## wpiea2024163-print-pdf - 5.5 times the value of a food supplier’s collateral.

## Source details

**Canonical URL:** [wpiea2024163-print-pdf - 5.5 times the value of a food supplier’s collateral.](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024163-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024163-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024163-print-pdf.pdf.json)

---

### Counterfactual simulations and policy interventions
- Three main policy interventions considered:
  - (i) improving the quality of connectivity,
  - (ii) reducing the cost of internet,
  - (iii) relaxing credit constraints.
- Main simulation findings:
  - Increasing the quality of the network (proxied by internet area coverage and safety) by 35 percentage point can yield up to 100% adoption rate and 180% increase in average productivity.
  - Reducing the cost of internet to zero, making it unrealistically free for everyone, would increase the adoption rate only by 4 percentage points.
  - Allowing suppliers to borrow up to 4 times the value of the average wealth in the data increases the adoption rate by only 1 percentage point and lifts average productivity by 3.2%.
  - No amount of credit is able to shift adoption rate beyond 58%.
- Interpretation:
  - The quality of internet is the most critical incentive for adoption given its impact on profits.
  - Gains from reducing the cost of access to the internet or access to credit will not be fully reaped unless the quality of internet connectivity is satisfactory.
  - Results align with literature that adoption of digital technologies, if complementary to input factors, can boost productivity and output, but complementary investment in skills and capital is required for large gains.

### Key quantitative results and limits on adoption
- Quality upgrade scenario:
  - 35 percentage point improvement in network quality → up to 100% adoption rate and 180% increase in average productivity.
- Cost reduction scenario:
  - Internet cost reduced to zero → adoption rate increases by 4 percentage points.
- Credit relaxation scenario:
  - Borrowing allowed up to 4 times average wealth → adoption rate increases by 1 percentage point; average productivity lifts by 3.2%.
  - Adoption rate cannot exceed 58% through credit alone.
- Implication:
  - Internet quality has a larger effect on adoption and productivity than cost reductions or credit expansion in these simulations.

### Data and sample (survey design and scope)
- Survey period: May to July 2023.
- Locations:
  - Bohicon: small town of 171,781 people located about 124 kilometers north of Cotonou.
  - Ouando: town of 264,320 people located 9 kilometers away from Porto-Novo; market sprawls 3.5 hectares.
  - Population sizes taken from the 2013 Fourth General Population and Habitat Census.
- Sample and focus:
  - 451 grains and legumes suppliers surveyed across Bohicon and Ouando markets.
  - Targeted value chains: beans, maize, peanuts, cowpeas, melon seeds.
  - Survey collected business characteristics and outcomes; access and usage of digital technologies; personal characteristics; risk aversion; access to credit; bad business experiences.
  - KasuwaGo app referenced as digital trading platform.

### Descriptive statistics and digital usage patterns
- Sample composition and demographics:
  - 81% are women.
  - 87.4% are married.
  - Average number of children: 5.1.
  - Average age: 45.8 years.
  - Average number of monthly transactions per trader: 11.6.
  - 90% of respondents have more than 10 years of experience in the business.
  - 52% do not have a formal training or education.
  - 49.9% employ 6 to 9 employees.
- Digital usage and spending:
  - 95.4% use mobile phones for calls.
  - 45% use phones to navigate on social media.
  - Nearly 30% use phones to trade products and/or make mobile money payments.
  - 23.5% use digital technologies for other usages (sending text messages, browsing internet, watching YouTube).
  - About 30% spend between 2,000 and 5,000 CFA Francs monthly on broadband services.
  - 54.1% spend more than 5,000 CFA per month on broadband services (represents about one-fifth of the minimum monthly wage in Benin).
- Adopter definition and partition:
  - “Non-Adopters”: use mobile phones only for basic calls and SMS; typically do not use smartphones.
  - “Adopters”: use mobile phones for digital applications and social media; use android or smartphones.
  - Partition in sample: 51% Adopters and 49% Non-Adopters.
- Stylized facts from stratification:
  - Adopters are on average more educated, perform more business transactions, operate larger businesses, are relatively younger, and have less business experience than non-adopters.
  - Two-thirds of non-adopters (66%) do not have formal education; none attended college.
  - Adopters spend more on mobile broadband than non-adopters; 53.2% of non-adopters complain about affordability of mobile broadband.
  - Respondents geographically close to many users (within 500 meters) tend to adopt more — indicating peer or network effects.
  - Motivations and concerns:
    - 89.4% motivated by better access to market information.
    - 7.98% motivated by ease of communication.
    - For not using digital technologies: 53.22% report “Too Expensive”; 24.83% say “Not Needed”; 18.63% cite “Privacy Concerns”.

