This paper estimates the arbitrage costs from international relative prices, and studies the economic determinants of implied arbitrage costs. We find that the magnitude of arbitrage costs depends on the characteristics of both the type of good and set of locations under examination. More specifically, higher share of non-traded input and lower tradability of good lead to larger arbitrage cost, as does a lower proximity of geographic distance between locations. The role of location-specific characteristic in accounting for the arbitrage cost is more significant as we move beyond an economic geography, while good-specific characteristics matter relatively more if we move to the interior of this geography. We also show that arbitrage cost plays an important role in explaining the puzzling behavior of real exchange rate. We view our evidence as complementary to those that emphasize the role of nominal price rigidities. 

I. Introduction 

II. Direct Measurement of Arbitrage Costs 

III. Arbitrage Costs and Relative Prices 

IV. Empirical Analysis 
1. Threshold Autoregressive (TAR) Model 
2. Results 

V. Implications for the PPP Puzzle 

VI. Conclusions 

References