## Annex 1. Corporate Income Tax in Selected Countries: A Brief Overview

## Source details

**Canonical URL:** [Annex 1. Corporate Income Tax in Selected Countries: A Brief Overview](https://www.imf.org/-/media/files/publications/tnm/2023/english/tnmea2023006.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/tnm/2023/english/tnmea2023006.pdf.md)
- [Structured JSON version](/-/media/files/publications/tnm/2023/english/tnmea2023006.pdf.json)

---

### Document structure and navigational pointers
- Annex sequence and exact page references:
  - Annex 1: "Corporate Income Tax in Selected Countries: A Brief Overview" — page 22
  - Annex 2: "Corporate Income Tax Gap Definitions" — page 23
  - Annex 3: "Corporate Income Tax Gap Estimation Techniques" — page 25
  - Annex 4: "Bottom-Up Corporate Income Tax Gap Estimation Techniques in Revenue Administrations with Little Experience and Limited Resources" — page 28
  - Annex 5: "Random Audit Programs versus Extreme Values Techniques" — page 31
- Annex 1 precedes annexes covering definitions, estimation techniques, resource-constrained implementation, and audit program methodologies.

### I. Purpose, challenges, and complementary approaches
- Purpose of CIT gap analysis:
  - Estimate the amount of CIT revenues not collected because of noncompliance to assess potential revenue mobilization through improved tax administration.
- Core challenge:
  - Noncompliance is unobservable; indirect methods must be used.
- Two main estimation approaches:
  - Bottom-up: uses RA microdata (random audits, operational audits, administrative data) to directly estimate noncompliance and diagnose root causes.
  - Top-down: uses third-party or macroeconomic aggregates (for IMF RA-GAP, operating surplus from national accounts) to estimate potential CIT and derive the gap by subtracting actual revenue.
- Complementarity:
  - Bottom-up techniques identify root causes and segment-level gaps; top-down provides comprehensive assessment of all revenue foregone.
- RA-GAP context:
  - Since the RA-GAP on the CIT gap began in 2017, about six technical assistances have been delivered to estimate CIT gap. Demand is growing, particularly among emerging market economies.
- Bottom-up technique types:
  - (1) Random audit program results.
  - (2) Operational audit results (risk-selected).
  - (3) General administrative data exploiting non-audit information.
- Limitations and corrective tools:
  - Undetected noncompliance is pervasive. Adjustment methods include uplift factors, auditors’ best guesses, Delphi methods, and upper/lower estimate bounds.

### II. Definitions and measurement scope (RA-GAP and bottom-up practice)
- RA-GAP definitions:
  - Overall tax gap = potential revenue of the underlying tax base − actual revenue.
  - Compliance gap = potential revenue within current tax policy framework − actual revenue collected.
  - Policy gap = potential revenue if all corporate incomes taxed at the current standard rate − potential revenue under current policy framework.
- Bottom-up RA definitions (seven RAs surveyed):
  - Focus limited to the compliance gap (excluding policy gap).
  - Gap = theoretical CIT revenue under current policy − actual CIT revenue.
  - Gross vs net gap:
    - Gross CIT gap = initially observed gap before RA/taxpayer corrective actions.
    - Net CIT gap = observed gap after corrective actions; RA-GAP definition is equivalent to net CIT gap.
    - Some RAs (example: Canada) present both gross and net to illustrate revenue yield from RA actions.
- Nonfilers and nonregistered taxpayers:
  - Bottom-up definitions typically exclude nonfilers and nonregistered gaps; RA-GAP top-down captures all forms of noncompliance via potential-collection estimation.

### III. Bottom-up techniques — methods, segments, and practical issues
- Technique categories:
  - Random audit program: unbiased sample if selection is truly random; requires correction for nondetection (uplifts, multipliers).
  - Operational audits: biased risk-based samples; inference techniques include extreme values (EV), econometric (Heckman two-stage), clustering, and expert judgment.
  - General administrative data approaches: model theoretical tax liability for each business (example: Brazil’s frontier production function approach).
- Segment-specific guidance:
  - Small businesses:
    - Preferred technique: random audit program (most statistically robust for this segment).
  - Mid-sized businesses:
    - Often estimated jointly with small or large segments; Australia estimates mid-sized independently using logistic/linear regressions and Monte Carlo simulations.
  - Large businesses:
    - Operational-audit–based techniques and EV are favored due to heterogeneous populations and high audit-resource intensity.
    - High existing audit coverage (example: historically close to 33 percent audit rate in the United States for large corporations, coverage reaching 80 percent of the largest of the large corporations) reduces efficiency of random sampling.
- Extreme Values (EV) technique (large businesses):
  - Method: assume detected audit adjustments follow a Pareto distribution (log NAR inversely related to log rank); extrapolate audited adjustments to nonaudited population.
  - Key assumptions:
    - Underreported amounts for large businesses fit a Pareto distribution.
    - Audits identify the highest-value adjustments (audit identification is never 100 percent).
  - Data and timing:
    - Multiyear audit databases needed because large-company audits can take many years to close.
    - United States looks at seven years of audit results; Canada uses eight years.
    - Use of imputation to project open cases (United Kingdom example).
  - Scale example:
    - United States contemplates about 8,000 entities in the large segment, of which up to 2,700 are annually audited, and only about 100 of the audited entities are included as the EV for estimation.
  - Practical choices:
    - RAs differ on number of EVs to include and on treatment of positive adjustments; auditors’ initial assessments often used rather than final litigation outcomes.
- Operational-audit econometric/statistical techniques:
  - Heckman two-stage procedure: models selection probability then outcome controlling for selection bias.
  - Clustering techniques: group audited and nonaudited taxpayers by characteristics, scale audit results within clusters (Canada uses two-step clustering and propensity score matching for validation).
  - Expert judgment: experienced auditors’ subjective risk assessments to extrapolate.
- General-administrative-data techniques (example: Brazil):
  - Frontier production function (Cobb–Douglas) treating gross tax liability as production, explained by purchases, salaries and reported employees, bank transactions, and fixed capital; model runs over about 2.7 million active taxpayers per year.

