An approximation approach to dynamic programming with unbounded returns
研究了无界回报下递归效用的随机动态规划，通过截断逼近序列的收敛性来筛选相关不动点，为从有界到无界回报的扩展提供自然选择标准。
We study stochastic dynamic programming with recursive utility in settings where multiplicity of values is only attributed to unbounded returns. That is, we consider Koopmans aggregators that, when artificially restricted to be bounded, satisfy the traditional Blackwell’s discounting condition (as it certainly happens with time-additive aggregators). We argue that, when the truncation is removed, the sequence of truncated values converges to the relevant fixed point of the untruncated Bellman operator, whenever it exists, and diverges otherwise. The experiment provides a natural selection criterion, corresponding to an extension of the recursive utility from bounded to unbounded returns.