Stochastic Financial Analytics for Cash-flow Bullwhip, Cash-flow Forecast, and Optimizing Working Capital

Open Access
Tangsucheeva, Rattachut
Graduate Program:
Industrial Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
April 15, 2014
Committee Members:
  • Vittaldas V Prabhu, Dissertation Advisor
  • Arunachalam Ravindran, Committee Member
  • Tao Yao, Committee Member
  • Douglas J Thomas, Committee Member
  • Stochastic
  • Analytics
  • Cash flow
  • Optimization
  • Working Capital
  • Forecast
  • Bullwhip
  • Supply chain
  • Decision support
  • Finance
Managing modern supply chains involves dealing with complex dynamics of materials, information, and cash flows around the globe, and is a key determinant of business success today. One of the well-recognized challenges in supply chain management is the inventory bullwhip effect in which demand forecast errors are amplified as it propagates upstream from a retailer, in part because of lags and errors in information flows. Adverse effects of such bullwhip effect include excessive inventory, stock-outs, backorders, and wasteful swings in manufacturing production. In this dissertation we theorize that inventory bullwhip also leads to cash-flow bullwhip (CFB). Specifically, this research focuses on studying CFB by developing mathematical and simulation models to analyze the relationship between inventory and cash-flow bullwhip by using Cash Conversion Cycle (CCC) as a metric. CFB predicted by the proposed mathematical models approximately differ 14% from detailed simulation models. We find that increasing variability increases inventory and cash-flow bullwhip along with lead time, whereas increasing the demand observation period has the opposite effect. The average marginal impact of the bullwhip effect on the CFB is approximately 20%. Additionally, the CFB is also an increasing function of an expected value of inventory and a decreasing function of an expected value of demand. Next, we develop stochastic financial analytics for cash flow forecasting for firms by integrating two models: (1) Markov chain model of the aggregate payment behavior across all customers of the firm using accounts receivable aging and; (2) Bayesian model of individual customer payment behavior at the individual invoice level. As the stochastic dynamics of cash flow evolves every day, the forecast can be updated every time an invoice is paid. The proposed model is back-tested using empirical data from a small manufacturing firm and found to differ 3%-6% from actual monthly cash flow, and differs approximately 2%-4% compared to actual annual cash flow. The forecast accuracy of the proposed stochastic financial analytics model is found to be considerably superior to other techniques commonly used. Furthermore, in computer simulation experiments, the proposed model is found to be largely robust to supply chain dynamics, including when subjected to severe bullwhip effect. The proposed model has been implemented in Excel, which allows it to be easily integrated with the accounts receivable aging data, making it practicable for small and large firms. Lastly, we identify a potential strategy to engineer a solution for dynamic financial decisions. This part focuses on maximizing profit of two types of manufacturing firms: firms with non-recurring customers and firms with recurring customers. The proposed model determines the optimal pricing of products sold to different customers using an integer programming model in which Friedman’s model is used to estimate bid winning probability and the model is constrained by several operational factors including working capital and customer credit risk. Customer credit risk is modeled as the probability of payment delay or default, which is used to add a risk premium into the bid price. The model can be used for decision-support in business development to select an optimal portfolio of customer projects or bids to pursue. Detailed industrial case studies used to test the efficacy of the proposed model show that the Price/Cost ratio has an inversely proportional relationship to the winning probability, and the winning probability has the inversely proportional relationship to the risk premium. However, at a high winning probability, the firm may not be able to make a profit due to the unrealistically low bidding price. The results also show the bidding price at which the firm is expected to maximize its profit.