Mixing stochastic dynamic programming and scenario aggregation

This paper describes a serial and parallel implementation of a hybrid stochastic dynamic programming and progressive hedging algorithm. Numerical experiments show good speedups in the parallel implementation. In spite of this, our hybrid algorithm has difficulties competing with a pure stochastic dynamic programming approach on a given test case from macroeconomic control theory.This research has been conducted with financial support from the Norwegian Research Council. As most of this work was conducted under the TRACS program at the University of Edinburgh, we want to thank Ken McKinnon and all other helpful people at the Department of Mathematics and Statistics of Edinburgh University and at the Edinburgh Parallel Computing Centre. We are also very grateful to our colleague Stein W. Wallace for his continuing support of our work. Without him, this research would probably never have taken place. We would also like to thank an anonymous referee for helpful corrections and comments.     

Scenario reduction in stochastic programming

 Given a convex stochastic programming problem with a discrete initial probability distribution, the problem of optimal scenario reduction is stated as follows: Determine a scenario subset of prescribed cardinality and a probability measure based on this set that is the closest to the initial distribution in terms of a natural (or canonical) probability metric. Arguments from stability analysis indicate that Fortet-Mourier type probability metrics may serve as such canonical metrics. Efficient algorithms are developed that determine optimal reduced measures approximately. Numerical experience is reported for reductions of electrical load scenario trees for power management under uncertainty. For instance, it turns out that after 50% reduction of the scenario tree the optimal reduced tree still has about 90% relative accuracy. Received: July 2000 / Accepted: May 2002 Published online: February 14, 2003 Key words. stochastic programming – quantitative stability – Fortet-Mourier metrics – scenario reduction – transportation problem – electrical load scenario tree Mathematics Subject Classification (1991): 90C15, 90C31     

Barycentric scenario trees in convex multistage stochastic programming

This work deals with the approximation of convex stochastic multistage programs allowing prices and demand to be stochastic with compact support. Based on earlier results, sequences of barycentric scenario trees with associated probability trees are derived for minorizing and majorizing the given problem. Error bounds for the optimal policies of the approximate problem and duality analysis with respect to the stochastic data determine the scenarios which improve the approximation. Convergence of the approximate solutions is proven under the stated assumptions. Preliminary computational results are outlined. This work has been supported by Schweizerischen Nationalfonds Grant Nr. 21-39 575.93.     

Parameter estimation in stochastic scenario generation systems

Scenario analysis offers an effective tool for addressing the stochastic elements in multi-period financial planning models. Critical to any scenario generation process is the estimation of the input parameters of the underlying stochastic model for economic factors. In this paper, we propose a new approach for estimation, known as the integrated parameter estimation (IPE). This approach combines the significant features of other well-known estimation techniques within a non-convex multiple objective optimization framework, with the objective weights controlling the relative importance of the features. We solve the non-convex optimization problem using adaptive memory programming – a variation of tabu search. Based on a short interest rate model using UK treasury rates from 1980 to 1995, the integrated approach compares favorably with maximum likelihood and the generalized method of moments. We also evaluate performance with Towers Perrin's CAP:Link scenario generation system.     

Solution sensitivity-based scenario reduction for stochastic unit commitment

A two-stage stochastic program is formulated for day-ahead commitment of thermal generating units to minimize total expected cost considering uncertainties in the day-ahead load and the availability of variable generation resources. Commitments of thermal units in the stochastic reliability unit commitment are viewed as first-stage decisions, and dispatch is relegated to the second stage. It is challenging to solve such a stochastic program if many scenarios are incorporated. A heuristic scenario reduction method termed forward selection in recourse clusters (FSRC), which selects scenarios based on their cost and reliability impacts, is presented to alleviate the computational burden. In instances down-sampled from data for an Independent System Operator in the US, FSRC results in more reliable commitment schedules having similar costs, compared to those from a scenario reduction method based on probability metrics. Moreover, in a rolling horizon study, FSRC preserves solution quality even if the reduction is substantial.     

Multi-stage scenario generation by the combined moment matching and scenario reduction method

We describe an opportunity to speed up multi-stage scenario generation and reduction using a combination of two well-known methods: the moment matching method (Høyland and Wallace, 2001) and the method for scenario reduction to approximately minimize a metric (Heitsch and Römish, 2009). Our suggestion is to combine them rather than using them in serial by making use of a stage-wise approximation to the moment matching algorithm. Computational results show that combining the methods can bring significant benefits.     

A clustering approach for scenario tree reduction: an application to a stochastic programming portfolio optimization problem

This paper deals with the problem of scenario tree reduction for stochastic programming problems. In particular, a reduction method based on cluster analysis is proposed and tested on a portfolio optimization problem. Extensive computational experiments were carried out to evaluate the performance of the proposed approach, both in terms of computational efficiency and efficacy. The analysis of the results shows that the clustering approach exhibits good performance also when compared with other reduction approaches.     

A note on scenario reduction for two-stage stochastic programs

We extend earlier work on scenario reduction by relying directly on Fortet-Mourier metrics instead of using upper bounds given in terms of mass transportation problems. The importance of Fortet-Mourier metrics for quantitative stability of two-stage models is reviewed and some numerical results are also provided.     

Observational data-based quality assessment of scenario generation for stochastic programs

Computational Management Science - In minimization problems with uncertain parameters, cost savings can be achieved by solving stochastic programming (SP) formulations instead of using expected...     

Solving stochastic complementarity problems in energy market modeling using scenario reduction

In this paper, we analyze market equilibrium models with random aspects that lead to stochastic complementarity problems. While the models presented depict energy markets, the results are believed to be applicable to more general stochastic complementarity problems. The contribution is the development of new heuristic, scenario reduction approaches that iteratively work towards solving the full, extensive form, stochastic market model. The methods are tested on three representative models and supporting numerical results are provided as well as derived mathematical bounds.     

