gilaie-Dotan et al. (2014)⁴⁵ found that greater GMV in the right PPC was associated with reduced risk aversion, suggesting that individuals with higher GMV in this region may exhibit a greater tolerance for risk. Accordingly, we defined the PPC ROI as a 15 mm sphere centered at MNI coordinates 27, − 79, 48, which corresponds to the peak voxel reported by Gilaie-Dotan et al. (2014)⁴⁵. The other three regions—DMPFC, right anterior insula, and left anterior insula—were identified as areas of activation during risk-related tasks in a coordinate-based meta-analysis that assessed the convergence of activation foci across 30 studies⁴⁶. These three regions corresponded to the largest clusters reported in the meta-analysis and exhibited the most robust activation in risk-related tasks. Each of these ROIs was defined as a 10 mm sphere located at MNI coordinates − 4, 30, 32; 32, 20, 8; and − 30, 20, 8, respectively. To limit the number of statistical tests, we did not define ROIs for smaller clusters reported in the meta-analysis⁴⁶. For each of the resulting four regions, we extracted GM volume from individual participants’ preprocessed files using MarsBaR routines⁴⁷  (https://marsbar-toolbox.github.io/). Additionally, we obtained each participant’s total GM volume.

We controlled stimulus delivery using Presentation software (Neurobehavioral Systems Inc., Albany, California, USA). Visual stimuli were back-projected onto a semitransparent screen and reflected to participants via a mirror inside the scanner. Participants used a two-button control device to move the on-screen cursor. We preprocessed and analyzed functional data using SPM12 (Wellcome Department of Imaging Neuroscience, London, UK)⁴⁸. Preprocessing included motion correction, slice-time correction, and indirect normalization to the MNI template through co-registration and segmentation procedures. Data were then spatially smoothed with a Gaussian kernel (FWHM 7 mm × 7 mm × 7 mm) and high-pass filtered with a cutoff of 0.008 Hz.

The fMRI data analysis involved two levels. At the single-subject level, we analyzed the fMRI data using the general linear model (GLM) and a blocked design approach. We presented the experiment conditions in separate blocks, where each evaluated lottery constituted a block. We had three types of blocks: risky, low ambiguity, and high ambiguity. The order of the 96 blocks was randomized for each participant. We modeled the blocks using boxcar functions with a duration of 5 s (Fig. 1). Additionally, we used the expected values of the lotteries as a single parametric modulator for each block. To address known sources of variability, we included six motion parameters (head motion) as covariates of no interest in the single-subject level model.

The second-level analysis studied activation for all participants (one-sample t-test) and group differences in activation (two-sample t-test). First, we conducted whole-brain analyses to identify brain areas with significant differences in the neural activity between risky and ambiguous trials, as well as brain areas showing significant group differences in expected value-related neural activity. To reflect the task participants performed, which involved assessing different risky and uncertain lotteries, we conducted a region of interest (ROI) analysis on specific brain areas associated with outcome valuation, in addition to the whole-brain analysis. The centers for our ROIs—the ventromedial prefrontal cortex (vmPFC), ventral striatum, and anterior insula—were selected based on the comprehensive meta-analysis by Bartra et al. (2013)⁴⁹, which examined convergence of activation foci from 206 studies on the subjective value of outcomes. These regions are also crucial for decision-making under risk and uncertainty. From Bartra et al.’s study (Table 1 in that study)  1 in, we selected the five regions with the highest meta-analytic statistics. For each of these regions, we defined a 10 mm radius sphere centered at the peak of the activation cluster. The sphere centers were located at the following coordinates: vmPFC (2, 46, − 8), left striatum (− 12, 4, 2), right striatum (12, 10, 2), left anterior insula (− 30, 22, − 6), and right anterior insula (32, 20, − 6). To limit the number of statistical tests, we did not define ROIs for the smaller clusters with lower meta-analytic statistics reported in Bartra et al. study⁴⁹. We defined the ROIs and calculated mean parameter values (betas) across voxels in the ROIs for each participant and each condition using the MarsBaR software package.

