Equitable railway corridor investment under demand uncertainty: A two-stage distributionally robust bi-objective framework for sustainable regional development

Purpose. This study aims to synthesise empirical and modelling evidence on inventory optimisation methods for raw materials, work-in-process, and finished goods in production and trading enterprises, and to translate that evidence into a practical, class-differentiated implementation framework deployable within standard warehouse management and enterprise resource planning systems. Methodology. A systematic review and meta-analytic synthesis of 31 peer-reviewed studies published between 2004 and 2025 was conducted following the PRISMA 2020 protocol. A random-effects model estimated by restricted maximum likelihood was applied to pool percentage cost-reduction effect sizes across 18 studies admissible to quantitative synthesis, complemented by a narrative synthesis of the remaining 13 studies. Pre-specified subgroup and moderator analyses examined the role of inventory class, demand pattern, and network complexity as effect-size moderators. Results. Distributional safety stock methods outperform classical normal approximations by a pooled mean of 9.3% (95% CI: 5.8–12.7%) at equivalent service levels, with the advantage being largest for high-variability SKU segments. Multi-echelon coordination yields a pooled mean cost reduction of 11.4% (95% CI: 6.9–15.9%), increasing significantly with network complexity and lead-time variability. Learning-based control methods deliver up to 16% cost reductions under complex network conditions but require substantial data and governance infrastructure. Commercial demand drivers systematically distort finished-goods inventory targets and require integration with sales-and-operations planning for accurate calibration. Theoretical contribution. The study provides the first cross-class synthesis covering raw materials, work-in-process, and finished goods within a unified evaluative framework, positioning machine learning and deep reinforcement learning methods alongside classical policy families and quantifying the boundary conditions for each approach. Practical implications. A six-phase, stepwise implementation framework is proposed, covering ABC-XYZ segmentation, forecast model selection, safety stock calibration, replenishment policy assignment, simulation-based parameter tuning, and KPI governance, enabling enterprises to achieve 9–16% reductions in inventory costs within existing WMS and ERP architectures.

Summary

Main Finding

The authors propose a two-stage, distributionally robust bi-objective optimisation framework for allocating railway corridor investments that jointly (1) minimises spatial inequality (via a dynamic connectivity Gini coefficient) and (2) preserves budgetary efficiency (via ROI constraints), while protecting decisions against demand uncertainty using Wasserstein ambiguity sets. The model is reformulated into a tractable mixed-integer/convex program (via linearisations and DRO duality), solved with an ε‑constraint scalarisation to generate Pareto-optimal equity–efficiency trade-offs suitable for policy use.

Key Points

Objectives and trade-offs
- Primary equity objective: minimise a dynamic, demand-responsive connectivity Gini coefficient across n regions.
- Efficiency constraint/objective: enforce minimum return-on-investment (ROI) thresholds per corridor and/or maximise an efficiency metric combining ridership potential and economic spillovers.
- Bi-objective solved with ε‑constraint method to produce Pareto frontier of equity vs efficiency.
Uncertainty modelling
- Demand uncertainty handled via distributionally robust optimisation (DRO) with Wasserstein ambiguity sets around an empirical distribution of demand samples.
- Wasserstein radius θ is calibrated adaptively using a bootstrap-based empirical CDF of Wasserstein distances (user confidence level α controls conservativeness).
Mathematical formulation and tractability
- Two-stage structure: first-stage (here-and-now) budget allocations x; second-stage recourse y(ξ) after demand realisation, with stability constraint ‖y−x‖2 ≤ δ.
- Connectivity metric ci integrates investment, adaptive adjustments, and network spillovers; spillovers modelled by h_ij = min((x_i+y_i)/2, (x_j+y_j)/2) and linearised.
- Gini objective linearised via pairwise difference variables and Charnes–Cooper transformation to avoid division by variable mean connectivity.
- DRO worst-case expectation reformulated using duality (Esfahani & Kuhn style result) to a finite convex program: sup_{P∈P} E_P[Q] = inf_{λ≥0} {λθ + (1/N) ∑k sup{ξ∈Ξ} [Q(x,ξ) − λ‖ξ−ξ_k‖]}.
- Many nonlinear pieces (min, g_i spillover/value functions) approximated piecewise-linearly to retain mixed-integer convex tractability.
Policy design features
- Dynamic ROI constraints: expected returns (fares + monetised spillovers) over ambiguity set must exceed corridor-specific thresholds τ_i.
- Parameters allow policy tuning: weights η_i balancing ridership vs spillovers, β weighting first- vs second-stage investments, θ (ambiguity), α (confidence), δ (recourse stability).
- Authors position the method for developing economies and SDG objectives (reduce inequalities while ensuring fiscal sustainability).
Practical considerations and limitations noted
- Requires historical demand samples and credible estimates to build the reference distribution and spillover monetisation α.
- Computational burden scales with number of regions n and sample size N (inner supremums and linearisations increase problem size).
- Sensitivity to calibration choices (θ, ROI monetisation) and to quality/representativeness of historical demand data.
(Meta note) The file excerpt provided opens with an abstract that appears focused on inventory optimisation (likely a mismatched abstract); the body of the paper is about equitable railway corridor investment. Summary above follows the railway-investment content.

