On biases in the measurement of foreign exchange risk premiums