(a) Testosterone and risk-taking
Testosterone mediates sexual behaviour as well as competitive encounters, so there are prima facie
reasons for believing it could also affect financial risk-taking. Research into how it may do so is, however, in its infancy. Much of the work on the cognitive and behavioural effects of androgens has instead studied humans taking anabolic steroids, studies that are pharmacological rather physiological because the steroids are taken in supra-physiological doses (Kashkin & Kleber 1989
); or the work has studied animal behaviour, thus leaving open the question of the results’ applicability to humans (Sapolsky 1997
). The animal studies, besides those examining sexual behaviour, have focused largely on the effects of testosterone on mating–guarding and territorial aggression, and on competitions for rank within a social hierarchy. This research has been elegantly synthesized by the biologist John Wingfield in his highly influential challenge hypothesis.
According to the challenge hypothesis, testosterone in males rises to a minimum level required for sexual behaviour; it will continue to rise beyond this level only when males are confronted with an intruder or a social challenge, the increased testosterone promoting aggressive behaviour (Wingfield et al. 1990
). The insights gained from the challenge hypothesis, and from animal hormone studies more generally, have been applied to human behaviour (Archer 2006
), but often with questionable success. Many studies, for example, could not determine whether testosterone caused aggression or the other way round; others found testosterone levels were poor predictors of who subsequently became aggressive (Sapolsky 1997
; Monaghan & Glickman 2001
); still others did not distinguish between aggressive and non-aggressive risk-taking (Vermeersch et al. 2008
). One problem with these studies stems from the fact that in humans, as in some non-human primates, higher cognitive functions refract the effects of testosterone, effects which in smaller brained animals are more deterministic. Furthermore, the dependent variables in these studies, such as aggression, dominance, or status seeking, often cannot be defined or measured in humans with any objectivity, leading to marginally significant experimental results and contradictory findings between papers (Archer et al. 1998
Studies of steroids and financial risk-taking promise to overcome many of these difficulties. To begin with, financial variables, such as profit, variance of returns, volatility of the market, can be defined objectively and measured precisely. Furthermore, the competitive behaviour Wingfield and his colleagues observed in animals may manifest itself in humans, not so much in aggressive encounters as in competitive economic behaviour. Through its known effects on dopamine transmission in the nucleus accumbens, testosterone may well have its most powerful effects in humans by shifting their utility functions, state of confidence or financial risk preferences.
We began testing this hypothesis by setting up a series of experiments on a trading floor in the City of London (Coates & Herbert 2008
). We chose to study professional traders because real risk-taking, with meaningful consequences, seemed most likely to trigger large endocrine reactions. Our hypothesis and predictions were based on the challenge hypothesis as well as a closely related model, the winner effect (see below). Biologists working with these models have noticed that two males entering a fight or contest experience androgenic priming in the form of elevated testosterone levels. Moreover, the winning male emerges with even greater levels of testosterone, the loser with lower ones. The orders of magnitude of these hormone swings can be large: Monaghan & Glickman (2001)
report that in a competition for rank among recently introduced rhesus monkeys, the winning male emerged with a 10-fold increase in testosterone, while the loser experienced a drop to 10 per cent of baseline levels within 24 h, and these new levels for both winner and loser persisted for several weeks. This reaction may make sense from an evolutionary point of view: in the wild, the loser of a fight is encouraged to retire from the field and nurse his wounds while the winner prepares for new challenges to his recently acquired rank.
A similar result has been found in experiments with humans (Gladue et al. 1989
). Athletes, for example, experience the same androgenic priming before a sporting contest, and a further increase in testosterone after a win. This experiment has been repeated for a number of different events, including tennis (Booth et al. 1989
) and wrestling (Elias 1981
), as well as less physical contests such as chess (Mazur et al. 1992
). It has also been found that the rising and falling levels of an athlete's testosterone can be mimicked by fans: Bernhardt et al. (1998)
took testosterone samples from fans during a World Cup match in which Brazil defeated Italy. Both sets of fans went into the game with elevated testosterone, but afterwards the Brazilian fans’ testosterone rose while the Italians’ fell.