Estimation of risk attitude and probability weighting parameters

We estimated participants’ risk attitudes and probability weighting using the data from the in-scanner task. To derive the estimating equation, we begin with the fundamental assumption that participants adjust their willingness-to-sell prices to maximize their expected utility. Let L represent a lottery. The participant’s expected utility from a lottery L is given by the formula:

Here, \(\:w\) represents the participant’s wealth, reward denotes the win amount of the lottery, and \(\:f\left(p\right)\) represents the probability of winning. We assume a utility function \(\:u\left(x\right)={x}^{\alpha\:}\), where α measures the individual’s risk attitude. A risk-loving individual has α > 1, a risk-averse individual has α < 1, and α = 1 indicates risk neutrality. For simplicity, we assume that \(\:w=0\) when \(\:u\left(w+0\right)=0\). While the standard expected utility model assumes linear probability weighting with \(\:f\left(p\right)=p\), we incorporate nonlinear probability weighting using the Prelec (1998) probability weighting function \(\:f\left(p\right)={e}^{{-(-lnp)}^{\gamma\:}}\:\:\:\gamma\:>0\).

We assume the participant’s sets the willingness-to-sell price of a lottery by maximizing the following function:

such that \(\:r\in\:\left[0,reward\right]\). As the buying price is randomly drawn from a uniform distribution between 0 and the winning amount (reward), the participant sells the lottery ticket and receives the price r with probability \(\:\frac{r}{reward}\). Otherwise, with probability \(\:1-\frac{r}{reward}\), the participant cannot sell the ticket and receives the expected return of the lottery ticket.

The first-order condition of the maximization problem can be expressed as:

Taking the natural logarithm of both sides and rearranging the equation yields:

We estimated the risk parameter α and probability weighting parameter γ using data from all participants, the equation above, and nonlinear least-squares estimation.

We assume that the risk parameter α is linearly dependent on gray matter volume, represented as \(\alpha =\) \({\beta }_{0}+{\beta }_{1}ROIgmv\) \(+{\beta }_{2}entrepreneur\) \(+{\beta }_{3}ROIgmv\) \(\times entrepreneur\). Here, the variable \(ROIgmv\) refers to the participant’s gray matter volume in a predefined region of interest (ROI). To account for variations in overall brain size, we normalized the ROI variables by dividing the gray matter volume in each ROI by the total gray matter volume centered at its mean. The variable \(entrepreneur\) takes a value of one for entrepreneurs and zero otherwise, while \(ROIgmv\times entrepreneur\) represents the interaction between the two variables. The model includes the constant  β₀, and we adjusted for clustering standard errors on the participant level to address correlations between observations from the same participant. For the probability weighting parameter, we assume \(\:\gamma\:={\beta\:}_{0}\), indicating a constant value.

Predicting participants’ entrepreneurial status

To conduct our prediction models, we employed a nested cross-validation procedure. The dataset was divided into holdout and training samples. Five participants from each group were set aside, and the remaining participants’ data were used to train a linear model with a least absolute shrinkage and selection operator (LASSO) penalty. The tuning parameter λ was optimized through ten-fold cross-validation for each training set. Ultimately, we generated out-of-sample predictions for the test samples using the trained models.

We evaluated the accuracy of our out-of-sample predictions using receiver operating characteristic (ROC) curves. ROC analysis is a widely used statistical tool for quantifying the performance of a binary classifier at different trade-offs between false positives and false negatives. ROC curves were computed by comparing our out-of-sample predictions for participants’ entrepreneurial status across various decision threshold values,\(t \epsilon [0, 1]\). The true-positive rate was plotted against the false-positive rate at these threshold values. Predicted values below t were classified as non-entry, while values greater than or equal to t were classified as entry. Each point on the ROC curve represents the empirical false-positive and true-positive rates at that threshold value. ROC analysis allows us to assess the performance of alternative classifiers across their entire operating range. A random classifier yields identical true-positive and false-positive rates, resulting in a 45-degree diagonal line on the ROC curve. A well-performing classifier increases the true-positive rate (moving up on the y-axis) and decreases the false-positive rate (moving left on the x-axis). The area under the ROC curve (AUC) is the most commonly used measure for evaluating prediction model performance. We utilized AUC to compare the performance of our prediction models, where an AUC value of 0.5 corresponds to a random classifier without predictive power.

Finally, we investigated the optimal predictors of participants’ entrepreneurial status. The nested cross-validation procedure produced 10 candidate models by fitting a logistic LASSO regression model to 10 different training samples. We used the following procedure for selecting the best model: First the performance of each model is evaluated by calculating ROC curves and their corresponding AUC values using the test samples. The model with the highest AUC is chosen as the most effective predictor^50,51. Subsequently, a logistic regression model is applied to the entire dataset using the selected set of covariates with non-zero LASSO penalized coefficients⁵². All behavioral analyses and statistical estimations were performed using the Stata software package (version 18).

