A vast literature documents that willingness to pay (WTP) is less than willingness to accept (WTA) a monetary amount for an object, a phenomenon called the endowment effect. Using data from three incentivized studies with a total representative sample of 4,000 U.S. adults, we add one additional finding: WTA and WTP for a lottery are (essentially) uncorrelated. In contrast, independent measures of WTA (or WTP) are highly correlated, and relatively stable across time. Leading models of reference dependent preferences are compatible with a zero correlation between WTA and WTP, but only for specific parameterizations and ruling out popular special cases. These models also predict a relationship between the endowment effect and loss aversion, which we do not find.