Analysing the relations between FTSE100 and self-reported measures of emotional well-being we confirmed that market ups (higher FTSE100 scores) were associated with higher scores of “happiness” and lower scores in self-reported “negative emotional facets”: irritability, hurt and nervous feelings, anxiety (Fig. 1; Table 1). The identified association also held true for the 5.5-years of the MRI subsample (Supplement Table S2). We further explored non-imaging variables that are associated with mood changes, i.e. alcohol intake (overall intake frequency and a composite score reflecting weekly intake of all alcoholic beverages) and diastolic blood pressure (automatic readings in mmHg measured at rest), and showed that they were also highly correlated with the FTSE100 (Fig. 1A) in that both measures increased when the stock market decreased in value. Several of these effects (relation between stock market and negative emotions, blood pressure or alcohol-intake) were reproduced in the My Connectome data-set consisting of one single subject whose measurements were taken at 81 timepoints during a period or 1,5 years (Fig. 1B).
Table 1
Subjective well-being and FTSE100 scores: 14 years period. βstd - standardized β coefficients, pfdr – p-values corrected for multiple testing with false discovery rate. Subcomponents of negative emotions are binary variables (-), Day/MonthAVG – data averaged by days and months. The analyses leveraged random linear mixed effects framework with subject as a random effect, as a subset (n = 1427) of the study subjects was assessed twice. * – corresponding effect-sizes estimated with mutual information criterion are reported in the supplement (Supplement Table S11).
|
Linear Mixed-Effects
|
Effect-sizes*, Pearson r (95% C.I.)
|
|
βstd
|
T(df)
|
pfdr
|
Raw
|
DayAVG
|
MonthAVG
|
NegEm (total)
|
-0.03
|
-24.33(37671)
|
< 0.001
|
-0.034(-0.037,-0.031)
|
-0.396(-0.428,-0.362)
|
-0.591(-0.692,-0.467)
|
Irritability
|
-0.01
|
-5.86(37671)
|
< 0.001
|
-
|
-0.117(-0.155,-0.078)
|
-0.266(-0.418,-0.099)
|
Sensitivity/hurt
|
-0.04
|
-24.75(37671)
|
< 0.001
|
-
|
-0.379(-0.412,-0.345)
|
-0.516(-0.632,-0.378)
|
Nervous
|
-0.01
|
-8.51(37671)
|
< 0.001
|
-
|
-0.264(-0.300,-0.228)
|
-0.508(-0.625,-0.368)
|
Worrier/anxious
|
-0.02
|
-10.9(37671)
|
< 0.001
|
-
|
-0.287(-0.322,-0.25)
|
-0.444(-0.572,-0.294)
|
Happiness
|
0.04
|
19.81(15633)
|
< 0.001
|
0.052(0.047,0.056)
|
0.247(0.204,0.288)
|
0.556(0.406,0.677)
|
We then tested and confirmed our main hypothesis by showing that FTSE100 oscillations exhibited significant associations with the morphometry of the affective brain circuits. The most notable result was that bilateral amygdala, involved in threat detection and anxiety processing 16–20, showed a negative relation with the UK economic performance (Fig. 2A, and Table 2, whole-brain analysis revealing that the effects are not limited only by the preregistered regions is reported in Supplement Fig. S2). Importantly, this effect was replicated in an independent set of 424 individuals from the PPMI database, an independent clinical study targeting the US population (www.ppmi-info.org), and conceptually also in “My Connectome” single-subject longitudinal study 32. In “My Connectome”, structural data was not publicly available, however, using BOLD-signal variability 33 in the amygdala as a proxy biological measure demonstrated that our results also generalise to functional characteristics of the fear network (Fig. 2B). It is worth noting, however, that unlike the main results, detrending the Dow Jones index in these two (PPMI and MyConnectome) datasets reduced effect-sizes without reversing the direction of the associations (Supplement Fig. S13).
Splitting the study timeline into 6 equal periods (11 months each) we showed that the correlations are strongest during and following phase transition events, i.e. when the change and variability of stock market dynamics is most pronounced (Supplement Fig S8).
Table 2
Associations between FTSE100 and structural characteristics of the fear network: cortical and subcortical volumes. Day/MonthAVG – data averaged over days and months. Intracranial volume (ICV) was selected as a reference measure, which was not expected to exhibit significant associations with global stock market behaviour. βstd - standardized β coefficients, pfdr – p-values corrected for multiple testing with false discovery rate. The analyses leveraged random linear mixed effects framework with subject as a random effect, as a subset (n = 1427) of the study subjects was scanned twice. * – corresponding effect sizes estimated with the mutual information criterion are reported in the supplement (Supplement Table S11)
|
Linear Mixed-Effects
|
Effect-sizes*, Pearson r (95% C.I.)
