The biology of risk-taking: why teenagers, traders, and soldiers make the same mistakes

In a neuroimaging lab at Temple University, a sixteen-year-old lies in an MRI scanner playing a driving simulation. The task is simple: as a yellow light appears, brake and lose a few seconds, or run it and gain a few seconds. Two of his friends are watching through an observation window. They don’t say anything. They don’t give instructions. They just watch. And during the blocks when the friends are present, the adolescent’s ventral striatum and orbitofrontal cortex show measurably elevated activation compared to the blocks when he drives alone. Not after he runs a red light. Not during the decision. Before it. The reward circuitry is already primed by the time the choice arrives.

What Jason Chein and his colleagues were measuring in that 2011 study wasn’t a response to risk. It was a response to the audience.

Here is where the standard story of teenage risk-taking starts coming apart. The standard account goes: the prefrontal cortex — the brain’s executive center, responsible for planning, impulse control, and consequence assessment — doesn’t fully mature until the mid-twenties, and adolescents do dangerous things because they lack a complete regulatory system. Plausible. Widely repeated. And undercut by what happens when you run the same risk-tolerance tests on teenagers sitting alone. Under solitary, controlled conditions, adolescent risk profiles are barely distinguishable from adults. The “immature brain” explanation predicts a consistent cognitive deficit. The data show a context-dependent one.

If the problem were simply cognitive capacity, the size of the effect wouldn’t depend on whether anyone was watching. But it does. Peer presence roughly doubles adolescent risk-taking in the same experimental task, producing no comparable effect in adults tested under identical conditions. The answer implicates a specific neural circuit responding to social salience — not general immaturity of judgment. And that circuit, it turns out, doesn’t fully retire with adolescence. A version of it is running in a 38-year-old bond trader during a bull market, and in a naval officer whose ship is taking fire. Different triggers, different stakes. The same underlying hardware.

The development gap

The model that best fits the data is Laurence Steinberg’s dual-systems framework. Two neural systems mature on different schedules. The socioemotional system — centered on the ventral striatum, orbitofrontal cortex, and other limbic structures — peaks in sensitivity around mid-adolescence, then gradually declines. The cognitive control system — built around the lateral prefrontal cortex, anterior cingulate, and parietal cortex — continues developing into the early-to-mid twenties. What makes this more than a scheduling curiosity is the direction of the asymmetry. During the years when the gap is widest, it isn’t simply that the brakes are weak. The accelerator is running hot.

The socioemotional system is actively more sensitive to reward signals during adolescence than it will be at any later point. Steinberg drew this distinction in a 2008 paper in Developmental Review and formalized it as the dual systems model in a 2010 paper in Developmental Psychobiology — two distinct papers making two distinct arguments, the first providing the peer-influence framework and the second articulating the full developmental model. The critical point — and it’s one the pop-science version consistently flattens — is that this isn’t a deficit model. It’s an amplification model. The adolescent brain isn’t failing to process reward the way an adult brain would. It’s processing reward more intensely. Casey, Getz, and Galvan documented the same asymmetry in Developmental Review in 2008: the problem isn’t faulty software in the cognitive control system, it’s a temporary timing mismatch between two systems that don’t fully balance each other until the second half of the third decade.

That mismatch is what makes social context so potent. Margo Gardner and Steinberg showed in 2005, in Developmental Psychology, that adolescents — but not adults — took significantly more risks in a simulated driving task when peers were watching, using a design that controlled for direct group pressure by having each subject drive alone while knowing friends were present. Early adolescents scored roughly twice as high on the risky driving index with peers in the room as without. Adults showed no equivalent effect. Chein’s 2011 study went inside the skull to show the mechanism: peer observation produced selectively elevated striatum and OFC activation in adolescent subjects, and that neural priming predicted the number of risks they went on to take within those same blocks. The adults’ reward circuitry simply didn’t respond to the audience in the same way.

Dopamine receptor density in the striatum reaches a developmental peak during adolescence — which is why the social salience signal hits as hard as it does. A friend behind an observation window provides no information about whether it’s safe to run a yellow light. But the brain isn’t processing it as information. It’s processing it as reward context. In an adolescent striatum, that context rewrites the parameters of the decision before the decision is made.

The thing the individual-blame framework misses entirely: this is adaptive. Adolescence is the developmental window for establishing peer bonds, finding a mate, transitioning from family to peer-group social structure. A nervous system that weighted social salience heavily would have been enormously useful for anyone navigating those tasks. The risk-taking is a side effect of a calibration that served the species well — in an environment that calibration was never designed for.