### Theoretical framework and modeling approach
- Purpose:
  - Structural model to formalize micro-foundations of technology adoption among small-scale food suppliers.
  - Model maps digital skills (driven by education, age and proximity to digital technology users), network effects and wealth to digital technology adoption.
  - Structural model used for counterfactual policy tracing via parameters fitted to survey data.
- Modeling choices and focus:
  - Focus on intermediaries (suppliers) in agricultural value chains.
  - Account for market frictions: wealth-related credit constraints, quality and reliability of broadband.
  - Anchored in observed adoption patterns, usage, and market frictions to analyze policy implications for productivity and output.

### Model set up (summary of 3.1)
- Agent heterogeneity:
  - Digital-specific productivity η ∈ (0, ̄η).
  - Initial wealth/assets z ∈ (0,+∞) usable as collateral.
- Capital constraint: maximum capital ̄k = λz, λ ∈ (0,∞).
- Market parameters: gross interest rate r, wage rate w.
- Production exhibits decreasing returns with α ∈ (0,1), β ∈ (0,1), 0 < β + α < 1.
- Non-adopter output: y_o = A k^α l^β with closed-form profit expressions π_o depending on whether z ≥ z^*.
- Adopter output: y_d = A k^α (η l)^β, adoption costs c, output loss probability p, profit π_d with closed forms for z ≥ z^* and z < z^*.
- Adoption decision: adopt iff Π_d(η,z) ≥ Π_o(η,z); threshold η^* given by η^*_u and η^*_c(z).
- Selection characterization:
  - η^*_c(z) > η^*_u so constrained suppliers need higher digital skills to adopt.
- Taxonomy and inefficiencies:
  - Four categories: unconstrained adopters, unconstrained non-adopters, constrained adopters, constrained non-adopters.
  - Two inefficiency sources: credit constraints; cost and quality of digital technologies.

### Impact of Technology Adoption (reduced-form and heterogeneous effects)
- Estimation framework:
  - Outcome_i = β0 + β1 Adopt_i + X'_i γ + u_i; main outcome = number of major monthly business transactions (≥ 5,000 CFA).
  - Instrument: NearUsers (number of nearby users).
  - Controls: Wealth, Age, Education, Experience, Female, Married, Native, Member.
  - Count-data specifications: OLS/2SLS and Poisson/IV Poisson.
- Reduced-form estimates (effects of Adopt on transactions):
  - OLS: 2.2236 (0.8117) — ** (p<0.01)
  - Poisson: 2.2507 (0.3652) — *** (p<0.001)
  - 2SLS IV: 4.3454 (2.1323) — ** (p<0.01)
  - IV Poisson: 4.5681 (0.9649) — *** (p<0.001)
  - Interpretation: Adopters perform on average 2 to 5 transactions more than non-adopters; given non-adopters average 10 transactions monthly, adoption could lift monthly trades by up to 50%.
- Heterogeneous effects by gender:
  - Adoption rates: Women 51.1%; Men 48.8%.
  - Adopt (men baseline) estimated average marginal effects:
    - OLS: 5.2736 (2.3161) — ** (p<0.05)
    - Poisson: 5.7015 (0.7914) — *** (p<0.001)
    - 2SLS: 8.3605 (2.9726) — ** (p<0.05)
    - IV Poisson: 8.9158 (4.3049) — ** (p<0.05)
  - Adopt × Female negative and significant across specifications:
    - OLS: -3.5928 (1.3806) — ** (p<0.05)
    - Poisson: -4.1074 (0.8391) — ** (p<0.05)
    - 2SLS: -5.4415 (2.2457) — ** (p<0.05)
    - IV Poisson: -5.8541 (2.6092) — ** (p<0.05)
  - Implication: Men adopters gain more transactions from adoption than women adopters; women adopt more but benefit less per adopter.

