This thesis investigates the measurement of predictor performance as applied to foreign exchange speculation. It outlines the development of key ideas and techniques over the course of the last 120 years, and examines the datasets and metrics used within a representative sample of the academic corpus. In this examination two problems are identified: first, there is a lack of consistency in the datasets used to test researchers’ algorithms; and second, a large variety of metrics are used, most of which are either inappropriate for or inappropriately applied to FOREX speculation. To address these issues, this thesis presents two solutions: a Python library, Hokohoko, which provides a consistent dataset and interface for testing FOREX prediction algorithms; and a new metric, Speculative Accuracy, which it argues provides a more appropriate measure of usefulness with regards to speculation. Hokohoko is then used to test a series of hypotheses regarding the usefulness of various metrics, alongside Speculative Accuracy.