## 4.4 Within-Country Identification. ## Source details **Canonical URL:** [4.4 Within-Country Identification.](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025255-source-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025255-source-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025255-source-pdf.pdf.json) --- ### Introduction and context - Mobile money expanded payments and financial access across low- and middle-income countries, especially in Africa, enabling P2P transfers, wage payments, and utility bill payments using basic phones. - After Kenya’s 2013 adoption of a mobile money tax, at least fifteen Sub-Saharan African countries had introduced similar levies by 2024. - Mali implemented a mobile-money tax effective March 5, 2025, and Senegal’s National Assembly adopted legislation in September 2025 to tax mobile-money transfers. - Debate: proponents emphasize revenue and ease of collection; critics emphasize efficiency losses, risks to financial inclusion, and potential regressivity where formal banking is scarce. ### Model setting and identification objects - Transfer media: cash, bank accounts, and mobile money; transfer size normalized to one. - Costs and tax: - Cash transaction cost: c > 0. - Mobile-money tax per unit: t; effective mobile-money cost: M(ω) + t. - Assume c − t > 0. - Sender types: - Unbanked (cannot make bank transfers). - Banked (can use bank transfers at cost B, holding account costs φ with cdf G(φ)). - Define: - W = min{M + t, c} (unbanked least-cost, tax-inclusive). - Z = min{M + t, B, c} (banked least-cost, tax-inclusive). - ∆ ≡ E[W] − E[Z]; equilibrium banked share = G(∆). - Key analytical objects: d∆/dt, dP_W/dt, dP_Z/dt and associated elasticities, where - P_W ≡ P(W = M + t) = P(M + t < c) - P_Z ≡ P(Z = M + t) = P(M + t < min(B, c)) ### Theoretical comparative statics and propositions - Behavioral derivatives: - dP_W/dt = − f_M(c − t) < 0, where f_M(m) = ∫_0^∞ f_{M,B}(m,b) db. - P_Z(t) = ∫_0^{c−t} ∫_{m+t}^{∞} f_{M,B}(m,b) db dm. - dP_Z/dt = − ∫_0^{c−t} f_{M,B}(m,m + t) dm − ∫_c^{∞} f_{M,B}(c − t,b) db < 0. - For small t and when c lies in the upper tail, |dP_Z/dt| > |dP_W/dt| (banked more responsive). - Key formal statements: - Proposition 1: d/dt ∆(t) = P(B ≤ M + t < c) > 0. The gap ∆ increases with the tax. - Corollary 1: The tax increases the proportion of the population with bank accounts. - Proposition 2: If the tax rate is small and c is in the upper tail (i.e., ∫_0^c f_{M,B}(b,b) db > ∫_0^c f_{M,B}(c,b) db), the proportional response to the tax is larger (more negative) for banked individuals than for unbanked individuals. - Proposition 3 (efficiency): The mobile money tax creates economic inefficiency. Excess burden equals the product of the tax rate and its effect on the incidence of mobile money transfers: EB = −t d/dt P (positive). - Proposition 4 (informality): The mobile money tax increases the use of cash for transfers; d/dt P(Z = c) > 0 and d/dt P(W = c) > 0. - Proposition 5 (average-cost effect): Allowing transfers to cease when U < transaction cost yields |d/dt P_UW(t)| > |d/dt P_UZ(t)| — reductions in transfer incidence are relatively larger for unbanked individuals. ### Within-country empirical identification (Cameroon and CAR) - Identification approaches: - Within-country difference-in-differences: compare taxed transactions (P2P transfers and cash-outs) with untaxed transactions (merchant payments and other exempt types) within the same country using user fixed effects and time fixed effects. - Event-time specification: y_{i,j,t} = ∑_{k=-6}^{5} β_k τ_{k j,t} + μ_i + λ_t + ε_{i,j,t} with pre-treatment detrending of each transaction type’s slope. - Data and exemptions: - Tax exemptions in Cameroon and CAR: merchant payments and some minor transaction types are untaxed, enabling within-country treated vs. untreated contrasts. ### Within-country empirical findings (event-study, detrended) - Pre-tax: taxed and untaxed transactions follow parallel trends. - Post-tax effects (taxed vs. untaxed): - Cameroon: drop of around 50,000 XAF in monthly value of taxed transactions relative to untaxed transactions — about 40% of the pre-treatment mean of 125,347.5 XAF. - CAR: decline similar in magnitude — corresponds to 36% of pre-treatment mean of 113,993.2 XAF. - Number of transactions: tax reduces number by roughly 10% in Cameroon and about 8.5% in CAR. ### Identification via matching with Mali (untreated benchmark) - Rationale: compare taxed transaction dynamics in Cameroon and CAR with those in Mali to mitigate SUTVA concerns. - Matching procedure: 1. Restrict Cameroon sample to users in majority-Muslim regions for benchmarking (not applicable to CAR). 2. Propensity-score (logit) using gender, age, urban/rural, and pre-tax average monthly taxed transaction value (with transformations and interactions). 3. Nearest-neighbor matching to Mali users with similar propensity scores. - Matched event-study results: - Both Cameroon and CAR exhibit sharp and persistent declines after the tax. - Absolute declines: - Cameroon: total monthly transaction value per user falls by more than 50,000 XAF (≈ 40% relative to pre-tax mean 125,347.5 XAF). - CAR: drop slightly larger in absolute terms (≈ 47% relative to pre-tax mean 113,993.2 XAF reported in matching benchmark). - Transaction counts: - Pre-treatment means: 6.2 transactions (Cameroon) and 5.77 transactions (CAR) per user per month. - Cameroon's counted decline: 33% (roughly a reduction of 2 transactions). - CAR's counted decline: 51% (roughly a reduction of 2.9 transactions). - Both P2P transfers and cash-outs decline; untaxed types (cash-in deposits, merchant payments) show weaker or no responses. ### Heterogeneity (Cameroon focus) - Splits examined: 1. Regions with low vs. high bank penetration (bottom quartile vs. top quartile). 