This paper contributes to the limited-information literature on savings in a stochastic environment. In particular, it contributes techniques and concepts to the question of state verification (or filtering), by including learning about aggregate income shocks, based on signals. As a seminal contribution to the extant literature, a "conviction function" is introduced, which takes into account histories of past prediction errors in determining how rational agents internalize such information in taking personal investment decisions. For purpose of a more transparent illustration, a numerical rendition of the posited model is provided for five consecutive time periods. We also perform a series of Monte Carlo simulations to demonstrate how the posited approach could potentially outperform traditional forward-looking models in the presence of sudden large extraneous shocks reminiscent of the recent Global Financial Crisis.