|
Region
|
βstd
|
T883
|
pfdr
|
Raw, n = 30,775
|
DayAVG, n = 1299
|
MonthAVG, n = 66
|
L Amygdala
|
-0.054
|
-9.51
|
< 0.001
|
-0.055(-0.066,-0.043)
|
-0.253(-0.304,-0.202)
|
-0.615(-0.746,-0.439)
|
R Amygdala
|
-0.062
|
-10.91
|
< 0.001
|
-0.063(-0.074,-0.052)
|
-0.282(-0.332,-0.231)
|
-0.644(-0.767,-0.477)
|
L Accumbens
|
-0.054
|
-9.54
|
< 0.001
|
-0.055(-0.066,-0.044)
|
-0.232(-0.283,-0.18)
|
-0.623(-0.752,-0.449)
|
R Accumbens
|
-0.062
|
-10.89
|
< 0.001
|
-0.064(-0.075,-0.052)
|
-0.259(-0.309,-0.207)
|
-0.662(-0.779,-0.5)
|
L LOFC
|
-0.026
|
-4.68
|
< 0.001
|
-0.031(-0.042,-0.02)
|
-0.141(-0.193,-0.087)
|
-0.443(-0.619,-0.225)
|
R LOFC
|
-0.019
|
-3.49
|
0.001
|
-0.023(-0.034,-0.012)
|
-0.082(-0.136,-0.028)
|
-0.292(-0.499,-0.054)
|
L Insula
|
0.037
|
6.62
|
< 0.001
|
0.042(0.031,0.053)
|
0.21(0.157,0.261)
|
0.494(0.286,0.657)
|
R Insula
|
0.032
|
5.86
|
< 0.001
|
0.037(0.026,0.048)
|
0.187(0.134,0.239)
|
0.413(0.19,0.595)
|
L Subcallosal
|
0.018
|
3.18
|
0.002
|
0.02(0.009,0.032)
|
0.134(0.08,0.187)
|
0.322(0.087,0.524)
|
R Subcallosal
|
0.015
|
2.74
|
0.008
|
0.019(0.007,0.03)
|
0.129(0.075,0.182)
|
0.305(0.068,0.509)
|
L Anterior Cingulate
|
0.025
|
4.67
|
< 0.001
|
0.035(0.024,0.046)
|
0.178(0.124,0.23)
|
0.408(0.184,0.591)
|
R Anterior Cingulate
|
0.024
|
4.44
|
< 0.001
|
0.033(0.022,0.044)
|
0.169(0.116,0.222)
|
0.428(0.207,0.607)
|
L Paracingulate
|
0.004
|
0.8
|
0.426
|
0.001(-0.01,0.013)
|
0.044(-0.01,0.098)
|
0.106(-0.14,0.339)
|
R Paracingulate
|
0.005
|
0.83
|
0.426
|
0.003(-0.008,0.014)
|
0.052(-0.003,0.106)
|
0.122(-0.124,0.353)
|
Intracranial Volume
|
0.004
|
0.87
|
0.426
|
-0.009(-0.02,0.002)
|
-0.016(-0.07,0.038)
|
-0.098(-0.332,0.147)
|
Similar findings were observed for nucleus accumbens and lateral orbitofrontal cortex (lOFC) that also increased in volume when market decreased (Fig. 3, A and B). While nucleus accumbens is mostly known for being involved in reward anticipation, it is equally important for processing losses 13,14. lOFC has been suggested to be involved in processing expectations within the emotional domain 34–36, including losses and rewards 37,38. Further supporting this, a significant interaction (β = -0.01, t776 = -2.87, p = 0.004, pfdr = 0.05) between FTSE100 and income index was found on the right lOFC volume (Supplement Table S5). Post-hoc analyses revealed the highest effects in individuals with the lowest and highest income, suggesting that right lOFC of those subjects is particularly sensitive to the capital market swings. Insula and anterior cingulate showed the opposite effect, i.e. the size correlated positively with the market (Fig. 3, A and C).
All regions mentioned above, except for lOFC, are involved in both positive and negative emotional processing 13–15. Out of those regions, volume of the subcortical nuclei (amygdala and nucleus accumbens) correlated negatively with the stock market. In contrast, the cortical regions (insula and anterior cingulate) seem exhibit a positive relation with the stock market moves. The magnitude of the identified effects varied depending on time scale with median Pearson correlation |r| = 0.033 (0.001–0.064) for the raw data, |r| = 0.169 (0.017–0.282) for the day-averaged measures, and |r| = 0.492 (0.09–0.73) when brain and market data were averaged over months (Table 2). Importantly, all of the reported associations changed very little after detrending the FTSE100 time-series. Deconvolving FTSE100 time-series into low- and high-frequency domains using fast Fourier transform, showed that low-frequency oscillations mostly drive the effect, although, a similar pattern of associations was observed for the high frequency band (Supplement Fig. S3, Table S6).