2. Urban vs. rural regions. - Findings: - Declines in mobile money use larger in regions with greater banking access — consistent with substitution toward bank transfers. - Effects weaker and often statistically insignificant in low-bank-penetration regions. - Reductions stronger in urban than rural areas. - Interpretation: banked and urban users (with outside options) adjust more; users in financially less developed and rural areas bear a disproportionate burden. ### Price elasticity estimates - Cameroon: - Average increase in consumer cost for taxable transactions: 19.45 percent (tax plus operator fees relative to pre-tax fee). - Observed 40% decline in average monthly transaction value per user for taxed transactions. - Estimated short-term price elasticity ≈ −2.1 (computed as 40% decline divided by 19.45 percent price increase). - Interpretation: a 10% increase in tax-inclusive consumer price associates with about a 21% decline in average monthly transaction value for taxed transactions. - CAR: - Pre-tax transfer price for P2P was zero (no operator fee); the 1% levy raises price from 0 to positive, so percentage change is undefined and no comparable elasticity reported. ### Within-country and matched-sample policy-relevant magnitudes (selected exact figures) - Cameroon tax parameters and fee impacts: - Tax levied at 0.2% of amount transferred and on amount withdrawn; combined cashout rate effectively 0.4%. - Application increased total transaction costs by approximately 20% at the lower end and up to 400% at the upper end of operator fee schedule. - Examples: - Sixth bracket (50,001–80,000 XAF): tax = 100–160 XAF; original fee = 100 XAF. - Highest bracket (500,001–1,000,000 XAF): tax = 1,000–2,000 XAF; original fee = 500 XAF. - Cameroon later introduced (January 1, 2025) a 4 XAF flat fee per mobile money transaction in addition to the proportional tax (applies to banks, credit institutions, microfinance entities). - CAR tax parameters: - Tax rate: 1% of transaction value applied to P2P transfers and cash withdrawals; deposits and payments to registered merchants exempt. - Tax amounts examples: 10–20 XAF in lowest bracket (1,000–2,000 XAF); 2,500–5,000 XAF in highest bracket (250,005–500,000 XAF). ### Within-country identification: interpretation and links to theory - Empirical patterns consistent with theoretical predictions: - Support for Proposition 2: mobile money usage declines after tax introduction, with larger proportional responses among banked users. - Support for Proposition 4: substitution toward cash observed in survey and transaction data. - Combined margins: evidence suggests substitution effect (toward bank transfers and cash) dominates average-cost effect in short run; unbanked users are less able to substitute and thus bear more of the tax burden. ### Policy implications (within-country perspective) - Mobile money taxation (value-based): - Unambiguously reduces mobile-money use. - Empirically elicits larger proportional reductions among banked users but raises effective transaction costs disproportionately for unbanked and poorer users. - Triggers substitution into cash, increasing informality and undermining formalization and traceability. - Generates deadweight loss: efficiency costs exceed revenue gains at the margin (excess burden computed elsewhere equals tax rate times fall in taxable base). - Short-run welfare-relevant estimates (linked calculations elsewhere in the source): - Estimated short-term price elasticity in Cameroon ≈ −2.1. - Within-sample predicted reduction in taxable transactions: 40 percent. - These within-country impacts imply substantial revenue–welfare trade-offs and distributional consequences to be considered when designing transaction taxes. *Source: wpiea2025255-source-pdf - 4.4 Within-Country Identification. (excerpt provided).* ### 4.4 Within-Country Identification. . . . . . . . . . . . . . . . . . . . . . . . . . . . ### 4.4 Within-Country Identification. ### Introduction - Mobile money has expanded payments and financial access across low- and middle-income countries, especially in Africa, enabling P2P transfers, wage payments, and utility bill payments using basic phones. - After Kenya’s 2013 adoption of a mobile money tax, at least fifteen Sub-Saharan African countries had introduced similar levies by 2024. - Mali implemented a mobile-money tax effective March 5, 2025, and Senegal’s National Assembly adopted legislation in September 2025 to tax mobile-money transfers. - Debate: proponents emphasize revenue and ease of collection; critics emphasize efficiency losses, risks to financial inclusion, and potential regressivity where formal banking is scarce. ### Model: setting and motives - Transfer media considered: cash, bank accounts, and mobile money. - Transaction costs vary by channel and depend on transfer circumstances ω ∈ Ω; transfer size normalized to one. - Cash transaction cost is fixed at c > 0 and assumed high relative to typical realizations of B and M. - Tax per unit on mobile money is t; effective mobile-money cost is M(ω) + t. Assume c − t > 0. - Two sender types: (i) unbanked (cannot make bank transfers) and (ii) banked. - Define: - W = min{M + t, c} (tax-inclusive least-cost for unbanked) - Z = min{M + t, B, c} (tax-inclusive least-cost for banked) - Holding a bank account has fixed cost φ with cdf G(φ); equilibrium banked share is G(∆) where ∆ ≡ E[W] − E[Z]. ### Theoretical framework and key objects - Transaction-cost distributions: (M, B) random vector with joint density f_{M,B}(m,b). - Probabilities that mobile money is least-cost: - P_W ≡ P(W = M + t) = P(M + t < c) - P_Z ≡ P(Z = M + t) = P(M + t < min(B, c)) - Analytical objects of interest: d∆/dt, dP_W/dt, and dP_Z/dt (and associated elasticities). ### Expected transaction costs and partitioning - W(t) = min(M + t, c) with W(t) = M + t if M + t < c, c otherwise. - Z(t) = min(M + t, B, c) with Z(t) = M + t if M + t < min(B, c), min(B, c) otherwise. - The (m, b) space partitions into six regions determining Z values (Table 1 and Figure 2 in source). - E[Z] ≤ E[W] with strict inequality when bank transfers are least-cost with positive probability; define ∆ ≡ E[W] − E[Z] > 0. ### Behavioral responses and comparative statics - dP_W/dt = − f_M(c − t) < 0, where f_M(m) = ∫_0^∞ f_{M,B}(m,b) db. - P_Z(t) = ∫_0^{c−t} ∫_{m+t}^{∞} f_{M,B}(m,b) db dm. - dP_Z/dt = − ∫_0^{c−t} f_{M,B}(m,m + t) dm − ∫_c^{∞} f_{M,B}(c − t,b) db < 0. - For small t and when cash cost c lies in the upper tail of the distribution, dP_Z/dt < dP_W/dt in magnitude: banked individuals are relatively more responsive in reducing mobile-money usage. ### Key propositions and implications (preserve terminology) - Proposition 1: d/dt ∆(t) = P(B ≤ M + t < c) > 0. The gap between expected tax-inclusive transaction costs of unbanked versus banked increases with the tax. - Corollary 1: The tax increases the proportion of the population with bank accounts. - Proposition 2: If the tax rate is small and c is in the upper tail of the transaction-cost distribution (i.e., ∫_0^c f_{M,B}(b,b) db > ∫_0^c f_{M,B}(c,b) db), then the proportional response to the tax is larger (more negative) for banked individuals than for unbanked individuals. - Intuition: bank accounts provide an additional untaxed substitute (substitution effect), whereas unbanked users face a larger average-cost effect that can reduce the net benefit of a transfer below its transaction cost. ### Empirical evidence summarized in source text - Cross-country survey evidence (IMF Financial Access Survey and World Bank Global Findex Database) shows taxation lowers number of active mobile money accounts and likelihood of using mobile money. - Heterogeneity: reductions are weaker among lower-income and unbanked individuals (consistent with limited access to formal banking); suggestive evidence of substitution into cash. - Transaction-level microdata from a mobile operator in Cameroon and Central African Republic (CAR) with Mali as control (propensity score matching) find: - Average monthly value of taxed transactions declines by 40% in Cameroon and 47% in CAR relative to Mali. - Number of transactions declines by 33% in Cameroon and 51% in CAR relative to Mali. - Reductions strongest among banked users; lower-income and rural users are less responsive and face higher effective taxation due to weaker bank access. - Overall conclusion from theory and evidence: mobile money taxes are inefficient and regressive, undermine financial inclusion, and raise little revenue. - Excess burden estimate: excess burden of approximately 33% of mobile-money tax revenue. ### Policy-relevant mechanisms and takeaways - Taxing mobile money: - Unambiguously reduces mobile-money use. - Is theoretically ambiguous whether reduction is larger for banked or unbanked, but empirically banked users respond more elastically due to access to untaxed substitutes. - Is regressive because it increases transaction costs disproportionately for unbanked users. - Triggers substitution into cash, increasing informality. - Generates deadweight loss: at the margin, the increase in users’ transaction costs exceeds the increase in tax revenue. *Source: wpiea2025255-source-pdf - 4.4 Within-Country Identification. (excerpt provided).* ### 2.6 Efficiency ### 2.6 Efficiency ### Excess burden (economic inefficiency) - The economic cost of the tax is computed as the difference between the increase in the expected transaction cost, inclusive of the tax, and the increase in expected tax revenues arising from a marginal increase in the tax rate. - For banked individuals: - d/dt E[Z(t)] = P(M + t < min(B, c)) = P_Z(t).(11) - ER(t) = t P_Z(t).(12) - d/dt [ER(t)] = P_Z(t) + t d/dt P_Z(t) - Excess burden: EB(Z) = d/dt [E[Z(t)] − ER(t)] = −t d/dt P_Z(t) > 0.(13) - For unbanked individuals: - EB(W) = −t d/dt P_W(t) > 0.(14) - Proposition 3: - The mobile money tax creates economic inefficiency. - The size of the excess burden equals the product of the tax rate and its effect on the incidence of mobile money transfers. - Interpretation: - The efficiency cost reflects the resource cost of individuals switching away from mobile money transfers as a result of the tax, which causes revenues to increase by less than the cost to taxpayers. - Thus, while the tax mobilizes revenues, it generates economic losses. ### 2.7 Informality ### Impact on cash use and informality - Informality is closely associated with the use of cash for transactions; cash is relatively expensive due to risks and inconveniences (e.g., carrying and storing cash and the risk of theft), and it facilitates avoidance of electronic traceability (e.g., for tax evasion). - The tax’s impact is studied via d/dt P(Z = c) for banked individuals and d/dt P(W = c) for unbanked individuals. - Proposition 4: - The mobile money tax increases the use of cash for transfers. - Implication: - The tax raises revenue from informal mobile-money transactions but also pushes more activity back into the cash economy, where informality is harder to detect. ### 2.8 Average Cost Effect of the Tax ### Extension: allowing the tax to reduce the number of transfers (extensive margin) - Model extension: - Each potential transfer has a utility draw U (net of the transfer amount but before transaction costs) with density f_U(u), independent of (M, B), supported on (0, ∞). - Transfer occurs iff U ≥ Z (banked) or U ≥ W (unbanked). - P_UZ(t) = P(U ≥ Z) and P_UW(t) = P(U ≥ W) denote transfer incidence. - Proposition 5: - d/dt P_UW(t) < d/dt P_UZ(t) in absolute terms, i.e., - |d/dt P_UW(t)| > |d/dt P_UZ(t)|. - The reduction in the incidence of transfers induced by the tax (the average-cost effect) is relatively larger for individuals without bank accounts. - Intuition: - Banked individuals have an additional substitute (bank transfers), so the set of realizations under which mobile money is least-cost is smaller; their transfer incidence declines less sharply with the tax. - Because the reduction in transfers carries a direct utility cost—and is more severe for the unbanked—this average-cost effect further reinforces the regressive and inefficient nature of the tax. - Combined margins: - The tax reduces mobile money transfers through (i) substitution effect (mobile money as least-cost method) and (ii) average-cost effect (net utility of transfer exceeding transaction cost); both probabilities fall with the tax. - Relative decline in mobile money transfers between banked and unbanked is ambiguous: the unbanked exhibit a smaller substitution effect but a larger average-cost effect. - Because the tax leaves unchanged the cost of bank and cash transfers, the average-cost effect is neutral for bank and cash transfers. - Cash usage increases unambiguously due to the substitution effect away from mobile money. - Data strategy note: - The propositions span multiple behavioral margins that no single dataset captures. The authors combine cross-country survey evidence and transaction-level data from Cameroon and the CAR to confront the model with data, focusing on Proposition 2 and using estimates to gauge implications for Propositions 3 and 5, with preliminary evidence consistent with Proposition 4. ### 3 Evidence from Cross-Country Surveys ### Data sources and treatment - Outcome variables: - IMF Financial Access Survey: annual country-level information on number of registered and active mobile money accounts in African countries from 2004 to 2023. - World Bank Findex database: surveys 127,602 individuals across African countries in four waves (2011, 2014, 2017, and 2021–2022); respondents report whether they used mobile money in the past year. - Treatment variable: - Equals one if country c has a mobile money tax in place at time t. - Timing and location of tax introductions compiled from national budget laws, identifying 15 adopting countries. - The 15 adopting countries are Benin, Burkina Faso, Cameroon, Central African Republic, Chad, Democratic Republic of Congo, Côte d’Ivoire, Ghana, Kenya, Nigeria, Republic of Congo, Tanzania, Uganda, Zambia, and Zimbabwe.14 ### Empirical strategy - Concern: - Staggered introduction of mobile money taxes across Africa can bias standard two-way fixed effects (TWFE) estimators. - Estimators used: - IMF panel (country-year): Callaway and Sant’Anna (2021) estimator to construct cohort- and period-specific average treatment effects and aggregate into event-study coefficients. - Findex repeated cross-sections: imputation-based estimator of Borusyak et al. (2024), modeling untreated outcomes using never-treated and not-yet-treated units, then imputing counterfactuals for treated units. - Controls: - Country-level: GDP per capita in constant 2015 USD (log) and number of ATMs (log). - Findex regressions include individual-level controls for age, age squared, education, gender, and income quintile. - Standard errors clustered at the country level. ### Key empirical findings - Event-study (IMF Financial Access Survey): - Mobile money taxation significantly reduces active accounts per 1,000 adults and transactions per 1,000 adults. - Declines become statistically significant in the second year after tax introduction and persist into the third year. - By the second year: - Number of active accounts per 1,000 adults is about 20% lower relative to the counterfactual without the tax. - Number of transactions per 1,000 adults is over 10% lower relative to the counterfactual without the tax. - Individual usage (Findex; Table 2): - Introduction of a mobile money tax reduces the probability that individuals report using mobile money in the previous year. - Average treatment effect: around a 8 percentage points reduction—relative to a pre-treatment mean of about 22% (Mean (Pretreat.) = 0.215). - Estimates: - Column (1) TWFE: Mobile Money Tax = -0.089 (0.037) ∗∗ - Column (2) TWFE with controls: -0.082 (0.038) ∗∗ - Column (3) BJS estimator: -0.080 (0.047) ∗ - Column (4) BJS estimator with controls: -0.075 (0.007) ∗∗∗ - Observations: columns (1)-(2): 87,823; columns (3)-(4): 85,938. - Standard errors in parentheses; significance levels reported. - Heterogeneity (Figure 4): - Stronger decline in mobile money use among individuals with bank accounts (those with outside options); banked users reduce probability of using mobile money by about 20 percentage points. - By income quintile: larger responses for higher-income groups; middle-income and poorest households show little adjustment and therefore are more likely to bear the incidence of the tax. - Substitution toward cash (Table 3; cash withdrawals): - Probability of actively withdrawing cash from a bank account rises by about 5 percentage points following the introduction of a mobile money tax (Table 3). - Table 3 estimates (selected columns): - Column (1): Mobile Money Tax = 0.274 (0.006) ∗∗∗, Controls: No, Country FE: Yes, Observations: 24,779, Mean (Pretreat.) = 0.398. - Column (2): 0.044 (0.020) ∗∗, Controls: No, Country FE: Yes, Year FE: Yes. - Column (3): 0.051 (0.014) ∗∗∗, Controls: Yes, Country FE: Yes, Year FE: Yes. - Columns (4)-(5) report alternate samples with Mean (Pretreat.) = 