Multi-stage asset allocation using SDDP

Specialeforsvar ved Malene Gregers Jensen 

Title: Multi-stage asset allocation using SDDP

 Abstract: In this thesis the problem of asset allocation is considered. We will consider investments within exchange-traded funds (ETF's) and will represent the uncertainty associated with future returns using scenarios generated by a bootstrap algorithm. We examine both two-and three-stage programs for a risk-averse and risk-neutral investor and compare them to the 1/N benchmark strategy. As for measuring risk, we will incorporate conditional value at risk (CVaR), which estimates the worst losses above some confidence level. Given the intractable computational challenges associated with multi-stage programs, we examine whether the stochastic dual dynamic programming (SDDP) solution approach, will improve the computational aspects and whether the method is suitable in the asset allocation frame-work. We find that the performance of the two-stage models manage quite better than the three-stage strategies. Furthermore, we find that the SDDP solution approach in theory has many advantages, but in practice it is not worth the trouble, since many trials solutions are needed for the algorithm to converge towards the true optimal solution gained from the multi-stage model

Vejledere:   Trine Krogh Boomsa, Kourosh M. Rasmussen, DTU
Censor:       Niklas Kohl, DTU