Power or loss aversion? Reinterpreting the bargaining weights in search and matching models