Behavioral responses to risk and ambiguity

Because both the risk level and reward amounts varied between the lotteries, we first normalized the lottery values the participants had provided, that is, their subjective values for the lotteries. This normalization was achieved by calculating, for each participant, the value relative to the expected value of a lottery (relative subjective value, RSV, where the expected value is normalized to one). In the BDM mechanism, the higher a participant sets the willingness-to-sell price, the lower the probability that the sell will be realized. That is, assigning a high price is riskier than assigning a low price. The average RSV therefore also reflects participants’ underlying risk attitude: the higher a participant prices the lotteries on average, the more willing he is to take risk.

Full size image

Behavioral results. Mean subjective values relative to the expected value of a lottery (RSV, y = 1, represents the expected value). On average, entrepreneurs assessed the subjective value of risky lotteries 18.28% above and slightly ambiguous lotteries 16.27% above their expected value, and the difference is not significant (two-sided t-test, p = 0.21). The control group of managers/employees assessed the subjective value of risky lotteries slightly above (1.05%) the expected value, and the small ambiguity significantly decreased their RSV (risk = 1.05 vs. low ambiguity = 0.97, two-sided t-test, p < 0.001). High ambiguity significantly decreases both entrepreneurs’ and the control group’s RSV (entrepreneurs: risk = 1.18 vs. high ambiguity = 1.02, two-sided t-test, p < 0.001; control group: risk = 1.02 vs. high ambiguity = 0.80, two-sided t-test, p < 0.001).

On average, participants priced the lotteries 5% above their expected value. However, a marked distinction in decision-making behavior emerged between entrepreneurs and managers/employees. Entrepreneurs priced risky lotteries 18% above their expected value—significantly higher than the control group of managers/employees, who priced them 8% higher (Fig. 2; p < 0.001, p-value for two-sided z-test, clustered). This disparity in RSV implies that entrepreneurs have a greater propensity for risk-taking than managers/employees. The positive correlation observed between participants’ RSV and general risk preference scores further supports this finding (pairwise correlation r = 0.31, p = 0.05)³⁸. Moreover, entrepreneurs displayed considerable resilience to ambiguity in their valuation decisions. When faced with low levels of ambiguity, entrepreneurs marginally reduced their willingness-to-sell prices, and even under high ambiguity, they continued to price lotteries above their expected values (Fig. 2; for descriptive statistics, see Supplementary Tables S1-S3). In stark contrast, managers/employees demonstrated a significant reduction in valuation under low ambiguity and priced lotteries well below their expected values in high ambiguity scenarios. 1 summarizes our generalized least squares (GLS) panel regressions explaining the relative subjective value (RSV) by entrepreneurial background and lottery ambiguity levels. The regression results confirm that entrepreneurs assess the subjective value of lotteries statistically significantly higher than managers/employees, and that ambiguity affects their valuation decisions significantly less than managers’/employees’ valuation decisions. In addition, individual characteristics, such as participants’ risk attitude, affect intensity, optimism, or confidence, do not directly explain the difference between entrepreneurs and managers/employees in subjective values and attitudes toward uncertainty. The findings remain consistent when we account for participants’ age and years of study as control variables (Supplementary Material, Table S4). While there is some evidence that entrepreneurs with children assign a higher subjective value to lotteries during risky trials compared to those without children (Supplementary Material, Table S3), including parenthood status as a control variable in the models does not substantially alter the results (Supplementary Material, Table S5).

fMRI results

In the whole-brain analysis, the comparison of all the risky trials versus all the ambiguous trials revealed activation in the bilateral caudate nucleus, and a wide network of other brain regions, including the insula, dorsomedial prefrontal cortex (DMPFC), and dorsolateral prefrontal cortex (DLPFC), all bilaterally (Table S6). Contrasting ambiguity versus risk, however, did not reveal any regions where activation was greater for the ambiguous trials than for the risky trials in the whole-brain analysis.

Next, we conducted a parametric modulation analysis using the expected value of a lottery as a parametric modulator in the first-level model. Focusing exclusively on ambiguous trials, our findings indicate a statistically significant difference between the groups (entrepreneurs��> managers/employees) in the processing of expected value within the right cuneus and anterior cingulate cortex (Table 2; Figure S1, Panel A). When we calculated the average parameter estimates around the peak activation for the right cuneus and anterior cingulate cortex ROIs during the ambiguous trials, it turned out, that, on average, parameter values for entrepreneurs in both ROIs are positive and for managers/employees negative. That is, on average, entrepreneurs have increasing brain activation in these areas with the parametric modulator, indicated by positive parameter values, in contrast to managers/employees who exhibit decreasing activation, suggested by negative parameter values.

