Abstract: The risk-reward model of the time series search problem is promoted under the assumption that the future can be partially forecasted, where an optimal strategy is presented. Moreover, the variations of the reward function with the parameters are studied by numerical computation, which show that the reward first increases and then is convergent as the risk tolerance and the lower limit of the expectation interval rise, respectively, and increase as the expectation probability rises and the upper limit of the expectation interval declines, respectively. The results enrich the theory of online time series search and are valuable in application.

Key words: online problem, time series search, risk-reward, probabilistic forecast