## 1. Refinements and Extensions to the Baseline NIC Measure

## Source details

**Canonical URL:** [1. Refinements and Extensions to the Baseline NIC Measure](https://www.imf.org/-/media/files/publications/wp/2018/wp18195.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18195.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18195.pdf.json)

---

### Objectives and scope
- Construct big data, cross-country news flow indices of corruption (NIC) and anti-corruption (anti-NIC) by running country-specific search algorithms over more than 665 million international news articles.
- Offer methodological improvements including a more transparent and reproducible approach to measuring corruption-related news flow.
- Explore theoretical channels by which corruption-related news can affect macro-outcomes and test empirical evidence for those channels.
- Highlight how institutional strengthening through technology and capacity development can reinforce anti-corruption strategies.

### Definitions and measurement approach
- Corruption defined as “the abuse of public office for private gain.”
- NIC measures news flow about corruption; it does not directly measure corruption.
- Measures constructed via country-specific search algorithms applied to an extensive international news corpus, producing cross-country news flow indices (NIC and anti-NIC).

### Theoretical motivation
- Asset-price responses to public information: asset prices respond to newly public and relevant information (Fama (1970)); financial variables expected to respond to corruption-related news as measured by the NIC.
- Macro models with belief persistence (Kozlowski et al. (2017)): large shocks revise agents’ beliefs persistently; large NIC shocks can propagate through beliefs to affect financial variables and longer-term real macro-outcomes.

### Key empirical findings
- Corruption-related news has macro-relevance, especially when paired with institutional strengthening.
- Evidence of negative impacts from NIC shocks on:
  - borrowing costs
  - exchange rates
  - market outcomes
  - capital flows
  - fiscal balances
  - growth
- NIC shocks lower real GDP per capita growth by roughly 3 percentage points, cumulatively, over a two-year period.
- Panel estimates: NIC Shock coefficient: -0.019** (0.007); NIC Shock t-1: -0.011* (0.005); Anti-NIC Stock coefficient: -0.001 (0.003); Observations 596; Adjusted R-squared 0.937.
- Anti-NIC shocks alone do not produce statistically significant impacts on growth in panel estimations.
- Anti-NIC shocks coupled with capacity development (CD) can increase real GDP per capita growth:
  - At a 1-year lag, an additional FTE of CD increases growth by an average of 0.8 percentage points.
  - At a 2-year lag, growth increases by 3 percentage points if additional CD is coupled with anti-corruption efforts.
- Effects concentrated in EMDEs and “high” NIC economies; Advanced Economies and “low” NIC economies show smaller or insignificant impacts.

### Data, sample, and algorithmic construction
- Newspaper database: more than 650 million articles dating back to the mid-1980s; Factiva aggregator covering 36,000 sources, over 450 continuously updated newswires / 700 wires; 665+ million articles in English.
- Sample covers 30 countries from 1995 to 2017. Sample selection criteria included minimum journalistic coverage of 100 country-related articles per month, on average (normalizer cutoff metric of 1200 articles per year).
- Five-part metric for counting articles (country-specific search algorithms):
  - mention of the country within 8 words of a corruption-related term (baseline algorithm includes 18 corruption-related terms);
  - mention of government: include 22 government-related terms;
  - no mention of an anti-corruption effort for the NIC to ensure unidirectionality;
  - article length of at least 99 words;
  - limited to major newspapers with reputable credentials; own-country sources excluded.
- Normalization and indexing:
  - Counts divided by a normalizer: count of all articles that reference the country, are from major news sources, and are longer than 99 words.
  - Normalizers smoothed using moving averages.
  - Raw measures indexed from [0,100] across countries and time.
- Algorithm dictionaries (preserved terms): Baseline corruption list includes corrupt*, kleptoc*, nepotism, favoritism, rent-seeking, bribe*, graft, extortion, kickback*, grease payment*, facilitation payment*, facilitation fee*, illegal pact, solicitation, double-dealing, siphon* near2 funds, crony*, scam*. Core list: corrupt*, kleptoc*, nepotism, graft. Lobbying additions: lobbying, money politic*, private donation*, campaign financ*, influence peddling, revolving door, financial secrec*, state capture, clientelism. Government-related terms: government, regime, authorities, public sector, bureaucra*, agenc*, ministr*, public off*, administrat*, civil service, bureau or committee, cabinet, president, minister*, military, state-owned, SOE, parliament*, legislature, assembly, central bank, reserve bank.
- Anti-Corruption Algorithm centers on anti-corrupt*, anti-graft plus proximity to corrupt* or graft with qualifying reduction terms.

### Advantages of the NIC approach
- Uses publicly available and “vetted” news articles from major news sources; improves on big-data approaches relying on voluntary self-reporting or government procurement data.
- Higher-frequency variation than perception- or expert-based measures; timely and salient.
- Mitigates indicator-based identification problems where poor performance may reflect capacity gaps rather than corruption.
- Human audits inform algorithm revisions and assess accuracy.

### Human audits, robustness, and correlations
- Human audits: multiple rounds; final round includes four countries and 780 articles; altogether well over 1,500 article audits. Random sampling rounds reference 490 and 290 article samples with 95% confidence levels and +/- 5 confidence intervals.
- NIC audit summary (percent):
  - Correct Country? 96.1
  - About Corruption? 87.6
  - Of correct country & about corruption: Decrease? 3.1; Actual (vs. Perceptions)? 74.8; Domestic Government as Source? 95.5; Foreign Entity as Source? 3.5; Related to a Political Purge? 9.4
- Anti-NIC audit summary (percent):
  - Correct Country? 92.4
  - About Anti-Corruption? 82.0
  - Of correct country & about anti-corruption: Actual (vs. Perceptions)? 87.9; Specific (vs. General Efforts)? 49.3; Related to a Political Purge? 3.1
- Correlations with established measures (5-year averages, 2012-16):
  - NIC — CPI: -0.69
  - NIC — WGI Control of Corruption (CoC): -0.69
  - NIC — World Bank Bribery Incidence: 0.44 (significant at the 10 percent level)
- Correlation matrix highlights (Pearson’s correlations; * significance at 1% level):
  - NIC — Anti-NIC: 0.325*
  - NIC — Lobbying NIC: 0.996*
  - NIC — Core NIC: 0.973*
  - NIC — Civil Liberties Adjusted NIC: 0.888*
  - NIC — Press Freedom Adjusted NIC: 0.951*
  - Civil Liberties Adjusted NIC — Press Freedom Adjusted NIC: 0.965*
- Granger causality tests (selected):
  - CPI does not Granger cause NIC — Chi-Squared 5.135** Prob > Chi-Squared 0.023
  - CoC does not Granger cause NIC — Chi-Squared 2.713* Prob > Chi-Squared 0.100
  - NIC does not Granger cause CPI — Chi-Squared 0.009 Prob > Chi-Squared 0.924
  - NIC does not Granger cause CoC — Chi-Squared 0.001 Prob > Chi-Squared 0.977

### Stylized facts and time/level heterogeneity
- Time trends:
  - NIC increased on average over the last decade; Anti-NIC also trended upwards but lags NIC; gap widened.
- Levels by income group:
  - NIC higher in lower development levels: NIC considerably lower in Advanced Economies (AEs) compared to Emerging Markets (EMs) and Low-Income Developing Countries (LIDCs).
  - Civil Liberties Adjusted NIC ranges from 0 to 100.
- Sample descriptive statistics (selected exact values):
  - Overall NIC: Mean 7.9; Median 7.7; Std. Deviation 5.8; Maximum 24.8; Minimum 0.6.
  - AE's NIC: Mean 3.0; Median 1.9; Std. Deviation 3.0; Maximum 10.2; Minimum 0.6.
  - EM's NIC: Mean 10.0; Median 9.1; Std. Deviation 4.3; Maximum 18.1; Minimum 3.4.
  - LIDC's NIC: Mean 14.5; Median 10.2; Std. Deviation 9.0; Maximum 24.8; Minimum 8.4.
  - Anti-NIC average annual standard deviation: 4.33 (vs. 8.90 for NIC).
  - NIC average annual standard deviations: 11.20 for LIDCs; 7.29 for EMs; 1.79 for AEs.

### Theoretical framework: news, beliefs, and macro-outcomes
- Foundations: semi-strong efficient market hypothesis and behavioral economics (persistence of media impacts).
- Large and rare NIC shocks lead agents to revise beliefs persistently; agents construct a kernel density estimator of underlying corruption distribution; larger shocks produce more persistent belief shifts.
- Implication: large corruption shocks can generate long-run changes in investor behavior and macro outcomes via persistent belief shifts.

### Empirical strategy and event-study evidence
- Three-stage empirical approach:
  - Stage 1: correlations between NIC and institutional indicators.
  - Stage 2: event studies at monthly, quarterly, and annual frequencies using NIC Stock and NIC Shock (shocks defined as increases in NIC Stock more than two standard deviations above economy-specific means). NIC Stock constructed as cumulative sum of ∆NIC with a 2-year lead-in; changes in NIC are first differences in logs. Lead-in: 1995-96; sample for event studies and panels: 1997-2017.
  - Stage 3: dynamic panel fixed effects models controlling for country and time fixed effects, lagged dependent variables, Anti-NIC Stock, and standard logged controls; standard errors clustered by country.
- Event study highlights (selected pre/post t-test values reported in source):
  - Stock returns: abnormal returns negative initially; volatility spikes remain above pre-period average for at least six months.
  - Bond spreads (two- and five-year): pre/post t-test values: 0.35 and 2.61, respectively.
  - Nominal effective exchange rates: pre/post t-test value: 3.40 (depreciation).
  - Capital flows (quarterly): pre/post test values: 0.097 for portfolio flows, 0.546 for FDI.
  - Fiscal outcomes (annual): revenues, tax revenues, expenditures, and debt pre/post t-test values: 3.512, 2.331, 2.499, and 2.888 respectively.
  - Real GDP per capita growth: pre/post t-test value significant at 3.146; growth negative in the year following a NIC shock and persistently subdued thereafter.