### Structural estimation: parameters, identification, and fit
- Digital skills specification:
  - ln η_i = κ0 + κ1 Educ_i + κ2 Age_i + κ3 NearUsers_i + ε_i, ε_i ∼ N(0,1).
- Structural parameter vector ψ = (κ0, κ1, κ2, κ3, α, β, λ).
- Institutional calibrations:
  - Gross interest rate r = 1.053 (official average monthly market rate).
  - Wage rate w = 30,000 CFA Francs (official minimum monthly wage).
  - Discount factor 1−p computed via p = 1 − Cover × Safe:
    - Broadband penetration (Cover) in Benin: 42% in 2022.
    - Proportion not reporting privacy concerns (Safe): 95.6%.
    - Implied discount factor 1−p = 40.3%.
- Structural estimates (Table 5):
  - κ0 = 3.4419 (Std. Error 0.4551)
  - κ1 = 0.1031 (Std. Error 0.0137)
  - κ2 = -0.1551 (Std. Error 0.0206)
  - κ3 = 0.0034 (Std. Error 0.0012)
  - α = 0.1426 (Std. Error 0.0368)
  - β = 0.8377 (Std. Error 0.3565)
  - λ = 5.5013 (Std. Error 0.0177)
  - Log-Likelihood = -102.42; Observations = 451.
- Interpretations:
  - κ1 = 0.103 → one unit of education associated with 10.3 percent increase in digital skills.
  - κ2 = -0.155 → one additional year of age associated with 15.5 percent decrease in digital skills.
  - κ3 = 0.0034 → one additional nearby user within 500 meters associated with 0.34 percent increase in digital skills.
  - λ = 5.5013 → total investment bounded up to 5.5 times initial wealth; adjust for data: about 70% of initial investment financed by personal sources and 30% by loans → estimated multiplier discounted by about 30% to reflect formal credit market = about 1.65.
  - α and β imply 1 percent increase in capital → 0.14 percent increase in output; 1 percent increase in hired labor → 0.84 percent increase in output; sum close to 1.
- Model fit:
  - Predicted adoption rate: 49.8%; Observed: 50.6%.
  - Predicted average monthly transactions: 11.2 overall; 13.6 for adopters; 10.01 for non-adopters.
  - Observed: 11.6 overall; 13.05 for adopters; 10.09 for non-adopters.
  - Structural model treatment effect = 4.15; reduced-form 2SLS = 4.35.
  - Deviations between data and model moments do not exceed 5% (Table C2, Appendix C).
- Robustness:
  - Alternative skill distributions (including wealth) produced worse fits; wealth not a proxy for digital skills (correlation insignificant).