Narrowing our analysis to risky trials alone, we identified a statistically significant group difference (entrepreneurs < managers/employees) in value encoding within the bilateral cerebellum (Table 2; Figure S1, Panel B). For entrepreneurs, the average parameter estimates in the left and right cerebellum ROIs were negative, on average, indicating a decrease in brain activation as the expected value increased. Managers/employees showed the opposite trend, with positive parameter values, indicating an increase in brain activation with increasing expected values.

Furthermore, our correlation analysis revealed a statistically significant negative association between brain activation patterns in the left and right cerebellum ROIs and participants’ self-reported general risk attitudes (Figure S2; pairwise correlations: right, r =  − 0.37, p = 0.02; left, r = − 0.34, p = 0.03). Specifically, individuals who reported lower risk-taking tendencies (often managers/employees) had brain activation in these regions that increased with the expected value of a lottery, as reflected by positive parameter estimates. Conversely, participants with a higher risk-taking propensity (often entrepreneurs) showed decreasing brain activation as the expected value increased, as indicated by negative parameter estimates.

When analyzing risky trials only, we observed a negative association between risk attitudes and brain activation also in regions typically associated with subjective value processing in the literature, such as the striatum, anterior insula, and ventromedial prefrontal cortex (vmPFC)^49,53. In all of these areas, there was a negative association between the average GLM coefficients for the expected value of a lottery and participants’ individual risk attitude scores. However, the associations for the vmPFC and left anterior insula did not reach statistical significance (pairwise correlations: left striatum, r = − 0.36, p = 0.02; right striatum, r = − 0.35, p = 0.03; left anterior insula, r = − 0.19, p = 0.23; right anterior insula, r = − 0.35, p = 0.02; vmPFC, r = − 0.22, p = 0.12). When analyzing ambiguous trials only, no similar association was found between risk attitudes and parameter estimates.

GMV results

In the following step, we investigated the relationship between brain gray matter volume (GMV) and participants’ risk attitudes. We employed data from the Becker-DeGroot-Marschak (BDM)³³ decision-making task and examined whether GMV in specific, predefined brain regions (rPPC, DMPFC, bilateral anterior insula) predicts participants’ risk attitudes.

We estimated individual risk attitudes by assuming that participants set the willingness-to-sell price at the level that maximizes their expected utility. We assume that the function \(\:u\left(x\right)={x}^{\alpha\:}\) represents the utility derived from the outcome \(\:x\), and allow for nonlinear probability weighting. In our analysis, we estimated two key parameters: the risk-attitude parameter α and the probability weighting parameter γ. The exponent α in the utility function captures an individual’s risk attitude, where α > 1 indicates risk-loving behavior, α < 1 suggests risk aversion, and α = 1 reflects risk neutrality. Additionally, we used the Prelec⁵⁴ probability weighting function:

In the function (1), the parameter γ governs the convexity and concavity of the Prelec probability weighting function, determining its S-shaped nature. When γ < 1, the Prelec function exhibits strict concavity for low probabilities and strict convexity for high probabilities, indicating an overweighting of low probabilities and an underweighting of high probabilities. Conversely, if γ > 1, the opposite pattern emerges.

Table 3 displays the estimation results when analyzing only the data from risky trials. Notably, despite the Prelec probability weighting function being known to align with decision-making behavior in various contexts, including our study, our findings deviated from the typical pattern. Here, the probability weighting parameter γ consistently demonstrated a magnitude of approximately 2.4 across all three models, each utilizing a different ROI. This parameter maintained high significance across all ROIs and within all three models (p < 0.001). This consistency suggests that our participants exhibited a consistent tendency to underweight low probabilities and overweight high probabilities. This behavior contrasts the usual pattern of γ < 1, where individuals would typically overweight low probabilities and underweight high probabilities.

As shown in the upper part of Table 3, we observe no statistically significant association between risk attitudes and PPC anatomy (ROI 1). However, in the case of the next ROI (ROI 2), a highly significant negative association is evident for DMPFC. Furthermore, we find a positive correlation between participants’ risk attitudes and their gray matter volume (GMV) in both the right and left anterior insula (ROI 3 and ROI 4), although the parameter values were weakly significant only in the case of the right anterior insula. Of particular interest, the interaction term was positive and significant both for the right and left anterior insula, suggesting that a higher GMV in the insula may be linked to a propensity for risk-taking among participants, especially entrepreneurs.