### Panel results, heterogeneity, and robustness
- Full-sample panel (with lagged dependent variables, select coefficients preserved):
  - NIC Shock (Real GDP Per Capita Growth): -0.0187** (0.00740)
  - NIC Shock t-1: -0.0110* (0.00541)
  - Observations: 596; Adjusted R-squared: 0.937.
- Sub-sample (by income/NIC level) — Dependent Variable: Real GDP Per Capita Growth (selected columns):
  - Advanced (Observations 220, Adjusted R-squared 0.890): NIC Shock: -0.009 (0.007); NIC Shock t-1: -0.007 (0.007).
  - EMDEs (Observations 376, Adjusted R-squared 0.944): NIC Shock: -0.023* (0.012); NIC Shock t-1: -0.012 (0.008).
  - High NIC (>75th Percentile) (Observations 140, Adjusted R-squared 0.967): NIC Shock: -0.028*** (0.007); NIC Shock t-1: -0.019** (0.007).
  - Low NIC (<25th Percentile) (Observations 140, Adjusted R-squared 0.924): NIC Shock: -0.002 (0.006); NIC Shock t-1: -0.006 (0.010).
- Robustness checks:
  - Results robust to excluding lagged dependent variable, controlling for rule of law and government effectiveness, and controlling for level effects.
  - Using simple NIC Stock (marginal changes) instead of NIC Shock produces insignificant panel results — consistent with theory that large shocks drive effects.
  - Some standard errors computed with as few as seven clusters; bootstrapped or jackknifed standard errors also yield significance at the 5 percent level in some specifications.
- Additional main-panel coefficients (select from Appendix V):
  - NIC Shock (Revenues/GDP): -0.00308 (0.00408)
  - NIC Shock (Portfolio Investment/GDP): 0.0337 (0.0798)
  - NIC Shock (Direct Investment/GDP): -0.0318 (0.0665)
  - NIC Shock t-1 (Real GDP Per Capita Growth): -0.0110* (0.00541)

### Anti-NIC evidence and dynamics
- Event study evidence:
  - Anti-NIC shocks show positive impacts on financial variables: decreasing stock market volatility and appreciating exchange rates. Pre/post t-test results: stock market volatility 4.946; exchange rate movements 1.713; change in debt/GDP 1.352.
  - Change in portfolio flows/GDP insignificant in immediate post-shock period.
- Panel estimation (Table 5, Dependent Variable: Real GDP Per Capita Growth):
  - Anti-NIC Shock: -0.005 (0.008)
  - Anti-NIC Shock t-1: 0.002 (0.005)
  - NIC Stock: -0.004 (0.004)
  - Observations: 596; Adjusted R-squared: 0.936.
- Interpretation:
  - Anti-NIC shocks have an insignificant immediate impact on growth in panels.
  - Persistent anti-corruption news flow correlated with lower NIC levels over time: 3-year episodes of anti-corruption news flow split into quintiles show that fourth and fifth quintiles yield a reduction in the NIC five years later (magnitude limited; causality not established).
  - Indirect channel: persistent anti-corruption news flow reducing NIC may be more relevant for medium-term macro benefits than immediate orthogonal macro effects of anti-corruption measures.

### Capacity development (CD), IMF FTEs, and interaction with anti-NIC
- Empirical evidence on CD x Anti-NIC interaction (Box 2 and Table 6):
  - Table 6 — Capacity Development x Anti-NIC Nexus (Full Sample Results), Dependent Variable: Real GDP Per Capita Growth:
    - Total FTEs t-1: 0.008** (0.004)
    - Total FTEs t-2: -0.004 (0.004)
    - Total FTEs x Anti-NIC Shock t-1: -0.012 (0.008)
    - Total FTEs x Anti-NIC Shock t-2: 0.031* (0.018)
    - Observations: 229; Adjusted R-squared: 0.899.
- Department-level FTE panel highlights (Table 5 examples):
  - FAD FTEs t-1: 0.00259 (0.00438)
  - MCM FTEs t-1: 0.00639 (0.0105)
  - STA FTEs t-1: 0.0777 (0.0550)
  - LEG FTEs t-1: 0.00579* (0.00303)
  - All FTEs t-1: 0.00817** (0.00366)
  - Observations: 229.
- Interpretation and caveats:
  - Capacity development, when consistently supported by authorities and embedded in a broader reform package, can positively impact AC efforts and growth.
  - Empirical estimates sensitive to specification; FTE measure does not distinguish quality of CD.

### Policy implications and anti-corruption interventions
- For beliefs to shift with associated economic gains, governments must go beyond cheap talk; AC efforts must be supported by meaningful policy changes and remedies to governance vulnerabilities.
- Traditional AC recommendations (derived from principal-agent literature):
  - Increase likelihood and severity of detection and penalties.
  - Reduce discretion of agents.
  - Enhance transparency.
  - Address underlying economic, structural, or political problems where corruption may function as “greasing the wheel.”
- Modern avenues emphasized:
  - Adoption of modern technology and capacity development as complements to traditional AC measures.
  - Digitalization channels: enhancing accountability, increasing transparency, removing middlemen, enabling faster detection via open data, improving public service delivery, enhancing financial access and inclusion.
- Digitalization caveats:
  - Examples show initial NIC drops post-technology adoption but NIC tends to rise again; pre/post t-test statistic for selected technological initiatives: t-stat: 2.186.
  - Successful digitalization requires access to technology, technological literacy, social and technology readiness, citizen participation, legal frameworks, organizational processes, leadership and campaigns.

### Selected country case studies — qualitative lessons
- Indonesia:
  - Post-Suharto reforms; Anti-Corruption Commission (KPK) founded 2003 with broad independent powers; anti-NIC picked up significantly higher anti-corruption coverage; NIC improved after major AC gains.
  - Conditions facilitating KPK included severe crisis, commitment of reformers, societal backlash, and donor-supported capacity development.
- Malaysia:
  - MACC established under Malaysian Anti-Corruption Commission Act 2009; complementary initiatives include e-government, National Integrity Plan (2004), Government Transformation Program (2009), Whistleblower Protection Act 2010, MyProcurement portal.
  - NIC lower than sample average for most of sample period; 1MDB scandal increased both NIC and anti-NIC coverage.
- Singapore:
  - Holistic anti-corruption strategy: independent judiciary, stringent laws, effective enforcement, responsive administration; CPIB established 1952; PCA enacted 1960 and regularly revised.
  - Public administration reforms: high salaries benchmarked to private sector, strict conduct regulations, mandatory asset declarations, caps on unsecured debt, streamlined procedures, whistle-blowing policies.
  - Result: sustained low NIC relative to sample.
- Case study CD examples (Box 4):
  - Philippines: IMF FAD-supported revenue administration reforms; by 2016 tax revenue returned to pre-GFC levels; NIC and anti-NIC measures show little improvement over time.
  - Indonesia: large IMF TA in revenue administration, PFM, banking supervision; NIC declined over 2010-16 while institutional strength and perceptions improved.
  - Nepal: FAD TA improved tax revenue from 8.9 percent of GDP in 2006 to 18.9 in 2016; institutional strength remained low and perception of corruption unchanged.

### Limitations, methodological caveats, and future research
- Limitations:
  - Comparability affected by press freedom/repression; NIC may understate corruption in restricted speech environments. Mitigations: remove own-country sources; adjust using Freedom House Civil Liberties and Press Freedom indices (Civil Liberties Adjusted NIC; Press Freedom Adjusted NIC).
  - Algorithms may pick up private crime or legal (“lobbying”) activity; Core and Lobbying-adjusted NICs explore tradeoffs.
  - English-language only articles; own-country sources excluded.
  - Duplicate reprints retained as signal of story importance.
  - Single scandals in low-coverage countries can generate spikes (high noise-to-signal).
- Identification concerns:
  - Event studies cannot establish causality due to omitted variables; panels control for fixed effects but endogeneity and measurement error concerns remain.
  - Measurement error likely attenuates panel estimates (conservative bias).
  - Nickell bias from dynamic panels; Arellano-Bond GMM inappropriate given small N (30).
  - Traditional instruments for corruption weak for news-flow proxying.
- Future research avenues:
  - Causal identification of channels between corruption news shocks and growth.
  - Classification of NIC and anti-NIC shock types.
  - Analysis of technological and social media roles in anti-corruption.
  - More nuanced CD measures and quality assessments.

### Conclusions (paper-wide)
- Constructed first cross-country news flow measures of corruption and anti-corruption (NIC and anti-NIC) capturing high-frequency variation, scandals and longer-term relative levels without relying on local experts.
- NIC shocks negatively impact financial outcomes and longer-term real outcomes (growth) persistently; impacts more pronounced in EMDEs with capacity weaknesses.
- Anti-NIC shocks alone are insufficient for sustained positive macro outcomes; persistent anti-NIC news flow may reduce NIC over time, and coupling anti-NICs with capacity development yields larger growth gains.
- Policy implication: anti-corruption plans must be paired with meaningful institutional reform and long-term capacity development frameworks, especially in EMDEs and fragile states; modern technology and technical assistance can support these efforts if tailored to country context.

*Source: wp18195*

### 1. Refinements and Extensions to the Baseline NIC Measure ________________________13

### 1. Refinements and Extensions to the Baseline NIC Measure

### Objectives and scope
- Construct big data, cross-country news flow indices of corruption (NIC) and anti-corruption (anti-NIC) by running country-specific search algorithms over more than 665 million international news articles.
- Offer methodological improvements including a more transparent and reproducible approach to measuring corruption-related news flow.
- Explore theoretical channels by which corruption-related news can affect macro-outcomes and test empirical evidence for those channels.
- Highlight how institutional strengthening through technology and capacity development can reinforce anti-corruption strategies.

### Definitions and measurement approach
- Corruption is defined as “the abuse of public office for private gain.”
- The NIC measures news flow about corruption; it does not directly measure corruption.
- Measures are constructed via country-specific search algorithms applied to an extensive international news corpus, producing the first cross-country news flow indices of corruption and anti-corruption (NIC and anti-NIC).

### Theoretical motivation
- Two relevant strands of literature motivate the empirical approach:
  - Asset-price responses to public information (Fama (1970)): asset prices respond immediately and with the correct magnitude to newly public and relevant information; financial variables are expected to respond to corruption-related news as measured by the NIC.
  - Macro models with belief persistence (Kozlowski et al. (2017)): large shocks affect agents’ beliefs persistently; large shocks to the NIC revise assessments of corruption and propagate through beliefs, affecting financial variables and longer-term real macro-outcomes.

### Key empirical findings
- Corruption-related news has macro-relevance when paired with institutional strengthening.
- Evidence of negative impacts from NIC shocks on:
  - borrowing costs
  - exchange rates
  - market outcomes
  - capital flows
  - fiscal balances
  - growth
- NIC shocks lower real GDP per capita growth by roughly 3 percentage points, cumulatively, over a two-year period.
- Anti-NIC shocks, on their own, do not produce statistically significant impacts on growth.
- When anti-NIC shocks are coupled with capacity development efforts, real GDP per capita growth significantly and cumulatively increases over a three-year period.
- The positive interaction between anti-NIC shocks and capacity development is sensitive to model specification but suggests capacity development can reinforce anti-corruption efforts.

### Methodological and conceptual contributions
- Introduces big-data news-flow measures (NIC and anti-NIC) that are transparent and reproducible.
- Provides a framework linking news flow about corruption to financial market reactions and longer-term macro outcomes via belief formation and persistence.
- Uses country-specific search algorithms and extensive international news coverage to overcome limitations of perception-based and expert-based corruption indicators.

### Organization of the chapter
- Section II surveys the literature on the macro-relevance of corruption and summarizes the measures’ contributions to the literature.