The 18-year threshold

The prefrontal cortex is among the last brain regions to complete myelination, with full structural maturity not reliably reached until the mid-to-late twenties. The socioemotional/cognitive control gap doesn't close overnight — it narrows gradually across the late teens and early twenties, with significant individual variation. Voting, military service, criminal accountability as an adult: all conferred at a developmental moment when the balance between reward sensitivity and executive control is not yet at its adult equilibrium. The neuroscience doesn't invalidate the age of majority. It notes, without sentiment, that the brain recognizes no such anniversary.

The hormonal dial

The development window closes. What doesn’t close is the hormonal sensitivity of the adult reward circuit.

In 2008, John Coates and Joe Herbert published in PNAS what remains one of the most striking naturalistic studies in behavioral neuroscience: a real-world examination of 17 male traders at a mid-sized City of London trading floor, observed across eight consecutive trading days. Saliva samples collected twice daily measured testosterone and cortisol levels. The headline finding: a trader’s morning testosterone level — measured at 11 a.m. — predicted his afternoon profitability. On days when his 11 a.m. testosterone was above his personal median, he took greater risks and made significantly greater returns over the following five hours. The hormonal state he arrived with shaped the decisions he made across the rest of the day.

Coates, who had spent years at Goldman Sachs and Deutsche Bank before becoming a neuroscientist at Cambridge, extended this finding in his 2012 book The Hour Between Dog and Wolf into what he called the winner effect — a well-documented phenomenon in animal behavior in which a competitive win raises testosterone, which raises confidence and risk appetite, which increases the probability of winning again. The loop is established in non-human animal research; Oyegbile and Marler documented the neurochemical baseline in Hormones and Behavior in 2005. Coates’s inference that something structurally similar operates in human traders was explicit about its limits: the 17-trader study is observational, not experimental, and too small for confident generalization. Cueva and colleagues provided experimental support in 2015 in Scientific Reports, administering cortisol or testosterone to 142 participants in an asset-market simulation and finding that both hormones increased risk-taking and price instability — testosterone largely via elevated optimism about future price movements.

The cortisol finding in Coates & Herbert is, if anything, more significant than the testosterone result. Cortisol didn’t peak when losses were largest. It rose when profit variance was highest — when the market was volatile in either direction. The implication: cortisol isn’t a loss-response system. It’s a risk-detection system, calibrating the body’s alert state to an unpredictable environment. And elevated cortisol correlates with increased risk aversion.

So the hormonal system creates two distinct failure modes. Testosterone-driven winner effects produce runaway risk appetite in conditions of success. Cortisol-driven stress responses produce paralytic caution in conditions of volatility. Neither is irrational given the environment that shaped these systems. Calibrating behavior to competitive dominance and environmental unpredictability made sense for most of the period in which these hormones evolved. What these systems weren’t calibrated for was a screen displaying bid-ask spreads on credit default swaps.

One limitation the data will not carry: Coates & Herbert studied male traders exclusively. Nothing in the study speaks to women in equivalent roles, and subsequent research on sex differences in hormonal modulation of financial risk-taking is more complex and less settled than the secondary literature usually acknowledges. That’s not a footnote — it means the model has hard data for roughly half the relevant population and nothing for the other.

When the floor moves together

If individual traders' testosterone levels are modulated by the same underlying market signals, and all traders on a floor are exposed to those signals simultaneously, then the winner effect could synchronize across a trading floor — amplifying collective risk appetite during rising markets, amplifying collective risk aversion during crashes, in a feedback loop that operates at the level of a market rather than an individual. Coates raises this possibility explicitly in The Hour Between Dog and Wolf, and it connects to the systemic risk literature: not individual traders making uncoordinated errors, but a coordinated biological response to shared environmental inputs, producing correlated behavior precisely when correlated behavior is most dangerous. Whether this is measurable at market scale is an open research question. Whether it's worth taking seriously is not.

What money does to the brain

The behavioral data says what hormones do to risk appetite. What it doesn’t account for is why financial loss registers the way it does — why a stop-loss rule that a trader agrees to in the morning becomes psychologically impossible to execute in the afternoon. Brian Knutson at Stanford put people in fMRI scanners to find out.

