A multi-period multi-product stochastic inventory problem with order-based loan

Technical Summary: A Multi-Period Multi-Product Stochastic Inventory Problem with Order-Based Loan

Problem Definition
This paper addresses the operational decision-making of a cash-constrained online retailer selling multiple products over a finite planning horizon. The retailer operates on Chinese e-commerce platforms (e.g., Tmall, JD.com) where a specific transactional friction exists: customer payments are held by the platform and transferred to the retailer only after a receipt delay (typically at least two weeks). This delay creates a cash flow gap, limiting the retailer's ability to restock inventory, particularly for small and medium-sized enterprises (SMEs) with limited initial capital.

To mitigate this, the study investigates the adoption of "order-based loans," a financing service offered by these platforms. Under this scheme, the platform advances the payment to the retailer immediately upon order delivery, allowing the retailer to use these funds for immediate replenishment. The retailer must repay the principal plus interest after an agreed period. The core problem is to determine optimal ordering quantities ($Q_{n,t}$) and the extent of loan utilization ($g_{n,t}$) for $N$ products over $T$ periods to maximize expected final cash, subject to cash constraints, inventory dynamics, and demand uncertainty.

Methodology
The authors develop a mathematical framework progressing from deterministic to stochastic formulations:

1. Deterministic Models: Two baseline models are constructed assuming demand is known (represented by mean values):

   • SO-D (Self-Owned Cash): The retailer operates solely on initial cash and sales revenue.
   • OL-D (Order-Based Loan): The retailer utilizes the order-based loan facility up to a credit limit ($BU$).
     Both models incorporate linear constraints for inventory flow, cash flow, and the specific revenue recognition logic involving the receipt delay length ($L$) and loan repayment terms.

2. Stochastic Programming Models: Recognizing that customer demand is uncertain, the authors formulate stochastic counterparts (SO-S and OL-S).

   • Scenario Representation: Demand uncertainty is modeled using scenario trees, where each node represents a discrete demand realization with an associated probability.
   • Non-Anticipativity: Constraints are imposed to ensure that decisions made at any time $t$ depend only on the information available up to that time (shared history), not on future scenario realizations.
   • Scenario Generation: The study employs the moment matching method (Høyland and Wallace, 2001) to generate a limited set of scenarios that satisfy specific statistical properties (mean, variance, skewness) of the underlying log-normal demand distributions.
   • Scenario Reduction: To address computational intractability with large trees, the Fast Forward Selection (FFS) algorithm (Heitsch and Römisch, 2003) is used to reduce the number of scenarios while preserving the statistical structure.

3. Numerical Analysis: The models are solved using real-world data crawled from an online store selling computer peripherals. The planning horizon is set to 6 periods (two weeks each), covering both normal and booming demand periods (e.g., Singles' Day). Stability tests are conducted to validate the scenario generation and reduction techniques.

Key Contributions
The paper claims three primary contributions to the literature:

1. Novel Financing Mechanism: It is among the first to integrate "order-based loans"—a specific financing tool prevalent in Chinese e-commerce but under-researched in academic inventory literature—into a multi-period stochastic inventory model.
2. Stochastic Modeling Framework: It constructs a comprehensive stochastic multi-product inventory model that explicitly accounts for cash constraints, receipt delays, and the mechanics of order-based loans, utilizing scenario trees and reduction techniques to solve the problem.
3. Empirical Insights: Through numerical analysis based on real transaction data, the study identifies specific operational conditions under which order-based loans are beneficial, moving beyond theoretical assumptions to data-driven managerial insights.

Results
Numerical experiments yield the following findings:

• Stochastic vs. Deterministic: The stochastic model consistently outperforms the deterministic model in terms of expected final cash. The value of the stochastic solution (VSS) is particularly significant when the retailer is less cash-constrained (i.e., has higher initial cash or lower overhead costs).
• Loan Adoption Triggers: The retailer is more likely to benefit from using order-based loans when:
  • Initial Cash is Low: Small initial capital ($C_0$) makes the liquidity provided by the loan critical.
  • Receipt Delay is Long: Longer delays ($L$) exacerbate cash flow gaps, increasing the value of immediate financing.
  • Overhead Costs are Moderate: High overhead costs increase the need for liquidity, though if costs are too high, profitability is lost regardless of financing.
  • Demand is High: In booming demand periods, the ability to restock quickly via loans yields higher profits.
• Stability: The scenario generation and reduction methods demonstrated stability, with objective values varying by less than 5% across different scenario trees, validating the robustness of the solution approach.

Significance
The paper posits that while supply chain finance is a growing market, the specific mechanism of order-based loans has been overlooked in inventory optimization literature. By modeling this mechanism, the study provides a quantitative basis for retailers to decide when to leverage platform financing. The results suggest that order-based loans are not a universal solution but a strategic tool most effective for retailers facing liquidity constraints due to low initial capital or long payment cycles. The study validates that incorporating demand uncertainty through stochastic programming provides a distinct advantage over deterministic planning, particularly in the context of cash-constrained operations. Future research directions suggested by the authors include modeling demand substitution and exploring other supply chain financing behaviors within similar stochastic frameworks.