The role of these elevated testosterone levels is further explored in an animal model known as the ‘winner effect’. In this model, winning in an agonistic encounter can itself contribute to a later win (Chase et al. 1994
; Oyegbile & Marler 2005
), an effect that is independent of (i) an animal's resource-holding potential (RHP), i.e. the physical resources it can draw on in an all-out fight, (ii) its motivation, i.e. the value of the resource in dispute, or (iii) its aggressiveness (Hurd 2006
). It is not known if the win imparts information to winner and loser about their respective resources (Hsu & Wolf 2001
; Rutte et al. 2006
) or whether it has physiological effects. This latter possibility is suggested by experiments in which elevated testosterone has been found to contribute to further wins (Trainor et al. 2004
; Oyegbile & Marler 2005
). Another possibility not fully considered in the literature is that higher testosterone, through its beneficial effects on the cardiovascular system and muscle mass, may effectively increase an animal's RHP, or, through its effects on confidence and risk-taking, may increase an animal's motivation or aggressiveness (Neat et al. 1998
). Whatever the mechanism, a winner, with heightened testosterone levels, may proceed to the next round of competition with an advantage. This positive feedback loop, in which victory raises testosterone which in turn raises the likelihood of later victories (), may help account for winning and losing streaks in round-robin animal competitions that establish a social hierarchy (Dugatkin & Druen 2004
Schematic representation of a winner effect mediated by testosterone.
We examined the relevance of the challenge hypothesis and winner effect models to the financial markets (Coates & Herbert 2008
) by looking for evidence that traders experience an increase in testosterone when they enjoy an above-average win in the markets. To do so, we sampled steroids from 17 young male traders, taking saliva samples twice a day, at 11.00 and 16.00, over a period of eight consecutive business days. Hormone readings are notoriously noisy owing to the pulsatile nature of their production and release into the blood stream, hence our protocol of repeated sampling to help separate ‘signal’ from ‘noise’. The traders were engaged in high-frequency trading, meaning that they positioned securities, mostly futures contracts in European and US bond and equity markets, in sizes up to £1 billion, but held their positions for a short period of time—several minutes, and sometimes mere seconds. They rarely positioned trades overnight, and they did not let winning or losing positions run for long.
We discovered that these traders did indeed have significantly higher testosterone levels on days when they made an above-average profit. We could not determine from this correlation whether the profits were raising hormone levels or vice versa, but since we took two samples per day, we could examine how morning testosterone levels were related to afternoon profits and losses (P&Ls). To do so, we looked at the days when each trader's 11.00 testosterone levels were above his median level during the study, these days showing testosterone levels a modest 25 per cent higher than on the other days. We found that on days of high morning testosterone, the traders returned an afternoon profit (a) that was almost a full standard deviation higher than on ‘low-testosterone’ days. Interestingly, this relationship was even stronger among experienced traders (b), i.e. those who had traded for longer than 2 years, suggesting that testosterone, at moderate levels, was not having its effect by encouraging overly risky behaviour but was instead optimizing performance, at least with respect to high-frequency trading.
Figure 3. P&L on low- and high-testosterone days. (a) P&L made between 11.00 and 16.00 for 17 traders on days when their testosterone levels were above their median level during the study (‘high T’) and on the rest of the days (‘low (more ...)
The effects of androgens on high-frequency trading were also evident in a second experiment, one that looked at a surrogate marker of pre-natal androgen exposure—the second to fourth digit (finger length) ratio (2D : 4D) (Coates et al. 2009
). As mentioned above, there are two distinct periods and types of hormone action—organizational effects of pre-natal steroids on the foetus and activational effects of circulating steroid on the adult. Androgens surge between the ninth and 18th week of gestation, masculinizing the foetus and exerting developmental changes on the body and brain that are permanent (Cohen-Bendahana et al. 2005
). After the 19th week, androgen production subsides, spikes again briefly in the neonate and then drops back to low levels until the onset of puberty. At puberty, androgen production increases, activating the circuits created earlier in life by pre-natal hormone exposure. According to the organizational/activational model of hormone action (Phoenix et al. 1959
), the sensitivity of adults to changes in circulating testosterone is a function of the amount of pre-natal androgen to which they were exposed (Meaney 1988
; Breedlove & Hampson 2002
Importantly, the amount of pre-natal androgen an individual was exposed to can be estimated because it leaves traces throughout the adult body, traces often measured by paediatricians looking for effects of environmental hormone disruptors on newborn infants. 2D : 4D is the most convenient measure for behavioural studies (McIntyre 2006
). A lower 2D : 4D ratio is thought to indicate higher levels of pre-natal testosterone exposure (Manning et al. 1998
; Brown et al. 2002
). Consistent with this, men on average have lower ratios than women. We sampled 2D : 4D from a total of 44 traders, including 14 from the first study, and found that it predicted both the traders’ P&Ls over a 20-month period and the number of years they had survived in the business. It also predicted, in line with the organizational/activational model, the sensitivity of the trading performance of the original 14 traders to increases in circulating testosterone: the lower the trader's 2D : 4D, the more money he made when his testosterone levels rose.