### IV. Random audit program design, sample sizes, stratification, and conduct
- Main methodological requirement:
  - Correctly extract a random sample of entities to be audited and extrapolate audit results to the population.
- Sample size and cost considerations:
  - Sample size defined by required statistical confidence and subject to resource limits.
  - Small-business segment is largest; random sample size necessary to guarantee confidence is proportionally lower than for other segments.
  - Individual random audit cost is lower for small businesses due to simpler operations.
- Reported sample sizes:
  - Canada: about 4,500 cases in each estimation process, including mid-sized businesses.
  - United Kingdom: about 330 cases per process.
  - Other RAs: sample sizes of fewer than 1,000 cases.
- Exclusions before sampling (examples):
  - Sweden excludes dormant entities, entities with no employees, low-income entities, certain small-business forms, partnership types, publicly owned companies, financial corporations, and nondomestic companies.
  - Australia excludes small businesses under a legal vehicle different from an incorporated company.
  - United Kingdom intends to consider small-business population without exclusions, including inactive and dormant companies.
- Stratification and pooling:
  - Stratification common (typically by turnover and industry).
  - Examples:
    - Australia: four strata from two turnover categories and two industry types.
    - Canada: stratifies into one of 21 economic sectors.
    - Sweden: limited companies into six strata by annual salary total; sole traders into three strata by annual turnover.
  - Sample pooling to improve efficiency:
    - Denmark moved from sampling every three years to annual sampling with pooling.
    - Australia bundles audit samples from different years under short-term noncompliance constancy assumption.
    - Canada conducts period random audits (not annual) and pools for trend analysis.
- Standardization and preprocessing:
  - Use standardized audit checklists and profiling to optimize resources (Sweden example: 600 firms audited with a standardized tool reviewing about 100 aspects; about 50 percent show no indication of noncompliance; auditor average 10 to 12 days per firm; largest companies average 17 days; 60 to 70 auditors work on program).
  - Preprocessing can improve efficiency though may affect pure randomness.

### V. Accounting for nondetection, uplift factors, and bounds
- Uplift factor approaches:
  - United States developed a methodology to estimate an “uplift factor” for detected audit adjustments in individual income tax; not applied to business CIT gap yet.
  - Australia uses uplift factors over CIT base by assurance category: 3 percent (not assured), 2 percent (medium assurance), and 1 percent (high assurance).
- Upper/lower bound approaches:
  - Canada produces upper and lower bounds: upper from clustering technique, lower from EV technique.
  - United Kingdom shifted from adapted uplift factors to lower/upper bound framework: upper bound from EV treating risk-based audits as representative; lower bound from EV under standard Pareto assumption.
- Delphi method:
  - Sweden and the United Kingdom started formalizing estimation of a CIT uplift factor using the Delphi method based on surveys of experienced auditors (Sweden results: Sweden should receive about 27 percent more in taxes from incorporated companies' tax gap and 17 percent from sole proprietorships' gap, plus an additional projected amount; a remaining nonverifiable gap exists).

### VI. Practical lessons, timing, and resourcing for bottom-up programs
- Time horizon:
  - Developing bottom-up CIT gap estimates requires several years; RAs without prior experience should plan medium- and long-term programs (example: at least three years from start of audits until first results for a new RAP).
- Specialized staff and institutional arrangements:
  - Staff should combine CIT legislation knowledge, micro-database management, and statistical/economic skills.
  - Specialized teams produce guidance manuals, control audits for quality, and identify IT support needs.
- Data management:
  - Detailed audit databases are critical (audited periods, adjustments, execution dates, dedicated auditors, error types).
  - EV techniques require multiyear detailed records (United States manages at least 10 years of detailed completed CIT audit information).
- RAP timing implications:
  - RAPs require design, execution, and compilation stages; results typically available two or three years after program start.
  - Example timeline (Denmark):
    - Income year: 2021 (January 1–December 31).
    - Filing date: July 1, 2022 (note pandemic postponements).
    - Preparing data: July/August 2022.
    - Audits: September 1, 2022–August 31, 2023.
    - Finalizing audits and error registration: September–October 2023.
    - Analysis and report writing: November–December 2023.

### VII. Comparative advantages and disadvantages: Random Audit Programs versus Extreme Values
- Advantages of Random Audit Programs:
  - Suitable for small and mid-sized businesses.
  - Straightforward extrapolation to population.
  - Obtain unbiased estimates for current year given confidence interval.
  - Break down results by causes of noncompliance and demographic factors.
  - Assess and recalibrate risk-based processes; useful for auditor training.
- Disadvantages of Random Audit Programs:
  - Not efficient for large businesses.
  - Resource- and time-intensive; at least three years to produce results for a single income year.
  - Internal resistance risk because yields are lower than risk-based audits.
  - Broader and shallower audits increase nondetection probability.
  - May fail to capture some noncompliance practices if audits focus narrowly.
- Advantages of Extreme Values Techniques:
  - Suitable for large businesses; extendable to mid-sized.
  - Simpler and less expensive than other statistical techniques.
  - Obtain results for several years; assumes Pareto distribution tail for largest firms.
  - Exploits existing audit results without requiring additional audits.
  - Based on deeper risk-based audits, lowering nondetection probability.
- Disadvantages of Extreme Values Techniques:
  - Not applicable to small businesses.
  - Requires more assumptions, including tail selection for Pareto distribution.
  - Large-business audits can take up to 10 years to close.
  - Results sensitive to a few large cases; production of adjustments may vary.

### VIII. Reporting practices, uncertainty, and empirical estimates
- Communication of uncertainty:
  - United Kingdom declares an uncertainty rating per CIT gap estimate: small and mid-sized businesses assessed “moderate level" of uncertainty; large businesses assessed “high level” of uncertainty.
  - Australia uses a similar segment-level rating.
- Public net CIT gap levels (most recent periods available) reported across RAs range from 3 to 15 percent of potential:
  - Canada 2018 (lower bound): 5.6
  - Canada 2018 (upper bound): 9.1
  - Denmark 2011–13 annual average: 14.5
  - Sweden 2011–13 annual average: 15.0
  - UK 2020–21 projection: 3.2
  - US 2011–13 annual average: 9.0
  - Australia 2018–19: 11.0
- Gross vs net gap reductions from RA actions:
  - Canada example: estimated compliance gap reduced by more than 11 percent points of potential after accounting for compliance actions.
  - United States: reduction of more than 3 percentage points.
- Share of gap attributable to large companies (selected jurisdictions, shares of net CIT gap):
  - Canada 2018 (lower bound): 50.5
  - Canada 2018 (upper bound): 69.2
  - Sweden 2011–13 annual average: 69.8
  - UK 2020–21 projection: 75.9
  - US 2011–13 annual average: 70.3
  - Note: United Kingdom is an outlier due to differences in technique, segment definitions, segment weighting in CIT revenues, and relative small/mid-sized business compliance risk.

### IX. Usefulness of audit-based estimates for diagnosis and policy design
- Audit-based estimates provide information on:
  - Composition of the CIT gap by type of error and sector concentrations.
  - Differentiation between unintentional errors (addressable by outreach/education) and deliberate noncompliance (requiring enforcement).
- Country practices:
  - Denmark publishes CIT gap broken down by economic sector and type of error.
- Transparency and mandates:
  - Almost all RAs surveyed publish CIT gap estimates; some have government mandates to estimate and publish the tax gap.
  - Internal communication of results supports understanding root causes and evaluating compliance actions.

