Modeling Risks in Infrastructure Asset Management

The goal of this dissertation research is to model risk in delivery, operation and maintenance phases of infrastructure asset management. More specifically, the two main objectives of this research are to quantify and measure financial risk in privatizing and operational risks in maintenance and rehabilitation of infrastructure facilities. To this end, a valuation procedure for valuing large-scale risky projects is proposed. This valuation approach is based on mean-risk portfolio optimization in which a risk-averse decision-maker seeks to maximize the expected return subject to downside risk. We show that, in complete markets, the value obtained from this approach is equal to the value obtained from the standard option pricing approach. Furthermore, we introduce Coherent Valuation Procedure (CVP) for valuing risky projects in partially complete markets. This approach leads to a lower degree of subjectivity as it only requires one parameter to incorporate user's risk preferences. Compared to the traditional discounted cash flow analysis, CVP displays a reasonable degree of sensitivity to the discount rate since only the risk-free rate is used to discount future cash flows. The application of this procedure on valuing a transportation public-private partnership is presented. %and demonstrate that the breakeven buying price of a risky project is equal to the value obtained from this valuation procedure.

Secondly, a risk-based framework for prescribing optimal risk-based maintenance and rehabilitation (M&R) policies for transportation infrastructure is presented. These policies guarantee a certain performance level across the network under a predefined level of risk. The long-term model is formulated in the Markov Decision Process framework with risk-averse actions and transitional probabilities describing the uncertainty in the deterioration process. Conditional Value at Risk (CVaR) is used as the measure of risk. The steady-state risk-averse M&R policies are modeled assuming no budget restriction. To address the short-term resource allocation problem, two linear programming models are presented to generate network-level polices with different objectives. In the first model, decision-maker minimizes the total risk across the network, and in the second model, the highest risk to the network performance is minimized.