This paper is the first to study the cross-section of currency anomalies to explore alternative explanations for their existence. Using real-time data, currency anomalies are profitable during in-sample and out-of-sample periods, both before and after transaction costs, but trading profits decrease substantially after the publication of the underlying academic research. The decline is greater for anomalies with larger in-sample profits and lower arbitrage costs, and signal ranks and performance decay quickly, suggesting that currency anomalies reflect mispricing rather than compensation for risk or statistical bias. Mispricing is systematically related to mistakes and changes in analysts’ currency forecasts. In particular, analysts expect anomaly payoffs that are too low compared with actual anomaly profits. However, analysts update their forecasts to incorporate lagged anomaly information. These results are consistent with a behavioral explanation for currency anomalies.