Data & Methods

Data requirements
- Historical demand samples {ξ1,...,ξN} for corridors/regions to form the empirical reference distribution P0.
- Corridor attributes: baseline connectivity measures, capacity/feasibility constraints A(ξ), fare structures p_i, parameters for economic spillovers g_i(·), budget B, ROI thresholds τ_i.
- Network topology (neighbour relations N(i)), policy weightings (w_k, η_i), and planner-specified confidence α for DRO calibration.
Core methods
- Two-stage stochastic programming with distributional robustness (Wasserstein ambiguity sets).
- DRO reformulation via strong duality (Esfahani & Kuhn 2018 type result) to convert worst-case expectation to tractable finite-dimensional convex subproblems.
- Linearisation/approximation techniques:
  - Pairwise absolute differences linearised with auxiliary variables for Gini.
  - Charnes–Cooper transform to handle division by mean connectivity.
  - Piecewise-linear approximations for spillover/value functions and min-operators.
- ε‑constraint scalarisation to convert bi-objective problem to a parametric single-objective optimisation (vary ε to trace Pareto front).
- Calibration: bootstrap resampling of historical demand to obtain empirical distribution of Wasserstein distances; choose θ as quantile FW^{-1}(1−α).
Computational aspects
- Resulting formulation is a mixed-integer convex program (linear/MILP components plus convex DRO terms) solvable with modern commercial solvers, but problem size and inner maximisations depend on N and polyhedral description of support Ξ.
- Authors report evaluation on synthetic and real-world networks (details not included in excerpt); empirical testing compares proposed DRO-bi-objective approach against deterministic and robust benchmarks.

Implications for AI Economics

Robust, data-driven investment decisions
- The framework demonstrates how DRO (Wasserstein) can be integrated with economic policy objectives (inequality, ROI) to make infrastructure investment decisions resilient to distributional shifts — a growing concern when historical data are poor predictors of the future.
- Useful for planners who must trade off efficiency and equity under model/data uncertainty; DRO gives a principled way to control conservativeness.
Opportunities for ML/AI integration
- Demand forecasting: machine learning models can supply richer empirical reference distributions P0 (or scenario generators) that feed into the DRO ambiguity set.
- End-to-end pipelines: GeoAI and spatial ML could estimate spillover functions g_i(·), connectivity-performance functions f_k(·), and network externalities from high-dimensional data, improving model fidelity.
- Adaptive policy/recourse: reinforcement learning or adaptive control could be used for second-stage recourse design (learning optimal y(·) policies over time as more data accumulate), complementing the two-stage formulation.
Economic and policy evaluation
- Provides a replicable tool to quantify and visualise equity–efficiency trade-offs (Pareto frontier), aiding transparent policymaking and stakeholder negotiation.
- Monetisation of spillovers (α) is central: careful economic valuation is needed because results and ROI constraints hinge on how externalities are converted to monetary terms.
Limitations and governance considerations
- DRO outcomes depend on sample representativeness and calibration choices (θ, α). Overly large θ yields conservative plans; too small θ risks fragility to distribution shifts.
- Requires institutional capacity: data infrastructure, model governance, and computational resources to implement DRO-based optimisation in practice.
- Ethical/political choices (how to weigh equity vs efficiency, how to monetise spillovers) remain normative and must be transparent.
Research directions relevant to AI economics
- Jointly learn P0 (or generative model of demand) with optimisation in a data-driven DRO loop; quantify value-of-data for reducing ambiguity radius θ.
- Study welfare implications across socioeconomic groups (beyond Gini) and integrate multi-dimensional equity metrics (access to jobs, services).
- Explore stochastic/control formulations with online learning for multi-period investments and dynamic reallocation as new demand data arrive.

If you want, I can: (a) extract the model’s mathematical formulation into a compact notation sheet; (b) draft a short policy brief translating the Pareto frontier outputs into decision rules for planners; or (c) outline how to couple this DRO framework with a specific ML demand forecasting pipeline. Which would you prefer?

