Decision-making and probabilistic forecasting of conflict outcomes

This work presents a new method for forecasting the outcomes of conflicts that measures the uncertainty of the outcome, is based on a Markovian model, and has shown significant predictive capabilities.; The Probabilistic Forecasting Method analyzes one issue at a time and assumes that we fully identify the outcome space and have a known finite number of actors. Each actor is characterized by its Power (its capability to influence the outcome of the conflict), Salience (the relative importance that an issue has for the actor) and Desirability Function (which measures how much the actor wants each possible outcome to happen).; This method models conflicts as Markovian processes, where the states are the possible outcomes and the transition probabilities are a function of power, salience and desirability function for the group of actors. The forecast is the steady-state probability distribution, except in some special cases. The computer implementation of this method is compact and fast.; The modeling flexibility of the method is attested by the thirty actual conflicts presented in this work. The set of conflicts includes issues on the National Information Infrastructure, the North American Free Trade Agreement, the economy and politics of Russia, the selection of a Mexican presidential candidate, the exchange rate US Dollar/German Mark, the European liberalization of civil aviation, the permissible level of radiation in food, and the setting of European emission standards for cars.; To test the calibration of the forecasted probability distribution, I divided the outcome space of each issue in nine probability intervals of size ranging from 10% to 90%. The actual outcomes fall in those intervals with a frequency close to the size of the intervals, proving the predictive capability of the method.; The work presents examples of how to take full advantage of probabilistic forecasts by using them in decision analysis. The method has a number of additional modeling capabilities (presented in this work) and provides the basic framework for expansions and further research.