Pre-natal testosterone appears, therefore, to predict long-term success in high-frequency trading, a style of trading requiring quick physical and cognitive reactions. However, there are grounds for believing that in other types of trading, especially those permitting more time for analysis and a longer holding period, or ones that do not make such physical demands, the correlation may weaken and even reverse sign (Coates et al. 2009
). The market, it appears, selects for biological traits but these traits may vary between market segments.
The two trading floor experiments described here raise troubling questions about the efficient markets hypothesis. If, as this hypothesis assumes, markets are random, then we should not be able to predict relative trading performance by means of biological traits. Yet, our results suggest that higher levels of circulating testosterone predict short-term profitability and higher levels of pre-natal testosterone predict long-term profitability, at least in the segment of the market inhabited by high-frequency traders. The implication seems to be that the markets are not efficient or that they select for traits other than rational expectations (De Bondt & Thaler 1987
; Shiller 2005
; Blume & Easley 2006
This leads us to another important question: how could testosterone exert its effects on profitability? Field studies such as those reported above do not allow us to establish a causal relationship between testosterone and profits, merely a predictive relationship, albeit a strong one. To establish causality, one needs pharmacological manipulation. Some studies administering testosterone esters to eugonadal males have found significant but weak effects on mood and aggressiveness (Bhasin et al. 2001
; O'Connor et al. 2004
), although they were not examining financial tasks. However, converging evidence from other lines of research suggests that androgen may affect confidence and risk preferences. For example, administered testosterone promotes confidence and fearlessness in the face of novelty, a result observed in both animals (Boissy & Bouissou 1994
) and humans (Hermans et al. 2006
). Furthermore, in a between-subjects study of male students playing an investment game, testosterone levels correlated with risk preferences (Apicella et al. 2008
). This study also examined 2D : 4D and risk preferences, finding a significant correlation among Swedish Caucasians but not in a more ethnically heterogeneous population, the difference in results being accounted for by the fact that ethnic population is an important confound for 2D : 4D.
Intriguingly, there is another potential path of causation between testosterone and trading profits. Trading, it is not often appreciated, is a physical activity, a demanding one, so the important effects of testosterone may be physical rather than cognitive. High testosterone levels or increased androgenic effects, for example, can increase vigilance and visuomotor skills such as scanning and speed of reactions (Salminen et al. 2004
; Falter et al. 2006
), qualities that may help traders to spot and trade price discrepancies before others arbitrage them away (Coates et al. 2009
). Elevated testosterone levels have also been found to increase an animal's search persistence (Andrew & Rogers 1972
) and, during search, to focus visual attention while decreasing distraction by irrelevant stimuli (Andrew 1991
). These last traits may be of particular importance in high-frequency trading because this form of trading requires lengthy periods of visuomotor scanning and quick reactions.
An increase in confidence or risk preferences, as found in some studies, would tend to increase a trader's position size; an increase in search persistence the frequency of trading; an increase in reaction times the chances of getting to a trade before others. Given that the traders in our study had a positive expected return, i.e. they usually made money, larger positions or more frequent trades would translate into higher daily profits. However, we cannot at this point say by which route these effects travelled, that is, whether testosterone was having its effect by augmenting the effort, speed, confidence or risk preferences of the traders.
