“…Hence, if estimated, the coefficient˛1 would provide a biased estimate of ex ante moral hazard. Second, if RT * (Arrow, 1985) and PE * (Winter, 1992) were perfectly observable, Dionne (2013) argues that when the task is to estimate an empirical model of insurance that contains a potentially endogenous variable, it is often better to instrument this variable (see Dionne, St-Amour, and Désiré, 2009;Dionne, La Haye, and Bergéres, 2010;Rowell 2011, for more details). We consider two methods, previously described by Rowell 2011, to obtain a consistent estimate of˛1 from Equation (1).…”