This feature encouraged subjects to form an estimate about the mean and variance of the INCB024360 ic50 individual outcomes and continually update their assumption about the correlation strength.

Subjects participated in three consecutive experimental blocks, each corresponding to a 21 min long session in an fMRI scanner (Siemens Trio 3T). They were instructed that the correlation would probabilistically change over the course of the study but were not given further details about specific parameters used. We also told subjects that the mean and variance of the two resources would remain constant over one block of the experiment, a simplification to an otherwise quite complex task that enabled subjects to perform well within the settings of this experiment. As our goal was to assess covariance learning (in contrast to learning the values and risk) this did not adversely impact on any mechanism we wanted to observe. However, mean and variance values were different for each block. To give subjects the

opportunity to learn these basic statistical parameters (mean and variance) before making portfolio choices, we presented them with a 20-trial observation phase at the beginning of each session. In this phase, which immediately preceded the start of fMRI data acquisition, subjects only observed the individual outcomes of the two resources and did not make any choices. There was no change in the GPCR Compound Library order ground truth correlation during this phase. Data from pilot studies and Acesulfame Potassium model simulations confirmed that 20 observations of a time series were sufficient to form an estimate of its mean and variance. The observation phase was followed by 84 choice trials, consisting of a 5 s choice period and a 3 s outcome period, separated by a blank gray screen of 1–6 s duration (uniform distribution). The

intertrial-interval was also 1–6 s (Figure 1A). The portfolio weights (wsun, wwind) indicate how much of a fraction the portfolio contains from both resources rsun and rwind (portfolio outcome value Vp = wsun∗rsun + wwind∗rwind). Subjects were allowed to set the portfolio weight wsun within a range between −1 and 2. Setting negative weights allowed subjects to trade-in a fraction of the trials output from one resource in exchange for multiplying the other output by the same fraction. This concept echoes the possibility of short selling in financial markets and is important for this task as it permits risk minimizing for positively correlated resources (see the section on variance minimizing strategies in the Supplemental Information for further details). The constraint that both weights always add up to 1 automatically determined the weight of the other resource (wwind = 1 − wsun). A horizontal line on the choice screen represented the slider during the choice period and icons of a solar and wind plant on both ends indicated which resources were mixed in the portfolio.