### X. Annex highlights — revenue, rates, and definitions (selected)
- Annex highlights on revenue and rates:
  - Average CIT revenue for 2016–20 varies between 1.4 and 5.1 percent of GDP in the selected countries.
  - For Australia, Canada, Denmark, and Sweden, CIT revenue levels increased compared with averages for 2011–15.
  - For Brazil, the United Kingdom, and the United States, CIT revenue levels show slight decreases compared with averages for 2011–15.
  - Average CIT top combined rate for 2016–20 ranges between 19.3 and 34.0 percent in the selected countries.
  - Australia, Brazil, and Canada: CIT top combined rates for 2016–20 do not show changes compared to averages for 2011–15.
  - Denmark, Sweden, the United Kingdom, and the United States: decreases in CIT top combined rates for 2012–20 observed compared to averages for 2011–15.
- Selected definitions (country excerpts):
  - Australia:
    - Gross CIT Gap: Difference between amount voluntarily reported to the ATO and amount that would have been collected if every taxpayer were fully compliant (theoretical tax liability).
    - Net CIT Gap: Difference between amount voluntarily reported plus amendments from compliance activities/voluntary disclosures and the theoretical tax liability.
  - Brazil:
    - Gross CIT Gap: Difference between potential tax liability under current tax system and actual liability as declared.
    - Net CIT Gap: Gross gap minus enforced and late payments.
  - Canada:
    - Gross CIT Gap: Difference between CIT that would be paid if all obligations fully met and CIT actually paid/collected before compliance actions.
    - Net CIT Gap: Gross gap after subtracting compliance and collection activities.
  - United Kingdom:
    - Gross CIT Gap: Difference between theoretical tax liability (TTL) and “voluntary” receipts.
    - Net CIT Gap: Difference between TTL and “total” receipts (voluntary plus compliance yield receipts).
  - United States:
    - Gross CIT Gap: Difference between total true CIT liability (TTCL) and CIT paid voluntarily and timely.
    - Net CIT Gap: Difference between TTCL and total CIT payments (voluntary and timely plus enforced and late payments).
    - Note: TTCL includes underreported (net of overreported) CIT liabilities. The CIT gap definition does not include tax avoidance.

### XI. Annex methodological boxes — summary of method steps
- Annex Box 3.1 (EV for large US corporations) — condensed steps:
  - Collect operational audit databases; extract NARs; truncate top N NARs; compute ratios p = N/S; rank extremes; estimate Log10(NAR) on Log10(rank) to obtain slope a and intercept c; extrapolate to whole population F; M = p × F; estimate total underreported CIT U = 10^c × sum_{r=1 to M} r^a. Largest corporations defined as assets > $250 million.
- Annex Box 3.2 (Logistic/Linear Regression for mid-sized businesses) — condensed steps:
  - Run logistic regression for probability of noncompliance; run linear regression for share of unreported tax among noncompliant entities; correct for selection bias using propensity score matching; combine results and, in Australia’s case, use Monte Carlo simulation with more than 20,000 iterations; apply nondetection uplift factors by turnover category; consolidate estimates.
- Annex Box 3.3 (Random Audit Program for small companies) — condensed steps:
  - Estimate average amendment and amendment rate from sampled taxpayers; extrapolate to population; estimate nondetection via uplift factor; compute gross and net gap (gross = Steps 1 + 2; net = gross minus compliance outcomes/voluntary disclosures); theoretical tax liability = net gap + net tax paid.
- Annex Box 3.4 (Stochastic Frontier Method — Brazil) — condensed steps:
  - Obtain RA population data; extract quality sample; treat outliers/missing data; calculate declared tax gross operating surplus and tax liability; estimate business-level CIT gap; aggregate results.
- Annex Box 3.5 (Delphi Method — Sweden) — condensed approach:
  - Recursive questionnaires to experienced auditors and senior coworkers to converge on assessed nondetection multipliers; Sweden results indicated uplift of about 27 percent for incorporated companies and 17 percent for sole proprietorships.

### XII. Practical recommendations and operational guidance (drawn from practices)
- For RAs with limited experience or resources:
  - Consider bottom-up techniques to cross-check top-down results and to diagnose noncompliance composition.
  - Plan medium- to long-term programs (expect at least three years to obtain first RAP results).
  - Assign specialized teams combining tax law expertise, microdata management, and statistical skills.
  - Establish robust audit-data management systems preserving multiyear detailed audit records.
  - Use stratification, preprocessing, and pooling to reduce RAP costs while preserving statistical validity.
  - Combine techniques by segment: random audits for small businesses; operational audits/EV for large businesses; tailored econometric approaches for mid-sized where feasible.
  - Transparently report uncertainty, maintain consistency in methods over time, and document method changes with comparability notes.
- When accounting for nondetection:
  - Move toward domestic uplift/multiplier estimates (instead of importing factors from other countries).
  - Consider Delphi and upper/lower bound frameworks to capture nondetection uncertainty.
- Use RAPs to:
  - Diagnose causes of noncompliance, inform risk selection, and recalibrate risk-based audit models.
  - Identify nonfilers/nonregistered entities when sample preprocessing or audit findings reveal transactions with such entities.

*Italic source attribution: IMF Technical Note tnmea2023006 — Corporate Income Tax Gap Estimation by Using Bottom-Up Techniques in Selected Countries*

### Annex 1. Corporate Income Tax in Selected Countries: A Brief Overview . . . . . . . . . . . . . . . . . . . . . . . . . 

### Annex 1. Corporate Income Tax in Selected Countries: A Brief Overview

### Document structure and related annexes
- Annex 1 title: "Corporate Income Tax in Selected Countries: A Brief Overview" — page 22
- Annex 2 title: "Corporate Income Tax Gap Definitions" — page 23
- Annex 3 title: "Corporate Income Tax Gap Estimation Techniques" — page 25
- Annex 4 title: "Bottom-Up Corporate Income Tax Gap Estimation Techniques in Revenue Administrations with Little Experience and Limited Resources" — page 28
- Annex 5 title: "Random Audit Programs versus Extreme Values Techniques" — page 31

### Key navigational pointers (as listed in the content unit)
- The annexes are organized sequentially with exact page references: 22, 23, 25, 28, 31.
- Annex 1 is positioned ahead of annexes addressing definitions, estimation techniques, resource-constrained implementation, and audit program methodologies.

