Integrity Review: "金融监管对中国实体经济增长的影响研究" (Research on the Impact of Financial Regulation on China's Real Economy Growth) — Highly Suspicious

Academic fraud report · Geng Detector

Summary

This report reviews the 2021 empirical economics paper "金融监管对中国实体经济增长的影响研究" by 王博峰, published in 财经论丛 (DOI: 10.13762/j.cnki.cjlc.2021.08.003). The overall verdict is "highly suspicious" (🟠), an upgrade from low risk due to three independent anomalies and one unresolved forensic clue. First, Sargan test p-values of exactly 1.0000 recur across multiple model specifications (Tables 2 and 3), which is unusual unless the models are exactly identified — unexplained in the methods section. Note that an earlier claim regarding Table 4 values (1.0004/0.7453) was found to be inaccurate and was not accepted. Second, Figure 1(b) shows two perfectly symmetric inverted-U curves differing only by horizontal shift; this has a benign explanation as standard theoretical diagramming, corroborated by a benign copy-move signal. Third, the sample size is internally inconsistent: 41 provinces over 2009–2019 implies 451 observations, yet all variables in Table 1 report n=410, with no explanation. Confidence is limited: no direct evidence of systematic data fabrication was found; the Bayesian posterior probability of fraud is 22% (95% CI: 6.9%–44.8%), BF = 5.35. Author clarification is recommended before any formal action.

Verdict

Highly suspicious (🟠). No direct evidence of systematic data fabrication was found, but three independent anomalies (Sargan test uniformity, theoretical diagram duplication pattern, sample size inconsistency) plus one unresolved image-forensic clue (noise variance) justify an elevated rating.

Key findings

Sargan test p-values uniformly 1.0000 across specifications (Tables 2 and 3) — Table 2, column (2) reports Sargan p-value = 1.0000, and all five rows of Table 3 report 1.0000. Repeated exact 1.0000 values across different specifications typically indicate exact identification or severely insufficient instruments; the paper does not state the identification setup. ⚠️ Note: an earlier citation of Table 4 values "1.0004" and "0.7453" does not appear in the paper (Table 4 actually reports ~0.3987 and ~0.4598); that portion of the original claim was rejected.
Figure 1(b) theoretical diagram: two perfectly symmetric inverted-U curves differing only by horizontal shift (page 5, moderating effect of financial efficiency). Forensics detected same-offset copy-move matching on this page; however, drawing two identical curves via horizontal translation of one function is standard theoretical practice, so this is flagged as having a benign explanation.
Panel sample size inconsistent — abstract and Section 3 state "41 provinces, 2009–2019"; Table 1 reports n = 410 for all variables. 41 × 11 years = 451, not 410 (410 would correspond to 41 × 10 years). The paper does not explain the discrepancy (missing year, sample attrition, or typo).
Unresolved forensic clue — noise variance inconsistency (anomalous blocks: 4, CV = 0.61) in img-000.png could not be cleared under review rules.

Evidence highlights

Sargan p-value = 1.0000 verified in Table 2, column (2) and all five rows of Table 3 (this portion verified against the source text).
Table 4 Sargan p-values in the paper are approximately 0.3987 and 0.4598; the previously cited 1.0004/0.7453 values do not exist in the original.
Table 1: all variables report n = 410; sample period stated as 2009–2019 for 41 provinces.
Copy-move forensics on Figure 1(b) page: offset (0,32), 51 matched block pairs; matches regular layout elements (table rows, axes), supporting a benign interpretation.
Bayesian synthesis: posterior 22% (prior 5%), 95% CI [6.9%, 44.8%], BF = 5.35; one clue excluded as benign. Top contributors: diagram symmetry (69.8%), noise variance (30.2%).
Dimensions cleared: no high-risk biomedical imagery; descriptive statistics plausible; no regression coefficient/SE contradictions; publication timeline consistent with data availability; citations (Barth et al. 2004; Aghion et al. 1990/2005) verified as consistent.

Notes

Recommended actions: contact the author to (1) clarify the GMM identification setup and instrument count, and (2) explain the 41-province/410-observation sample composition. No journal report is warranted at this stage; a public query could be considered if the author cannot provide raw GMM output and sample documentation.
AI-assisted report; false positives and negatives are possible. Final determination of misconduct requires official institutional investigation.