*Source: wp18195 - 1. Refinements and Extensions to the Baseline NIC Measure*

### Section III describes the construction of the news flow indicator and its strength relative to

### wp18195 - Section III describes the construction of the news flow indicator and its strength relative to

### Overview of document structure
- Section III describes the construction of the news flow indicator and its strength relative to existing indicators.
- Section IV provides the basic theoretical framework for assessing the macro-relevance of corruption news as well as some empirical findings on the link between the NIC and financial and real macro variables.
- Section V focuses on anti-corruption efforts, using our complementary anti-NIC indicator as well as country case studies and a review of the use of technology and capacity development.
- Section VI concludes.

### Channels through which corruption affects the macroeconomy
- Public finances
  - Revenue collection—in particular tax revenues—can be lower through corruption’s impact on tax compliance and tax evasion.
  - Corruption can weaken quality of public expenditure and public investment via distorted procurement processes, off-budget financing, misallocated budgets, and cost inflation.
  - Taken together, corruption can increase fiscal deficits and lead to substantial debt accumulation; lower government revenues and wasteful public spending can harm investment, distort economic activity, and could lead to debt crises.
- Monetary and financial stability
  - Corruption can raise inflation by fostering fiscal dominance and eroding central bank operations and monetary policy.
  - Corruption can weaken financial market policies and oversight, negatively affecting financial development and inclusion, lending practices and banking supervision.
  - Perception of corruption can increase borrowing costs via a higher default risk.
- Growth
  - Corruption can reduce incentives for productive activity, encourage rent-seeking, and stimulate emigration of the highly educated workforce (brain drain).
  - Distorted public spending can undermine social services, education and health spending; lower productivity and reduced innovation; hinder new firm entry and force market exits.
  - Corruption has been shown to reduce private investment, including FDI.
  - Countervailing “greasing the wheel” hypothesis exists in the literature (corruption as a second-best solution), but most empirical literature does not give strong support; some studies show non-linear relationships depending on institutional quality.
- Trust
  - Corruption weakens citizens’ trust in institutions and processes, which can reduce financial inclusion and stock investment.
  - Low cross-country trust is associated with lower trade, direct investment, and portfolio investment.
  - In extreme cases, corruption can undermine trust in government, incite civil unrest and conflict, increasing uncertainty that harms investment and economic activity.

### Measuring corruption — three generations of measures
- First generation
  - Relied on surveys measuring expert and citizens’ perceptions of prevalence and nature of corruption (examples: Transparency International’s CPI; World Bank’s Control of Corruption).
- Second generation
  - Used victimization surveys and bureaucratic quality indicators (e.g., tax collection efficiency or fiscal transparency).
  - Limitations: still subjective/sticky, some providers not transparent, may reflect bureaucratic capacity rather than corruption.
- Third generation (big data)
  - Uses large databases and online sources (examples: India’s IPaidABribe.com; procurement analyses identifying contractual outliers).
  - Advantages: less subjective, transparent, reproducible; higher frequency; less sticky.
- NIC and anti-NIC
  - Contribute to the third generation: first big data, cross-country news flow indices of corruption (NIC) and anti-corruption (anti-NIC).
  - Improve on other big data approaches by not relying on government procurement data or unvetted citizen reports.
  - Allow construction of anti-NIC to assess magnitude and persistence of anti-corruption efforts.
  - Algorithms ensure greater nuance and unidirectionality and are vetted by human audits.

### Construction of the NIC (measurement details)
- Data sources and sample
  - Newspaper database of more than 650 million articles dating back to the mid-1980s.
  - Sample covers 30 countries from 1995 to 2017.
  - Uses Dow Jones’s Factiva news aggregator, which covers over 36,000 sources, including almost 700 newswires and major news sources.
  - Sources include digitized newspapers dating to the 1980s and newspapers’ online websites.
- Country-specific search algorithms and five-part metric
  - Mention of the country within 8 words of a corruption mention. Baseline algorithm includes 18 corruption-related terms.
  - Mention of government: include 22 government-related terms to capture abuse of public office for private gain.
  - No mention of an anti-corruption effort to ensure unidirectionality (i.e., pick up incidents of corruption, not responses).
  - Article length of at least 99 words to exclude “ticker” articles and to allow human audits.
  - Limited to major newspapers with reputable credentials.
- Anti-corruption algorithm
  - Similar to NIC but allows mention of both anti-corruption and corruption.
  - Algorithms are in English.
- Normalization, smoothing, indexing
  - Counts of articles meeting algorithm divided by a normalizer: count of all articles that reference the country, are from major news sources, and are longer than 99 words.
  - Normalizers smoothed using moving averages to reduce influence of temporary spikes in world interest.
  - Raw measures are indexed from [0,100] across countries and time; higher levels imply more news flow on corruption or anti-corruption.
- Sample of economies included (as reported)
  - Argentina, Australia, Bangladesh, Brazil, Cambodia, China, Colombia, Egypt, Germany, Hong Kong SAR, India, Indonesia, Italy, Japan, Korea, Lebanon, Malaysia, Mexico, New Zealand, Philippines, Russia, Singapore, South Africa, Spain, Sri Lanka, Thailand, Turkey, United Kingdom, United States, and Vietnam.
  - Sample selection criteria included minimum journalistic coverage of 100 country-related articles per month, on average.

### Advantages of the NIC approach
- Uses publicly available and “vetted” news articles from major news sources, improving on big data approaches that rely on voluntary self-reporting or government-provided procurement data.
- Reflects higher-frequency changes in measured corruption than perception- or expert-based measures.
- Mitigates indicator-based identification problems where poor performance may reflect capacity gaps rather than corruption.
- Timely: press exposes corruption quickly; news stories are released promptly when vetted.
- Salient: daily newspaper stories can influence economic actors’ beliefs and decisions in real time.
- Broader pool of assessments: draws on many sources, reducing overreliance on a small number of experts or local sources.
- Accuracy can be assessed via human audit that examines whether counted articles reflect actual or perceived corruption, country relevance, magnitude, and type of corruption, and political purges; audits informed algorithm revisions.

### Limitations and robustness concerns
- Comparability across countries may be affected by differences in press freedom or repression; NIC may understate corruption in countries with restricted speech.
  - Mitigations: removal of own-country sources; adjustment of baseline measure using time-varying measures of civil liberties and press freedom.
- Algorithms may pick up private criminal activity rather than public corruption due to overlapping vocabulary.
- Distinction between legal and illegal corruption may reduce measurement of phenomena such as state capture or excessive lobbying in advanced markets.
- In countries with low baseline coverage, single scandals can generate spikes in NIC (high noise-to-signal ratio).
  - Theoretical framing focuses on impact of shocks to corruption coverage, which can affect beliefs and behavior regardless of whether coverage perfectly reflects underlying corruption.
- Additional robustness steps and refinements summarized in Box 1 (not reproduced here).
- Language and source restrictions
  - Only English-language articles used.
  - Own-country sources excluded due to concerns about domestic bias.
- Duplicate articles (wires republished across outlets) are not removed because proliferation of reprints reflects story importance.
- Compositional shifts in database coverage over time addressed cautiously: moving averages smooth but do not remove statistically identifiable outliers that may reflect new sources.

*Source: wp18195 (Section III and related sections as provided).*

### Box 1. Refinements and Extensions to the Baseline NIC Measure

### Box 1. Refinements and Extensions to the Baseline NIC Measure

### Refinements to the baseline index
- Press-coverage bias addressed by two adjustments applied after the baseline index is constructed:
  - Remove “own country” sources from each country’s search algorithm.
  - Use Freedom House’s Civil Liberties and Press Freedom index data to upweight the baseline measure for years when countries are characterized by low press freedom and/or high repression.
  - Resulting adjusted measures: Civil Liberties Adjusted NIC and Press Freedom Adjusted NIC.
- Rationale for removing domestic sources:
  - In repressed societies citizens are unlikely to feel free to express themselves to domestic or foreign press.
  - Domestic media can over-report rare instances of corruption when country norm is absence of corrupt incidents (Rizzica and Tonello (2018) evidence).
- Search algorithm refinements:
  - Dictionaries made context-specific by adding words based on country experts and removing words with different cultural meanings (e.g., “patronage”) based on a human audit.
  - International coverage advantage: journalists tend to use similar language in describing corruption, reducing sensitivity to local language norms.
- Excluding false positives and private crime:
  - Removed “wobbler” words that might reference ethically problematic but not corrupt activities (e.g., tax haven, offshore financial center, illicit financial flows).
  - Human audit results indicate the algorithm largely captures public corruption rather than solely private criminal activity.
- Minimum information threshold:
  - Exclude countries with limited press coverage defined by 100 generic articles per month on average (normalizer cutoff metric of 1200 articles per year).

### Extensions
- Core Corruption NIC:
  - Narrow corruption dictionary to: corrupt*, kleptoc*, graft, and nepotism to explore tradeoff between coverage breadth and false positives.
- Lobbying Adjusted NIC:
  - Add terms related to lobbying and state-capture to capture “legal” corruption more likely in advanced markets.
- Purpose:
  - Extensions buttress and test robustness of the baseline index.

### Robustness of the NIC (human audits, comparisons, correlations)
- Human audits:
  - Multiple rounds performed; final round includes four countries and 780 articles; altogether well over 1,500 article audits.
  - Random sampling for audit rounds: hundreds of articles per selected country to ensure results have a 95% confidence level.
  - Specific audit sample referenced in Table 1: random sampling of 490 articles; 95% confidence level with +/- 5 confidence interval.
- Human audit findings (Table 1; in percent):
  - Correct Country? 96.1
  - About Corruption? 87.6
  - Of correct country & about corruption...
    - Decrease? 3.1
    - Actual (vs. Perceptions)? 74.8
    - Domestic Government as Source? 95.5
    - Foreign Entity as Source? 3.5
    - Related to a Political Purge? 9.4
- Summary of audit conclusions:
  - For the NIC: 96 percent of articles are both about the correct country and 88 percent are about corruption. Of articles that are about the correct country and about corruption, only 3 percent are about declines in corruption and articles largely pick up coverage of actual corruption—not simply perceptions. Domestic government tends to be the source of corruption and corruption coverage is only alleged to be related to political purges in 9 percent of audited articles.
  - For the anti-NIC: over 80 percent of articles are both about the correct country and about anti-corruption efforts; of those, 88 percent are about actual efforts (vs. perceptions); only 3 percent referenced political purges.
- Cross-checks and comparisons:
  - Spikes and relative levels in the NIC correspond with country experts’ priors.
  - Ordering across baseline and extensions remains largely the same; close correlations across measures (Appendix 3, Table 1).
- Correlations with established corruption measures (5-year averages, 2012-16):
  - Correlation between NIC and Transparency International’s Corruption Perceptions Index (CPI): -0.69.
  - Correlation between NIC and WGI Control of Corruption (CoC) measure: -0.69.
  - NIC is also correlated with World Bank’s Bribery Incidence measure: correlation .44 (significant at the 10 percent level).
- Time variability:
  - NIC reflects far more variability across time relative to CPI and CoC (NIC registers upward and downward shocks that CPI and CoC may not capture).