0.431. - Interpretation: - Returning to cash is counterproductive to formalization efforts that leverage transaction records. - Overall interpretation of cross-country evidence: - Findings lend support to Proposition 2 (mobile money usage declines after tax) and Proposition 4 (substitution toward cash). - The substitution effect—away from mobile money—appears to dominate the average-cost effect in observed data. - Cross-country IMF survey shows systematic declines in mobile money accounts; Findex data show distribution across households and substitution toward cash. ### 4 Micro-Evidence from Transaction Data ### 4.1 Background: Country cases used for transaction-level analysis - Cameroon - Tax introduction: January 1, 2022 (Finance Law No. 2021/026, Section 228). - Tax description: - Levied at 0.2% of the amount transferred and on the amount withdrawn via any traceable electronic or mobile money platform (including mobile operators). - Applies to both sending transfers and withdrawals; combined rate for a transfer followed by withdrawal ("cashout") is effectively 0.4%. - Excludes bank transfers, deposits into e-wallets/virtual wallets, payments to registered merchants, and transfers for paying taxes, customs, or duties. - Tax calculated on actual transaction amount, not including operator fees or VAT. - Context: - Rapid growth in mobile money: registered and active accounts grew dramatically between 2015 and 2020; transaction values reached close to 50% of GDP. - Tax enacted quickly with visible public reaction (e.g., hashtag #EndMobileMoneyTax). - Fee impacts: - Pre-existing operator fee schedule for transfers remained: 1% of transaction value for amounts between 50 and 6,500 XAF, progressive schedule from 50 to 500 XAF for transactions between 6,501 and 1,000,000 XAF. - Application of 0.2% tax substantially increased total transaction costs (operator fees plus tax)—by approximately 20% at the lower end and up to 400% at the upper end of the fee schedule. - Examples: - Sixth bracket (50,001–80,000 XAF): tax = 100–160 XAF, effectively doubling original fee of 100 XAF. - Highest bracket (500,001–1,000,000 XAF): tax = 1,000–2,000 XAF, representing two to four times the original fee of 500 XAF. - For cash withdrawals (cash-outs): - Fees generally higher than for transfers but slightly reduced in two brackets after tax introduction—from 180 to 175 XAF in the second bracket (6,500–10,000 XAF) and from 1,350 to 1,300 XAF in the sixth (50,001–80,000 XAF). - Overall cost of withdrawals increased by 4% to 28%, depending on transaction size. - Note: - On January 1, 2025, Cameroon introduced a 4 XAF flat fee per mobile money transaction, in addition to the existing proportional tax; this charge now also applies to banks, credit institutions, and microfinance entities, which were previously exempt.16 - Central African Republic (CAR) - Tax introduction: April 2024. - Tax description: - Applies to peer-to-peer transfers and cash withdrawals. - Tax rate is 1% of the value of each transaction. - Deposits into accounts and payments to registered merchants are exempt. - Context: - Adoption occurred amid rapid expansion of mobile money as an alternative to a weak formal banking sector. - Public debate prior to introduction was limited; civil society groups voiced concerns post-implementation about regressivity. - Fee impacts: - Transfer fees of operator remained unchanged and untaxed; total transaction costs increased mechanically by the tax amount. - Tax ranged from 10 to 20 XAF in the lowest bracket (1,000–2,000 XAF) and from 2,500 to 5,000 XAF in the highest (250,005–500,000 XAF). - Withdrawal fees ranged from 30 XAF to 5,000 XAF before the tax and were not revised but compounded by the tax amount; total withdrawal costs rose from 40 XAF to 10,000 XAF, corresponding to increases of 33% to 100%, depending on transaction value. - Mali - No mobile money tax in place during the sample period; serves as a control country. - Francophone country with relatively high mobile money penetration; usage patterns (peer-to-peer transfers, withdrawals, merchant payments) closely resemble those in Cameroon and CAR.17 *Italic: Source: wpiea2025255-source-pdf - 2.6 Efficiency* ### 4.2 Transaction Data ### 4.2 Transaction Data ### Data coverage and design - Anonymized transaction-level data from a major mobile money operator covering Cameroon, CAR, and Mali. - Random sample of approximately 20% of all users in each country. - Observation windows: - Cameroon: July 2021–June 2022 (balanced six-month window before and after tax introduction; July 2021 excluded from estimation to avoid Eid al-Adha effects; estimation covers 11 months: August 2021–June 2022). - CAR: October 2023–September 2024 (estimation covers 12 months). - Mali: both periods (used as contemporaneous untreated benchmark). - For each user: age, gender, registration location, and full universe of transactions for the sample period. - Transactions classified by type (see Appendix B.3 for detailed categories and descriptive statistics). ### Transaction-type composition - Transfers between users (P2P transfers): about 27% of total transaction volume in both Cameroon and CAR. - Cash-out withdrawals: above 30% of volume. - Cash-in deposits: above 30% of volume. - Merchant payments ("Merchpay"): 1–2.6% of volume. - Granularity allows analysis of volume and value across categories and assessment of substitution among them. ### Per-user averages and comparison to expenditures - Average monthly transaction value per user: - Cameroon: 141,310.23 XAF (approximately 252 USD18) - CAR: 181,989.17 XAF - Mali (07/2021–06/2022): 201,477.93 XOF - Mali (10/2023–09/2024): 383,830.36 XOF - Average annual per capita expenditures used for