DoesDaylightSavingTimeSaveEnergy-Universityof.ppt

* * * * * * * * * * * * * * * Strange result we don’t believe in Problem with standard DID Specification: Temporal variations (seasonality by industry) lead to unstable estimates * We augment the standard DID model by estimating a “triple-DID” specification because, as outlined in section 4.1, our control structure is three-fold: (a) cross-sectional over states (using VIC as the treated state and SA as the control) (b) temporal over years (using the untreated years in SA and VIC as controls) (c) temporal within days (using afternoon hours as “within” controls) Our linear specification is (1) The dependent variable for each observation is the difference in logs between demand, q, in state i in day d in half-hour h, and , the average electricity demand between 12:00-14:30 in the same state and day. The reference case model uses data from VIC and SA during 27 August to 27 October in 1999, 2000, and 2001, whereby i ? {VIC, SA}, d?=?{1,2,…,186}, and h = {1,2,…,48}. The covariates of primary interest are the indicator variables Tidh for the treatment period, unpooled by half-hour that are active from 27 August to 14 September 2000 and from 2 October to 28 October 2000, thereby excluding the Olympic Games from the treatment. Dummy variables Xidh include 48 half-hour dummies, and interactions of these dummies with indicator variables for the following: state, year, day of week, holidays, school vacations, transition days, the interaction of state with week of year, and the interaction of state with a flag for the Olympic period. The weather variables Widh are also interacted with half-hour dummies[1] and include a quadratic in hourly heating degrees,[2] daily sunlight hours, the interaction of sunlight with temperature, hourly precipitation, the interaction of precipitation with temperature, and the average of the afternoon heating degrees. All weather variables enter the model lagged by one hour. In sp