We also show that accounting for age-related differences does not significantly alter the observed relationship between risk attitude and brain structure in decision-making during risky trials (see Supplementary Material, Table S7). However, when controlling for participants’ parenthood status, a notable difference emerges: higher GMV in the right PPC is significantly associated with a lower risk attitude parameter in entrepreneurs (see Supplementary Material, Table S8). Finally, when using the full dataset (risky and ambiguous trials) but without additional controls, the findings remain consistent with those reported in Table 3 (see Supplementary Material, Table S9).

Prediction

Next, we investigate how well participants’ self-rated characteristics, immediate, average, neural responses to risk and ambiguity, and neuroanatomical features predict their entrepreneurial status. We utilized machine learning algorithms to evaluate the predictive power of our behavioral, fMRI and GMV data in relation to entrepreneurship. To assess the predictive capability, we developed four distinct models for out-of-sample prediction of participants’ entrepreneurial status. In all models, participants’ pricing decisions were included as a critical behavioral variable (RSV, see Fig. 3). The baseline model further incorporated other relevant behavioral variables, such as participants’ self-rated general risk attitude, confidence, affect intensity, and optimism scores. Additionally, age and years of education were included in the baseline model. The second model included all fMRI variables measured during risky trials, while the third model included all fMRI variables measured during ambiguous trials. The fourth model focused on the inclusion of all GMV variables. This modeling approach allowed us to systematically compare the predictive power of each category of variables and identify which factors provide the most significant insight into entrepreneurial tendencies.

Full size image

Predictive models for entrepreneurship and biometric variable effects. Panel (A) ROC curves for predicting entrepreneurship. The dashed line (45-degree lines) represents the false-positive and true-positive rates of a random classifier. Panel (B) The figure illustrates the average marginal effects and their corresponding 95% confidence intervals from logistic regression for outcome prediction, where entrepreneurial status is explained by pricing decisions and brain responses to the lotteries during risky trials (fMRI variables, clustered standard errors).

Figure 3 (Panel A) displays ROC curves, providing insights into the relative trade-offs in predicting entrepreneurial status. All models exhibit greater predictive accuracy compared to a random classifier. Model 1, utilizing behavioral data alone, achieves significantly better predictive accuracy (AUC = 0.728, 95% confidence interval 0.712–0.745) compared to a classifier combining participants’ valuation decisions with fMRI ROI data during ambiguous trials (Model 3; AUC = 0.665, 95% confidence interval 0.647–0.683), as well as a classifier incorporating participants’ valuation decisions and their GMV in selected ROIs (Model 4; AUC = 0.613, 95% confidence interval 0.594–0.631). Remarkably, Model 2, integrating participants’ valuation decisions with fMRI ROI data during risk trials, demonstrates the highest AUC (AUC = 0.769, 95% confidence interval 0.754–0.785) among all alternative models.

Figure 3 (Panel B) presents the marginal effects extracted from the logistic regression model with the highest predictive accuracy. This model explains entrepreneurial status through pricing decisions and brain responses to the expected value of lotteries during risky trials (for marginal effects using brain responses to lotteries during ambiguous trials, see Figure S3). Notably, a significant positive correlation emerges between participants’ brain responses in the left anterior insula and left striatum during risky trials and their likelihood of being entrepreneurs. This means that a stronger brain response to the expected value of a lottery in these regions increases the likelihood of a participant being classified as an entrepreneur. In the cerebellum, the marginal effect is negative, but it is only significant on the left side.

We investigated the interplay between self-rated behavioral traits, brain structure, neural responses and behavior during decision-making under risk and ambiguity, comparing entrepreneurs with employees/managers. Using a machine learning approach, we aimed to predict participants’ entrepreneurial status—their classification as entrepreneurs or non-entrepreneurs—and to assess the relative predictive power of self-reported traits, brain responses (fMRI), and brain structure (GMV). The method identifies patterns in the data and evaluates their generalizability by training on one portion and predicting outcomes on a separate, unseen set. Immediate brain responses to risk, measured in fMRI, emerged as the strongest predictor of entrepreneurial status.

Decision-making in uncertain contexts has been studied before using both fMRI and structural brain data^45,46,53. In general, fMRI captures dynamic brain activity by measuring blood-oxygen-level-dependent (BOLD) signals, which reflect neural activity in response to specific tasks or conditions. In this study, fMRI data allowed us to examine how the brain responds in real-time during ambiguous and risky decision-making tasks, capturing cognitive processes such as value evaluation, risk processing, and ambiguity tolerance^31,55. These responses are task-specific and can vary significantly based on context and individual differences.

