Interoceptive Ability Predicts Survival on a London Trading Floor

Traders have enhanced interoceptive ability compared to non-trading individuals

As a first step in assessing the importance of interoception in financial risk taking, we tested simply if traders scored higher on heartbeat detection tasks than non-trading controls. We compared the average score on the heartbeat counting task for the traders with the average score from our control group. We found that the traders had significantly higher scores, indicating greater interoceptive accuracy, than the controls, with a mean score of 78.2 for traders and 66.9 for controls (p = 0.011, N = 66, Fig. 1).

Interoceptive ability predicts trader profitability

To test the hypothesis that interoceptive accuracy predicts trading performance, we collected data on each trader’s average daily profit and loss (P&L) over the previous year. One of the traders was acting as trading manager, so did not have his own P&L; another trader had a P&L more than 5 standard deviations higher than the mean of other traders, so even with data transformations he distorted the distribution. Both traders were therefore omitted from analyses that used ‘raw’ P&L. We found that the traders’ scores on the heartbeat counting task predicted their P&L (coeff = 2.61, R² = 0.27, p = 0.007, N = 16). To further validate this association, we also rank-ordered the P&L, which permitted the inclusion of the trader with the outlying P&L. Again we found that heartbeat detection score predicted relative trading performance (coeff = 17.84, R² = 0.22, p = 0.01, N = 17, Fig. 2).

Interoceptive ability predicts survival in the financial markets

If heartbeat detection score predicts traders’ P&L, does it also predict how long traders survive in the financial markets? To answer this question we plotted heartbeat detection scores against years of experience in the financial markets and found that a trader’s heartbeat counting score predicted the number of years he had survived as a trader (coeff = 21.64, R² = 0.344, p = 0.001, N = 18, Fig. 3).

Solid red line is regression plot. A regression model with conditional mean and conditional standard deviation (std) estimated jointly is used to assess the significance of changes in the heartbeat detection mean and std over years of trading. Light dashed horizontal lines are +/−1 std. Vertical dotted lines show distributions of residuals for each bucket of trading experience. These distributions show a declining variance of heartbeat detection as years of experience increase.

Does this result mean the markets select for traders with greater interoceptive ability? To pursue this analysis we reasoned as follows: i) If firms know nothing about interoception (and the hosting firm of our study did not) they will not hire traders on the basis of heartbeat detection scores. We should find, therefore, that beginner traders and the non-trading population have mean heartbeat detection scores and a standard deviation of scores that do not differ significantly. ii) If the market selects for traders with good gut feelings (i.e. high heartbeat detection scores), then what we should find is that as the traders’ careers progress and the market selects for traders with better gut feelings, those with low heartbeat detection scores will be eliminated from the markets. Average heartbeat detection score for the traders will rise, while the standard deviation of scores will fall as the lower end of the range drops out.

To test this possibility, we partitioned the group according to experience: junior traders (1–4 yrs experience), mid (5–8 yrs), and senior (>8 yrs) and then calculated the mean and standard deviation of detection scores for each bucket (Table 1). We found that beginner traders did not differ significantly from controls in the mean (p = 0.852, N = 53) nor standard deviation (p = 0.614, N = 53) of their heartbeat detection score. Over time, however, the traders’ mean heartbeat detection score did indeed rise, from 68.7 for beginners to 85.3 for experienced traders. This latter level differed significantly from controls (T test, p = 0.02, N = 56). In addition, the standard deviation decreased, from 16 for beginners to 8.6 for experienced traders, and this latter standard deviation also differed significantly from controls (F test, p = 0.018, N = 56) (Table 1). This result is represented graphically in Fig. 3, where the variance of residuals declines with the number of years of trading (coeff = −0.063, p = 0.042, N = 18), the pattern of these residuals forming a wedge shape rather than the customary parallelogram.