### Stylized facts from NIC and anti-NIC
- Time trends:
  - On average, the NIC has increased over the last decade.
  - Anti-NIC has also trended upwards but lags greatly behind NIC; the gap between NIC and anti-NIC has widened.
- Levels by income group:
  - NIC is higher in lower development levels: NIC considerably lower in Advanced Economies (AEs) compared to Emerging Markets (EMs) and Low-Income Developing Countries (LIDCs).
  - NIC ranges from 0 (low corruption news flow) to 100 (high corruption news flow) in Civil Liberties Adjusted NIC.
- Heterogeneity across groups and time variation:
  - LIDCs: large improvements in NIC levels in recent period; marked decrease compared to 1990s, hitting a 15 year low in 2016, with heterogeneity across LIDCs.
  - AEs: consistently low NIC across period with slight uptick starting in 2016.
  - EMs: moderate corruption-related news flow with major spikes from 1999-2002 and 2009-2015; upward trend 2009-2015 coinciding with period of large capital inflows into EMs.
- Commodity exporters:
  - NIC on upward trend in commodity exporters following the global financial crisis (increase 2009-2012).
- Anti-NIC:
  - Levels remain low across the sample period irrespective of income levels.
  - Exception: large uptick in LIDCs between 2006-2008 driven primarily by one country’s anti-corruption news flow (arrest of thousands of officials).
  - EMs saw a slight increase in anti-NIC between 2012 and 2016.
- Average annual standard deviations (sample averages noted in text):
  - NIC average annual standard deviations: 11.20 for LIDCs; 7.29 for EMs; 1.79 for AEs.
  - Commodity and non-commodity exporters average annual standard deviations: 7.50 and 8.63, respectively.
  - Anti-NIC average annual standard deviation: 4.33 (vs. 8.90 for NIC).

### Theoretical framework: news, beliefs, and macro-outcomes
- Theoretical foundations:
  - Draws on semi-strong efficient market hypothesis (Fama, 1970) and behavioral economics (persistence of media impacts).
  - Expected return equation linking changes in NIC to security prices provided (equation notation preserved in source).
- Belief formation:
  - Large shocks to news (NIC) can have persistent effects on beliefs about underlying corruption prevalence.
  - Agents construct kernel density estimator of underlying corruption distribution; larger and rarer shocks produce more persistent belief shifts.
- Implication:
  - Large corruption shocks can generate long-run changes in investor behavior and macro outcomes via persistent belief shifts.

### Empirical approaches and key results
- Empirical strategy overview:
  - Stage 1: correlations between NIC and institutional indicators.
  - Stage 2: event studies at monthly, quarterly, and annual frequencies using NIC Stock and NIC Shock (shocks defined as increases in NIC Stock more than two standard deviations above economy-specific means).
    - NIC Stock constructed as cumulative sum of ∆NIC with a 2-year lead-in; changes in NIC are first differences in logs. Lead-in: 1995-96; sample period for event studies and panels: 1997-2017.
  - Stage 3: panel estimations (dynamic panel fixed effects models), controlling for country and time fixed effects, lagged dependent variables, Anti-NIC Stock, and standard logged controls; standard errors clustered by country.
- Stage 1 institutional correlations:
  - Higher NIC correlated with lower Global Competitiveness Index, lower Regulatory Quality (WGI), lower Ease of Doing Business, less judicial independence (Linzer & Staton), and worse Legatum Prosperity Index; relationships significant based on Pearson’s correlation tests and robust to controlling for income per capita (see Appendix 3).
- Stage 2 event study findings (selected):
  - Monthly frequency:
    - Abnormal stock returns negative in initial period and show mean reversion to zero once shock is priced in.
    - Stock market volatility spikes significantly after a NIC shock and remains above pre-period average for at least six months.
  - Bond yields:
    - Borrowing costs tend to rise after a NIC shock, followed by considerable volatility (two- and five-year bond spread pre/post t-test values: 0.35 and 2.61, respectively).
  - Exchange rates:
    - Nominal effective exchange rates tend to significantly depreciate after NIC shocks (pre/post t-test value: 3.40).
  - Capital flows (quarterly):
    - Portfolio and FDI flows fall following NIC shocks, with more negative level impact on portfolio flows and increased volatility for both (pre/post test values: 0.097 for portfolio flows, 0.546 for FDI; immediate period for portfolio flows shows significantly lower post-shock values).
  - Fiscal outcomes (annual):
    - NIC shocks correlated with weakening fiscal positions: revenues, tax revenues, expenditures, and debt show pre/post t-test significant changes (3.512, 2.331, 2.499, and 2.888 respectively).
  - Real GDP per capita growth:
    - Real GDP per capita growth is negative in the year following a NIC shock and persistently subdued thereafter (pre/post t-test value significant at 3.146).
- Stage 3 panel estimation (focus on growth):
  - Panel specification:
    - y_it = α_i + δ_t + β1 NIC Shock_it + β2 Anti-NIC Stock_it + ∑ θ_k X_it^k + ε_it
    - Standard errors clustered by country.
  - NIC shocks have significantly negative impact on real per capita growth:
    - NIC Shock coefficient: -0.019** (standard error (0.007)).
    - NIC Shock t-1 coefficient: -0.011* (standard error (0.005)).
    - Anti-NIC Stock coefficient: -0.001 (standard error (0.003)).
    - Observations 596.
    - Adjusted R-squared 0.937.
    - Country & Year FE: Yes.
    - Full Set of Controls & Lagged Dependent Variables: Yes.
    - Standard errors clustered by country in parentheses; constant included; * p<.10, ** p<.05, and *** p<.01. There are 30 clusters.
  - Interpretation:
    - Negative NIC shocks lower real GDP per capita growth by roughly 2 percentage points concurrently and by 1 percentage point at a one-year lag, reflecting persistent effects.
    - Results robust to excluding lagged dependent variable, controlling for rule of law and government effectiveness, and controlling for level effects.
    - Using simple NIC Stock (marginal changes) instead of NIC Shock produces insignificant panel results, consistent with theory that large shocks (scarring) drive the effects rather than small marginal changes.
  - Heterogeneity:
    - NIC shocks have smaller and insignificant impact in Advanced Economies compared to EMDEs.
    - Non-linearity: negative impacts driven by “high” NIC economies (mean NIC above sample’s 75th percentile). Effects insignificant in “low” NIC economies (mean NIC below sample’s 25th percentile).
  - Other outcomes:
    - NIC shocks do not have significant effects on revenue/GDP and capital flows once additional variables are controlled for in panels (appendix results).
    - Coefficients on Anti-NIC stock are insignificant, likely reflecting heterogeneity in size and quality of anti-corruption efforts across countries.
- Methodological notes and caveats:
  - Event studies cannot establish causality due to omitted variable bias; panels help control for omitted variables but concerns remain (measurement error, endogeneity).
  - Measurement error in NIC attenuates panel estimates (conservative bias).
  - Dynamic panels introduce Nickell bias; Arellano-Bond GMM inappropriate due to small N (30); OLS fixed effects with and without lagged dependent variables reported.
  - Traditional instruments for corruption (e.g., ethnic division) weak in this setting for proxying news flow.

*Source: wp18195 - Box 1. Refinements and Extensions to the Baseline NIC Measure*

### Section II, when shocks occur in high NIC countries, agents may have less trust that

### Section II, when shocks occur in high NIC countries, agents may have less trust that governments will successfully counter the issue leading to larger belief shifts and lingering growth effects

### NIC shocks and macroeconomic effects — key empirical findings
- Table 4 (NIC Sub-Sample Panels Results, by Income Classification and NIC Level) — Dependent Variable: Real GDP Per Capita Growth
  - Advanced (column 1)
    - NIC Shock: -0.009 (0.007)
    - NIC Shock t-1: -0.007 (0.007)
    - Observations: 220
    - Adjusted R-squared: 0.890
    - Country and Year FE: Yes
    - Full Set of Controls & Lagged Dependent Variables: Yes
    - Standard errors clustered by country in parentheses; constant included; * p<.10, ** p<.05, and *** p<.01. Column’s standard errors based on 11 clusters.
  - EMDEs (column 2)
    - NIC Shock: -0.023* (0.012)
    - NIC Shock t-1: -0.012 (0.008)
    - Observations: 376
    - Adjusted R-squared: 0.944
    - Country and Year FE: Yes
    - Full Set of Controls & Lagged Dependent Variables: Yes
    - Column’s standard errors based on 19 clusters.
  - High NIC (>75th Percentile) (column 3)
    - NIC Shock: -0.028*** (0.007)
    - NIC Shock t-1: -0.019** (0.007)
    - Observations: 140
    - Adjusted R-squared: 0.967
    - Country and Year FE: Yes
    - Full Set of Controls & Lagged Dependent Variables: Yes
    - Column’s standard errors based on 7 clusters.
  - Low NIC (<25th Percentile) (column 4)
    - NIC Shock: -0.002 (0.006)
    - NIC Shock t-1: -0.006 (0.010)
    - Observations: 140
    - Adjusted R-squared: 0.924
    - Country and Year FE: Yes
    - Full Set of Controls & Lagged Dependent Variables: Yes
    - Column’s standard errors based on 7 clusters.
- Interpretation:
  - Corruption belief shocks (NIC shocks) are macro-relevant for growth prospects in EMDEs and economies with corruption vulnerabilities.
  - Advanced, low NIC economies do not appear to face sizable growth implications from country-specific NIC shocks.
  - Possible explanations include stronger/resilient institutions in advanced, low NIC economies; lower frequency/size of NIC shocks; and countervailing anti-corruption news flow.
  - Insignificance of capital flow results likely related to use of annual data; capital markets respond quickly to news shocks.
  - Results are robust to excluding the lagged dependent variable (see Appendix 5).
- Data and estimation notes:
  - Standard errors in some specifications computed based on seven clusters, which may introduce a “too few cluster” issue. Using bootstrapped or jackknifed standard errors also produces significant results (at the 5 percent level).
  - Quarterly capital flow specifications also run; results similarly insignificant.

### Anti-corruption news flow (anti-NIC) — effects and dynamics
- Event study evidence:
  - Shocks to the anti-NIC show positive impacts on financial variables:
    - Decreasing stock market volatility and appreciating exchange rates (see Figure 16).
  - Pre/post t-test results for stock market volatility, exchange rate movements, and change in debt/GDP are significant at 4.946, 1.713, 1.352, respectively.
  - Change in portfolio flows/GDP are insignificant, even when comparing the immediate post-shock period to the rest of the event window.
  - Longer-run responses (e.g., portfolio investment-to-GDP, debt-to-GDP) are more muted.
- Panel estimation (Table 5) — Dependent Variable: Real GDP Per Capita Growth
  - Anti-NIC Shock: -0.005 (0.008)
  - Anti-NIC Shock t-1: 0.002 (0.005)
  - NIC Stock: -0.004 (0.004)
  - Observations: 596
  - Adjusted R-squared: 0.936
  - Country & Year FE: Yes
  - Full Set of Controls & Lagged Dependent Variables: Yes
  - Standard errors clustered by country in parentheses; constant included; * p<.10, ** p<.05, and *** p<.01. There are 30 clusters.
- Interpretation:
  - Anti-NIC shocks have an insignificant impact on real GDP per capita growth in panel estimations; the concurrent sign runs counter to theoretical expectations.
  - Persistent anti-corruption news flow is correlated with lower NIC levels over time: 3-year episodes of anti-corruption news flow split into quintiles show that fourth and fifth quintiles (most persistent anti-corruption news flow) yield a reduction in the NIC five years later.
  - The magnitude of NIC reduction is relatively limited and does not establish causality; reductions could be driven by factors other than the anti-NIC.
  - Suggests that indirect channel—persistent anti-corruption news flow reducing NIC, thereby improving macro outcomes—may be more relevant than immediate orthogonal macro effects of AC measures.