comparison: - Cameroon: 1,118,128.58 XAF (around 93,177.38 XAF per month) - Mali: around 723,900.18 XOF (corresponding to 60,325.02 XOF per month) - Note: ratio of transaction value to monthly expenditures is high partly because incoming and outgoing P2P transactions are counted twice. ### Transaction counts and dynamics - Number of transactions per user (monthly averages): - Cameroon: 13.93 (slightly higher than CAR) - CAR: 12.11 - Mali: increases from 15.5 (first sample period) to 22.58 (second sample period), reflecting strong growth of mobile money use in Mali. - Dataset balanced-panel sizes (Table 4 & 5 notes): - Cameroon: N = 574,845 users. - CAR: N = 54,441 users. - Mali (control for Cameroon, 07/2021 - 06/2022): N = 459,384 users. - Mali (control for CAR, 10/2023 - 09/2024): N = 633,477 users. ### Selected summary statistics (as reported in Tables 4 and 5; selected entries) - Individual characteristics (means): - Cameroon: Age 36.73; Female 0.39; Access to Banks 0.32; Urban 0.73; Yearly Expenditures (scaled) 1,118,128.58. - CAR: Age 35.96; Female 0.29; Access to Banks 0.24; Urban 0.90. - Mali (control for Cameroon): Age 38.39; Female 0.11; Access to Banks 0.31; Urban 0.49; Yearly Expenditures (scaled) 723,900.18. - Mali (control for CAR): Age 38.23; Female 0.12; Access to Banks 0.30; Urban 0.47; Yearly Expenditures (scaled) 718,035.49. - Mobile money use (selected means from tables): - Cameroon: - Transaction Volume (untaxed) 17,689.67 - Transaction Volume (taxed) 123,620.56 - Number of Transactions (untaxed) 7.74 - Number of Transactions (taxed) 6.19 - CAR: - Transaction Volume (untaxed) 67,995.49 - Transaction Volume (taxed) 113,993.68 - Number of Transactions (untaxed) 6.95 - Number of Transactions (taxed) 5.16 - Mali (control for Cameroon): - Transaction Volume (untaxed) 85,252.31 - Transaction Volume (taxed) 116,255.62 - Number of Transactions (untaxed) 9.25 - Number of Transactions (taxed) 5.80 - Mali (control for CAR): - Transaction Volume (untaxed) 171,986.47 - Transaction Volume (taxed) 211,843.89 - Number of Transactions (untaxed) 13.14 - Number of Transactions (taxed) 9.44 *Italicized notes in source: Mali introduced a tax of 1% on mobile money withdrawals, effective on March 5, 2025. Using an exchange rate of 1 USD = 559.08 XAF as of October 1, 2025.* --- ### 4.3 Household Survey Data - Complementary household survey data used for Cameroon and Mali to: - Compare changes in mobile money use across regions with different bank access. - Construct effective tax rates using expenditure data (see subsection 4.5 and section 5). - Surveys used: - Cameroon: Cinquième Enquête Camerounaise sur les Ménages (ECAM 5). - Mali: Enquête Harmonisée sur les Conditions de Vie des Ménages 2018–2019 (WAEMU2019). - Cannot conduct heterogeneity analysis for CAR because usage is concentrated in Bangui. - Derived regional variables: - Consumption proxied by total monthly household expenditures, scaled using FAO adult-equivalence scale. - Bank availability defined as share of household heads in a region reporting an account with a financial institution (commercial bank, microfinance, or caisse rurale), using survey sampling weights. - Matching: users in transaction data matched to regional characteristics based on registration location, using a fuzzy-matching algorithm for town names; manual assignment for unmatched towns. --- ### 4.4 Within-Country Identification - Tax exemptions: merchant payments and some other minor transaction types are exempt from the mobile money tax in Cameroon and CAR. - Primary within-country specification: difference-in-differences comparing taxed transactions (P2P transfers and cash-out withdrawals) with untaxed ones (merchant payments and other untaxed types), within each country. - Aggregation: monthly panel tracing taxed vs. untaxed transactions up to six months post-tax. - Event-study specification (Equation 15): - y_{i,j,t} = sum_{k=-6}^{5} β_k τ_{k j,t} + μ_i + λ_t + ε_{i,j,t} - y_{i,j,t}: outcome (aggregate transaction values or number of transactions) of user i for transaction type j in month t. - τ_{k j,t}: event-time dummies equal to one if month t is k months relative to tax introduction and transaction type j is taxed (P2P or cash-out). - μ_i and λ_t denote individual and time fixed effects. - Detrending: following Goodman-Bacon (2021), each transaction type detrended by its pre-treatment slope to improve comparability. ### Within-country findings - Pre-tax: taxed and untaxed transactions follow parallel trends. - Post-tax: - Cameroon: drop of around 50,000 XAF in monthly value of taxed transactions relative to untaxed transactions, equivalent to about 40% of the pre-treatment mean of 125,347.5 XAF. - CAR: decline similar in magnitude, corresponding to 36% of pre-treatment level of 113,993.2 XAF. - Number of transactions: - Tax reduces number of transactions by roughly 10% in Cameroon and about 8.5% in CAR (Appendix Figure B.4). --- ### 4.5 Identification Based on Matching with Mali #### Rationale and approach - Second identification strategy: compare taxed transaction dynamics in Cameroon and CAR with those in Mali (untreated), mitigating SUTVA concerns from within-country comparisons. - Matching idea: compare taxed transaction types of users in treated country to users with similar observed characteristics and pre-tax mobile money use patterns in Mali. - Suitability of Mali as control: - All three countries are francophone, low- or lower-middle-income, agriculture-dominated, and use CFA francs (XOF in Mali; XAF in Cameroon and CAR), pegged at a fixed rate to the euro. - Mobile money prevalence: Mali and Cameroon rank high; CAR medium (GSMA (2025) mobile money prevalence index). - Matching procedure (three steps): 1. For Cameroon benchmark, restrict sample to users in majority-Muslim regions to mitigate seasonal Christmas effects; retain 25 of 69 regions (~54% of original sample). This step not applicable to CAR (no Muslim-majority regions). 