In many markets, changes in the spot price are partially predictable. We show that when this is the case: (1) although unbiased, traditional regression estimates of the minimum variance hedge ratio are inefficient, (2) estimates of the riskiness of both hedged and unhedged positions are biased upward, and (3) estimates of the percentage risk reduction achievable through hedging are biased downward. For natural gas cross hedges, we find that both the inefficiency and bias are substantial. We further find that incorporating the expected change in the spot price, as measured by the futures-spot price spread at the beginning of the hedge, into the regression results in a substantial increase in efficiency and reduction in the bias.