### Mechanisms, channels, and unresolved empirical questions
- Belief formation channel:
  - Large shocks to the NIC shift beliefs about the prevalence of corruption persistently; anti-corruption news could accelerate re-adjustment of beliefs following a shock.
  - Anti-corruption news often reports concrete actions against individuals/institutions; if AC efforts reduce corruption over the medium- to long-term, this could lower NIC levels and reduce frequency/amplitude of NIC shocks.
- Heterogeneity and institutional context:
  - Prior literature documents non-linear relationships between growth and corruption due to differences in institutional quality and setting (e.g., democratic vs. autocratic regimes).
  - EMDEs and high NIC countries may face larger standard deviations in NIC Stocks and larger persistent belief shifts when shocks occur.
- Empirical limitations and research needs:
  - More work needed to detect channels through which belief changes (proxied by NIC shocks) impact growth outcomes in EMDEs with higher NIC levels.
  - Use of higher-frequency NIC in panel settings limited by poor control data availability for emerging and low-income countries at higher frequencies.
  - Ideally, one could directly control for trust; current trust databases (e.g., OECD’s Trust Database) do not cover many sample countries.

### Anti-corruption interventions and policy implications
- Conceptual guidance:
  - For beliefs to shift (with associated economic gains), governments must do more than engage in cheap talk; AC efforts must be supported by meaningful policy changes and governance vulnerability remedies.
  - AC strategies should reflect country-specific context.
- Traditional AC framework recommendations (derived from principal-agent and related literatures):
  - Increase likelihood and severity of detection and penalties.
  - Reduce discretion of agents.
  - Enhance transparency.
  - Address underlying economic, structural, or political problems if corruption functions as a “greasing the wheel” response to overregulation or weak institutions.
- Role of news flow and policy quality:
  - Heterogeneous nature of anti-corruption news and AC policies matters; quality and duration of AC policies are important.
  - Persistent anti-corruption news flow, reflecting steady AC efforts, appears linked to lower NIC levels over time, suggesting indirect macro benefits.
- Modern avenues emphasized:
  - Adoption of modern technology and capacity development as promising complements to traditional AC measures.

### Selected country case studies — qualitative lessons
- Indonesia:
  - Post-Suharto reforms from 2002 onward included founding the Anti-Corruption Commission (KPK) in 2003.
  - KPK is fully independent with a broad mandate to investigate, prosecute, prevent, and educate; it has powers including tapping communications, banning individuals from leaving the country, freezing bank assets, and conducting wealth checks.
  - KPK’s success and high-profile arrests made it one of the most credible and trusted institutions in Indonesia.
  - Indonesia’s anti-NIC measure picked up significantly higher anti-corruption news coverage relative to the sample average; after major AC gains, NIC improved (after an initial steep rise) and the gap between Indonesia’s NIC and the NIC sample average narrowed.
  - Conditions facilitating KPK’s implementation included a severe economic and political crisis, commitment of reformers, societal backlash against corruption, and donor-supported capacity development.
- Malaysia:
  - Anti-corruption laws and a commission established in the 1960s; the current commission (MACC) created under the Malaysian Anti-Corruption Commission Act 2009.
  - MACC has investigative, preventive, and educational powers but no independent prosecutorial power; prosecution decisions lie with the politically-appointed Attorney General.
  - Broader initiatives include e-government, National Integrity Plan (2004), Government Transformation Program (2009), Whistleblower Protection Act 2010, Special Courts on Corruption, and the electronic “MyProcurement” portal.
  - Malaysia regarded as having reduced corruption over time; NIC measure lower than sample average for most of the sample period, though recent 1MDB scandal increased both NIC and anti-NIC coverage.
- Singapore (Box 2 summary):
  - Holistic anti-corruption strategy with four pillars: independent judiciary, stringent laws, effective enforcement, responsive administration.
  - Corrupt Practices Investigation Bureau (CPIB) established in 1952; CPIB reports to the Prime Minister, investigates and prosecutes corruption, and has considerable powers.
  - Prevention of Corruption Act (PCA) enacted in 1960 and regularly revised; high penalties and requirements to prove lawful acquisition of wealth.
  - Public administration reforms: high salaries benchmarked to private sector, strict conduct regulations, mandatory asset declarations, caps on unsecured debt, streamlined administrative procedures, public outreach, Complaints Evaluation Committee, and whistle-blowing policies.
  - Singapore’s overall approach moved corruption from a low risk/high reward to a high risk/low reward activity and produced sustained low NIC relative to the sample.

*Source: wp18195 - Section II, when shocks occur in high NIC countries, agents may have less trust that*

### Box 2. Singapore: The Development of a Holistic Anti-Corruption Strategy (concluded)

### Box 2. Singapore: The Development of a Holistic Anti-Corruption Strategy (concluded)

### Key Lessons
- Any AC framework requires constant effort and ongoing refinements of the anti-corruption strategy to keep corruption low and develop an anti-corruption culture within society.
- Singapore’s specific characteristics—colonial institutions, its small size, independence from Britain and a new Constitution at the time of implementation of first anti-corruption measures, and a conservative culture—limit the transferability of their reform efforts to other countries, and point to the necessity of country-specific tailored measures.
- Addressing petty corruption within the civil service by paying fair salaries and removing red tape could be relatively low-hanging fruits that might not only reduce corruption, but could also improve the overall business environment; modern technology could provide support.

### Closing the Gap: 21st Century Anti-Corruption Interventions
- Information technology is increasingly viewed as the new frontier for a successful AC framework; digitalization in the public sector can modernize, increase efficiency, and provide dynamic working methods that could be a major force against corruption.
- Capacity development (CD) can support AC frameworks by focusing development agendas on institutional strengthening relevant to anti-corruption (areas include fiscal governance, AML/CFT and financial sector supervision).
- The IMF’s CD is used as a proxy for capacity development initiatives, noting it is not directly designed to address corruption but intervenes in relevant areas.

### Empirical Evidence on CD and Anti-NIC Interaction
- There is a positive correlation between the amount of IMF CD that countries have received in the last decade and the improvement in the news flow of corruption (Figure 20).
- Case studies (Indonesia, Philippines and Nepal) with intensive IMF CD over the last decade show CD intervention is closely associated with improvements in broad macroeconomic aggregates and indicators of institutional strength and perceptions of corruption.
- Panel estimations using IMF Technical Assistance Field Time Estimates (FTEs) for 2007-17 indicate that anti-NIC shocks coupled with CD reinforce one another:
  - At a 1-year lag, an additional FTE of CD increases growth by an average of 0.8 percentage points.
  - At a 2-year lag, growth increases by 3 percentage points if additional CD is coupled with anti-corruption efforts.
- Caveats: the measure does not distinguish quality of CD; results are sensitive to specification; results are encouraging but not conclusive. Future research should explore more nuanced CD measures.
- Supporting literature: Improving institutional quality in developing countries to the average level of advanced economies increased annual growth in GDP per capita by 1.5 percentage points, on average (IMF, 2003).

### Table 6 — Capacity Development x Anti-NIC Nexus (Full Sample Results)
- Dependent Variable: Real GDP Per Capita Growth
- Coefficients and significance (standard errors in parentheses):
  - Total FTEs t-1: 0.008** (0.004)
  - Total FTEs t-2: -0.004 (0.004)
  - Total FTEs x Anti-NIC Shock t-1: -0.012 (0.008)
  - Total FTEs x Anti-NIC Shock t-2: 0.031* (0.018)
- Observations: 229
- Adjusted R-squared: 0.899
- Country & Year FE: Yes
- Full Set of Controls, including Anti-NIC Shocks & Lagged Dependent Variables: Yes
- Standard errors clustered by country; constant included; * p<.10, ** p<.05, and *** p<.01.

### Interpretation of Empirical Results
- Capacity development, when consistently supported by authorities and embedded within a broader package of mutually reinforcing reforms, can positively impact AC efforts and growth.
- The IMF and other international organizations can act as catalysts by improving fiscal transparency, market regulation, financial sector oversight, central banking governance, and rule of law, contributing to more inclusive and transparent institutions.
- No single instrument is sufficient; a comprehensive, holistic anti-corruption package is essential.

### Conclusions (Paper-wide)
- Constructed a first cross-country news flow measure of corruption and anti-corruption; measures capture high-frequency variation, scandals and longer-term relative levels, and do not rely on local experts.
- NIC shocks negatively impact financial outcomes and longer-term real outcomes (growth) in a persistent manner; impacts are more pronounced in EMDEs with capacity weaknesses.
- Anti-NIC shocks alone are insufficient for sustained positive outcomes; persistent anti-NIC news flow might positively impact the NIC, and coupling anti-NICs with capacity development yields larger gains to growth.
- Policy implication: anti-corruption plans must be paired with meaningful institutional reform and long-term capacity development frameworks, especially in EMDEs and fragile states; modern technology and technical assistance can provide supporting roles if tailored well.
- Future research avenues include: causal identification of channels between corruption news shocks and growth; classification of NIC and anti-NIC shock types; analysis of technological and social media roles in anti-corruption.

### Box 3 — The Role of Digitalization in Combatting Corruption
- ICT and digital tools (e-filing of taxes, e-governance) can modernize public sector processes and improve service delivery.
- Channels through which digitalization could prevent and reduce corruption include:
  - Enhancing accountability in the public sector.
  - Increasing transparency in dealings with the public sector.
  - Removing the role of the “middleman” public officer.
  - Enhancing faster detection of corruption through open data techniques.
  - Improving public service delivery.
  - Enhancing financial access and inclusion.
- Successful examples referenced: Estonia (e-governance), Tanzania and Kenya (M-Pesa and iTax), Bangladesh (digital land services, electronic/mobile ticketing), India (Aadhaar, Bhoomi, corruption reporting), Indonesia and South Korea (procurement systems), Russia (digitalization of legal and law enforcement systems).
- Documented technological initiative dates analyzed for NIC impact: Bangladesh 2010, India 2001; 2011, Indonesia 2012, Korea 1997.
- Empirical observation: after an initial drop in the NIC post-technology adoption, NIC tends to rise again after a few periods; pre/post t-test statistic is significant (t-stat: 2.186).
- Possible explanations for lack of sustained NIC decline: corruption may adapt and bypass technological measures; necessary conditions (access to technology, technological literacy, social and technology readiness, citizen participation, legal frameworks, organizational processes, leadership and campaigns) may be lacking.
- The “government usage” sub-index of the networked readiness indicator (NRI) shows a positive relationship with the NIC; note that a lower NRI score implies more readiness.
- Digitalization is not a panacea; critical complementary measures include:
  - Ensure equal public access to technology and active citizen engagement.
  - Increase citizens’ technological literacy and social and technology readiness.
  - Enhance citizens’ broader participation in e-government services.
  - Improve legal frameworks, organizational processes, leadership and campaign strategies.