In contrast, structural brain data, such as GMV, reflect long-term anatomical differences that develop over time due to the interplay of genetic, environmental, and experiential factors^56,57,58. Structural differences are relatively stable and may underpin broader traits or capacities, such as general cognitive abilities, personality, or predispositions relevant to entrepreneurship. For example, structural data may indicate baseline neural resources or capacity in regions critical for decision-making but do not provide information about neural engagement during specific tasks.

Notably, immediate brain responses to risk, measured in fMRI, demonstrated the highest predictive power in distinguishing entrepreneurs from non-entrepreneurs. This finding is intriguing as no previous study has investigated these neural mechanisms in the context of entrepreneurial decision-making, highlighting the novel contribution of our research. The higher predictive power of brain activity during risky decision-making in this study could reflect the fact that fMRI captures transient, task-specific processes more directly tied to the experimental paradigm. Entrepreneurs, who are often exposed to risky decision-making in their professional lives, may show distinct task-related neural patterns that contribute to the predictive strength of fMRI data. Structural brain data, while informative, may provide a broader, less task-specific perspective, capturing traits that are relevant but not directly engaged during the decision-making tasks.

While brain structure measures provide insight into baseline individual traits that may underlie cognitive and decision-making abilities, self-reported behavioral traits offer a more direct understanding of how individuals approach risk and uncertainty. We measured a range of self-reported behavioral traits—risk preferences, affect intensity, confidence, and optimism—since these traits have been shown to influence how individuals approach risks and opportunities, also in entrepreneurial contexts. Including these behavioral traits in the prediction model, however, did not lead to the highest predictive accuracy. Additionally, when included as control variables in the model explaining participants’ behavior in the task, these traits did not alter the basic results, and none of the traits significantly explained participants’ choices. Considering the wide variety of cognitive abilities, such as divergent thinking⁵⁹, and personality traits, like the Big Five dimensions—openness to experience, extroversion, agreeableness, emotional stability, and conscientiousness—commonly associated with entrepreneurship^5,6,8, we acknowledge that other traits or combinations of traits may influence entrepreneurial behavior and could potentially improve the predictive accuracy of the model.

Our results highlight the significance of anterior insular activation and, to a lesser extent, anterior insular GMV, as predictors of entrepreneurial status, with the insular GMV also correlating with entrepreneurs’ general risk attitude. These findings are consistent with existing research linking the anterior insula to entrepreneurial behavior, particularly its role in processing risk and decision-making under uncertainty. The anterior insula has been implicated in various aspects of risk-related cognition, including risk perception, risk aversion, and the detection of risky situations^60,61,62,63. In line with these findings, our study also shows that patterns of activation in the right anterior insula correlate with participants’ risk attitudes. These results suggest that the anterior insula may serve as a neural indicator of long-term entrepreneurial experience, with real-time activation providing deeper insights into entrepreneurial behavior. Our research further supports the idea that entrepreneurs are more inclined to engage in risk-taking than employees or managers^11,12,22, underscoring the importance of specific brain responses in risk-related decision-making.

The model with the highest predictive accuracy shows a significant positive correlation between participants’ brain responses in the left anterior insula and left striatum during risky trials and their likelihood of being entrepreneurs. These regions, linked to reward processing and positive emotions⁶², may reflect an entrepreneurial mindset in risk-related situations. The anterior insula plays a crucial role in emotion regulation, overseeing, evaluating, and adjusting emotional responses. Its activation is consistently observed, irrespective of the regulation strategy employed⁶⁴. It is a key component of the salience detection network⁶⁵, involved in processing emotional valence, both positive and negative⁶⁶, and has been linked to awareness of bodily sensations, such when seeing something disgusting or during orgasm⁶⁷, suggesting that it might enable the “gut feeling” that guides intuitive decision-making. Additionally, the insula plays a significant role in supervisory attentional control, which is essential for tasks requiring cognitive flexibility, such as go/no-go or Stroop tasks^68,69. Prior fMRI studies have linked insular activity to greater cognitive flexibility, particularly in entrepreneurs⁷⁰, indicating that they may have a superior ability to adjust and switch their behavior in response to changing circumstances⁷¹. This cognitive flexibility is beneficial in entrepreneurial environments, where risk-taking and the ability to adapt to new situations are crucial^28,72.

