This thesis presents theoretical work on armed group activity and empirical work on post-conflict recovery. In chapter two, I develop a general equilibrium model of violence to explain observed variation in coercive practices in conflict zones. Armed groups own land in the resource sector and allocate military resources between conflict and coercion, which assign de facto ownership over land and labour respectively. I find that coercion is higher if labour is scare relative to land, production is labour-intensive, or if one group is dominant relative to others. Furthermore, contrary to other studies, I find that coercion could decreases with price if military power is sufficiently decentralised, since conflict draws resources away from coercion. In chapter three, I evaluate a reconciliation program in post-conflict Sierra Leone. The program provides a forum for villagers to air war-time grievances, and also forges institutions designed to improve conflict resolution and build social capital. I find that respondents who received the intervention are more forgiving and are more charitable in their views of ex-combatants. Furthermore, conflict resolution improved and involvement in village groups and activities increased. However, psychological health---depression, post-traumatic stress disorder and anxiety---deteriorated. This study has direct implications for the design of transitional justice programs, as well as programs that aim to promote institutional change. In chapter four, I experimentally vary foreigner presence across behavioural games conducted in 60 communities in Sierra Leone, and assess its effect on standard measures of generosity. I find that foreigner presence substantially increases player contributions in dictator games, by an average of 19 per cent. Furthermore, the treatment effect is smaller for players who hold positions of authority; and subjects from villages with greater exposure to development aid give substantially less and are more inclined to believe that the behavioural games were conducted to test them for future aid. In chapter five, I use a model of repeated bargaining with one-sided asymmetric information to investigate the difficulties of reaching and sustaining power sharing agreements. I show that asymmetric information can explain the persistence of conflict, since learning slows down when there are future opportunities for bargaining.