We can develop an individualized risk tolerance, h, characterized by a utility function over time, then h(u(y), t_p), where we can let t_p=0 correspond to the present time, and where the utility function is learned from the result of responses, y, to a set of questions asked of the individual. The questions that inform the utility function are generated as needed based on the responses to previous questions, in order to maximize the information gain of the next question with respect to t_p, thus the next question is q_prime{i+1} = argmax_{g} of g(y, q_{i+1}). The question generating function for possible q_{i+1} takes as inputs the set of previous responses, y to y_{i-1}, as well as a return generating function, which would be basically an auto-correlative daily Monte Carlo sampling of a LaPlacian function distribution, r(L(m,v), c, t), for t ranging in size by a number of increments relative to t_p, utilizing an auto-correlative constant, c, as the variance of white Gaussian noise used to transform L, which would be on the order of multiples of v, the variance of L. The function for q_{i+1} would be called recursively, with the best estimate for q_prime_{i+1} used to suggest the t_p and the return involved in q_{i+1}. The form of the questions would be combinations of return, risk, and time from the hypothetical present, t_h, are presented, such as, [ y_i = q_i(r_h1, r_h2, s_h1, s_h2, t_h1, t_h2) ], where r_h1 is generated from r(L(m,v), c, t_h1), and the risk is that s_h1 and s_h2 are the probabilities of experiencing h1 and h2, and s_h1 + s_h2 = 1. Thus, y is essentially a Sharpe ratio.
Net worth can be incorporated in the risk tolerance calculation, as part of the function aggregating u(y), to more appropriately explain some common actions that seem to go against utility theory - e.g. playing the lottery (even though it's a negative utility action, humans may not be able to distinguish such small changes to net worth such as an instance of a lottery playing loss).
The difference between t_p and t_h is relevant due to the individual's mortality. Thus, a relationship between t_p and t_h can be mapped, eg. from actuarial tables, which could, for instance, incorporate the current age of the respondent at t_p to account any possible indifference to a rational choice that would be likely to payoff over a sufficiently long time to t_h for a sufficiently old respondent.
Since people have a hard time abstractly judging probabilities that are close to either 0 or 1, most questions can utilize probabilities close to 0.5, where s_h1 ~= s_h2, or simply defined as absolutely 0 or 1, such that s_h1 != s_h2.