At the same time, in entrepreneurs, cerebellar areas involved in emotional processing and regulation^73,74,75,76 showed decreased activation as expected value increased, tentatively. This suggests that entrepreneurs may become excited when faced with risky decision-making situations, leading to a reduction in neural mechanisms responsible for emotion regulation. Prior research utilizing behavioral data has posited that entrepreneurs are inherently emotional and may find it hard to control their emotions in exhilarating circumstances^77,78,79. Such lack of self-regulation may make them less cautious, as indicated by our behavioral results that show entrepreneurs valuing lotteries higher than non-entrepreneurs.Thus, our fMRI findings may offer insights into the neural mechanisms that could underlie the impulsive, thrill-seeking, and action-oriented decision-making tendencies sometimes observed in entrepreneurs, as well as potential difficulties in emotion regulation during periods of heightened excitement^{77,79,80,81,82}.

Our findings align with previous research indicating that entrepreneurs have a higher risk-taking propensity than non-entrepreneurs^8,13,14, while contrasting with studies that did not find such a difference^9,10. However, it is important to recognize that this tendency — observed in our results — might lead to both benefits and drawbacks. Research on cognitive biases in entrepreneurship suggests that higher risk-taking propensity can sometimes lead entrepreneurs to take excessive risks, particularly when decisions are made quickly and without thorough assessment^83,84. Often, such decisions are driven by “System 1 thinking”, a rapid, unconscious mode of thinking that leads to automatic responses to stimuli^85,86. Coupled with the excitement of high potential rewards, this tendency can make entrepreneurs more willing to take on physical, social, legal, and financial risks^80,84.

Nonetheless, entrepreneurs often prioritize the potential positive outcomes over the associated risks. For instance, choosing to pursue an entrepreneurial career with uncertain income, as opposed to a more secure salaried position, can lead to serious consequences if not carefully considered¹⁰. That said, the ability to make quick decisions can be advantageous, particularly in competitive markets where brief windows of opportunity arise, often tied to new business models or technologies^81,85,87. Our research suggests that this willingness to take immediate risks may play a role in facilitating entry into entrepreneurship — a tendency that may be less common among individuals who adopt a more cautious and analytical mindset^81,82.

We have also identified a negative and highly significant association between participants’ estimated risk attitudes and the structural characteristics of the DMPFC. Additionally, we observe a similar negative correlation between GMV in the right posterior parietal cortex (PPC) and participants’ inclination toward risk-taking, although this relationship does not reach statistical significance. Interestingly, when controlling for participants’ parenthood status, we find that higher PPC GMV is significantly associated with lower risk-taking in entrepreneurs. This result contrasts with previous findings by Gilaie-Dotan et al. (2014), who reported that greater GMV in the right PPC was linked to reduced risk aversion⁴⁵. One possible explanation for this discrepancy is that our study focuses on individuals at the higher end of the risk-taking spectrum, as entrepreneurs, as a group, are generally more inclined to take risks than the general population. Notably, our findings align with those of Aydogan et al. (2021), who reported a consistent inverse relationship between GMV in several brain regions—including the amygdala, ventral striatum, and dorsolateral prefrontal cortex (DLPFC)—and real-world risky behaviors⁵⁸. What sets our results apart is that, in our study, entrepreneurs exhibit a positive correlation between gray matter volume in the right and left anterior insula and their inclination for risk-taking. These findings suggest that unique structural characteristics of the entrepreneurial brain may contribute to differences in risk-taking preferences and further underscore the pivotal role of the anterior insula in entrepreneurs’ risky choices.

The estimation results of the structural models suggest that the participants employ a different approach to probability weighting compared to typical adults. This is interesting because, unlike many adults who tend to overestimate low-probability outcomes and underestimate high-probability ones, our participants display the opposite pattern. Relevant to our discussion is the study by Harbaugh et al. (2002), which examines risk attitudes across different age groups and suggests that the probability weighting function is not fixed but evolves with factors such as age⁸⁸. That study indicates that younger individuals tend to display more impulsive behavior by underweighting low probabilities and overvaluing high probabilities, affecting their inclination toward risky actions. Similarly, prior research in entrepreneurship has found that entrepreneurs perceive risk differently from other individuals⁸⁹. Our findings add insight into the origins of these perceptual differences, by showing that entrepreneurs’ risk-taking behavior resembles that of younger adults, suggestive of a youthful appetite for risk. This observation supports the hypothesis that entrepreneurial risk-taking may stem from an innate form of “impulsivity”, or that it develops with experience in entrepreneurial activities²⁸. Regardless of the cause, it could be concluded that just like children who tend to be impulsive and attracted by novelty and exploration regardless of associated risks, entrepreneurs do not shy away from risky situations but rather seem to embrace them⁸⁸.