*Source: tnmea2023006 - Annex 1. Corporate Income Tax in Selected Countries: A Brief Overview.*

### References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

### Corporate Income Tax Gap Estimation by Using Bottom-Up Techniques in Selected Countries

### I. Why Estimate the Corporate Income Tax (CIT) Gap and What Approaches Exist?
- Purpose
  - CIT gap analysis estimates the amount of CIT revenues not collected because of noncompliance, allowing ministries of finance and revenue administrations (RAs) to assess potential revenue mobilization through more effective tax administration.
- Challenges
  - Noncompliance is unobservable; indirect methods must be used.
- Two main estimation approaches
  - Bottom-up: uses RA microdata (random audits, operational audits, administrative data) to directly estimate noncompliance and diagnose root causes.
  - Top-down: uses third-party or macroeconomic aggregates (for IMF RA-GAP, operating surplus from national accounts) to estimate potential CIT and derive the gap by subtracting actual revenue (see Ueda 2018).
- Complementarity
  - Bottom-up techniques better identify root causes and segment-level gaps; top-down provides comprehensive assessment of all revenue foregone.
- RA-GAP context
  - Since the RA-GAP on the CIT gap began in 2017, about six technical assistances have been delivered to estimate CIT gap. Demand is growing, particularly among emerging market economies.
- Bottom-up CIT gap technique types
  - (1) Random audit program results.
  - (2) Operational audit results (risk-selected).
  - (3) General administrative data exploiting non-audit information.
- Limitations and corrective tools
  - Undetected noncompliance is pervasive. Methods to adjust include uplift factors, auditors’ best guesses, Delphi methods, and upper/lower estimate bounds (see notes 6, 7).

### II. CIT Gap Definitions and Measurement Scope
- RA-GAP definition
  - Overall tax gap = potential revenue of the underlying tax base − actual revenue.
  - Split into compliance gap and policy gap.
  - Compliance gap = potential revenue within current tax policy framework − actual revenue collected.
  - Policy gap = potential revenue if all corporate incomes taxed at the current standard rate − potential revenue under current policy framework; affected by policy structure and tax-base composition shifts.
- Bottom-up RA definitions
  - The seven RAs limit the analysis to the compliance gap (excluding policy gap).
  - All consider the gap as theoretical CIT revenue (what would be collected if taxpayers declared and paid amounts due under current policy) − actual CIT revenue.
  - Gross vs net gap nuance:
    - Gross CIT gap = initially observed gap before RA/taxpayer corrective actions.
    - Net CIT gap = observed gap after corrective actions; the RA-GAP definition is equivalent to net CIT gap.
    - Some RAs (example: Canada) present gross and net to illustrate revenue yield from RA actions.
  - Nonfilers and nonregistered taxpayers
    - Bottom-up definitions typically exclude nonfilers and nonregistered gaps; RA-GAP top-down captures all forms of noncompliance via potential-collection estimation.

### III. CIT Gap Bottom-Up Techniques — Methods, Segments, and Practical Issues
- Technique categories (summary)
  - Random audit program: unbiased sample if selection is truly random; requires correction for nondetection (uplifts, multipliers).
  - Operational audits: biased risk-based samples; inference techniques include extreme values (EV), econometric (Heckman two-stage), clustering, and expert judgment.
  - General administrative data: modeling theoretical tax liability for each business (example: Brazil’s frontier production function approach).
- Segment-specific considerations
  - Small businesses
    - Preferred technique: random audit program (most statistically robust for this segment).
  - Mid-sized businesses
    - Often estimated jointly with small or large segments; only Australia estimates mid-sized independently using logistic/linear regressions.
  - Large businesses
    - Most RAs favor operational-audit–based techniques for cost-effectiveness and practicality given smaller, heterogeneous populations and high audit resource intensity.
    - High existing audit coverage in some RAs (example: historically close to 33 percent audit rate in the United States for large corporations, and coverage reaching 80 percent of the largest of the large corporations) makes random sampling less efficient.

- EV (Extreme Values) technique for large businesses
  - Method: assume detected audit adjustments follow a Pareto statistical distribution (log known noncompliance inversely related to log rank); extrapolate audited adjustments to nonaudited population.
  - Key assumptions:
    - Underreported amounts for large businesses fit a Pareto distribution.
    - Audits identify the highest-value adjustments (audit identification is never 100 percent, implying potential underestimation).
  - Data requirements and timing:
    - Multiyear audit databases are needed because large-company audits can take many years to close.
    - United States looks at seven years of audit results; Canada uses eight years.
    - Use of imputation to project open cases (United Kingdom example).
  - Scale examples:
    - United States contemplates about 8,000 entities in the large segment, of which up to 2,700 are annually audited, and only about 100 of the audited entities are included as the EV for estimation.
  - Practical choices:
    - RAs differ on number of EVs to include and on treatment of positive adjustments (standard truncation vs variations).
    - RAs typically use initial auditors’ assessments rather than final litigation outcomes.

- Operational-audit econometric/statistical techniques
  - Heckman two-stage procedure: first stage models selection probability; second stage models audit outcome controlling for selection bias.
  - Clustering techniques: group audited and nonaudited taxpayers by relevant characteristics, then scale audit results within clusters (Canada uses two-step clustering; also uses propensity score matching for validation).
  - Expert judgment: use experienced auditors’ subjective risk assessments to extrapolate.

- Other country practices
  - Sweden: ad hoc method dividing gap into (1) inadequate internal business practices (from audits/controls) and (2) tax avoidance schemes (estimated from a few operational audits; method kept confidential).
  - Australia: uses audit coverage levels and expert judgment plus adjustment factors; projects amendments over years and applies selection-bias discounts and assurance-level discounts (discounts by assurance levels: high assurance 95.45 percent, medium assurance 68.27 percent; uplift for unreported amounts by business assurance category described below).
  - Brazil: benchmark approach using average effective rate of low-risk taxpayers by sector, removing outliers, then estimating dispersion.

- Accounting for undetected noncompliance (corrections and bounds)
  - Uplift factor approaches
    - United States: developed a methodology to estimate an “uplift factor” for detected audit adjustments in individual income tax; not applied to business CIT gap yet.
    - Australia uses uplift factors over CIT base by assurance category: 3 percent (not assured), 2 percent (medium assurance), and 1 percent (high assurance).
  - Upper/lower bound approaches
    - Canada explores an uplift factor and produces upper and lower bounds: upper from clustering technique, lower from EV technique.
    - United Kingdom shifted from adapted uplift factors to lower/upper bound framework: upper bound from EV treating risk-based audits as representative; lower bound from EV under standard Pareto assumption.

- Mixing techniques across segments
  - RAs choose techniques by segment characteristics, internal capability, and population structure.
  - Table summary (text)
    - Large businesses: EV techniques and operational audits used extensively across Australia, Brazil (trial EV), Canada, Denmark (random audits of large), Sweden, United Kingdom, United States.
    - Mid-sized businesses: mixed methods; Australia estimates mid-sized separately; others often combine with large or small segments.
    - Small businesses: random audit programs common across Australia, Brazil (trial EV), Canada, Denmark, Sweden, United Kingdom, United States (though the US uses other techniques for this segment).

