Potential Pitfalls for the Purchasing-Power-Parity Puzzle? Sampling and Specification Biases in Mean-Reversion Tests of the Law of One Price
指出，现有购买力平价谜题的证据可能源于方法选择偏差：使用低频数据和线性模型会掩盖实际存在的快速价格收敛，对研究套利和市场效率的学者有重要警示。
The PPP puzzle is based on empirical evidence that international price differences for individual goods (LOOP) or baskets of goods (PPP) appear highly persistent or even non-stationary. The present consensus is these price differences have a half-life that is of the order of five years at best, and infinity at worst. This seems unreasonable in a world where transportation and transaction costs appear so low as to encourage arbitrage and the convergence of price gaps over much shorter horizons, typically days or weeks. However, current empirics rely on a particular choice of methodology, involving (i) relatively low-frequency monthly, quarterly, or annual data, and (ii) a linear model specification. In fact, these methodological choices are not innocent, and they can be shown to bias analysis to-wards findings of slow convergence and a random walk. Intuitively, if we suspect that the actual adjustment horizon is of the order of days then monthly and annual data cannot be expected to reveal it. If we suspect arbitrage costs are high enough to produce a substantial band of inaction' then a linear model will fail to support convergence if the process spends considerable time random-walking in that band. Thus, when testing for PPP or LOOP, model specification and data sampling should not proceed without consideration of the actual institutional context and logistical framework of markets.