### Box 4 — IMF’s Capacity Development Work and Anti-Corruption Efforts (Case Studies)
- Philippines:
  - 2010 election of President Aquino on an anti-corruption platform; IMF FAD supported a 5-year Revenue Administration reform program and a PFM program; IMF MCM provided TA on banking supervision.
  - Revenue Administration objectives: reduce discretion of revenue collection officers, improve predictability and impartiality of revenue laws and regulation; priorities included institutionalizing arrears management and VAT audit procedures, enhancing taxpayer compliance, integrating new IT systems.
  - Assessments (FTE, TADAT) noted improvements in tax administration (increased tax base, collection performance), VAT audit, arrears management and cash flow management.
  - By 2016, tax revenue had increased back to its pre-GFC level, and the C-efficiency ratio was improved.
  - Some planned projects to improve fiscal governance and transparency were yet to be implemented.
  - Overall: reforms had positive fiscal impacts but little direct effect on perception of corruption and institutional strength based on selected measures; NIC and anti-NIC measures show little improvement over time.
- Indonesia:
  - Large TA from IMF FAD (revenue administration and PFM) and MCM (banking supervision); most IMF LEG TA in the region in last 10 years.
  - Fiscal reform objectives included improving legal framework for tax administration, enhancing VAT compliance, improving governance (code of conduct, special compensation, complaint hotline), implementing taxpayer services IT systems, and improving audit function.
  - Between 2008 and 2011, IMF-assisted design and implementation of AML/CFT framework and a National Legal Reforms program.
  - Over 2010-16, despite large fiscal TA, non-oil tax revenue continued to decline; however measures of institutional strength and corruption perception improved significantly and the NIC measure declined over the period.
- Nepal:
  - Large amount of FAD TA in tax and customs administration improved tax revenue from 8.9 percent of GDP in 2006 to 18.9 in 2016.
  - IMF provided TA for implementing the AML/CFT framework.
  - Over the period, measures of institutional strength remained low and perception of corruption was unchanged at a high level.

*Source: wp18195 - Box 2. Singapore: The Development of a Holistic Anti-Corruption Strategy (concluded)*

### Appendix I. Search Algorithm Dictionaries

### Appendix I. Search Algorithm Dictionaries

### Search algorithms and dictionaries
- Three primary algorithms: Baseline Corruption Algorithm, Core Corruption Algorithm, Anti-Corruption Algorithm.
- Additional extension: Lobbying Algorithm (adds lobbying-related terms; indicated as Bolded Terms in source).
- Common structural rules across algorithms:
  - Corruption term must appear within 8 words of a country identifier (COUNTRY NAME or DEMONYM).
  - Government-related term must appear within 8 words (examples below).
  - Must not mention anti-corruption campaigns (for corruption algorithms).
  - Article length filter: Articles over 99 words (removes ticker articles).
  - Limited to Major news sources and Articles about the country (as characterized by news aggregator).

- Corruption-term lists (examples preserved exactly as in source):
  - Baseline and Lobbying lists include: corrupt*, kleptoc*, nepotism, favoritism, rent-seeking, bribe*, graft, extortion, kickback*, grease payment*, facilitation payment*, facilitation fee*, illegal pact, solicitation, double-dealing, siphon* near2 funds, crony*, scam*.
  - Core list is smaller: corrupt*, kleptoc*, nepotism, graft.
  - Lobbying Algorithm additionally searches for: lobbying, money politic*, private donation*, campaign financ*, influence peddling, revolving door, financial secrec*, state capture, clientelism.
  - Anti-Corruption Algorithm centers on: anti-corrupt*, anti-graft plus proximity (near2) to corrupt* or graft with qualifying reduction terms (see below).

- Government-related term list (used across algorithms), preserved exactly:
  - government, regime, authorities, public sector, bureaucra*, agenc*, ministr*, public off*, administrat*, civil service, bureau or committee, cabinet, president, minister*, military, state-owned, SOE, parliament*, legislature, assembly, central bank, reserve bank.

- Qualifying (reduction/anti) proximity terms (used to exclude or identify anti-corruption context; preserved exactly):
  - lower or decreas* or improv* or crack* down or crackdown or control of curb* or fight* or battl* or war or combat* or tackl* or root out or end to or counter* near2 corrupt* or graft
  - For Anti-Corruption Algorithm the sign anchor is anti-corrupt* or anti-graft or the reduction phrasing near2 corrupt* or graft.

### Implementation constraints
- Articles are limited to Major news sources and to those characterized by the news aggregator as Articles about the country.
- Article-length threshold: over 99 words.
- Algorithms were run on Factiva (details in Appendix II).

---

### Appendix II. Construction of the News Flow Measure of Corruption

### Five-part metric used to count articles (country-specific search algorithms)
- mention of the country within 8 words of corruption-related term
- mention of government-related term within 8 words
- no mention of anti-corruption efforts
- a length of at least 99 words
- limited to major newspapers and no own-country sources

### Indexation and extensions
- Normalization by total news coverage of the country over time.
- Index scaled from [0,100], with higher levels implying more corruption news flow.
- Extensions: Anti-corruption, lobbying and core indices.

### Factiva implementation details (exact figures preserved)
- Algorithms run on Factiva covering:
  - 36,000 sources in 28 languages
  - The Wall Street Journal, The New York Times, Washington Post, The Financial Times, The Globe and Mail, the Guardian, etc.
  - Over 450 continuously updated newswires / 700 wires (e.g., Associated Press and Reuters)
  - Website Publications: e.g., The NYT Online Edition
  - 665+ million articles in English

---

### Appendix III. NIC Robustness Checks

### Human-audit procedure for corruption articles (questions answered per article)
- Is the reference about the intended country?
- Is the reference about corruption (“public abuse for private gain”)?
- If the article is about corruption, is the reference about an increase or decrease in corruption, or is it ambiguous?
- Is this a current (within year) episode of corruption?
- What is the source of corruption?
- Are we picking up petty or grand corruption?
- Is it a perception of corruption (e.g., opinions of possible corruption) or actual/real corruption?
- Is the reference associated with a political purge?

### Human-audit procedure for anti-corruption articles (questions answered per article)
- Is the reference about the intended country?
- Is the reference about anti-corruption efforts?
- Is this a current (within year) episode of anti-corruption?
- Is it targeted at petty or grand corruption?
- Is it perception of anti-corruption or actual anti-corruption efforts?
- Is the reference associated with a political purge?
- Is this about a specific case (e.g., a trial) or about general anti-corruption efforts?

### Audit results (Table 1: Anti-NIC Search Algorithm Results Audit; values in percent)
- Correct Country? 92.4
- About Anti-Corruption? 82.0
- Of correct country & about anti-corruption...
  - Actual (vs. Perceptions)? 87.9
  - Specific (vs. General Efforts)? 49.3
  - Related to a Political Purge? 3.1
- Note: Based on a random sampling of 290 articles; 95% confidence level with +/- 5 confidence interval.

### Correlation structure (Table 2: Pearson's Correlation Coefficients; note: * significance at 1% level)
- Correlations (preserved matrix entries):
  - NIC — Anti-NIC: 0.325*
  - NIC — Lobbying NIC: 0.996*
  - NIC — Core NIC: 0.973*
  - NIC — Civil Liberties Adjusted NIC: 0.888*
  - NIC — Press Freedom Adjusted NIC: 0.951*
  - Anti-NIC — Lobbying NIC: 0.313*
  - Anti-NIC — Core NIC: 0.348*
  - Anti-NIC — Civil Liberties Adjusted NIC: 0.373*
  - Anti-NIC — Press Freedom Adjusted NIC: 0.402*
  - Lobbying NIC — Core NIC: 0.969*
  - Lobbying NIC — Civil Liberties Adjusted NIC: 0.884*
  - Lobbying NIC — Press Freedom Adjusted NIC: 0.947*
  - Core NIC — Civil Liberties Adjusted NIC: 0.886*
  - Core NIC — Press Freedom Adjusted NIC: 0.936*
  - Civil Liberties Adjusted NIC — Press Freedom Adjusted NIC: 0.965*

### Granger causality tests (Table: Granger Causality Tests for NIC & Alternative Measures)
- Null Hypothesis: CPI does not Granger cause NIC — Chi-Squared 5.135** Prob > Chi-Squared 0.023
- Null Hypothesis: CoC does not Granger cause NIC — Chi-Squared 2.713* Prob > Chi-Squared 0.100
- Null Hypothesis: NIC does not Granger cause CPI — Chi-Squared 0.009 Prob > Chi-Squared 0.924
- Null Hypothesis: NIC does not Granger cause CoC — Chi-Squared 0.001 Prob > Chi-Squared 0.977
- Note: * p<.10, ** p<.05, and *** p<.01.

---

### Appendix IV. Stylized Facts: NIC and Anti-NIC

### Summary statistics (table headings and values preserved)
- Panel headers include: Overall, AE's, EM's, LIDC's across indices: Anti-Corruption NIC, NIC, Lobbying NIC, Core NIC, Press Freedom Adjusted NIC, Civil Liberties Adjusted NIC.
- Selected exact descriptive statistics (preserved values):
  - Overall NIC: Mean 7.9; Median 7.7; Std. Deviation 5.8; Maximum 24.8; Minimum 0.6.
  - AE's NIC: Mean 3.0; Median 1.9; Std. Deviation 3.0; Maximum 10.2; Minimum 0.6.
  - EM's NIC: Mean 10.0; Median 9.1; Std. Deviation 4.3; Maximum 18.1; Minimum 3.4.
  - LIDC's NIC: Mean 14.5; Median 10.2; Std. Deviation 9.0; Maximum 24.8; Minimum 8.4.
  - Press Freedom Adjusted NIC: Mean 2.6; Median 1.6; Std. Deviation 2.5; Maximum 8.4; Minimum 0.1.
- Additional index-level panels include similar sets of means, medians, standard deviations, maxima and minima for other NIC variants (values preserved in source tables).

