Integrity Review: Systemic risk in the Angolan interbank payment system – a network approach

Academic fraud report · Geng Detector

Summary

Verdict: Serious statistical anomalies flagged as potentially fabricated descriptive statistics (Chinese academic-fraud report by 'Geng' reviewer). Applied Economics 2020 paper by Borges, Ulica & Gubareva reports in Table 1 (p. 4905) standard deviations that are mathematically incompatible with the reported minima, maxima, and means for daily transactions on the Angolan RGTS. For 'number of transactions' (mean 733.24; min 465; max 1,805) the reported SD is only 15.15, implying values lie tens to ~70 SDs from the mean, which is incompatible with any plausible real distribution. Similar inconsistencies appear for settlement amounts (mean 30,440.60; SD 128.84; range >72,000) and connectivity (range 14.63 percentage points; SD 0.153). A secondary issue is a textual inconsistency: the prose cites total settlement of 1,887,317.41 million kwanzas while Table 1 lists 1,887,317.43. The reviewer also criticizes calling 'RStudio' a program rather than an IDE. Confidence in the SD anomaly is high; the prose/table mismatch and IDE wording are moderate indicators and could reflect careless editing rather than fraud.

Verdict

🔴 Serious concerns. Table 1 descriptive statistics are mathematically implausible to the point of strongly suggesting fabricated or heavily corrupted summary statistics. Secondary issues include a prose-vs-table numerical mismatch and imprecise software description.

Key findings

Statistically impossible standard deviations in Table 1 (p. 4905). Reported SDs are orders of magnitude too small relative to the reported ranges:
Number of transactions: mean 733.24, min 465, max 1,805 → reported SD 15.15. The minimum is ~17.7 SDs below the mean and the maximum is ~70.7 SDs above it.
Settlement amounts: mean 30,440.60, min 12,032.97, max 84,899.78 → reported SD 128.84. Extreme observations lie hundreds of SDs from the mean.
Connectivity: range = 32.619% − 17.989% = 14.63 percentage points → reported SD 0.153 (~95.6× smaller than the range).
Prose/table numerical inconsistency (p. 4905). Text states totals of "1,887,317.41 million kwanzas" while the corresponding Table 1 cell reads 1,887,317.43 million Kwanzas (last digit differs).
Methodological wording issue (p. 4906). Authors write they "use the RStudio program" to process data; RStudio is an IDE, not a statistical engine — sloppy but not inherently fraudulent.

Evidence highlights

DOI: 10.1080/00036846.2020.1751052
Table 1 (p. 4905) descriptive statistics for daily RGTS transactions.
Section IV (p. 4905) total-settlement sentence vs. Table 1 total cell.
Section V (p. 4906) mention of "RStudio program".

Notes

The SD figures are so inconsistent with min/max that random typing errors are implausible; a unit-of-measurement or decimal-point error (e.g., an extra factor of 10 or misplaced decimal) would not fully reconcile the transactions or amounts columns.
A robustness check would be to recompute SDs from the underlying daily data; the authors should release raw transaction-level data.
The numerical mismatch on totals and the IDE/program wording are weak indicators on their own but reinforce concerns raised by the SD anomalies.
Reviewer self-identifies as an AI-assisted 'Geng' report (耿同学打假); institutional confirmation is required before any formal finding of misconduct.