Using Prediction Markets to Predict Flu Trends

My second paper for this class was originally intended to be about webidemiology (which I will discuss a bit in my next blog post) and prediction markets. However, as the paper would have been too long had I included both topics, I opted to focus on just webidemiology in the paper and write a bit about prediction markets here.
Anyways, one of the studies that I researched for the paper involved organizing a prediction market in which healthcare professionals would buy and sell "contracts" that represented different levels of flu activity. The participants would buy a contracts that they thought represented the correct flu activity level and would sell contracts that they thought represented the incorrect flu activity level. The market prices of the contracts would then reflect the participants' opinions on the level of flu activity that would be present in their state in two weeks from the start of the buying and selling period. New contracts were released once every two weeks (in accordance with how frequently the CDC releases flu activity data). Participants were also narrowed down by giving "money" to those who held contracts to the correct flu activity levels, not giving money to those carrying incorrect contracts, and not replenishing the amount of theoretical money in their theoretical accounts once they ran out.
The ability of the participants to predict flu activity during a flu season was compared to historical data. The results showed that historical data accurately predicted flu activity 36% of the time, while the participants did so 79% of the time. Also, the predictions of the participants noticeably improved as the buying/selling periods neared the end of their two week span, though their estimates at the beginning of the periods were already accurate 90% of the time. This makes sense, as the participants see the effects of flu in their communities firsthand and every day as part of their jobs, so one would think that they would have a the sort of experience and awareness that would enable them to predict flu activity. This also demonstrates the advantage of using prediction markets in this way.