Investigation Report: "Upstream subsidy or downstream subsidy? A quantitative analysis of credit subsidy in China" (Economic Modelling, 2023)

Academic fraud report · Geng Detector

Summary

This report assesses a 2023 paper in Economic Modelling (DOI: 10.1016/j.econmod.2023.106539) by Song, Yang, and Zhou for potential academic integrity concerns. The overall verdict is highly suspicious (orange). Key findings include: (1) an implausibly rapid review timeline, with only 1 day between revised submission and acceptance and 5 days from acceptance to online publication; (2) suspiciously perfect calibration matches between reported data and model moments in Tables 3 and 5, where values align to two decimal places (e.g., 0.64, 1.34, 0.18, 0.14), with Table 5 'additional moments' appearing to duplicate Table 3 figures; (3) mechanically linear numerical patterns in Table 4 across policy parameter variations, atypical for heterogeneous-firm GE models; (4) text artifacts suggesting copy-paste drafting and absent proofreading; and (5) a minor inconsistency between δ=0.069 in the text and δ=0.07 in Table 2. These issues collectively raise concerns about editorial process irregularities, possible data fabrication, and insufficient quality control. Limitations: this assessment is based on text extraction only, without access to source code, raw data, or peer review records.

Verdict

Highly suspicious. Multiple convergent indicators—including an anomalously rapid peer-review turnaround, suspiciously perfect moment matches, mechanically linear simulation outputs, and minor text inconsistencies—warrant formal investigation by the journal and institution. However, no single finding is independently conclusive, and absence of source code prevents definitive determination of data fabrication.

Key findings

Implausibly fast review timeline: Revised submission on 20 September 2023; accepted on 21 September 2023 (1 day); available online 26 September 2023 (5 days post-acceptance). The 1-day revise-to-accept interval suggests the revised manuscript may not have undergone external re-review.
Perfect calibration alignment: In Table 3 (Moments for calibration) and Table 5 (Additional moments), reported Data and Model values match to two decimal places (0.64 vs 0.64; 1.34 vs 1.34; 0.80 vs 0.80; 0.18 vs 0.18; 0.14 vs 0.14).
Cross-table data duplication: Table 5 'Model' entries for 'SD of change in log value added' (0.64) and 'SD of log value added' (1.34) replicate the real-world labor data in Table 3 rather than serving as independent out-of-sample validation.
Mechanically linear simulation results: Table 4 shows near-constant step sizes (≈ -0.0042 to -0.0044) across variations of θ/θ* from 1.00 to 1.20, inconsistent with the diminishing marginal effects expected in heterogeneous-firm GE models.
Text extraction artifacts: Multiple fragmented citations suggest unpolished drafting (e.g., "follow-ing the methodology of 2012)", "1983), followed by1998),1999)").
Parameter inconsistency: Section 3.1 states δ = 0.069, while Table 2 lists δ = 0.07.

Evidence highlights

Timeline: Received 2 Nov 2022 → Revised 20 Sep 2023 → Accepted 21 Sep 2023 → Online 26 Sep 2023. Δ(revise→accept) = 1 day; Δ(accept→online) = 5 days.
DOI: 10.1016/j.econmod.2023.106539
Table 3 / Table 5 identical-value pairs: 0.64/0.64, 1.34/1.34, 0.80/0.80, 0.18/0.18, 0.14/0.14.
Table 4: Downstream X_u/X_d column exhibits step sizes in the narrow range -0.0042 to -0.0044 across θ/θ* ∈ {1.00, 1.05, 1.10, 1.15, 1.20}.
Parameter discrepancy: δ = 0.069 (text) vs δ = 0.07 (Table 2).

Notes

These findings are based exclusively on the published text and table values; source code (likely Dynare/Matlab), raw micro-data, and editorial correspondence were not available for verification.
The near-perfect calibration matches could in principle arise from overfitting a flexible calibration routine, but exact duplication between Table 3 'Data' and Table 5 'Model' columns is not explained by standard practice.
Linear patterns in counterfactual tables could reflect a stylized partial-equilibrium numerical setup, though this is atypical for full GE/DSGE frameworks claimed in the paper.
Independent confirmation would require: (i) release of replication code and data, (ii) editorial disclosure of review history, (iii) verification of Table 5 computations.
No image-based manipulations (e.g., Western blot reuse) could be assessed in this text-only review.