1 9. Jahrgang (1 993), Heft 4 Wirtschaft und Gesellschaft If wages are set at the level of an industry, however, then Tf > Tf* and (} < 1 which implies that h < 1 . In this case, each union knows that an in­ crease in its nominal wage will increase its product price to a greater ex­ tent than it will raise the cost of living. The union is less sensitive to em­ ployment loss than with either purely local wage setting or fully centra­ lized wage setting. The important difference here between different degrees of centraliza­ tion is the ability of each union to increase its real consumption wage without an equivalent increase in the real product wage in its sector. At intermediate levels of centralization, each union is able to pass some of the cost of a wage increase on to others through a higher product price, rather than bearing all of the cost itself in the form of lower employment. Since the cost of a wage increase is reduced at intermediate levels of centralization, unians choose higher wages. However, one union's pro­ duct price increase is another union's consumer price increase. When all unians raise nominal wages, all prices rise and the result is higher real wages (whether in terms of the price of output or of consumption) and lower employment than would result from either purely local bargaining or wage bargaining that was centralized at the national level. The cen­ tral result is that the relationship between wages and bargaining level is hump-shaped with both very decentralized and highly centralized bar­ gaining systems producing greater wage restraint and lower unemploy­ ment than bargaining systems in between (5) . According to the model presented so far, either extreme of very decen­ tralized or completely centralized wage setting is equally good. If the as­ sumption of a completely closed economy is relaxed, however, then even perfectly centralized bargaining produces less wage restraint than local bargaining (6). The reason is that the purest form of centralization within national borders is still incomplete in an open economy. A wage increase in all sectors of an open economy can raise the real exchange rate (i. e. , the relative price of non-traded goods) thereby raising the real consumer wage more than the real product wage in the non-traded goods sector. This dampens the employment loss that results from a wage in­ crease at the national level. In an open economy, purely local bargaining results in lower wage demands than complete centralization. In contrast, if the assumption of perfect competition is relaxed, then highly centralized bargaining produces greater wage restraint than pu­ rely local bargaining. If individual firms have market power, then in­ creased wage costs will be passed on to prices to some extent even with firm-level bargaining (7). Thus, in open economies in which some firms are imperfectly competitive, the ranking of local and national bargain­ ing in terms of wage militancy is no Ionger clear. What remains is the conclusion that unions' wage demands are highest at the level of wage setting that corresponds to the level of maximum divergence between the effect of a wage increase on the real consumer wage-which the union wants to increase-and on the real product wage-which the union wants to prevent from rising. 429