We show that the ability of oil price changes to predict stock returns is limited to periods of extreme geopolitical unrest. Four events generate most of the predictability: the 1973 Arab-Israel war, the 1986 OPEC collapse, the 1990/91 Persian gulf war, and the 2003 invasion of Iraq. We also find that a market-timing trading strategy based on oil price changes typically generates insignificant abnormal returns, contradicting previously published results. Our findings serve as an example of how a significant predictor in a time series forecasting regression may not be a useful or profitable market-timing signal.