- Practical lessons on random audit programs
  - Random audit programs require strict sampling purity to be unbiased; stratified samples introduce varying probabilities and must be accounted for.
  - Random audits can reveal nature of noncompliance by segment:
    - Small businesses: omissions of income, overreporting of expenses.
    - Large businesses: sophisticated issues in depreciation/amortization bases, carryforward losses, nondeductible expenses, shareholder loans.
  - Random audit results can inform risk identification, service improvements, and treatment strategies.

- Special notes and operational caveats
  - Audit-based estimates typically use audit recommendations based on laws and regulations; estimates may include tax avoidance as interpreted by the RA and are not adjusted for subsequent legal appeal results.
  - Scope limitation: bottom-up estimates typically cover domestic CIT obligations; international compliance gap components may be excluded without detailed cross-jurisdictional data.
  - Nondetection varies by RA; some initially applied uplift factors from other countries but are moving toward domestic multipliers (Australia and the United Kingdom initially used US uplift factors, now developing their own).

_Italic source attribution: IMF Technical Note tnmea2023006 — Corporate Income Tax Gap Estimation by Using Bottom-Up Techniques in Selected Countries_

### Annex Box 3.3). This technique's main methodological requirement is to correctly extract a random sample

### Annex Box 3.3). This technique's main methodological requirement is to correctly extract a random sample of entities to be audited and extrapolate the audit results to the population.

### Random audit sampling: design, sample sizes, and exclusions
- Main methodological requirement: correctly extract a random sample of entities to be audited and extrapolate the audit results to the population.
- Sample size is defined by the level of statistical confidence required and is subject to resource limitations.
- The small businesses segment is the largest of the taxpayer segments under all the selected countries' definitions; therefore, the random sample size necessary to guarantee a minimum level of confidence is much lower in proportion to the population compared to other segments.
- Individual random audit cost: each audit involving small businesses is lower than for mid-sized and large businesses because small-business operations tend to be simpler.
- Reported sample sizes per estimation process:
  - Canada: about 4,500 cases in each estimation process, including mid-sized businesses.
  - United Kingdom: about 330 cases per process.
  - Other RAs: sample sizes of fewer than 1,000 cases.
- Some RAs exclude entities before sampling. Examples:
  - Sweden focuses on limited companies and sole traders and excludes dormant entities, entities with no employees, those with low levels of income, and certain small businesses (partnerships, limited partnerships, publicly owned companies, financial corporations, and nondomestic companies).
  - Australia excludes small businesses under a legal vehicle different from an incorporated company.
  - United Kingdom intends to consider the population of small businesses without exclusions, thus including inactive and dormant companies.

### Stratification and sample pooling
- Stratifying the population is widespread; most RAs stratify by turnover and other variables.
- Examples of stratification approaches:
  - Australia: four strata from two turnover size categories and two industry types (companies investing in financial assets versus the rest).
  - Canada: stratifies around sector of activity into one of 21 different economic sectors.
  - Sweden: limited companies into six strata based on annual salary total; sole traders into three strata based on annual turnover.
- Within each stratum, many firms are randomly selected; the number per stratum varies depending on resources and needs to improve estimates in different strata.
- Samples are randomly selected to each stratum; average tax gap for the sample is extrapolated to all other firms of the same stratum.
- Some RAs are analyzing new stratification variables and their effects on overall sample size.
- Sample pooling to improve efficiency:
  - Denmark changed from sampling every three years to sampling every year with a smaller number of cases, then pooling data for two or three years.
  - Australia selects cases each year and bundles audit samples from different years under the assumption that noncompliance is constant over the short term.
  - Canada conducts period random audits (not annual) to better understand compliance trends and enhance risk-assessment systems.

### Standardization, profiling, preprocessing, and audit conduct
- Use of standardized audit checklists, profiling, and preprocessing helps optimize resources.
- Sweden example:
  - Each of the 600 firms selected through the random audit program is audited with a standardized audit tool that reviews about 100 aspects of the firm and its tax returns.
  - If the audit tool finds mistakes or indications of noncompliance, that motivates further inquiries and deeper audit.
  - For about 50 percent of audited companies, there is no indication of noncompliance.
  - On average, an auditor takes about 10 to 12 days to audit a firm, with a large standard deviation; for the largest companies within the sample it takes on average of 17 days.
  - Between 60 to 70 auditors work on the random audit program.
  - The checklist and audit parameters do not change year over year, making data for different years comparable.
- Australia: for each taxpayer sampled, gathers information provided by the taxpayer and from third parties, identifies all tax risks and issues; if no material risks are found the company is considered compliant; preprocessing means audit cases are not strictly randomly selected but overall process efficiency improves.

### Scope and limitations of random audits
- Random audits are usually comprehensive and not risk-based; they can include multiple taxes (CIT, value-added tax, Social Security contributions, self-employment contributions).
- Characterization: random audits are broad but shallow compared to risk-based audits and are not expected to exhaustively cover all types of noncompliance practices.
- Denmark recognizes its random audit program includes a value-added tax gap but does not cover items such as transfer pricing issues, unregistered businesses, economic crime, tax havens, and undeclared work; those gaps are estimated separately by other techniques.

### Operational audits (econometric/statistical) and other techniques
- United States:
  - Uses operational audits for estimating small businesses’ CIT gap with the Heckman approach based on an econometric model of five simultaneous equations:
    1. probability of a CIT return being audited,
    2. probability of detecting underreported CIT conditional on an audit,
    3. amount of underreported CIT conditional on detected underreporting,
    4. probability of detecting overreported CIT conditional on an audit and no detected underreporting,
    5. amount of overreported CIT conditional on an audit and no detected underreporting (Heckman 1979).
  - Equations are estimated combining data from randomly selected audited and unaudited returns.
  - Small companies are categorized into five activity code levels based on reported assets.
  - Independent variables are items from the CIT returns; a risk score for each return received during the regular CIT declaration process is added as an independent variable.
- Brazil:
  - Estimates the CIT gap in micro and small businesses by adapting the stochastic frontier method using administrative information to model a Cobb–Douglas production function.
  - Production is assimilated to the gross tax liability and explained by factors: (1) purchases, (2) salaries and reported employees, (3) bank transactions, and (4) fixed capital, with controls for economic sector, region, and years.
  - Gap for each business is the “vertical distance” between actual revenue level and the estimated frontier based on the most compliant business operating with the same combined use of factors.
  - Model runs over about 2.7 million active taxpayers per year in Brazil and requires high information technology capacities and agile data management tools.
  - Caveat: the method assumes companies on the frontier exhibit perfect compliance; this may not hold in segments with widespread noncompliance practices.

