From the NARPS data descriptor paper: https://www.nature.com/articles/s41597-019-0113-7

We used a general linear model (GLM) with three regressors, all of them with the onsets of all trials and the mean RT of the condition (equal indifference/equal range) as duration. The first regressor was without modulation and used to analyze task versus baseline fMRI activity. The other two modelled the demeaned gains and the demeaned losses. These regressors were convolved with the canonical double-gamma hemodynamic response function (HRF), and their temporal derivatives were added to the model. In addition, we added the following nuisance regressors as confounds: standard deviation of DVARS, six aCompCor regressors41, framewise displacement and 24 motion regressors (“Friston24”: six motion parameters - translation and rotation in three directions, the square of the six motion parameters and their temporal derivatives).
We further modeled out volumes with extensive motion (i.e. scrubbing) by adding a single time-point nuisance regressor for each volume with framewise displacement value greater than 0.9 (an arbitrary threshold meant to serve as a relatively high threshold for motion exclusion, consistent with Siegel et al.49). One participant (sub-030) was excluded from analysis based on extensive motion, with more than 100 scrubbed volumes for each run.