An important advance in the study of reference-dependent preferences is the discipline provided by coherent accounts of reference point formation. K?szegi and Rabin (2006) provide such discipline by positing a reference point grounded in rational expectations. We examine the predictions of K?szegi and Rabin (2006) in the context of market experiments with probabilistic forced exchange. The experiment tightly tests the predictions of K?szegi and Rabin (2006), as when the probability of forced exchange increases, individuals should grow more willing to exchange. This mechanism has the theoretical potential to eliminate and even reverse the 'endowment effect' (Knetsch and Sinden, 1984; Knetsch, 1989; Kahneman et al., 1990). Our results uniformly reject these theoretical predictions. In a series of experiments with a total of 930 subjects, sellers' valuations exceed buyers' valuations under all probabilities of forced exchange. In robustness tests where attention is drawn specifically to the forced exchange mechanism, the results are directionally more promising for buyers, but still reject the main thrust of the theoretical predictions. Our findings suggest a potential path forward incorporating failures to completely forecast sensations of gain and loss into models of expectations-based reference dependence.