### Accounting for undetected noncompliance and uplift factors
- Random audit nondetection problem is addressed similarly to operational audit–based techniques.
- Some RAs apply an uplift factor over random audit programs’ detected amounts to account for likely undetected noncompliance.
- Recent practice: Sweden and the United Kingdom started formalizing estimation of a CIT uplift factor using the Delphi method based on survey results from an experienced group of auditors (see Annex Box 3.5).
- Canada, Denmark, and the United States do not currently apply an uplift factor for any segment of CIT.

### CIT gap estimates, uncertainty, and reporting practices
- RAs vary in how explicitly they communicate uncertainty.
  - United Kingdom declares an uncertainty rating for each CIT gap estimate: for the most recent years, small and mid-sized businesses CIT gap estimates were assessed a “moderate level" of uncertainty whereas large businesses estimates were considered to have “high level” of uncertainty.
  - Australia uses a similar rating for each population segment CIT gap estimate.
- Public estimated net CIT gap levels (most recent periods available) range from 3 to 15 percent of potential across the RAs:
  - Canada 2018 (lower bound): 5.6
  - Canada 2018 (upper bound): 9.1
  - Denmark 2011–13 annual average: 14.5
  - Sweden 2011–13 annual average: 15.0
  - UK 2020–21 projection: 3.2
  - US 2011–13 annual average: 9.0
  - Australia 2018–19: 11.0
- Difference between gross and net CIT gap corresponds to enforced CIT compliance and late payments after RAs execute compliance and collection actions:
  - Canada example: estimated compliance gap was reduced by more than 11 percent points of potential after accounting for compliance actions.
  - United States: reduction of more than 3 percentage points.
- Share of the CIT gap attributable to large companies exceeds half of the entire CIT gap in most surveyed RAs:
  - Large businesses’ shares of the net CIT gap shown for selected jurisdictions:
    - Canada 2018 (lower bound): 50.5
    - Canada 2018 (upper bound): 69.2
    - Sweden 2011–13 annual average: 69.8
    - UK 2020–21 projection: 75.9
    - US 2011–13 annual average: 70.3
  - Note: United Kingdom is an outlier for reasons including estimate techniques, segment definitions, relative weight of each segment in CIT revenues, and relatively greater compliance risk in small and mid-sized businesses than in large ones.
- Time series and consistency:
  - RAs seek consistency in estimate technique over time to ensure comparability.
  - United Kingdom reports extensive series of CIT gaps including 16 fiscal years by segment and applies a three-year moving average with double weighting for the current year to reduce noise.
  - When methods change, reports include comparisons to previous estimates and note method effects.

### Usefulness of audit-based estimates for diagnosing noncompliance
- Audit-based CIT gap estimates allow estimating distinct noncompliance practices and sectoral concentrations.
  - Denmark publishes CIT gap broken down by economic sector and type of error, enabling evaluation of activities with concentrated compliance risk and distinguishing unintentional errors from deliberate noncompliance.
- Public availability:
  - Almost all RAs in the survey make CIT gap estimates available to the public.
  - Some RAs have explicit government mandates to estimate and publish the tax gap.
  - Internal communication of results is crucial for understanding compliance causes and assessing compliance actions.

*Source: IMF staff based on information from the revenue administrations of the select countries.*

### 1. By Economic Sector2. By Type of Error

### tnmea2023006 - 1. By Economic Sector2. By Type of Error

### The Bottom-Up Techniques
- Bottom-up techniques provide information about the composition of the CIT gap and the types of practices behind the compliance gap, not just the general size.
- A random audit program can provide insight into the level of risk of various noncompliance behaviors in the general population and help identify types of errors and unintentional noncompliance that could be addressed by taxpayer outreach and education rather than audit and reassessment.
- Limitations:
  - Bottom-up techniques are limited in estimating the overall size of the compliance gap because they can work only with the noncompliance issues the administration is potentially addressing.
  - Top-down methods by design can capture the overall size of noncompliance, whereas bottom-up methods require adjustments (for example, uplift factors, Delphi methods, or estimating upper and lower bounds).
- Random audit programs:
  - Offer the greatest relative statistical advantages among bottom-up techniques but are the most expensive to execute (see Annex 5).
  - Require dedicating time and resources in parallel to operational audits, with a high opportunity cost because well-designed random audit programs will consume audit resources in unproductive cases.
  - Can be costly or unfeasible for large businesses due to expected high variability and reduced population requiring extremely high sample sizes for reliable results.
  - Cost-reduction strategies include stratification, preprocessing and profiling, and pooling of samples.
- Extreme values (EV) technique:
  - Low-cost approach but requires a comprehensive database of audit results spanning many years (see Annex 5).
  - Requires audit management systems to record audit results, fiscal periods covered, causes of audit changes, details of CIT adjustments, and ideally more than a decade’s worth of results.
  - EV-based estimates need reestimation given long delays in completing audits that change levels and trends in audit results over time.
- Combination approach:
  - A well-designed bottom-up program can combine techniques to target gap estimates for different taxpayer segments: random audit programs are generally most suitable for small businesses; techniques based on operational audit data work better for large businesses; both can be extended to mid-sized businesses.
  - Econometric or statistical approaches are used less frequently and targeted to particular business segments.

### Revenue Administration Strategies to Apply Bottom-Up Techniques
- Developing bottom-up CIT gap estimates typically requires several years of research; RAs without prior experience should plan medium- and long-term programs.
  - Example: An RA without previous random audit program experience should consider planning for at least three years from the start of audits until first results are obtained.
- Institutional and external support:
  - All RAs allocate some fraction of institutional resources to estimation processes and receive support from government authorities and, in most cases, from statistical offices and academic units.
- Specialized staff:
  - Most RAs assign bottom-up applications to specialized staff who accumulate knowledge and experience, produce guidance manuals, control completed audits for quality, provide feedback to case workers, and identify statistical or IT support needs.
- Audit data management:
  - Strong institutional orientation to the management of audit data is required. RAs gather detailed information on each audit, including audited periods, adjustments made, execution dates, dedicated auditors, and types of errors detected.
  - For EV techniques, RAs should accumulate and maintain detailed and reliable information from multiple years of audit results and not retain results for only a limited time.
- Public mandate and transparency:
  - A public mandate to estimate the CIT gap and maintaining a regular publication schedule helps foster internal and external support and resource allocation.
  - Transparency of results helps provide credibility and reduces the risk of misinterpretation.