The 2007 paper he coauthored with Peter Bossaerts in the Journal of Neuroscience found a clean anatomical dissociation: the ventral striatum — specifically the nucleus accumbens — activates in anticipation of financial reward. The anterior insula activates in anticipation of financial risk and potential loss.

The anterior insula is the same region activated by physical pain, disgust, and social exclusion.

And critically, the insula activation precedes the decision. The neural environment in which a financial choice is made is already shaped by what amounts to a pain signal before the choice is taken. Here is the mechanism behind behavior that looks most irrational from outside: the trader who holds a losing position past any defensible stop-loss point, adding to it as the loss grows. Closing the position makes the loss real — which fires the insula. Holding the position keeps the loss notional, on paper, and the brain treats those two states differently. Not rationally. Differently. The circuits engaged in this decision predate financial markets by millions of years and were not calibrated for abstract instruments.

When losses hurt more than gains feel good

Kahneman and Tversky's loss aversion finding — that losses register as psychologically more powerful than equivalent gains — has a neural substrate in the insula. The specific ratio most often cited (ranging across studies from roughly 1.5x to 2.5x, not a single settled figure) is a median over a literature with considerable variance; treat any single number as provisional. Richard Thaler's endowment effect — the tendency to overvalue what you already possess relative to what you might acquire — maps to the same anatomical address. Both look irrational in behavioral terms. From inside the insula, they are a pain-avoidance system operating exactly as designed, in a domain it was never meant to process.

The prefrontal cortex goes offline

What testosterone and cortisol do gradually, at the hormonal level, acute stress does suddenly, at the neurological level.

The cascade begins with the HPA axis — the hypothalamic-pituitary-adrenal stress response that floods the bloodstream with cortisol within minutes of a perceived threat. In the prefrontal cortex, cortisol binds to glucocorticoid receptors and weakens synaptic connections, reducing the PFC’s capacity to override automatic, subcortical responses. McEwen and Morrison’s 2013 review in Neuron documented the mechanism: the PFC is among the brain structures most sensitive to stress hormones, and the structural effects under acute cortisol elevation are rapid. The deliberative system doesn’t slow gradually. Under sufficient load, it goes offline.

Lars Schwabe and Oliver Wolf showed, beginning around 2009, that stress induces a shift from goal-directed behavior — flexible, outcome-sensitive, capable of updating when circumstances change — to habit-based behavior, driven by learned stimulus-response associations that execute independently of whether the current outcome is desirable. Subsequent replication attempts have produced mixed results. The shift appears to depend on cortisol reactivity, is clearest in individuals showing the strongest hormonal response to stress, and varies with baseline working memory capacity. Schwabe and Wolf’s hypothesis is well-supported, not settled. What the data consistently point toward is that under peak cortisol load, the brain runs its habits. The question is whose habits are good enough.

This is the right framing for what happens on a naval bridge when a contact classification turns ambiguous, in a firefight when a vehicle enters a kill zone at an unexpected angle, in a cockpit when instruments contradict each other. The USS Vincennes, a US Navy Aegis cruiser operating in the Persian Gulf on July 3, 1988, shot down Iran Air Flight 655 — a scheduled commercial Airbus A300 on a routine civilian flight — killing all 290 people aboard. The official investigation found that stress, task fixation, and unconscious distortion of data may have played a major role in the incident. The crew had classified the ascending commercial aircraft as a descending Iranian F-14, and continued processing incoming sensor data against that threat template despite signals inconsistent with it. The cortisol-mediated PFC suppression framework is consistent with this pattern: the shift from deliberate evaluation to pattern-matching under acute stress is precisely the mechanism the stress-cognition literature describes. Whether the specific neurobiological pathway was at work in the way the research models it has not been established in peer-reviewed military psychology literature — the incident is primarily analyzed through cognitive bias and naturalistic decision-making frameworks. The mechanism fits. That’s not the same thing as proof.

What the research does establish — through Starcke and Brand’s 2012 review in Neuroscience & Biobehavioral Reviews and the broader literature on high-stakes performance under stress — is that what separates better from worse decision-making under acute cortisol load is not the ability to reason rationally. Rational deliberation is compromised equally across individuals under sufficient stress. What differs is the quality of the habits the stress-driven shift defaults to. Repetitive scenario training works because it installs better material in the habit system before the acute stress event. Not so the brain can overcome cortisol when it comes. So that when cortisol takes deliberation offline, the system it hands control to was built deliberately, when the PFC was still functional enough to build it.