2. Propensity-score matching: estimate logit where dependent variable = 1 for users in Cameroon or CAR and 0 for users in Mali. Covariates: gender, age, urban/rural, and pre-tax average monthly transaction value of taxed transactions (for Cameroon average includes January 2022 for better balancing). Include transformations and interactions; generate propensity scores. 3. Nearest-neighbor matching of treated users to Mali users with similar propensity scores. Appendix Figure B.5 shows substantial overlap of scores. #### Event-study specification after matching (Equation 16) - y_{i,c,t} = sum_{k=-6}^{5} β_k τ_{k c,t} + μ_i + λ_t + ε_{i,c,t} - y_{i,c,t}: outcome of user i in country c in month t. - τ_{k c,t}: event-time dummies equal to one if user i resides in a treated country (c ∈ {Cameroon, CAR}) and month t is k months relative to tax introduction; for Mali dummies set to zero. - μ_i and λ_t denote user and month fixed effects. - Estimated separately for Cameroon–Mali and CAR–Mali matched samples; focus on transactions taxed in treated countries. --- ### Benchmark Results from Matching with Mali - Main findings (Figure 6): - Both Cameroon and CAR exhibit sharp and persistent declines in mobile money usage after the tax. - Cameroon: total monthly transaction value per user falls by more than 50,000 XAF. - CAR: drop slightly larger in absolute terms. - Relative declines correspond to: - Cameroon: 40% relative to pre-tax mean 125,347.5 XAF. - CAR: 47% relative to pre-tax mean 113,993.2 XAF. - Transaction-type breakdown: separate estimates for P2P transfers and cash-outs show both decline due to the tax (Appendix Figure B.7). - Untaxed transactions (merchant payments and cash-in deposits) either exhibit different pre-trends or do not show significant responses (Appendix Figure B.8). - Number of monthly taxed transactions (Figure 7): - Pre-treatment means: 6.2 transactions (Cameroon) and 5.77 transactions (CAR) per user per month. - Percentage declines in counts mirror volume declines: - Cameroon: 33% decline (roughly a reduction of 2 transactions from 6.2). - CAR: 51% decline (roughly a reduction of 2.9 transactions from 5.77). --- ### Heterogeneity - Focused on Cameroon (CAR concentrated in Bangui, 87% of users). - Two heterogeneity splits: 1. Regions with low vs. high bank penetration (bottom quartile vs. top quartile of regional bank penetration). 2. Urban vs. rural regions. - Methodology note: first split substantially reduces sample size and limits matching; use approach similar to within-country analysis with detrending of subsample pre-treatment slopes. - Findings (Figure 8): - Declines in mobile money use are markedly larger in regions with greater banking access, consistent with substitution toward alternative financial services. - In low-bank-penetration regions, effects are weaker and often statistically insignificant. - Reductions stronger in urban than in rural areas. - Evidence supports that the tax burden falls disproportionately on users in financially less developed and rural areas who face few substitutes. - Overall, substitution effects outweigh the average cost effect. --- ### Price Elasticity - Computation approach: for each observed mobile money transaction, compute pre-tax and post-tax consumer price. - Cameroon: - Tax resulted in average increase of 19.45% in the cost to the consumer for making a taxable transaction.29 - Although proportional tax rate is 0.2%, this is large relative to operator fees. - Dividing the 40% decline in average monthly transaction value per user for taxed transactions by the 19.45% tax-inclusive price increase yields an estimated short-term price elasticity of about −2.1 in Cameroon. - Interpretation: a 10% increase in the tax-inclusive consumer price (operator fee plus mobile money tax) associated with about a 21% decline in average monthly transaction value per user for taxed transactions. - CAR: - No operator fee for transfers (only withdrawals entail a fee), so pre-tax consumer price for a transfer is zero; the 1% levy raises the price from 0 to positive, making percentage change undefined. - Therefore, no comparable elasticity reported for transfers in CAR. *Footnote from source (29): In Cameroon, only P2P transactions and withdrawals from a mobile money account are charged a fee. For P2P transactions, either the sender or the receiver pays the fee. The authors assume the sender pays it and compute, for all transactions prior to January 1, 2022, the difference between the fee only and the fee including the tax. Computing the cost increase using all transactions through the end of the sample does not change the results.* --- *Italic: Source — 4.2 Transaction Data (wpiea2025255-source-pdf).