Assessment

Paper Typereview_meta Evidence Strengthmedium — Synthesises multiple empirical and modelling studies under PRISMA 2020 and produces pooled effect-size estimates, improving precision over single studies; however, primary studies are heterogeneous in design (simulations, field pilots, observational implementations), sample sizes and quality vary, and potential publication/small-study bias and limited real-world randomized evidence constrain causal claims. Methods Rigorhigh — Follows PRISMA 2020, distinguishes quantitative and narrative streams, uses a random-effects REML estimator appropriate for heterogeneous effects, and reports subgroup/moderator analyses and confidence intervals; limitations are acknowledged (heterogeneity, infrastructure needs) but core meta-analytic methods and pre-specification indicate strong methodological execution. Sample31 peer-reviewed studies (2004–2025) on inventory optimisation in production and trading firms; 18 studies provided quantitative effect sizes for pooling, covering raw materials, work-in-process and finished goods across rule-based, distributional safety-stock, multi-echelon coordination and learning-based control (including ML/RL) interventions; remaining 13 studies included qualitative, modelling or implementation accounts; geographic and sectoral distribution not uniformly reported. Themesproductivity adoption IdentificationSystematic review and random-effects meta-analysis (REML) of percentage cost-reduction effect sizes reported or calculable from primary studies (18 studies pooled), supplemented with narrative synthesis of 13 additional studies; pre-specified subgroup and moderator analyses (inventory class, demand pattern, network complexity) used to examine heterogeneity and boundary conditions. GeneralizabilityMany primary studies are simulation-based or small-scale pilots rather than large randomized field trials, limiting external validity, Sector and firm-size coverage is uneven (results may not translate from manufacturing to retail/wholesale or from large firms to SMEs), Rapid evolution of ML/RL methods means older studies may not reflect current capabilities or data requirements, Effect estimates assume equivalent service-level constraints; results may differ under different service targets or contractual contexts, Required data, IT, and governance capacity for learning-based methods may be absent in many firms, constraining real-world uptake

Claims (10)

Claim	Direction	Confidence	Outcome	Details
A systematic review and meta-analytic synthesis of 31 peer-reviewed studies published between 2004 and 2025 was conducted following the PRISMA 2020 protocol. Other	null_result	high	number and coverage of studies included in the review	n=31 0.4
A random-effects model estimated by restricted maximum likelihood was applied to pool percentage cost-reduction effect sizes across 18 studies admissible to quantitative synthesis. Other	null_result	high	pooled percentage cost-reduction effect sizes	n=18 0.4
Distributional safety stock methods outperform classical normal approximations by a pooled mean of 9.3% (95% CI: 5.8–12.7%) at equivalent service levels. Firm Productivity	positive	high	inventory cost reduction at equivalent service levels	n=18 9.3% (95% CI: 5.8–12.7%) 0.4
The advantage of distributional safety stock methods is largest for high-variability SKU segments. Firm Productivity	positive	high	relative cost-reduction advantage of distributional safety-stock vs normal approximation across SKU variability segments	n=18 0.24
Multi-echelon coordination yields a pooled mean cost reduction of 11.4% (95% CI: 6.9–15.9%). Firm Productivity	positive	high	inventory cost reduction from multi-echelon coordination	n=18 11.4% (95% CI: 6.9–15.9%) 0.4
The cost reduction from multi-echelon coordination increases significantly with network complexity and lead-time variability. Firm Productivity	positive	high	magnitude of multi-echelon coordination cost reduction as a function of network complexity and lead-time variability	n=18 0.24
Learning-based control methods deliver up to 16% cost reductions under complex network conditions but require substantial data and governance infrastructure. Firm Productivity	positive	high	inventory cost reduction achieved by learning-based control methods; infrastructure/data requirements	up to 16% 0.24
Commercial demand drivers systematically distort finished-goods inventory targets and require integration with sales-and-operations planning for accurate calibration. Organizational Efficiency	negative	high	accuracy/calibration of finished-goods inventory targets	0.24
This study provides the first cross-class synthesis covering raw materials, work-in-process, and finished goods within a unified evaluative framework, positioning machine learning and deep reinforcement learning methods alongside classical policy families and quantifying the boundary conditions for each approach. Other	null_result	high	breadth and novelty of synthesis across inventory classes and methods	n=31 0.12
A six-phase, stepwise implementation framework (ABC-XYZ segmentation, forecast model selection, safety stock calibration, replenishment policy assignment, simulation-based parameter tuning, KPI governance) enables enterprises to achieve 9–16% reductions in inventory costs within existing WMS and ERP architectures. Firm Productivity	positive	high	expected inventory cost reduction achievable by implementing the proposed framework	9–16% reductions in inventory costs 0.04