T HIS study develops an empirical model for the joint determination of the demand and supply for bilateral trade, in which both bilateral quantities and bilateral prices are endogenous. In theory each bilateral demand (supply) equation would contain income (output) variables together with all relevant bilateral prices. Given differences in the commodity composition of trade and given price discrimination, export prices of any country will differ over destination countries. In addition, transport costs and tariffs will drive a wedge between bilateral export and import prices. Thus, a bilateral demand and supply system for aggregate trade among n countries would require consideration of 2n2 price variables, an overwhelming task for empirical research. As a compromise between pure theory and practice, we use a two-tier construction in which total import demand and export supply are determined by aggregate real income (output) and relative aggregate prices, and then bilateral trade flows are determined by theory-based allocations featuring relative bilateral prices. This approach is more general than either specifying only bilateral demand behavior (assuming infinitely elastic supply at given prices) or including the supply side through stringent assumptions such as constant market shares.' For empirical application the model is extended to include proxies for nonprice competition and structural factors, a flowadjustment specification of dynamics, and a trend/cycle decomposition of income (output). The model is estimated from a panel of bilateral trade flows covering the United States, Japan, France, Germany, and United Kingdom for the years 1958-1974. The theoretical model is developed in section II, the empirical specification is presented in section III, and the empirical results are discussed in section IV.

T HIS study examines the direction and level of aggregate bilateral flows in a multi-country network. We are not so much concerned with an explanation of a country's total imports as with the geographical pattern of its imports from among its trading partners. We therefore incorporate economic variables for both the importing and exporting countries, including demand and supply conditions and trade resistance factors, in particular, the costs of transportation. The latter are incorporated into the analysis through an errorsin-variables specification. Our study begins on familiar ground. Similar models have been proposed and tested by Tinbergen (1962), Poyhonen (1963a, b), Pulliainen (1963), Linnemann (1966) and others. The key difference between our study and previous ones lies in the treatment of the costs of transportation. Previous studies have used distance as a proxy for transport costs, but that has some serious limitations. First, the cost of transportation is influenced by other factors such as the value of the commodity being transported.1 For example, Moneta (1959) has pointed out that the effect of distance is small for high-valued commodities. Second, the use of distance imposes the assumption that the cost of transportation is the same in either direction between any pair of trading countries. This is restrictive when analyzing aggregate flows, since the commodity composition of differs by direction. Third, estimated relationships (e.g., elasticities) between flows and static variables such as distance are not very helpful in predicting future levels and thus are not very helpful in policy analysis. Finger and Yeats (1976) have shown for the United States that effective protection due to international transport costs is at least as high as tllat due to tariffs. Moreover, they indicate that the importance of transport costs has been increasing rapidly in recent years (even before full consideration of the recent petroleum price increases).2 The need to move beyond the distance specification of transport costs is clear. The fact that reliable data on transport costs are unavailable has been the primary reason for the use of distance as a proxy in the previous studies. In principle, the difference between c.i.f. and f.o.b. values represents the costs of freight and insurance.3 However, due to notorious measurement errors, these figures cannot be used in traditional econometric procedures. Consequently, most studies dealing with this subject have not utilized the differences between c.i.f. and f.o.b. values.4 Though these differences are indeed highly inaccurate measures of transport costs, they are included in our empirical analysis by applying an errorsin-variables approach. This allows the estimation of the elasticity of bilateral flows with respect to transport costs, which is the key product of this study. The theoretical background is introduced in section II and the empirical model is specified