* ### 5.1 Calculation of Excess Burden (EB) ### 5.1 Calculation of Excess Burden (EB) ### Excess Burden Calculation - Framework: EB = τ × ∆B (tax rate times fall in taxable base). - Observed total value of taxable transactions in sample over six months: 444,653 million XAF. - Predicted reduction due to tax: 40% → drop of 177,861 million XAF in sample. - Sample represents 20% of customer base → scale fivefold to 889,306 million XAF. - Adjust for operator’s approximately 50% market share → economy-wide decline in taxable mobile money transactions = 1,778,612 million XAF. - Statutory rate per taxed operation: 0.2% (0.002). - EB over six months: EB = 0.002 × 1,778,612 = 3,557 million XAF. - Annualized EB: 7,114 million XAF. ### Comparison to Revenue - EB / Revenue ≈ 0.35 (i.e., about 35% of mobile money tax revenue) based on official statistics over the same horizon. ### Calculation Assumptions (explicit) - The 20% sample is representative. - The operator’s market share is 50% across periods. - The 40% decline applies uniformly across the taxable base. - Transfers and withdrawals by their recipients are treated as distinct transactions in the data. ### Welfare Threshold and Social Marginal Value of Public Funds - Necessary condition for desirability: λ ≥ 1 + EB/R. - With EB/R ≈ 0.35 → implies λ ≳ 1.35. - Accounting for additional losses: - Shift to cash (informality) equal to s percent of revenue and - Distributional penalty of d percent, - Threshold becomes λ ≥ 1 + 0.35 + s + d. - Example numeric scenario: if s = 15% and d = 15%, required λ ≥ 1.65. - Existing estimates for African countries place the marginal cost of public funds around 1.1–1.2 (median reported by Auriol and Warlters (2012) is 1.17). - Conclusion: required threshold (> 1.35) exceeds these typical values under usual normalization. ### Mobile Money Tax Incidence and Distributional Effects (summary of linked Section 5.2) - Effective mobile money tax rate defined as share of mobile money taxes in total household expenditures: τ_eff_i,t,MM = τ_t,MM × MM_i,t / E_i,t. - Pre-treatment and post-response comparisons indicate regressivity, with poorer and less financially developed regions facing higher effective rates. Key reported group statistics (from Table 6) - Bank Penetration / Location groups: Low / High / Rural / Urban - Pre-treatment Volume (CFA): 119,449 / 134,515 / 94,848 / 138,795 - Monthly Expenditures (CFA): 50,983 / 122,242 / 60,541 / 105,001 - Response in Volume (%): –21.4 / –26.8 / –20.6 / –29.5 - Tax Rate: 0.2% / 0.2% / 0.2% / 0.2% - Eff. Tax Rate (no response): 0.469% / 0.220% / 0.313% / 0.264% - Eff. Tax Rate (with response): 0.368% / 0.161% / 0.250% / 0.189% - Share of Tax Revenue (with resp.): 25.8% / 26.6% / 22% / 78% - Note: Shares from low and high bank penetration do not sum to 100%; remaining 47.6.5% of tax revenue stems from users with intermediate bank penetration (between the 25th and 75th percentile). ### Key Findings and Policy-Relevant Conclusions - Empirical elasticities: mobile money taxes sharply reduce usage—with an elasticity around −2. - Excess burden reaches about one third of revenue collected (EB / Revenue ≈ 0.35). - Tax is regressive: poorer users and those in financially less developed regions bear higher effective rates despite contributing less in absolute terms. - Tax induces substitution toward cash, undermining financial inclusion and formality by eroding transaction information useful for enforcement of other taxes and measurement of economic activity. - Overall assessment: mobile money taxation (on transaction value) is an inefficient and regressive policy instrument. - Evidence covers the first six months after commencement of the tax; longer-run responses remain for future work. - Theoretical model yields additional testable predictions (e.g., adjustments in bank transactions) pending suitable data. *Source: wpiea2025255-source-pdf - 5.1 Calculation of Excess Burden (EB)* ### References ### wpiea2025255-source-pdf - References ### References - Comprehensive bibliography of works cited, including research on mobile money, financial transaction taxes, taxation in developing countries, difference-in-differences methodology, and empirical studies of mobile money in Africa. Representative citations include: - Agrawal, D. R., and Hebous, S. (2025). Tax Policy in the Era of Digital Money and the Metaverse, Mimeo. - Ahmad, A. H., Green, C., and Jiang, F. (2020). Mobile Money, Financial Inclusion and Development: A Review with Reference to African Experience. Journal of Economic Surveys, 34(4), 753–792. - Ahnert, T., Hoffmann, P., and Monnet, C. (2025). Payments and Privacy in the Digital Economy. Journal of Financial Economics, 169, 104050. - Brunnermeier, M., Limodio, N., and Spadavecchia, L. (2023). Mobile Money, Interoperability, and Financial Inclusion, NBER Working Paper No. 31696. - GSMA. (2025). The State of the Industry Report on Mobile Money 2025. GSM Association. - Methodological and empirical sources cited include standard econometric references on event-study and difference-in-differences designs (e.g., Borusyak et al. (2024); Callaway and Sant’Anna (2021); Goodman-Bacon (2021)) and literature on transaction taxes and market effects (e.g., Matheson (2012); Colliard and Hoffmann (2017); Cai et al. (2020)). ### Appendix A: Proofs (A.1–A.5) - A.1 Proof of Proposition 1 and Corollary - Differentiation under the integral sign is applied (reference to Theorem 2.27 in Folland (1999)). - For W(t): d/dt E[W(t)] = P(M + t< c). (Equation (A.1)) - For Z(t): d/dt E[Z(t)] = P(M + t< min(B, c)). (Equation (A.2)) - d/dt ∆(t) = P(B≤ M + t< c). (Equations (A.3)–(A.7)) - Conclusion: since φ* = ∆ and ∆ increases with the tax, the proportion with bank accounts increases, dG(φ*(t))/dt> 0. - A.2 Proof of Proposition 2 - Expressions for d/dt P_W(t) and d/dt P_Z(t) are given in integral form. - Comparison highlights that for c in the upper tail, density along the diagonal exceeds density along (c, b), implying d/dt P_Z(t) > d/dt P_W(t). - Relative derivatives satisfy (dP_Z/dt)/P_Z > (dP_W/dt)/P_W. - A.3 Proof of Proposition 4 - For banked: P(Z = c) = ∫_{m=c−t}^∞ ∫_{b=c}^∞ f_{M,B}(m,b) db dm. - d/dt P(Z = c) = ∫_{b=c}^∞ f_{M,B}(c−t, b) db > 0. - For unbanked: P(W = c) = ∫_{m=c−t}^∞ f_M(m) dm; hence d/dt P(W = c) = f_M(c−t) > 0. - A.4 Proofs for the Average-Cost Effect - Defines survival function S_U(x) = P(U ≥ x) and rewrites transfer-incidence probabilities (Equation (A.8)). - Differentiation yields d/dt P_U_Z(t) = −E[f_U(M + t)·1{M+t