Relaxing CV(RMSE) Requirements for Option C M&V Regression Analysis

John M. Avina,



The Option C Measurement and Verification (M&V) methods for energy service companies (ESCOs) often involve performing regression analysis of utility bills against weather data. We have been advised by the International Performance Measurement and Verification Protocol (IPMVP) that our regressions should yield CV(RMSE)s (coefficients of variation of the root mean square of the error), below a certain level in order for the regression to be considered statistically significant. But what happens if you have a large portfolio, such as a school district? Is it necessary that every meter’s regression have a CV(RMSE) conforming to this rule? This paper suggests that individual meters’ CV(RMSE)s do not matter. What matters is the portfolio’s overall CV(RMSE). We tested this theory on a sample of 236 meters and found that the CV(RMSE) of the portfolio can be more than 50% lower than the average CV(RMSE) of the individual meters.