Opioid abuse is currently the most significant public health problem in the US. Many US states have implemented prescription drug monitoring programs (PDMPs) in response. In the first paper, I use a new micro-level medical claims database to exploit state-level and time-series variations in PDMP implementation and shed light on the impacts of these programs. My results show that PDMPs have led to an overall 14% reduction in the odds ratio of abuse/addiction. Also, there is evidence of substantial heterogeneity in impacts, with larger impacts for females and minorities. Another finding is that at least 23% of opioid abuse is a result of drug diversion to nonmedical opioid users. PDMPs were not successful in decreasing the rate of abuse for this group, and, in fact, there is some evidence that they increased the diversion to heroin. Finally, I show that PDMPs' effectiveness varies by type of insurance and that they are more effective in reducing abuse rates in the general population as compared with Medicare Part D recipients. I use my estimates to analyze the potential effects of modifying PDMPs to include giving insurance providers access to electronic databases, providing educational programs for less-educated people, and expanding their "must access" requirement. In the second chapter, I estimate dierent models for opioid demand and compare their performance. My results suggest that the NB2 and Poisson FE models best match the data. Using these models for calculating the marginal effect of insurance characteristics provides suggestive evidence of the best insurance design to reduce the demand for opioids.