A paper with the title "Giffen Behavior: Theory and Evidence" may not get your heart racing with excitement, but here is one of the most interesting pieces of economic research I have seen in a while. Many years in the making, this paper by my two Kennedy School colleagues Rob Jensen and Nolan Miller demonstrates for the first time the existence of something that economists knew was theoretically possible, but had never been observed to date: demand for a commodity that is upward sloping. That is, a good for which demand goes down when you lower its price, or for which demand goes up when you increase its price! Such a good is called a Giffen good, and anyone who has had to suffer through introductory economics has heard of it at least once as a theoretical curiosum. Economists' urban legend has it that potato was a Giffen good during the Irish potato famine of the 1840s, but as Jensen and Miller note in their paper, this is highly unlikely.
How can demand curves slope upward? Imagine a very poor household who spend a very large part of its income on some subsistence commodity such as rice. Now suppose that the price of rice goes down for some reason. How does the household alter its pattern of consumption? One response is to increase consumption of rice at the expense of other things because rice is now cheaper. This is the standard substitution effect that ensures demand curves are downward sloping. But another is that the household's overall purchasing power--its real income--has increased and therefore the family may choose to consume things, such as meat, that richer families tend to consume. If this decrease in demand for rice outweighs the substitution effect, we have a Giffen good. Put differently, the lower price of rice allows the family to satisfy its nutritional needs by a tastier combination of products, one that relies less on rice and more on meat.
As this example illustrates, a Giffen good is more likely to exist in a situation where households spend a large part of their income on a subsistence good. So Jensen and Miller went to Hunan and Gansu provinces in China, randomly subsidized the price of rice and wheat, respectively, for poor families for five months, and then observed the results. They found that the poorer families in both provinces did indeed exhibit Giffen behavior with respect to the subsistence crop. The price subsidy led to reduced consumption of rice or wheat, and its removal to more consumption.
OK, in case you were tempted to extrapolate from this result, let me be clear that it doesn't imply that the way to reduce demand for energy is to subsidize the price of oil, or that improving the U.S. trade balance will require an appreciation of the dollar.
UPDATE: Nolan Miller tells me that a 2005 article in The Nation, referring to an earlier version of the Jensen-Miller paper, makes the extrapolation to gasoline that I said in my last paragraph would be dicey.