### Annex Highlights — Actual Corporate Income Tax Revenue and Rates
- Average corporate income tax (CIT) revenue for 2016–20 varies between 1.4 and 5.1 percent of GDP in the selected countries.
- For Australia, Canada, Denmark, and Sweden, CIT revenue levels increased compared with averages for 2011–15.
- For Brazil, the United Kingdom, and the United States, CIT revenue levels show slight decreases compared with averages for 2011–15.
- The average CIT top combined rate for 2016–20 ranges between 19.3 and 34.0 percent in the selected countries.
- For Australia, Brazil, and Canada, the CIT top combined rates for 2016–20 do not show changes compared to the averages for 2011–15.
- For Denmark, Sweden, the United Kingdom, and the United States, decreases in the CIT top combined rates for 2012–20 are observed when compared to averages for 2011–15.

### Annex Highlights — Definitions (selected)
- Australia:
  - Gross CIT Gap: Difference between the amount voluntarily reported to the Australian Taxation Office (ATO), and the amount that would have been collected if every taxpayer were fully compliant with tax law (i.e., the theoretical tax liability).
  - Net CIT Gap: Difference between the amount voluntarily reported to the ATO plus amendments as a result of compliance activities and voluntary disclosures, and the amount that would have been collected if every taxpayer were fully compliant with tax law.
- Brazil:
  - Gross CIT Gap: Difference between the potential tax liability under the current tax system and the actual liability as declared by taxpayers.
  - Net CIT Gap: Gross CIT gap after subtracting enforced and late payments.
- Canada:
  - Gross CIT Gap: Difference between the CIT that would be paid if all obligations were fully met in all instances and CIT that is actually paid and collected before accounting for compliance and collection actions.
  - Net CIT Gap: Gross CIT gap after subtracting compliance and collection activities results.
- Denmark and Sweden: distinction between gross and net not always explicitly used but information available to estimate both.
- United Kingdom:
  - Gross CIT Gap: Difference between the theoretical tax liability (TTL) and the “voluntary” receipts.
  - Net CIT Gap: Difference between the TTL and the “total” receipts (voluntary plus compliance yield receipts).
- United States:
  - Gross CIT Gap: Difference between total true CIT liability (TTCL) and CIT paid voluntarily and timely.
  - Net CIT Gap: Difference between TTCL and total CIT payments (voluntary and timely plus enforced and late payments).
  - Note: In the United States, TTCL includes underreported (net of overreported) CIT liabilities. The CIT gap definition does not include tax avoidance.

### Annex Highlights — Estimation Techniques and Method Steps
- Annex Box 3.1. Extreme Values Technique for Large US Corporations (summary of steps):
  - Step 1: Collect operational audits databases for large corporations.
  - Step 2: Extract the amount of the NARs on each audited return.
  - Step 3: Truncate the data for the top N number of NARs from the large corporations. Sum net recommended tax change for all operational audit cases (S) with a refund amount; record as R; delete cases with refund amount or no tax change; sort remaining cases ascending by tax change; compute cumulative sum; identify audit case number (m) where cumulative sum ≤ R; delete cases up to m; let N be remaining cases; p = N/S.
  - Step 4: Order extreme NARs by rank r.
  - Step 5: Estimate linear relationship Log10(NAR) dependent on Log10(rank); let a and c be slope and y-intercept.
  - Step 6: Extrapolate remaining cases to whole population of large corporations (F); M = p × F.
  - Step 7: Estimate total underreported CIT U = 10^c × sum_{r=1 to M} r^a.
  - Note: Largest corporations defined as those with assets of more than $250 million.
- Annex Box 3.2. Logistic/Linear Regression Technique for Mid-Sized Businesses:
  - Step 1: Run logistic regression to determine probability of noncompliance for each company; dependent variable = 1 for noncompliant, 0 for compliant; characteristics include gross distributions from trusts, total turnover, expenses, age, effective tax rate, etc. Australia uses Monte Carlo simulation with more than 20,000 iterations instead of a probability threshold.
  - Step 2: Run linear regression to estimate share of unreported tax over last tax for noncompliant entities; correct for selection bias using propensity score matching; apply linear regression to each company.
  - Step 3: Combine results of Steps 1 and 2; apply potential size estimates to companies predicted noncompliant; results are average (including amendments) of 20,000 iterations.
  - Step 4: Apply nondetection uplift factor (small-business uplift factor for lower turnover mid-sized companies; large-business uplift factor for greater turnover mid-sized companies).
  - Step 5: Consolidate the tax gap estimates.
- Annex Box 3.3. Random Audit Program for Small Companies:
  - Step 1: Estimate unreported amounts by identifying average amendment and amendment rate in sampled taxpayers; extrapolate sampled amendment to whole population to obtain unreported tax liability base.
  - Step 2: Estimate nondetection using an uplift factor (independently determined or based on external estimates).
  - Step 3: Estimate net and gross gap: sum of Steps 1 and 2 = gross gap; deduct compliance outcomes and voluntary disclosures to obtain net gap.
  - Step 4: Estimate theoretical tax liability as sum of net gap and net tax paid.
- Annex Box 3.4. Stochastic Frontier Method (Brazil implementation):
  - Practical steps: (1) obtain population information from RA databases; (2) extract sample with good data quality for frontier calculation; (3) treat outliers and missing data to include them indirectly in the model; (4) calculate declared tax gross operating surplus and tax liability; (5) estimate the corporate income tax gap for each business; (6) treat and present aggregated results.
- Annex Box 3.5. Delphi Method for Nondetection Multipliers (Sweden):
  - Methodology based on questionnaires to experienced auditors and senior coworkers in a recursive process to converge on an assessed amount that could be corroborated if auditors had all instruments and time for an operational audit.
  - Results indicated Sweden should receive about 27 percent more in taxes from incorporated companies' tax gap and 17 percent from sole proprietorships' gap; an additional amount was projected over the found tax gap in both segments; a remaining nonverifiable tax gap exists that cannot be detected through audit methodologies.

*Source: IMF staff calculations based on Danish Customs and Tax Administration (2017) and compilations from revenue administrations of selected countries.*

### Annex 4. Bottom-Up Corporate Income

### Annex 4. Bottom-Up Corporate Income

### Rationale for using bottom-up techniques
- Top-down corporate income tax (CIT) gap estimates are typically based on the gross operating surplus macro variable from national accounts (Ueda 2018).
- Reasons a revenue administration (RA) with little experience or limited resources might explore bottom-up techniques:
  - To test results obtained via the top-down approach when its gap-level figures raise doubts or the macro variable is not updated or cannot reliably be broken down into required components.
  - To generate a deeper understanding of the composition of the CIT gap: what noncompliance practices are most frequent, what types of companies are more prone to errors or noncompliance, and what controls, audits, or tax policy adjustments could reduce the CIT gap.
- Bottom-up techniques provide insights into noncompliance composition that are unlikely to be obtained from a top-down approach.