The pharmacological question

Beta-blockers — propranolol and related compounds — reduce the physiological arousal associated with acute stress by blocking adrenergic receptors, dampening the physical escalation without eliminating alertness. They're studied in non-combat high-performance contexts: musicians, surgeons, air traffic controllers. The question of whether pharmacological management of the acute stress response could improve decision-making under combat conditions is active research, without clean answers and with significant ethical complications. A soldier whose cortisol response is chemically moderated may make better deliberative decisions at the moment of engagement. The acute stress response exists partly because the environment is genuinely dangerous. Suppressing it pharmacologically is not obviously the same thing as defeating it wisely.

Building for the brain we have

The advice that follows from two decades of converging neuroscience — if your response is to recommend that individuals be more aware of their biases — is neurologically incoherent. You cannot be aware of PFC suppression from inside a suppressed PFC. The cortisol that degrades deliberative capacity also degrades the metacognitive capacity to notice that your deliberation is degraded. “Think carefully” is the worst possible instruction to give someone whose thinking apparatus is currently offline.

The institutional responses that actually work share a structural feature: they compensate for predictable neural failure at the system level, rather than demanding that individuals perform differently under conditions where they biologically cannot.

Some were designed explicitly in response to the science. In Roper v. Simmons (2005), the US Supreme Court ruled capital punishment for juveniles categorically unconstitutional. Justice Kennedy’s majority opinion relied on scientific and sociological research establishing that juveniles characteristically lack maturity and a developed sense of responsibility, are more susceptible to outside pressure and negative influences, and have more transitory, less fixed character than adults — findings the Court drew from psychology and behavioral science without deploying explicit prefrontal-maturation language. Miller v. Alabama (2012) extended the same logic to mandatory life-without-parole sentences for juvenile offenders; Justice Kagan’s majority opinion made the neuroscience explicit, quoting Graham v. Florida’s language that “developments in psychology and brain science continue to show fundamental differences between juvenile and adult minds,” particularly in “parts of the brain involved in behavior control.” These are unambiguous cases of a legal institution updating its operative assumptions in direct response to neurobiological findings. The science changed what the law was willing to treat as culpability.

Others arrived at the same structural logic without the scientific framework. Commander’s Intent doctrine — formalized in US Army doctrine publication ADP 6-0 (current edition July 2019) — requires commanders to articulate the purpose and desired end-state of a mission before it begins, so that subordinates operating under acute stress can act within a pre-committed understanding of the goal rather than trying to deliberate under fire. The Army didn’t design this because someone briefed them on glucocorticoid receptor binding in the lateral PFC. The doctrine’s intellectual roots go back to Auftragstaktik — Prussian mission tactics developed in the nineteenth century from post-Napoleonic lessons about command under battlefield chaos. ADP 6-0 acknowledges those roots explicitly. What neuroscience now clarifies is why the doctrine works: the deliberative burden is offloaded before cortisol can suppress it. The pre-commitment to purpose happens when the PFC is functional. When cortisol takes it offline under fire, the habit system running in its place is executing against a goal that was chosen deliberately, in advance. The institution intuited the structural solution long before the mechanism had a name.

Market-wide circuit breakers follow the same logic. The NYSE and CME framework, rebuilt following the Securities and Exchange Commission’s investigation of the May 6, 2010 Flash Crash — in which automated selling and cascading panic temporarily erased hundreds of billions in market value within minutes — triggers mandatory trading halts when S&P 500 movements exceed thresholds of 7%, 13%, and 20% from the prior day’s close. These mechanisms are distinct from pre-existing single-stock breakers; the market-wide system was effectively redesigned from different premises after the event that exposed its predecessor’s inadequacy. Structurally, circuit breakers remove the decision from individual traders at precisely the moment when those traders are most likely operating under peak cortisol and degraded prefrontal function. Like Commander’s Intent, they weren’t designed because market architects read the stress-cognition literature. They were designed because uninterrupted markets under extreme stress had failed badly enough that a structural intervention could no longer be deferred.

What all three share: they compensate for predictable neural failure at the system level. They don’t ask individuals to overcome biology. They design around it.

Which makes the following question worth sitting with: these responses exist, they work where implemented, and the neuroscience explaining their logic has been accumulating across multiple disciplines for decades. Why do they remain exceptional rather than standard? The answer the evidence implies is not flattering. Individual accountability is a simpler, more satisfying framework. Blame requires a person. The neuroscience keeps pointing at a process. And institutions built around the assignment of individual blame don’t update easily when the evidence starts locating the failure somewhere else.