---

### Appendix V. Additional Panel Estimation Results

### Main panel findings (select coefficients preserved exactly)
- Table 1. Full Sample (with lagged dependent variables; standard errors in parentheses; * p<.10, ** p<.05, and *** p<.01.)
  - Dependent variables: Real GDP Per Capita Growth; Revenues/GDP; Portfolio Investment/GDP; Direct Investment/GDP.
  - NIC Shock coefficients:
    - Real GDP Per Capita Growth: -0.0187** (0.00740)
    - Revenues/GDP: -0.00308 (0.00408)
    - Portfolio Investment/GDP: 0.0337 (0.0798)
    - Direct Investment/GDP: -0.0318 (0.0665)
  - NIC Shock t-1 coefficients:
    - Real GDP Per Capita Growth: -0.0110* (0.00541)
    - Revenues/GDP: 0.00107 (0.00184)
    - Portfolio Investment/GDP: -0.0629 (0.0744)
    - Direct Investment/GDP: -0.0235 (0.0375)
  - Observations: 596, 545, 555, 584 (by column).
  - Adjusted R-squared: 0.937, 0.984, 0.351, 0.818.
  - Country & Year FE? Yes (all columns).

- Table 2. Full Sample, without Lagged Dependent Variables (select coefficients preserved)
  - NIC Shock:
    - Real GDP Per Capita Growth: -0.0187** (0.00759)
  - NIC Shock t-1:
    - Real GDP Per Capita Growth: -0.0109* (0.00546)
  - Anti-NIC:
    - Revenues/GDP: -0.00584** (0.00256)
  - Observations: 596, 550, 559, 587.
  - Adjusted R-squared: 0.936, 0.952, 0.310, 0.751.

- Table 3. Full Sample, Additional Robustness Checks (Dependent Variable: Real GDP Per Capita Growth)
  - NIC Shock: -0.0167* (0.00867); -.0182** (0.00723); -.01790** (0.00769); -.01796** (0.00769) across columns.
  - NIC Shock t-1: -0.0106* (0.00541); -.012227* (0.006660); -.00934* (0.00550); -.0090 (0.00560).
  - Observations: 596, 507, 576, 576.
  - Adjusted R-squared: 0.937, 0.934, 0.936, 0.937.

- Table 4. Sub-sample, Without Lagged Dependent Variables (Dependent Variable: Real GDP Per Capita Growth)
  - Subsamples: Advanced EMDCs; High NIC (>75th Percentile); Low NIC (<25th Percentile).
  - NIC Shock coefficients:
    - Advanced EMDCs: -0.00756 (0.00768)
    - High NIC: -0.0230* (0.0126)
    - Low NIC: -0.0181 (0.0179)
    - Full sample column: -0.00423 (0.00527)
  - NIC Shock t-1 coefficients:
    - Advanced EMDCs: -0.00887 (0.00691)
    - High NIC: -0.0116 (0.00842)
    - Low NIC: -0.0192** (0.00740)
    - Full sample column: -0.00372 (0.0103)
  - Observations: 220, 376, 140, 140.
  - Adjusted R-squared: 0.889, 0.944, 0.967, 0.924.

- Table 5. Capacity Development Panel Results, by IMF Department (Dependent Variable: Real GDP Per Capita Growth)
  - FTEs, by Type t-1 coefficients (examples preserved):
    - FAD FTEs: 0.00259 (0.00438)
    - MCM FTEs: 0.00639 (0.0105)
    - STA FTEs: 0.0777 (0.0550)
    - LEG FTEs: 0.00579* (0.00303)
    - All FTEs: 0.00817** (0.00366)
  - Observations: 229 (all columns).
  - Adjusted R-squared: 0.898, 0.900, 0.907, 0.898, 0.899.

- Standard errors clustered by country in parentheses; constant included; significance notation preserved: * p<.10, ** p<.05, and *** p<.01.

---

### Appendix VI. Data Description and Sources

### Key variable definitions and sources (preserved exactly)
- Real GDP — Gross domestic product, constant prices, USD — IMF
- Nominal GDP — Gross domestic product, current prices, USD — IMF
- Real GDP per capita — Gross domestic product, constant prices, per capita — IMF
- PPP GDP — Nominal GDP in PPP$ — IMF
- Consumer Price Index — Consumer Prices, period average — IMF
- Inflation — Consumer Prices, period average, percent change — IMF
- Exports — Exports of goods and services, constant prices — IMF
- Imports — Imports of goods and services, constant prices — IMF
- Terms of Trade — Terms of trade, total, US Dollars, percent change — IMF
- Revenue/GDP — General government revenue, percent of fiscal year GDP — IMF
- Tax Revenue/GDP — General government taxes, percent of fiscal year GDP — IMF
- Expenditure/GDP — General government total expenditure, percent of Fiscal year GDP — IMF
- Fiscal Balance/GDP — General government net lending/borrowing, percent of fiscal year GDP — IMF
- Debt/GDP — General government gross debt, percent of Fiscal year GDP — IMF
- Investment — Gross fixed capital formation, constant prices — IMF
- Unemployment — Unemployment rate — IMF
- Broad Money — Broad money — IMF
- NEER — Nominal Effective Exchange Rate, Trade Partners by Consumer Price Index, Index — IMF
- Bilateral Exchange Rate — US Dollar per National Currency, Period Average, Rate — IMF
- Oil Price — Brent — IMF
- Direct Investment — Financial Account, Direct Investment, Net Incurrence of Liabilities, US Dollars — IMF
- Portfolio Investment — Financial Account, Portfolio Investment, Net Incurrence of Liabilities, US Dollars — IMF
- Nominal Interest Rate — Financial, Interest Rates, Lending Rate, Percent per annum — IMF
- Real Interest Rate — Real interest rate — World Bank
- Money Market Rates — Money Market Rates — IMF, CEIC, OECD
- C-efficiency ratio — Proxy for tax-collection efficiency — IMF
- Portfolio Equity Assets / Liabilities, FDI Assets/Liabilities, Debt Assets/Liabilities, FX Reserves — External Wealth of Nations Mark II
- Credit/GDP, Private Credit/GDP — BIS
- Population, Population Growth, Primary/Tertiary Education, International Aid, Public Expenditure on Education, Agriculture Value Added, Inequality Gini Index — World Bank
- Institutional variables (Government Effectiveness, Rule of Law, Corruption Measure) — ICRG; Control of Corruption — World Bank; Corruption Perceptions Index — Transparency International
- Human Development Index — UNDP
- Proxy for regime — Polity IV: Center for Systemic Peace
- Proxy for ethnic/civil conflict — Major Episodes of Political Violence: Center for Systemic Peace
- Proxy for language diversity — Ethnologue
- Stock Market Price — Benchmark index closing price — Bloomberg
- Sovereign Bond spreads – 5Y and 2Y — Bloomberg

*Source: wp18195 - Appendix I. Search Algorithm Dictionaries (PDF chapter).*

### REFERENCES

### REFERENCES

### Institutions, Corruption, and Growth
- Acemoglu, D., S. Johnson, J. Robinson, and Y. Thaicharoen, 2003, “Institutional causes, macroeconomic symptoms: volatility, crises and growth.” Journal of Monetary Economics, Vol. 50 pp. 49–123.
- Campos, N., Dimova, R., and Saleh, A., 2010, “Whither Corruption? A Quantitative Survey of the Literature on Corruption and Growth,” Discussion Paper No. 5334 (Bonn: Institute for the Study of Labor).
- Heckelman, J. C., & Powell, B., 2010. “Corruption and the institutional environment for growth,” Comparative Economic Studies, Vol. 52(3), pp. 351-378.
- Huntington, S.P., 1968, “Political order in changing societies.” New Haven: Yale University Press.
- Leff, N.H., 1964, “Economic development through bureaucratic corruption.” American Behavioral Scientist Vol 8, pp.8-14. Reprint in Heidenheimer, A.J., M. Johnston and V.T. LeVine (eds), 1989, “Political corruption: A Handbook,” pp. 389-403. Oxford: Transaction Books.
- Marquette, H., and Peiffer, C., 1996, “The Effects of Corruption on Growth, Investment, and Government Expenditure,” IMF Working Paper No. 96/98, pp. 1-28. (Washington, DC: International Monetary Fund).
- Mauro, P., 1995, “Corruption and Growth.” The Quarterly Journal of Economics, Vol. 110 (3), pp. 681-712.
- Méon, P., and K. Sekkat, 2005, “Does corruption grease or sand the wheels of growth?” Public Choice Vol. 122, pp. 69-97.
- Persson, A., Rothstein, B., and Teorell, J., 2010, “The failure of Anti-Corruption Policies—A Theoretical Mischaracterization of the Problem,” QoG Working Paper Series 2010:19 (Gothenburg: The Quality of Government Institute, University of Gothenburg).
- Murphy, K., Shleifer, A., and Vishney, R., 1991, “The Allocation of Talent: Implication for Growth” The Quarterly Journal of Economics, Vol. 106 (2), pp. 503-30.
- Marquette, H., and Peiffer, C., 2015, “Corruption and Collective Action,” Research Paper No. 32, January (Developmental Leadership Program and U4 Anti-Corruption Resource Centre).

### Political Economy, Rent-Seeking, and Public Choice
- Krueger, A. O., 1974, “The political economy of the rent-seeking society,” The American economic review, Vol. 64 (3), pp. 291-303.
- Klitgaard, R., 1988, Controlling Corruption (Berkeley and Los Angeles: University of California Press).
- Klitgaard, R., 2015a, Addressing Corruption Together (Paris: OECD, Development Cooperation Directorate).
- Klitgaard, R., 2015b, Addressing Corruption Together, background paper prepared for the OECD Symposium on Anti-Corruption Development Assistance: Good Practices among Providers of Development Co-operation, which took place in Paris on December 11-12, 2014 (Paris: Organisation for Economic Co-operation and Development).
- Leff, N.H., 1964 (see above).

### Anti-Corruption Agencies, Case Studies, and Country Experiences
- Bolongaita, E. P., 2010, “An Exception to the Rule? Why Indonesia’s Anti‐Corruption Commission succeeds when others don’t – a comparison with the Philippines’ Ombudsman,” U4 Anti‐Corruption Resource Centre, U4 Issue No. 4, August (Chr. Michelsen Institute (CMI), Anti-Corruption Resource Center).
- Koh, T. H., 2011, “Corruption Control in Singapore,” United Nations Asia and Far East Institute for the Prevention of Crime and the Treatment of Offenders (UNAFEI), Resource Material Series No. 83, pp. 122-131.
- Lee, H. L., 2016, “Fight against corruption: Singapore's experience” in Essay for the Anti-Corruption Summit, London, May.
- Kuris, G., 2012a, Inviting a Tiger into Your Home: Indonesia Creates an Anti-Corruption Commission with Teeth, 2002-2007 (Princeton University: Innovations for Successful Societies).
- Kuris, G., 2012b, Holding the High Ground with Public Support: Indonesia’s Anti-Corruption Commission Digs In, 2007-2011 (Princeton University: Innovations for Successful Societies).
- Maxwell, A., 2017, “Indonesia’s Corruption Eradication Commission Law Weathers the Storm, For Now,” The Diplomat (September).
- Meagher, P., 2005, “Anti‐corruption agencies: Rhetoric Versus reality,” The Journal of Policy Reform, Vol. 8 (1), pp. 69-103.
- Khan, F., 2016, “Combating Corruption in Pakistan," Asian Education and Development Studies, Vol. 5 (2), pp.195-210.
- Hershman, M. J., 2016, “Anti-Corruption Program in Malaysia — A Comprehensive Approach,” The World Post.
- Siddiquee, N. A., 2009, “Combating Corruption and Managing Integrity in Malaysia: A Critical Overview of Recent Strategies and Initiatives,” Public Organization Review, Vol. 10 (2), pp. 153–171.
- Transparency International (TI), 2013, “Best Practices for Anti-Corruption Commissions,” TI Anti-Corruption Helpdesk – Providing On-Demand Research to Help Fight Corruption.
- USAID, 2016, Report on Combatting Corruption and Strengthening Integrity in Jamaica.