One limitation of our study is that it focuses solely on male entrepreneurs at a single time point. Extensive research in economics demonstrates that women are generally less inclined to take risks and are less likely to invest in higher-risk assets^90,91. Similarly, although neuroscience research on this topic is limited, findings suggest that women may process risk differently at the neural level, potentially leading to more cautious decision-making compared to men^92,93. Research also indicates that female entrepreneurs often exhibit a strong sense of responsibility, prioritizing the protection of family resources and, as a result, favoring safer choices for their ventures compared to their male counterparts^94,95. These findings highlight the importance of including both male and female entrepreneurs in future studies to examine potential gender differences in behavioral and brain responses to risky and ambiguous decision-making situations. While female entrepreneurs might exhibit more cautious behavior than their male counterparts, with corresponding differences in brain responses and possibly brain structure, we do not anticipate that these differences would reduce the predictive power of immediate brain responses to risk in classifying entrepreneurial status for female entrepreneurs.

Another potential limitation of this study is the non-randomized nature of our sample, which may affect the generalizability of our findings. While practical considerations shaped our recruitment strategy, this approach could have introduced confounding variables, such as parenthood status, that might influence the results. Because the managers in our study also participated in a parenthood-related experiment, all of them had children. In contrast, 40% of the entrepreneurs in our sample were parents. However, our analysis found no statistically significant differences in key background variables—such as risk attitude, confidence, optimism, and affect intensity—between entrepreneurs with and without children. These findings suggest that parenthood status is unlikely to have driven the observed group differences between entrepreneurs and non-entrepreneurs. Nevertheless, we acknowledge the importance of employing randomized sampling strategies in future research to enhance generalizability. It is also worth noting that very few neuroimaging studies recruit samples representative of the general population. Within this context, our study offers valuable insights into the neural and behavioral foundations of entrepreneurial and managerial decision-making.

Our study stands as an initial foray into the neural underpinnings of entrepreneurial risk-taking behavior. It makes several important theoretical contributions to the fields of entrepreneurship and decision-making. First, we advance the understanding of entrepreneurial cognition by demonstrating that entrepreneurs exhibit unique task-related neural activation patterns during risky decision-making. Second, we extend and validate prior research on the role of the anterior insula, implicated in valence processing and supervisory control supporting cognitive flexibility, in risk-related cognition. Specifically, we demonstrate that entrepreneurial status can be predicted based on insular activation. Third, we contribute to the literature on emotional regulation in entrepreneurship by suggesting that, as the expected value of a lottery increases, entrepreneurs show decreased activation in cerebellar regions associated with emotion regulation. This pattern implies that emotional arousal and excitement may impair their capacity for self-regulation, potentially overriding caution and leading to impulsive decisions. Supporting this interpretation, our results indicate that entrepreneurs’ probability weighting resembles that of younger adults, reflecting a youthful appetite for risk. Fourth, we add to previous research on entrepreneurial neuroanatomy and neuroplasticity⁹⁶ by showing that differences in risk-taking propensity may be reflected in structural brain features—specifically, increased GMV—potentially shaped by prior entrepreneurial experience. Finally, we contribute to the growing interdisciplinary field at the intersection of entrepreneurship and neuroscience, demonstrating that neurobiological insights can meaningfully enrich our understanding of entrepreneurial behavior.

While our research opens various avenues for future exploration, further investigation is particularly warranted to understand the neural response to risk in different subgroups of entrepreneurs. Our current focus on male entrepreneurs at a single time point lays the groundwork for longitudinal and comparative studies that can unravel the complex mechanisms we have begun to uncover. Thus, future research could compare the neural responses and structural brain differences of novice and serial entrepreneurs to examine how entrepreneurial experience shapes decision-making in situations involving risk and uncertainty. Since our study focused solely on male entrepreneurs, future studies could use a similar design to investigate potential gender differences in entrepreneurial behavior, particularly in how male and female entrepreneurs process risk. This would test the hypothesis that females may demonstrate more risk-averse behavior, which could lead to distinct patterns of brain activation in regions associated with risk processing.

As our study revealed, entrepreneurs show a strong appetite for risk-taking, as evidenced by both behavioral data and structural and functional MRI findings. While such risk tolerance can be advantageous in entrepreneurial settings, it also carries the danger of excessive risk-taking, which may contribute to venture failures or bankruptcies. Against this background, it is important for policymakers to ensure that the environment in which entrepreneurs operate does not have overly stringent bankruptcy regulations that increase the costs of failure. At the same time, fostering a societal culture that does not stigmatize failure is equally important, as it can encourage risk-taking and entrepreneurial activity without the fear of long-term negative consequences.