### Policy evaluations (summary of 5.2)
- Counterfactual setup: short-term simulations with wage and interest rates and distribution of wealth fixed.
- Outcomes: fraction of adopters, aggregate/average (log) output, differential responses between constrained and unconstrained adopters.
- Improving quality of connectivity:
  - Policy: ̃p = p − τp.
  - 35-percentage point decrement in p → adoption rate almost 100% and 180% increase in average productivity (see Counterfactual simulations).
  - Investment cost benchmark: extending 4G to entire population could amount to 2.5% of Benin’s projected average annual GDP between 2021 and 2030.
  - Average (log) output response two-speed: for τp < 0.35 rapid extensive-margin gains; for τp ≥ 0.35 adoption ≈ 100% and slower intensive-margin gains.
  - Unconstrained adopters’ log output response ≈ 10 times that of constrained adopters; constrained adopters have higher elasticity to connectivity improvement.
- Reducing costs of connectivity:
  - Policy: ̃ci = (1 − τc) × ci.
  - Adoption rate virtually unchanged for cost reductions < 45 percent; increases and plateaus around 53 percent when cost reduction reaches 90 percent.
  - Even cost = 0 increases adoption only by 4 percentage points (see Counterfactual simulations); 25% of suppliers reported not interested in adopting.
  - Log output follows similar step-wise pattern; constrained adopters’ elasticity higher than unconstrained adopters’.
- Improving access to credit:
  - Policy: ̃ki = max{k̄i, zi + τz k̄z}.
  - Adoption probability rises gradually from 50% up to 56% for 3 < τz ≤ 13; plateaus when τz > 13.
  - Maximum adoption in simulations: 56% irrespective of further top-up funding.
  - Average (log) output grows monotonically, driven by constrained suppliers borrowing more; unconstrained suppliers’ output remains flat.
- Comparative effectiveness and mixes:
  - Single-policy ranking: upgrading internet quality delivers largest adoption rate and output gains.
  - Two-pronged packages:
    - Quality improvement + large cost reduction → sizeable reinforcing effects.
    - Quality improvement + easier credit → higher adoption and output.
    - Cost reduction + easier credit → better than either alone, but weaker than packages including quality improvements.
  - Three-pronged package (better quality, lower costs, easier credit) expected to deliver largest benefits.
- Key contextual findings:
  - Suppliers predominately women: 80%.
  - Average age: 45.8.
  - Adoption rate about 51%.
  - One third use phones to trade or settle transactions.
  - 52% had no formal education.
  - 25% reported not being interested in using digital technologies.
- Limitations:
  - Counterfactuals are short-term; do not capture general equilibrium or aggregate welfare gains.
  - Digital nomads not included.
  - Demand-side behaviour absent; no cost-benefit analysis in simulations.

*Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024163-print-pdf.pdf*

### 5.5 times the value of a food supplier’s collateral.

### wpiea2024163-print-pdf - 5.5 times the value of a food supplier’s collateral.

### Counterfactual simulations and policy interventions
- Considered three main policy interventions:
  - (i) improving the quality of connectivity,
  - (ii) reducing the cost of internet,
  - (iii) relaxing credit constraints.
- Main simulation findings:
  - Increasing the quality of the network (proxied by internet area coverage and safety) by 35 percentage point can yield up to 100% adoption rate and 180% increase in average productivity.
  - Reducing the cost of internet to zero, making it unrealistically free for everyone, would increase the adoption rate only by 4 percentage points.
  - Allowing suppliers to borrow up to 4 times the value of the average wealth in the data increases the adoption rate by only 1 percentage point and lifts average productivity by 3.2%.
  - No amount of credit is able to shift adoption rate beyond 58%.
- Interpretation:
  - The quality of internet is the most critical incentive for adoption given its impact on profits.
  - Gains from reducing the cost of access to the internet or access to credit will not be fully reaped unless the quality of internet connectivity is satisfactory.
  - Results align with literature that adoption of digital technologies, if complementary to input factors, can boost productivity and output, but complementary investment in skills and capital is required for large gains.

### Key quantitative results and limits on adoption
- Quality upgrade scenario:
  - 35 percentage point improvement in network quality → up to 100% adoption rate and 180% increase in average productivity.
- Cost reduction scenario:
  - Internet cost reduced to zero → adoption rate increases by 4 percentage points.
- Credit relaxation scenario:
  - Borrowing allowed up to 4 times average wealth → adoption rate increases by 1 percentage point; average productivity lifts by 3.2%.
  - Adoption rate cannot exceed 58% through credit alone.
- Implication:
  - Internet quality has a larger effect on adoption and productivity than cost reductions or credit expansion in these simulations.

### Data and sample (survey design and scope)
- Survey period: May to July 2023.
- Locations: two semi-urban markets in Benin (Bohicon market and Ouando market).
  - Bohicon: small town of 171,781 people located about 124 kilometers north of Cotonou.
  - Ouando: town of 264,320 people located 9 kilometers away from Porto-Novo; market sprawls 3.5 hectares.
  - Population sizes are taken from the 2013 Fourth General Population and Habitat Census (latest census data available at the time of the survey).
- Sample and focus:
  - 451 grains and legumes suppliers surveyed across Bohicon and Ouando markets.
  - Targeted the grain and legume value chains (beans, maize, peanuts, cowpeas, melon seeds).
  - Survey collected information on business characteristics and outcomes; access and usage of digital technologies; personal characteristics; risk aversion; access to credit; and bad business experiences.
  - The KasuwaGo app is a digital trading platform referenced in the survey.