### ICT, E-Government, Transparency, and Civic Tech
- Andersen, T. B., 2009, “E-Government as an Anti-corruption Strategy,” Information Economics and Policy, Vol. 21 (3), pp. 201–210.
- Bertot, J. C., Grimes, J. M., and Jaeger, P. T., 2010, “Using ICTs to create a culture of transparency: E-government and social media as openness and anti-corruption tools for societies,” Government Information Quarterly, Vol. 98 (July), pp. 264-271.
- Chêne, M., 2016, “Literature review: The use of ICTs in the fight against corruption,” No. 6, November (U4 Anti-Corruption Resource Center).
- Dupuy, E.P., and Serrat, O., 2014, “Fighting Corruption with ICT: Strengthening Civil Society’s Role,” June, pp. 797-811 (Manila: Asian Development Bank).
- Davies, T., and Fumega, S., 2014, “Mixed incentives: Adopting ICT innovations for transparency, accountability, and anti-corruption,” No. 4, June (Chr. Michelsen Institute (CMI), Anti-Corruption Resource Center).
- Pina, V., Royo, S., and Torres, L., 2007, “Are ICTs Improving Transparency and Accountability in the EU Regional and Local Governments? An Empirical Study,” Public Administration, Vol. 85, No. 2, pp. 449–72.
- Shim, D. and T. Eom, 2008, “E-Government and Anti-Corruption: Empirical Analysis of International Data. International Journal of Public Administration, Vol. 13, pp. 298-316.
- Srivastava, B., 2016, Tackling Corruption With Agents & ICT: A Vision (IBM Research).
- Wickberg, S., 2012, Technological innovations to identify and reduce corruption (Transparency International).
- Gigler, B. S., 2015, Development as Freedom in a Digital Age: Experiences from the Rural Poor in Bolivia (Washington, DC: The World Bank).
- Dupuy, E.P., and Serrat, O., 2014 (see above).
- Chêne, M., 2008, “Overview of Corruption in Pakistan,” U4 Anti-Corruption Resource, Centre.

### Media, Perceptions, and Information Effects
- Goel, R. K., Naretta, M. A., and Nelson, M. A., 2012, “The internet as an indicator of corruption awareness,” European Journal of Political Economy, 28 (1), 64-75.
- Lu, J., Shi, T., and Zhu, J., 2013, “When Grapevine News Meets Mass Media: Different Information Sources and Popular Perceptions of Government Corruption in Mainland China,” Comparative Political Studies, Vol. 46 (8), pp. 920-946.
- Rizzica, L., and Tonello, M., 2018, “Persuadable Perceptions. The Effect of Media Content on Beliefs About Corruption,” Working Paper, (Rome: Bank of Italy, Statistics and Research, Structural Economic Analysis Directorate, Law and Economics Division.).
- Shapiro, A. H., Sudhof, M., and Wilson, D., 2018, “Measuring News Sentiment,” Working Paper No. 2017-01, March (San Francisco: Federal Reserve Bank of San Francisco).
- Lee, C. M., and Ng, D., 2009, “Corruption and International Valuation: Does Virtue Pay?,” The Journal of Investing, Vol. 18 (4), pp. 23-41.
- Costas-Pérez, E., Solé-Ollé, A., and Sorribas-Navarro, P., 2012, “Corruption scandals, voter information, and accountability,” European Journal of Political Economy, 28 (4), pp. 469-484.
- Lu, J., Shi, T., and Zhu, J., 2013 (see above).
- Rizzica, L., and Tonello, M., 2018 (see above).

### Financial Markets, Risk, and Policy Uncertainty
- Baker, S. R., Bloom, N., and Davis, S. J., 2016, “Measuring Economic Policy Uncertainty,” The Quarterly Journal of Economics, Vol. 131 (4), pp. 1593-1636.
- Brogaard, J., and Detzel, A., 2015, “The Asset-Pricing Implications of Government Economic Policy Uncertainty,” Management Science, Vol. 61 (1), pp. 3-18.
- Ciocchini, F., Durbin, E., and Ng, D. T., 2003, “Does corruption increase emerging market bond spreads?” Journal of Economics and Business, 55 (5-6), 503-528.
- Grigorian, D., and Manole, V., 2017, “Sovereign risk and deposit dynamics: evidence from Europe,” Applied Economics, Vol. 49 (29), pp. 2851-2860.
- Hlatshwayo, S., and Saxegaard, M., 2016, “The Consequences of Policy Uncertainty: Disconnects and Dilutions in the South African Real Effective Exchange Rate-Export Relationship,” IMF Working Paper No. 113, pp. 1-30. (Washington, DC: International Monetary Fund).
- Gunnemann, J., 2014, “The Impact of Economic Policy Uncertainty: New Evidence from Europe,” unpublished manuscript.
- Arbatli, E. C., Davis, S. J., Ito, A., Miake, N., and Saito, I., 2017, “Policy Uncertainty in Japan,” NBER Working Paper No. 23411, May (Massachusetts: National Bureau of Economic Research).
- Wolfers, J., and Zitzewitz, E., 2016, “What do financial markets think of the 2016 election,” Working Paper.
- Brogaard, J., and Detzel, A., 2015 (see above).

### Public Finance, Taxation, and Trade
- Alonso-Terme, R., Davoodi, H., and Gupta, S., 2002, “Does corruption affect income inequality and poverty?” Economics of Governance, Vol. 3 (1), pp. 23-45.
- Keen, M., and Baunsgaard, T., 2005, “Tax Revenue and (or?) Trade Liberalization,” IMF Working Paper No. 5-112 (Washington, DC: International Monetary Fund).
- Razinabc, A., Sadkaa, E., and Yuend, C. W., 1998, “A pecking order of capital inflows and international tax principles,” Journal of International Economics, Vol. 44 (1), pp. 45-68.
- Rajkumar, A., S., and Vinaya, S., 2002, ”Public Spending and Outcomes: Does Governance Matter?,” Policy Research Working Paper No.2840 (Washington, DC: The World Bank).
- Wei, S. J., 1997, “Why is Corruption So Much More Taxing Than Tax? Arbitrariness Kills,” NBER Working Paper No. 62 (Cambridge, MA: National Bureau of Economic Research).
- Wei, S. J., and Wu, Y., 2002, “Negative alchemy? Corruption, Composition of Capital Flows, and Currency Crises,” in Preventing Currency Crises in Emerging Markets, pp. 461-506 (Chicago: University of Chicago Press).

### Sectoral Vulnerabilities, Public Service Delivery, and Governance
- Engelschalk, M., Ferreira, C., and Mayville, W., 2007, “The Challenge of Combating Corruption in Customs Administrations,” in Campos, J., and Sanjay, P., (eds), The Many Faces of Corruption: Tracking Vulnerabilities at the Sector Level, pp. 367-86 (Washington, DC: The World Bank).
- Bold, T., Kimenyi, M., Mwabu, G., and Sandefur, J., 2011, “Why Did Abolishing Fees Not Increase Public School Enrollment in Kenya?,” Working Paper No. 271 (Washington, DC: Center for Global Development).
- Detragiache, E., Gupta, P., and Tressel,T., 2005, Finance in Lower-Income Countries: An Empirical Exploration, IMF Working Paper 05/167, International Monetary Fund, Washington.
- Reinikka, R., and Svensson, J., 2004, “Local Capture: Evidence from a Central Government Transfer Program in Uganda,” Quarterly Journal of Economics, 119 (2), pp. 679-705.
- Aucoin, P., and Heintzman, R., 2000, “The Dialectics of Accountability for Performance in Public Management Reform.”
- Rajkumar, A., S., and Vinaya, S., 2002 (see above).

### Measures, Indices, and Methodologies
- Dadasov, D., Dykes, V., Khaghaghordyan, A., Kossow, N., Kukutschka N. A., Martínez, R. B., and Mungiu-Pippidi, A., 2017, “Index of Public Integrity,” European Research Centre for Anti-Corruption and State-Building (ERCAS).
- Linzer, D. A., and Staton, J. K., 2015, “A global measure of judicial independence, 1948–2012,” Journal of Law and Courts, Vol. 3 (2), pp. 223-256.
- UNDP, 2015, A Users’ Guide to Measuring Corruption (Norway: United Nations Development Programme, Oslo Governance Centre).
- UNIFEM, 2008, Analysis of Transparency International Global Corruption Barometer database.
- CPIB, 2016, “Our Heritage.”
- World Bank, 2016, Report on Corruption.
- Dadasov et al., 2017 (see above).

### Experimental, Behavioral, and Cultural Studies
- Cameron, L., Chaudhuri, A., Erkal, N., and Gangadharan, L., 2009, “Propensities to Engage In and Punish Corrupt Behavior: Experimental Evidence from Australia, India, Indonesia and Singapore,” Journal of Public Economics 93 (7–9), pp. 843–851.
- Guiso, L., Sapienza, P., and Zingales, L., 2009, “Cultural biases in Economic Exchange?,” The Quarterly Journal of Economics, Vol. 124 (3), pp. 1095-1131.
- Sapienza, P., and Zingales, L., 2012, “A Trust Crisis,” International Review of Finance, Vol. 12 (2), pp. 123-131.
- Cameron et al., 2009 (see above).

### Miscellaneous Empirical and Theoretical Contributions
- Fama, E. F., 1970, “Efficient Capital Markets: A Review of Theory and Empirical Work,” The journal of Finance, Vol. 25 (2), pp. 383-417.
- Jordà, Ò., 2005, “Estimation and inference of impulse responses by local projections,” American Economic Review, Vol. 95 (1), pp. 161-182.
- Kozlowski, J., Veldkamp, L., and Venkateswaran, V., 2016, “The Tail that Wags the Economy: Belief-Driven Business Cycles and Persistent Stagnation,” CEPR Discussion Papers, No. 11352 (London: Centre for Economic Policy Research).
- Cavoli, T., and Wilson, J., 2015, Corruption, Central Bank (in)dependence and Optimal Monetary Policy in a Simple Model, Journal of Policy Modeling, 37 (3): 501–9.
- Leff, N.H., 1964 (see above).

*Source: wp18195 - REFERENCES (wp18195 - REFERENCES).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18195.pdf_