Individual differences in risk preference predict neural responses during financial decision-making

Abstract

We investigated the neural correlates of subjective valuations during a task involving risky choices about lotteries. Because expected value was held constant across all lotteries, decisions were influenced by subjective preferences, which manifest behaviorally as risk-seeking or risk-averse attitudes. To isolate structures encoding risk preference during choice, we probed for areas showing increased activation as a function of selected risk-level. Such response patterns were obtained in anterior (ACC) and posterior cingulate cortex (PCC), superior frontal gyrus, caudate nucleus, and substantia nigra. Behavioral results revealed the presence of risk-averse and risk-neutral individuals. In parallel, brain signals revealed modulation of activity by risk attitude during choice. Correlations between risk-seeking attitudes and neural activity during risky choice were obtained in superior and inferior frontal gyri, medial and lateral orbitofrontal cortex, and parahippocampal gyrus, while correlations with risk-averse attitudes were found in the caudate. The dynamics of neural responses relevant to each stage of the task (decision, anticipation, outcome) were investigated via timeseries and conjunction analyses. Though the networks engaged in each of the task stages were mostly distinct, regions within ACC, PCC and caudate were consistently activated during each decision-making phase. These results demonstrate (1) that subjective assessments of risk, as well as individual attitudes toward risk, play a significant role in modulating activity within brain regions recruited during decision-making, and (2) that ACC, PCC and caudate are relevant during each phase of a decision-making task requiring subjective valuations, strengthening the role of these regions in self-referential subjective valuations during choice.

Introduction

The majority of our everyday decisions involve some level of risk. Decision-making under risk is a complex process and both cognitive and emotional factors are involved in the selection of goal-directed responses (Kahneman and Tverksy, 1979, Loewenstein et al., 2001, Mellers, 2000). In order to maximize rewards, a decision maker needs to evaluate alternative courses of actions whose outcomes are uncertain. According to classical expected utility theory, the standard microeconomic model of choice under risk, such valuations involve estimations of the reward magnitude and likelihood with which some outcome can be obtained. The multiplicative product of objective reward magnitude and likelihood is referred to as expected value (Bernoulli, 1763/1958, von Neumann and Morgenstern, 1944). Expected Utility Theory (EUT) has proven to be a useful construct for predicting animal and human choices (Camerer, 2003). Importantly, the utility of obtaining a given outcome can be influenced by subjective factors determining the “moral”, or subjective, value of a lottery (Kable and Glimcher, 2007, Schunk and Betsch, 2006, Trepel et al., 2005). An influential factor is a person's preference for risk. Results from previous research indicate that most people show risk-averse attitudes when gambling for monetary gains, which is reflected by a concave utility function (Abdellaoui, 2000, Gonzales and Wu, 1999, Tversky and Fox, 1994). Various experiments, however, have reported heterogeneity in risk attitudes across individuals (e.g. Fetherstonhaugh et al., 1997, Huettel et al., 2006, Schunk and Betsch, 2006). Results from these experiments support the notion that individuals not only estimate expected value, but more importantly, they seem to weigh these estimates by subjective factors, such as risk preference (Kahneman and Tverksy, 1979, Schunk and Betsch, 2006) and the anticipated pleasure (Mellers, 2001) and dread (Berns et al., 2006) of the outcome. Due to the subjective nature of these variables, these estimates can vary from person to person and thus influence individual's decisions to different degrees. Such heterogeneity in risk attitudes across individuals has been demonstrated in behavioral economics and neuroeconomics (e.g. Barsky et al., 1997, Beetsma and Schotman, 2001, Huettel et al., 2006, Kachelmeier and Shehata, 1992, Schunk and Betsch, 2006).

