We compare the observed RTAs to the probability threshold optimizing the joint prediction of signed and unsigned agreements, and derive a counterfactual RTA geography by signing predicted but unsigned agreements. To assess the trade and welfare consequences of signing unsigned agreements in General Equilibrium, we follow Anderson, Larch & Yotov (2018) and use a General Equilibrium Poisson Pseudo Maximum Likelihood (GEPPML) method to solve the system of equations associated with the model. We proceed by estimating a probability model using data over the 1990-2014 period; the fiveyear lag for covariates implies that we will consider agreements that were signed between country pairs over the 1995-2014 period, adding to the usual controls for the indirect involvement of the country pair in the fragmentation of value chains. We evaluate model goodness-of-fit by considering the probability cut-off that provides the best percentage of correctly-predicted events (a countrypair being, or not, members at a given date of an RTA: RTA=1 and RTA=0).: the optimal cut-off probability maximizes the percentages of true positives and true negatives. We focus on the mis-classified RTAs that are classified as FP from the probabilistic model are be used in a structural gravity system to evaluate the welfare changes associated with alternative sets of trade agreements