We amended the preregistered protocol by adding additional possible confounding variables to confirm that the main results are robust and withstand correction for age, sex, presence of psychiatric diagnoses, seasonal effects (months) and intracranial volume (Supplement Table S3), as well as mixed generalized additive modelling 29 conducted under various assumptions for autocorrelation structure (Supplement Table S12). Moreover, robustness of our findings was also confirmed in the specification curve analysis 27 that showed stability of the effects with respect to different model specification strategies (Supplement Figure S10, S11).
When considering the indexes of the UK’s fifteen top trading partners 39, a similar pattern of associations to the one for FTSE100 was observed for the equivalent local European indexes (e.g. German GDAXI, Dutch АЕХ, French FCHI) but was of smaller magnitude (Fig. 4). The associations further declined or had different directions for markets that were more distant in a socioeconomic dimension (as also reflected in a weaker correlation with FTSE100), including the reference Shanghai Composite Index (SSEC). Importantly, the results also withstood correction for these indexes (Supplement Table S4), which implies that the local economic performance captured by the FTSE100 exhibits a specific association with the characteristics of the scanned UK population.
Regarding causality, the most widely accepted hypothesis states that population mood and well-being are impacted by market via effects on the socioeconomic environment 4,6,26. These effects, heavily reinforced by media, represent an index for variables like fluctuations in house prices 40 and unemployment 41. They can also be considered as threat signals that subsequently impact brains and emotional states of the population 10. Another related hypothesis from socionomics is currently growing in popularity. It puts forward the idea of “social mood” as a herding-driven emergent state that originates from population dynamics and subsequently drives global processes, including economic crises, wars, art and fashion 1,2. According to this hypothesis, social mood is an inherently hidden state of the society. It is related (but not identical) to the mood of individuals that such a group consists of. This hypothesis is conceptually supported by the data acquired in small-scale experimental studies demonstrating involvement of reward and fear circuits in future financial decisions 42–44. Of importance for the present discussion, this hypothesis considers stock market dynamics as a valuable “metric stick” of the social mood and global societal dynamics 1.
To begin to further investigate these relationships, we evaluated associations with time-lagged Pearson correlation. We identified that brain volumes correlate higher with earlier market prices. The correlation remains significant for approximately one year and then gradually decays (Fig. 5). While an autocorrelation, as expected, is present in the stock market time-series 45 (Supplement Table S7), the fact that earlier economic data peaks with the brain volume implies that the market events may be antecedent to the brain volume fluctuations, offering initial evidence that the market “impacts” the brain, mood, and well-being. The same analyses were carried out on the monthly scale yielding similar results (Supplement Fig. S4) and also for the mood data with the FTSE100, although no clear antecedent relationship could be drawn for the latter (Supplement Fig. S5).
Despite the fact that Toda-Yamamoto implementation of Granger Causality tests specifically designed for serially correlated data 31 provided somewhat stronger support in favour of a causal link “Market impacts Population Brain/Mood” (Supplement Tables S8 and S9), a caution is still advised when interpreting these analyses due to scale-free properties of the investigated time-series (Fig. 6). To illustrate this point, we first show the absence of any significant effects after shuffling the dates (Supplement Table S10, column 1), but appearance of residual associations for the time-shifted data (Supplement Table S10, column 2, also seen on Fig. 5). Importantly, simulating stock market data with 1/f noise is capable of producing effect-sizes of similar magnitude (Supplement Table S10, column 3), pooled effect of which, however, converges to zero due to inconsistency of directions in the estimated associations (Supplement Fig. S12), and, unlike the main results, also disappear after adjusting for other stock market indexes (Supplement Table S10, column 4), confirming that the main effect is not driven by a randomly-seeded 1/f noise. Moreover, we demonstrated that the magnitude of the brain-market links (measured as median squared root correlations) is related to economic and sociocultural ties of the UK to other countries 28,39 (Supplement Fig S6 and S7) and that no other global candidate metrics with 1/f properties (UK seismic activity and mortality rates) exhibit an equivalent level of specificity with respect to the investigated variables (Supplement Fig S9).
Due to self-similarity properties identified in the data (Fig. 6, right panel), we decided to conduct a follow-up series of noise simulation experiments. Simulating brain data with uniform and gaussian noise failed to induce the afore-mentioned correlations with FTSE100, but, as expected, they were more likely to be discovered for the brain data simulated with 1/f noise (Fig. 6, left panel).
Therefore, it appears so that scale-free properties are observed at different levels of population dynamics, which is reflected in fluctuations of stock markets, mood and brains. To confirm that the effects still hold after accounting for scale-free noise, we repeated the simulations of brain data with 1/f noise and matched it with the stock markets of the UK’s 15 top trading partners. We then subtracted the yielded Pearson correlations from the real ones (prior to calculating the medians) and, as expected, the effect sizes only became larger (Supplement Fig S7 B). Moreover, a negative association was also identified for a number of sociocultural distances of the UK from 17 countries using data from Liu et al, 2018 28 (Supplement Fig S7). All of the above supported Casti’s hypothesis of stock markets as a useful metric stick for global societal dynamics 1.