Our initial analysis of RGGI data shows that the price distribution between the primary and secondary markets are different by a significant margin – the primary auction settles at an average price of $3.70 per MtCO2, while the allowance transfer market settles at a price of $4.72 per MtCO2. This is intuitive, given that traders on the secondary market would be selling to those that didn’t purchase enough allowances in the primary auction. In addition to the official secondary market exchanges, we see a small volume of other exchanges that have price data but are not official. These are presumed to have been legal transfers of allowance rights through some program allowed by the RGGI system apart from the standard exchange and are sparse enough to ignore in our analysis.

​

IIt’s also clear that the primary auction is significantly larger market than the secondary exchange, at about a 3:1 ratio, so we focus the goal of our project on modeling the auction itself with an additional analysis of the secondary market. When cleaning and summarizing the RGGI data files, another data limitation presented itself – the data that RGGI releases to the public is anonymized to the point that linking records regarding emissions or offsets to price and allowance distribution data at the level of an individual auction participant would be difficult. Given the resulting lack of bid-level data apart from number of allowances transferred and uniform price, building a predictive model of bidding behavior using traditional regression techniques leads to unreliable results – both OLS and quantile regression (which is similar to standard OLS but with a specified quantile as the predicted outcome rather than the mean and is useful when an assumption of equal variance isn’t reasonable) have very poor predictive power when used in this context. Since it isn’t feasible to build a robust regression model to predict behavior with our given data, we turn to simulation to generate bidding behavior based on set, theory-based rules.