A new hybrid methodology is introduced which is a combination of multiple regression model and generalised autoregressive conditional heteroskedasticity (GARCH) model. Comparing the new approach and the vector auto-regression (VAR) model, this paper analyses the short-term dynamics of the energy efficiency index (EEI) in response to change in the five indicator variables for thermal power plants in China. The result indicates that: (i) The new hybrid model can directly calculate the EEIs of thermal power plants without artificial intervention. (ii) It can eliminate the disturbance of residual superposition. (iii) The new method will offer more direct information on the degree of volatility among determinants and operating inefficiency. (C) 2016 Elsevier Ltd. All rights reserved.