I show that the Koszegi and Rabin (2006, 2007) model of reference dependent preferences with loss aversion can rationalize myopic loss aversion (Benartzi & Thaler 1995, Gneezy & Potter 1997) : an agent's willingness to take risks is bigger (smaller) prior to (after) the realization of a background risk. Koszegi and Rabin's model however also predicts an endowment effect for risk : an agent's willingness to take risks prior to the realization of a background risk increases in the level of background risk. In a tight experimental setup, I investigate these predictions both in the presence of a negative and a positive background risk. While I cannot reject a myopic loss averse behavior when the background risk is positive, I unambiguously reject it when the background risk is negative. I also overwhelmingly reject the endowment effect for risk predicted by the K ̋oszegi and Rabin model. These results cast doubt on the ability of the Koszegi and Rabin model to neatly pin down the channels through which loss aversion strikes, and provide further evidence on the limits of expectation-based reference dependence.