Using a unique field experiment from Canada, we estimate individual preference over risk and time and show considerable heterogeneity in both dimensions and relatively stable distributions across our various specifications, which include hyperbolic, quasi-hyperbolic discounting as well as subjective failure probability over future payments. We investigate the predictive power (transportability) of the estimated preference parameters when used to explain the take-up decision of higher education grants where financial stakes are approximately seven to fifty times larger than the cash transfers used to elicit preferences. We find that both long-run discount factors and subjective payment failure risk parameters have a high degree of transportability across tasks, while parameters characterizing short-run discount preferences are irrelevant when considering higher-stakes decisions.