(b) Cortisol and risk-taking
A review of research on cortisol and financial risk-taking is necessarily brief as there is almost no work done on this subject. Van Honk et al. (2003)
looked at the cortisol levels of people playing the Iowa Gambling Task and found that they correlated with risk aversion. In our own studies, we hypothesized that cortisol, as a stress hormone, would increase as traders lost money. This seemed a reasonable assumption, but our experiment did not find evidence to support it, as we observed no relationship between trading losses, even above-average ones, and cortisol levels. However, caution is needed before extrapolating these findings, as the style of trading and the risk management practices on this trading floor prevented traders from losing large sums of money. Had they not done so, or had we sampled in a different setting, for example in an investment bank where traders position interest rate or credit risk for longer periods of time, and had these traders entered a sustained losing streak, it is likely they would have experienced high levels of stress and cortisol.
However, we did note a potentially more interesting finding—that cortisol was rising with uncertainty. Early research on stress and cortisol, especially the pioneering work of Hans Selye, focused on how cortisol production reacts to actual bodily harm. But later research found that the HPA axis can respond more robustly to expected harm and that the size of the response is an increasing function of the uncertainty over timing. For example, an animal receiving a shock at regular intervals or after a warning tone may have normal cortisol levels at the end of an experiment; in contrast, an animal receiving the same quantity of shock will experience rising cortisol levels as the timing of the shocks becomes more and more unpredictable, reaching a maximum when the timing becomes random (Levine et al. 1989
). Animals can have a similarly elevated HPA response when exposed to situations of novelty (Erikson et al. 2003
) or uncontrollability (Swenson & Vogel 1983
; Breier et al. 1987
). Uncertainty, novelty and uncontrollability can perhaps be reduced to a common denominator of uncertainty; all three describe a situation in which an animal finds it increasingly difficult to predict what may happen and what actions will be required. The necessity of being prepared for the unexpected signals to the body, via cortisol, that catabolic metabolism may be needed. As it transpires, ‘uncertainty’, ‘novelty’ and ‘uncontrollability’ aptly describe the financial markets and the environment in which traders find themselves on a daily basis.
To examine the effect of uncertainty on traders’ HPA axes, we looked at the risk faced by each trader, as measured by the variance of his P&L, over the course of the study (Coates & Herbert 2008
). We found a highly significant correlation with cortisol that once again displayed a large effect size. Variance in P&L is a measure of the uncertainty or uncontrollability a trader has just lived through; but we also wanted to measure how uncertain the traders were about upcoming events in the market, such as the release of important economic statistics. To do so, we used the implied volatility of the Bund futures contract (a future on German Government bonds), which was the security most widely traded by the traders in the study. Bond options require for their pricing the market's estimate of the future variance of the underlying asset, so option prices provide an objective measure of the market's collective uncertainty. Here, again we observed a very high and significant correlation between the traders’ daily cortisol levels, averaged from all traders, and the market's uncertainty regarding upcoming market moves. Our results raise the possibility that while testosterone codes for economic return, cortisol codes for risk.
Our experiment represents only the mere beginning of research into the role of cortisol in financial decision-making. To underline our belief in the critical importance of this hormone, we should point out that the cortisol fluctuations we observed were large. In the normal course of a day, cortisol, like testosterone, peaks in the morning and falls over the course of the day. Between our sampling times, cortisol levels would be predicted to fall by approximately 40 per cent, yet in many of our subjects it rose, in some cases by as much as 500 per cent. Similar-sized cortisol fluctuations were also observed between days. What purpose do changes of this magnitude serve? Cortisol, as highlighted above, marshalls glucose for immediate use, and it promotes anticipatory arousal and a focused attention (Erikson et al. 2003
). We speculate therefore that traders, when expecting a market move, would benefit from such an acute increase in cortisol, as it prepares them for the money-making opportunities that increased volatility brings.