Let that gap speak for itself.

That sixteen-year-old in the scanner wasn’t making an error in judgment. His reward circuitry was doing exactly what it was built to do — amplifying social salience, priming the decision environment for an audience — in a context it was never calibrated for. The fMRI captured a system operating precisely within its design specifications. The same architecture, differently dressed, is visible in the trader whose anterior insula fires before he decides whether to close a losing position, and in the naval officer whose acute stress response floods cortisol through the prefrontal connections responsible for updating a threat classification in real time.

We know enough of the mechanism. Specific circuits, documented failure conditions — the anatomical substrate of decisions that look like bad judgment. We know enough to build around it. The question is what it says about every institution still built on the assumption that better-informed individuals, properly warned, should be able to think their way past it.

The world built around it was not designed with that in mind.

Gen AI 免責事項

このページの一部のコンテンツは、ジェネレーティブAIの助けを借りて生成・編集されたものです。.

メディア

A sign on the lighthouse structure in the Milwaukee harbor – Wikipedia

主な情報源と参考文献

Chein, J., Albert, D., O’Brien, L., Uckert, K., & Steinberg, L. (2011). Peers increase adolescent risk taking by enhancing activity in the brain’s reward circuitry. Developmental Science, 14(2), F1–F10.

Steinberg, L. (2008). A social neuroscience perspective on adolescent risk-taking. Developmental Review, 28, 78–106.

Steinberg, L. (2010). A dual systems model of adolescent risk-taking. Developmental Psychobiology, 52(3), 216–224.

Casey, B. J., Getz, S., & Galvan, A. (2008). The adolescent brain. Developmental Review, 28, 62–77.

Gardner, M., & Steinberg, L. (2005). Peer influence on risk taking, risk preference, and risky decision making in adolescence and adulthood: An experimental study. Developmental Psychology, 41(3), 625–635.

Coates, J. M., & Herbert, J. (2008). Endogenous steroids and financial risk taking on a London trading floor. Proceedings of the National Academy of Sciences USA, 105(16), 6167–6172.

Oyegbile, T. O., & Marler, C. A. (2005). Winning fights elevates testosterone levels in California mice and enhances future ability to win fights. Hormones and Behavior, 48(3), 259–267.

Cueva, C., Roberts, R. E., Spencer, T., Rani, N., Tempest, M., Tobler, P. N., Herbert, J., & Rustichini, A. (2015). Cortisol and testosterone increase financial risk taking and may destabilize markets. Scientific Reports, 5, 11206.

Coates, J. (2012). The Hour Between Dog and Wolf: Risk-Taking, Gut Feelings and the Biology of Boom and Bust. Penguin Press.

Knutson, B., & Bossaerts, P. (2007). Neural antecedents of financial decisions. Journal of Neuroscience, 27(31), 8174–8177.

Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263–291.

Thaler, R. H. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior and Organization, 1(1), 39–60.

McEwen, B. S., & Morrison, J. H. (2013). The brain on stress: Vulnerability and plasticity of the prefrontal cortex over the life course. Neuron, 79(1), 16–29.

Schwabe, L., & Wolf, O. T. (2009). Stress prompts habit behavior in humans. Journal of Neuroscience, 29(22), 7191–7198.

Starcke, K., & Brand, M. (2012). Decision making under stress: A selective review. Neuroscience & Biobehavioral Reviews, 36(4), 1228–1248.

Fogarty, W. M. (1988). Formal investigation into the circumstances surrounding the downing of Iran Air Flight 655 on 3 July 1988. Department of Defense, United States Navy.

Roper v. Simmons, 543 U.S. 551 (2005). United States Supreme Court.

Graham v. Florida, 560 U.S. 48 (2010). United States Supreme Court.

Miller v. Alabama, 567 U.S. 460 (2012). United States Supreme Court.

U.S. Army (2019). ADP 6-0: Mission Command — Command and Control of Army Forces. Headquarters, Department of the Army. Washington, DC.

U.S. Commodity Futures Trading Commission & U.S. Securities and Exchange Commission (2010). Findings Regarding the Market Events of May 6, 2010. Report of the Staffs of the CFTC and SEC to the Joint Advisory Committee on Emerging Regulatory Issues.