UC Berkeley

In this dissertation, I present three essays that empirically study water and energy economics issues in California.

The objective of the first chapter is to investigate whether and to what extent farmers' crop choice decision is affected by the irrigation water salinity. Using a highly granular land use data and random coefficients logit method, the effect of irrigation water salinity on crop choice is studied in the context of Sacramento-San Joaquin River Delta--- California's major water source and home to prime agricultural farmlands. The results show that though the effect of salinity was statistically significant during the past decade, highest and most significant coefficients were those of crop class indicators and weather. This finding suggests that it is essential to reach out to the farmer community to ensure that they are fully capable of coping with expected salinity increases in medium to long run. Additionally, there is evidence for heterogeneity in farmers' response to salinity even though the area studied is relatively small. Ignoring the heterogeneity can result in misleading coefficient estimates especially for those researchers who wish to study farmer behavior in larger regions. Finally, revenue losses are simulated under baseline salinity and potential future salinity scenarios due to building a water conveying facility around the Delta, which suggests an expected revenue loss of about 19%.

In the second chapter, together with Steven Buck, I question the wisdom of selecting a forecast model based on a within-sample goodness-of-fit criterion in the context of commercial and industrial (C&I) water demand in the Southern California. Initially, a set of about 350 thousand regression models are estimated using retailer level panel data featuring water consumption, price, employment, weather variables, and GDP. Out-of-sample forecasting performances of those models that rank within the top 1 \% based on various in and out-of-sample goodness-of-fit criteria were compared. We found that the models that provide the best in-sample fit are not necessarily the most favorable ones when it comes to forecasting water demand. The results indicate that on average, these models have a significantly higher absolute forecast error and a larger gap between the highest and lowest forecasts that they generate compared to the models that rank high based on out-of-sample fit criteria we defined.

Finally, the third chpater investigates the effect of the 2000 California energy crisis on the take up of an engineering audit program funded by the Department of Energy, aiming operational improvements in various domains, including energy efficiency, at small and medium sized firms. Using a detailed data set containing information on both firm characteristics and the specifics of the recommendations made, a linear probability model is estimated using difference-in-difference strategy. In order to keep the treatment and the control groups as comparable as possible to ensure credible identification, the firms that applied to be audited and made the take up decision before the crisis are compared to those that applied right before the crisis and had to decide after the crisis started. The results show that the 2000 California energy crisis was associated with a 16% increase in the take up of the IAC energy efficiency recommendations. The coefficient estimate is statistically significant and robust to different model specifications.