### Costs, data, and human-resource requirements
- Specialized staff requirements:
  - Staff backgrounds include knowledge of CIT legislation, management of micro databases, and statistical and economic knowledge.
  - Training personnel with these characteristics can represent a significant effort for the RA.
- Data and IT requirements:
  - Reliable, updated, and comprehensive databases of CIT forms, audit reports, and third-party databases are necessary.
  - For the Extreme Values (EV) technique, data for several years are required.
  - Computer tools to process databases for analytical purposes are critically necessary.
  - Example: the United States manages at least 10 years of very detailed information for each completed CIT audit (adjustments made, execution dates, dedicated auditors, type of errors detected, and so forth), which is essential for applying the EV technique.
- Personnel costs and time for Random Audit Programs (RAPs):
  - Personnel costs rise compared to EV because auditor hours must be dedicated to executing randomly selected cases.
  - RAPs require a design stage, an execution stage, and a result compilation stage; results cannot be obtained until two or three years after the program starts.
  - Budgetary limitations and efficiency objectives often require adjusting RAPs to reduce audit costs.
  - The number of random audits executed over small businesses does not drop below 300 cases per process; pooling to complete execution in three years would partialize the effort to about 100 cases per year.
  - Random audits are usually comprehensive; unit cost in the small business segment may be less onerous than in mid-sized and large segments.
- Timing implications:
  - Many audits in EV applied to large businesses can take longer, sometimes up to 10 years, to close each audit when applied to large businesses.
  - RAPs often mean estimates for current years are based on projections since many audits would not yet be closed.

### Nonfilers and nonregistered businesses under bottom-up techniques
- Primary bottom-up techniques (EV and RAPs) do not intrinsically include nonfiling and nonregistration in potential revenue.
- In some countries, nonfiling/nonregistration is assessed by other gap estimation techniques (for example, methods for estimating the underground economy).
- The note did not compile estimates for nonfilers/nonregistered contribution to the CIT gap; some countries reported its contribution was not too significant, citing:
  - Off-registration economic activity strongly limited by various government agencies' control.
  - Nonfilers/nonregistered entities may have reduced potential to generate net CIT liabilities (but may have revenue potential in other taxes such as tax on wages, social security, municipal fees).
- In other countries analyzed, nonfilers and nonregistered businesses have a nonnegligible effect on the CIT potential base.
  - Countries starting RAPs that sample from highly incomplete returns or registration databases should note RAP results could significantly underestimate the CIT gap.
  - RAP results can still provide evidence about the phenomenon if audited businesses are detected having transactions with nonfilers or nonregistered suppliers or users.
  - Australia has used its RAP for small businesses to detect entities outside the tax system (for example, cash-only businesses operating without an Australian Business Number).
  - EV technique, usually applied to large businesses, is less affected by nondeclaration or nonregistration.

### Using bottom-up results to inform risk selection
- A key opportunity from implementing a RAP is comparing the audit yield of the RAP with yields from existing risk-based programs.
  - RAP yields are expected to be lower because the sample includes taxpayers with minimal compliance risk.
  - RAPs enable evaluating the effectiveness of the risk-based model and recalibrating it.
- Example from Denmark:
  - The CIT gap estimated by a RAP can be broken down into seven compliance risk levels.
  - An expected result is that CIT compliance gap increases with assigned risk level; failure to observe this would signal the need to recalibrate the risk model using RAP results.
  - Annex Figure 4.2 (Denmark, 2014) shows CIT gap amounts per business (DKK) by risk level: 9,200; 5,300; 5,600; 23,600; 98,200; 161,700; 210,000.
  - Annex Figure 4.2 also reports proportions of businesses (percent) by risk level with axis labels shown as 0; 50; 20; 10; 30; 60; 40 in the figure.

### Example timeline for a RAP (Denmark)
- Income year: 2021 (January 1–December 31).
- Filing date: July 1, 2022.
  - Note: Because of the pandemic the filing date was postponed from July 1 to September 1 for income years 2019 and 2020.
- Preparing data for random sample: July/August 2022.
- Audits are carried out: September 1, 2022–August 31, 2023.
- Finalizing the last audits + detailed error type registration: September 2023–October 2023.
- Analysis and writing report: November–December 2023.

### Comparative advantages and disadvantages: Random Audit Programs versus Extreme Values Techniques
- Advantages of Random Audit Programs:
  - Suitable for small and mid-sized businesses.
  - Like in any other random sample, extrapolation to the population is straightforward.
  - Better as a statistical tool for obtaining unbiased estimates of noncompliance in the current year, subject to a determined confidence interval.
  - Allows a breakdown of results identifying different causes of tax noncompliance (types of mistakes, intentional versus unintentional mistakes, demographic breakdowns, and so forth).
  - Assesses and feeds current risk analysis, allowing evaluation of the effectiveness of risk-based processes and recalibration.
  - Can be used by audit managers to train new auditors because the program is usually fixed every year.
- Disadvantages of Random Audit Programs:
  - Would not be efficient if applied to large businesses.
  - Consumes time and resources, requiring at least three years to produce results for a single income year and with resources related to sample size.
  - Could face internal resistance when starting because yield is normally lower than that of risk-based audits.
  - Involves broader and shallower audits, increasing the probability of nondetection.
  - Fails to capture some noncompliance practices when auditors focus only on a specific audit hypothesis.
  - Requires assessing practical issues: replacement of cases already audited, pool of years to achieve minimum sample size, variability of auditors' abilities to detect noncompliance, and so forth.
  - Results for current years may be based on projections since many audits would not yet be closed.
- Advantages of Extreme Values Techniques:
  - Suitable for large businesses and results could be extended to mid-sized businesses.
  - Simpler and less expensive in resources and time than other statistical techniques.
  - Allows obtaining corporate income tax gap results for several years.
  - The main assumption—underreported corporate income tax for largest businesses fitting a Pareto distribution—should be applicable to any tax system.
  - Exploits existing audit results, thus not requiring additional resources to program extra audits or controls.
  - Permits audits involving deeper control, thus lowering the probability of nondetection because of their basis in risks.
- Disadvantages of Extreme Values Techniques:
  - Is not applicable to small businesses.
  - Requires more assumptions than other techniques; requires setting the tail for the Pareto distribution.
  - Takes longer, sometimes up to 10 year, to close each audit when applied to large businesses.
  - Results could be sensitive to two or three important cases in a year; “production” of adjustments could vary and affect results.
  - Allows comparison of results from one income year to the other, being less sensitive to a few cases, except if audit methods systematically change.

*Source: Annex 4 and Annex 5, "Corporate Income Tax Gap Estimation by Using Bottom-Up Techniques in Selected Countries: The Revenue Administration Gap Analysis Program," TNM/2023/06.*

---


_Source: https://www.imf.org/-/media/files/publications/tnm/2023/english/tnmea2023006.pdf_