Chasing the key player: A network approach to Myanmar civil war

Project Active from to State

The proposed project has two key objectives: to provide an insight into the extent of Myanmar's expansive informal welfare economy and to offer analytical tools for incoming governments.

Civil war has been the most destructive form of violent military conflict since the end of World War II. In the last three years alone, direct casualties from civil wars exceeded half a million. Because of civil wars, human and physical capital is destroyed, people are displaced and economic development is impaired. The goal of our project is to understand the determinants and evolution of the longest civil war of our time, the Myanmar civil conflict.

Our approach is novel under several dimensions. First, we study the sequential decision of the Myanmar army to attack the various armed groups active in the country. In order to do so, we model the fighting decision of the Myanmar army using a game over network approach as in Ballester, Calvò-Armengol and Zenou (2006). Through the network representation, the model allows us to keep track of the military relationships between armed groups (i.e. alliances, hostilities, and neutral relationships). Given the complex structure of armed groups alliances and enmities over time, the model yields predictions on which groups the Myanmar army should attack in order to reduce the overall fighting intensity by every armed group in the country. Indeed, attacking a group has two effects: it reduces its ability to fight against the Myanmar army and it also weaken its allies that can no longer benefit from its help in the battlefield. Second, our study is new because the model's predictions are testable versus competing theories of conflict such as for instance those stressing the role of opportunity cost of fighting.

The core of our work follows a two-step empirical analysis to test whether the Myanmar’s army attacks to rebels’ groups are linked to the relative importance of each group within the network of armed groups over time. In the first step, we derive predictions from the model. That is, given the network structure we compute which groups are more likely to be targeted over time. In order to do so, we collect historical information on as many active armed groups as possible. As there is substantial heterogeneity at the rebels’ group level in terms of army size, numbers of allies, and geographical reach, we gather these from a variety of historical sources.

The second step of the empirical analysis is to use the model's predictions in order to explain violence outbreaks over time and space in the country. To this extent, we use dyadic information on conflict outbreaks in Myanmar available for the period 1989-2015 from the UCDP Geo-referenced Event Dataset as well as from newspaper records.  This dataset identifies fighting actors as well as the location of conflict outbreaks at a fine geographical level so that geographical grids are the units of observations. Furthermore, we enrich the dataset with information on ethnic and religious identity of the groups living within every grid as well with data on natural resources extracted in areas under the influence of armed groups. These data will allow us to controls for other potential channels that are found correlated with conflict outbreaks in the previous literature such as the appropriation of natural resources following exogenous price shocks or theories that stress the role of ethnic fractionalisation in causing conflict.