Distinguishing Correlation From Causation

How do you avoid assuming that one market movement caused another simply because the two instruments moved together?