(c) Steroids and impaired risk-taking
If market volatility or the variance in the traders’ P&L were to remain high, cortisol levels could also remain elevated for an extended period. Chronically elevated cortisol levels, as we have seen, can have the opposite effect on cognitive performance as acute levels. Cortisol displays an inverted U-shaped dose–response curve, according to which performance on a range of cognitive and behavioural tasks is optimized at moderate levels, while being impaired at lower and higher levels () (Conrad et al. 1999
). As cortisol levels rise past the optimal point on the dose–response curve, they may begin to impair trading performance, specifically by promoting irrational risk aversion. Chronically elevated cortisol levels increase CRH gene transcription in the central nucleus of the amygdala thereby promoting fear (Corodimas et al. 1994
), anxiety (Shephard et al. 2000
; Korte 2001
) and the tendency to find risk where perhaps none exists (Schulkin et al. 1994
; McEwen 1998
). They may also alter the types of memory recalled, causing a person to selectively recall mostly negative precedents (Erikson et al. 2003
). Lastly, chronic stress, as we have seen, downregulates dopamine transporters, receptors and downstream signalling molecules in the nucleus accumbens, and may thereby alter risk-related behaviours. All these effects would tend to decrease a trader's appetite for risk.
Inverted U-shaped dose–response curve relating cortsol levels to cognitive function, such as performance, on a spatial navigation or declarative memory task.
When might conditions of chronic stress occur in the markets? Bear markets and crashes are notable for their extreme levels of volatility, the protracted subprime mortgage crisis being a notable example, with the VIX, an index of implied volatilities on the New York Stock Exchange, rising from 12 per cent before the crisis to a high of 80 per cent 18 months later. It seems likely that cortisol levels among traders threatened for so long with historic levels of uncertainty would have increased and perhaps remained elevated for a prolonged period of time. Under such circumstances, the steroid may have contributed to the extreme levels of risk aversion observed among traders. Indeed, extended periods of uncertainty and uncontrollable stress can promote a condition known as ‘learned helplessness’, in which persons, and animals, lose all belief in their ability to control or influence their environment (Kademian et al. 2005
). Under these circumstances, traders could become price insensitive and fail to respond to lower asset prices or interest rates, thereby rendering monetary policy ineffective. In short, rising cortisol levels among traders and investors may promote risk aversion during a bear market, exaggerating the market's downward move.
Could testosterone work in the opposite direction, encouraging irrational risk-taking during a bull market? This is a difficult question. Moderate levels, as described above, may promote effective risk-taking among animals and high-frequency traders. But higher levels may indeed carry increased costs such as encouraging excessive risk-taking. In studies related to the challenge hypothesis and the winner effect, animal behaviourists have found that the higher a male's testosterone level (either on account of the breeding season, an agonistic encounter or an experimental implant), the more often he fights, the larger the area he patrols or the more often he ventures into the open (Marler & Moore 1988
; Beletsky et al. 1995
). These habits can lead to loss of fat stores (i.e. nutritional reserves), neglect of parenting duties, frequent wounds and increased predation (Dufty 1989
; Wingfield et al. 2001
). High-testosterone males end up paying a stiff price for their risk-taking in the form of a higher rate of mortality.
We do not know if traders can experience rises in endogenous testosterone sufficient to encourage analogous forms of over-confidence and irrational risk-taking. The traders we observed experienced only moderate increases, although one trader, who enjoyed a 5-day winning streak during which he made over twice his daily average P&L, experienced a 75 per cent increase in mean daily testosterone. It is known that cortisol can rise to extreme levels, and for extended periods of time; but research on the costs of high physiological levels of testosterone in humans is rare. Nonetheless, some studies have found that physiological levels of testosterone are indeed correlated with risky behaviour (Booth et al. 1999
), sensation seeking (Daitzman & Zuckerman 1980
) and the size of offers rejected in the Ultimatum Game, rejections often considered as violations of economic rationality (Van den Bergh & Dewitte 2006
; Burnham 2007
). Other studies with users of anabolic steroids, or subjects administered pharmacological doses of testosterone, have found evidence of manic behaviour (Pope & Katz 1988
; Pope et al. 2000
). In one study, researchers administered testosterone to a group of women playing the Iowa Gambling Task (van Honk et al. 2004
) and found that it shifted risk preferences to such an extent that the women switched from playing the low variance and positive expected-return decks of cards to the high variance but negative expected-return decks. A similar result was found in a physiological study in which the performance of young males on the Iowa Gambling Task was negatively correlated with their testosterone levels (Reavis & Overman 2001
). These study results suggest that elevated levels of testosterone could at some point begin to impair rational financial decision-making.