Intertemporal decision making under subjective uncertainty

Since Savage's seminal work, a state space has been used as the standard tool for modeling uncertainty, and it has been assumed as a primitive. This assumption implicitly requires the analyst to know all the uncertainties a decision maker (DM) perceives. This is problematic because states are in the mind of the DM and hence are not directly observable to the analyst---the state space should be derived rather than assumed as a primitive. The derivation of subjective state spaces is addressed by Kreps (1979, 1992) and refined by Dekel, Lipman and Rustichini (2001) (hereafter DLR) in static settings.;I provide several extensions of DLR to dynamic settings. Chapter 2 provides a foundation for subjective decision trees. In a dynamic setting, a decisions tree, that is, a pair consisting of a state space S and a filtration Ft Tt=0 over S, is the standard tool for modeling uncertainty. It has been taken as a primitive. This assumption is problematic as in the static setting. I derive from preference both a subjective state space S and a subjective filtration Ft Tt=0 over S. As in DLR, S is identified with the set of uncertain future preferences.;Chapter 3 extends DLR to an infinite horizon setting, and identifies the behavior that reduces all subjective uncertainties to those about future discount factors. In other words, I provide an axiomatic foundation for the random discounting model, where a DM believes her future discount factors change randomly over time. I show uniqueness of subjective belief about discount factors. Moreover, I provide the condition that agent 2 is more uncertain about discount factors than agent 1. The characterizing condition is that agent 2 values flexibility more than does agent 1.;Though states are in general subjective, some states may be directly observable to the analyst. In Chapter 4, I assume some objective states as primitives, and model a DM who is uncertain about future beliefs about objective states but certain about future risk preference. This additional assumption makes possible stronger results than in Chapter 2. I derive from preference not only a subjective decision tree (S, Ft Tt=0 ) but also a subjective belief P over S..