In the current experiment, we set out to investigate the neural correlates of individual differences in risk attitude employing a two-choice financial decision-making task between lotteries involving risk. This task is well-established in behavioral economics and allows for the extraction of factors that can influence choice, particularly risk preference. Here, we created a setting within which participants were asked to make binary choices between lottery pairs of equal expected value. Lotteries varied in reward magnitude and winning probability, such that one of the lotteries yielded a higher payoff at a comparatively large risk, while the other yielded a lower payoff, but involving less risk. Because in the context of the current paradigm decisions were not influenced by the expected value of lotteries, which was equal for all decisions, but by subjective factors involved in participant's valuations of lottery pairs, e.g. risk attitude, we created a setting ideal for investigating the neurobiological basis of individual differences in risk preference. We extracted risk preference from participants' choices during scanning using nonlinear logistic regression. This approach provided individualized risk preference parameters, which characterize the curvature of participants' utility functions. We then examined the neural basis of risk preference by regressing each participant's behavioral risk preference parameter against neural activations during risky choice. Finally, we investigated activation patterns in these regions separately for risk-averse and risk-seeking individuals to illustrate differential activation patterns as a function of risk attitude.

In the second part of this paper, we made use of the event-related fMRI design employed in the current experiment to investigate potential differences and similarities in networks recruited during component phases of decision-making under risk (Fig. 1). Component phases include (1) a selection phase during which choices between the two lotteries were made, (2) reward anticipation and (3) the outcome phase, during which the outcome of the chosen gamble was presented resulting in either receipt or omission of the anticipated cash reward (e.g. Ernst et al., 2004, Knutson et al., 2000, Knutson et al., 2001a). While previous research has dissociated different phases of decision-making under uncertainty (e.g. Liu et al., 2007; for review see also Knutson and Cooper (2005)), to our knowledge, this is the first study on decision-making under risk investigating the dynamics of all three task phases.

The selection phase requires the evaluation of sensory input based on prior knowledge and experiences made in similar situations (Platt, 2002). A course of action is then chosen from potential alternatives. During this evaluatory period, an anticipation of the potential outcome is formed on the basis of probabilities of expected outcomes and the magnitude of the reward in conjunction with previous experiences (Ernst et al., 2004, Mellers, 2001, Platt and Glimcher, 1999). Based on previous research on risky decision-making (e.g. Cohen and Ranganath, 2005, Huettel et al., 2005, Paulus and Frank, 2006, Preuschoff et al., 2006), we expected the choice evaluation network to include anterior insula, the anterior cingulate cortex, ventral and dorsal striatum, and posterior cingulate and orbitofrontal cortex. During the second phase, reward anticipation is maintained until the outcome period and typically involves activation of ventral striatum, whose importance in computing reward expectancy has been demonstrated by both monkey electrophysiology (Mirenowicz and Schultz, 1996, Schultz et al., 1992, Schultz et al., 1997) and human neuroimaging studies (e.g. Knutson et al., 2001a, Knutson et al., 2005). We also expected to observe activations in areas involved in sustained attention, such as parietal cortex (Corbetta et al., 2000, Engelmann et al., 2009, Hopfinger et al., 2000, Pessoa et al., 2002). Finally, during the outcome period, risk is resolved and an evaluation of the obtained result takes place. This involves a comparison between the expected outcome and the actual outcome. The discrepancy between expected and obtained outcomes forms the prediction error, which is a key term in computational models of reinforcement learning (Montague and Berns, 2002, Montague et al., 2004, Schultz and Dickinson, 2000). The reward prediction error provides the basis for adjustments of future behavior and reward expectation (e.g. Hollerman et al., 1998, McClure et al., 2003, O'Doherty et al., 2003b, Seymour et al., 2004) and has been shown to engage ventral striatum (McClure et al., 2003, Menon et al., 2007, O'Doherty et al., 2003b) and orbitofrontal cortex (O'Doherty et al., 2003b).

To extract the network encoding risk preference during choice, we sorted trials based on the decisions made by subjects and then probed for brain regions showing linear increases in activation as a function of relative risk-level selected by participants during decision and anticipation phases. We hypothesized that risky choices between lotteries of equal expected value would lead to differential brain activations in regions encoding subjective valuations concerning a trade-off between risk-level and gain magnitude of the chosen lottery. In order to examine the relative involvement of each area during different phases of our task, we investigated activation patterns within each of the task phases using deconvolved time courses spanning the entire trial length. This allowed us to examine whether regions activated during one phase also showed increased activations during other phases. Results from time course analyses were corroborated by conjunction analyses, which allowed us to inspect the relative spatial overlap between activation clusters observed in different task phases.