Synchronization task demonstrates a bi-modal distribution

We found, as have others, that the data from the heart beat synchronization task (asking if a series of tones was synchronised to heartbeat) produced a bimodal distribution: a few subjects excelled, with scores around the 90 level, while the remainder had scores tightly clustered around the 60 level^30,31. It is thought that the synchronisation task produces this bimodal distribution because it involves a more complex and difficult integration of interoceptive (heartbeat) signals with exteroceptive (audible tone) signals and crossmodal timing. This task may help identify high scoring individuals from relatively large populations. But in our sample we found no correlations between the traders’ scores on the synchronization task and their scores on the counting task; nor between the synchronization task and any of our dependent variables.

Interoceptive accuracy but not confidence informs trading performance

We next analysed the traders’ level of confidence in their estimates of their own heartbeats, as recorded on visual analogue scales during the heartbeat counting task^29,32. We found no significant correlations between confidence and heartbeat detection accuracy (coeff = 0.17, p = 0.51, N = 17); nor between confidence and P&L (coeff = −0.01, p = 0.97, N = 17); nor between confidence and years of survival (coeff = −0.16, p = 0.54, N = 17).

This disjunct between the traders’ heartbeat detection accuracy and their confidence in their accuracy may seem contradictory but, as with many physiological measures, objective performance and subjective appraisal often diverge. In the case of interoceptive ability, it is conceivable that a self-conscious awareness of interoceptive signals impairs the signal’s utility because people may dismiss the signals as ‘merely’ physical and distracting. Alternatively, self-consciousness may impair risk taking in much the same way that focusing self-consciously on, say, your tennis stroke can impair your game. Related to this point, it is noteworthy that the control group in our study included 16 medical students who, despite their more advanced understanding of the cardiovascular system, performed on average worse than the traders on the counting task (65.9 versus 78.2), although this difference only approached significance (p = 0.056, N = 34).

Physiological markers predict heartbeat detection score

If heartbeat detection ability predicts profitability and survival in the markets, can we offer any insights into what predicts heartbeat detection ability itself? To address this question we examined physiological data that was available for a subset of traders (N = 14). In a multiple regression model predicting heartbeat detection score, we found that a lower body mass index (BMI), a lower heart rate (HR), and a lower root mean square of successive differences (RMSSD) in R-R heartbeat intervals (a measure of heart rate variability) predicted higher heartbeat detection scores: R² = 63.7%, p = 0.006; HR (coef = −0.01, p = 0.012); BMI (coef = −0.11, p = 0.003); RMSSD (coef = −0.006, p = 0.009)^33,34,35,36. Previous research has found that heartbeat detection scores decline with age³⁷, but we found that age was not a significant predictor.

Given how strongly the physiology predicted heart beat detection scores (HBD), the question then arose: does the physiology itself predict the P&L and years of survival, independently of heart beat detection? To answer this question we combined all available independent variables (HBD, HR, RMSSD, BMI) in a multiple regression model and then employed Akaike Information Criteria to select the variables that best explain P&L and years of survival. The Akaike Information Criterion measures the quality of a model as a trade-off between its goodness of fit (including R² and p value) and its complexity, i.e., number of variables.

We found that heart beat detection and physiology jointly predicted P&L and years of survival. Specifically, P&L was best predicted by a model which included heart beat detection and a single physiological variable, heart rate variability: R² = 45.4%, p = 0.008; HBD (coeff = 16.08, p = 0.011); RMSSD (coeff = −0.09, p = 0.019). Years of survival was predicted by the same two variables: R² = 60.5%, p = 0.001; HBD (coeff = 15.5, p = 0.004); RMSSD (coeff = −0.12, p = 0.001). The other variables that predicted heart beat detection scores (HR, BMI) did not help predict either P&L nor years of survival, and may therefore have their main effects on these dependent variables through their intermediate effects on interoception. The results of this analysis suggest that the physiological variables we examined affect P&L and survival partly through their effects on interoception and partly through their direct effects.