SDDP for multistage stochastic programs: preprocessing via scenario reduction

Even with recent enhancements, computation times for large-scale multistage problems with risk-averse objective functions can be very long. Therefore, preprocessing via scenario reduction could be considered as a way to significantly improve the overall performance. Stage-wise backward reduction of single scenarios applied to a fixed branching structure of the tree is a promising tool for efficient algorithms like stochastic dual dynamic programming. We provide computational results which show an acceptable precision of the results for the reduced problem and a substantial decrease of the total computation time.     

The intuitive dynamic programming approach to optimal stochastic navigation

Summary This paper supplies an intuitive probabilistic interpretation of the necessary conditions on locally optimal solutions of the optimization problems treated in Optimal Navigation with Random Terminal Time in the Presence of Phase Constraints by means of a functional analytic approach. The dynamic programming approach yields rigorous results only under restrictive and hard-to-verify assumptions on the cost function, but it follows an interpretation of an optimal control as direction of fastest fecrease of the conditional expected residual costs.     

极大极小随机规划逼近问题最优解集和最优值的稳定性

研究了特殊的二层极大极小随机规划逼近收敛问题. 首先将下层初始随机规划最优解集拓展到非单点集情形, 且可行集正则的条件下, 讨论了下层随机规划逼近问题最优解集关于上层决策变量参数的上半收敛性和最优值函数的连续性. 然后把下层随机规划的epsilon-最优解向量函数反馈到上层随机规划的目标函数中, 得到了上层随机规划逼近问题的最优解集关于最小信息概率度量收敛的上半收敛性和最优值的连续性.     

Scenario generation and stochastic programming models for asset liability management

In this paper, we develop and test scenario generation methods for asset liability management models. We propose a multi-stage stochastic programming model for a Dutch pension fund. Both randomly sampled event trees and event trees fitting the mean and the covariance of the return distribution are used for generating the coefficients of the stochastic program. In order to investigate the performance of the model and the scenario generation procedures we conduct rolling horizon simulations. The average cost and the risk of the stochastic programming policy are compared to the results of a simple fixed mix model. We compare the average switching behavior of the optimal investment policies. Our results show that the performance of the multi-stage stochastic program could be improved drastically by choosing an appropriate scenario generation method.     

Scenario reduction in stochastic programming with respect to discrepancy distances

Discrete approximations to chance constrained and mixed-integer two-stage stochastic programs require moderately sized scenario sets. The relevant distances of (multivariate) probability distributions for deriving quantitative stability results for such stochastic programs are ℬ-discrepancies, where the class ℬ of Borel sets depends on their structural properties. Hence, the optimal scenario reduction problem for such models is stated with respect to ℬ-discrepancies. In this paper, upper and lower bounds, and some explicit solutions for optimal scenario reduction problems are derived. In addition, we develop heuristic algorithms for determining nearly optimally reduced probability measures, discuss the case of the cell discrepancy (or Kolmogorov metric) in some detail and provide some numerical experience.     

Financial scenario generation for stochastic multi-stage decision processes as facility location problems

The quality of multi-stage stochastic optimization models as they appear in asset liability management, energy planning, transportation, supply chain management, and other applications depends heavily on the quality of the underlying scenario model, describing the uncertain processes influencing the profit/cost function, such as asset prices and liabilities, the energy demand process, demand for transportation, and the like. A common approach to generate scenarios is based on estimating an unknown distribution and matching its moments with moments of a discrete scenario model. This paper demonstrates that the problem of finding valuable scenario approximations can be viewed as the problem of optimally approximating a given distribution with some distance function. We show that for Lipschitz continuous cost/profit functions it is best to employ the Wasserstein distance. The resulting optimization problem can be viewed as a multi-dimensional facility location problem, for which at least good heuristic algorithms exist. For multi-stage problems, a scenario tree is constructed as a nested facility location problem. Numerical convergence results for financial mean-risk portfolio selection conclude the paper.     

A stochastic programming model for the optimal issuance of government bonds

Sovereign states issue fixed and floating securities to fund their public debt. The value of such portfolios strongly depends on the fluctuations of the term structure of interest rates. This is a typical example of planning under uncertainty, where decisions have to be taken on the base of the key stochastic economic factors underneath the model.     

A multistage linear stochastic programming model for optimal corporate debt management

Large corporations fund their capital and operational expenses by issuing bonds with a variety of indexations, denominations, maturities and amortization schedules. We propose a multistage linear stochastic programming model that optimizes bond issuance by minimizing the mean funding cost while keeping leverage under control and insolvency risk at an acceptable level. The funding requirements are determined by a fixed investment schedule with uncertain cash flows. Candidate bonds are described in a detailed and realistic manner. A specific scenario tree structure guarantees computational tractability even for long horizon problems. Based on a simplified example, we present a sensitivity analysis of the first stage solution and the stochastic efficient frontier of the mean-risk trade-off. A realistic exercise stresses the importance of controlling leverage. Based on the proposed model, a financial planning tool has been implemented and deployed for Brazilian oil company Petrobras.     

Challenges in stochastic programming

Remarkable progress has been made in the development of algorithmic procedures and the availability of software for stochastic programming problems. However, some fundamental questions have remained unexplored. This paper identifies the more challenging open questions in the field of stochastic programming. Some are purely technical in nature, but many also go to the foundations of designing models for decision making under uncertainty. Research supported by grants of the National Science Foundation and the US-Israel Binational Science Foundation.