### Descriptive statistics and digital usage patterns
- Sample composition and demographics:
  - 81% are women.
  - 87.4% are married.
  - Average number of children: 5.1.
  - Average age: 45.8 years.
  - Average number of monthly transactions per trader: 11.6.
  - 90% of respondents have more than 10 years of experience in the business.
  - 52% do not have a formal training or education.
  - 49.9% employ 6 to 9 employees.
- Digital usage and spending:
  - 95.4% use mobile phones for calls.
  - 45% use phones to navigate on social media.
  - Nearly 30% use phones to trade products and/or make mobile money payments.
  - 23.5% use digital technologies for other usages (sending text messages, browsing internet, watching YouTube).
  - About 30% of respondents spend between 2,000 and 5,000 CFA Francs monthly on broadband services.
  - 54.1% spend more than 5,000 CFA per month on broadband services (represents about one-fifth of the minimum monthly wage in Benin).
- Adopter definition and partition:
  - “Non-Adopters”: use mobile phones only for basic calls and SMS; typically do not use smartphones.
  - “Adopters”: use mobile phones for digital applications and social media; use android or smartphones.
  - Partition in sample: 51% Adopters and 49% Non-Adopters.
- Stylized facts from stratification:
  - Adopters are on average more educated, perform more business transactions, operate larger businesses, are relatively younger, and have less business experience than non-adopters.
  - Two-thirds of non-adopters (66%) do not have formal education; none attended college.
  - Adopters spend more on mobile broadband than non-adopters; 53.2% of non-adopters complain about affordability of mobile broadband.
  - Respondents geographically close to many users (within 500 meters) tend to adopt more — indicating peer or network effects.
  - Motivations and concerns:
    - 89.4% of users motivated by better access to market information (customers, prices, etc).
    - 7.98% motivated by ease of communication.
    - For not using digital technologies: 53.22% report “Too Expensive”; 24.83% say “Not Needed”; 18.63% cite “Privacy Concerns”.

### Theoretical framework and modeling approach
- Purpose:
  - Develop a simple structural model to formalize micro-foundations of technology adoption among small-scale food suppliers.
  - Model maps digital skills (driven by education, age and proximity to digital technology users), network effects and wealth to digital technology adoption.
  - Structural model allows tracing out outcomes of counterfactual policies using structural parameters fitted with the survey data.
- Modeling choices and focus:
  - Focus on intermediaries (suppliers) in agricultural value chains rather than farmers/producers.
  - Account for market frictions observed in the data: wealth-related credit constraints, quality and reliability of broadband.
  - Model is anchored in observed adoption patterns, usage, and market frictions to analyze policy implications for productivity and output.

*Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024163-print-pdf.pdf*

### 3.1  Model set up

### 3.1  Model set up

### Model framework
- Continuum of heterogeneous agents (small-scale grain and legumes suppliers) differ by:
  - digital-specific productivity, η ∈ (0, ̄η), where ̄η is a finite upper bound;
  - initial wealth/assets, z ∈ (0,+∞), usable as collateral to secure credit.
- Agents choose between adopting digital technologies (accessing mobile broadband and using digital mobile phone applications) or remaining non-adopters.
- Digital-specific productivity η interpreted as ability to exploit digital mobile applications to publicize, build networks, settle deals, and/or learn new skills that increase trading effectiveness.
- Capital constraint: maximum capital used, ̄k = λz, for proportionality factor λ ∈ (0,∞).
- Market prices: gross interest rate r, wage rate w.
- Production technology exhibits decreasing returns to scale with parameters α ∈ (0,1), β ∈ (0,1), and assume 0 < β + α < 1.

### Non-Adopters
- Output (gross per-period margin/value added) for non-adopter:
  - y_o = A k^α l^β
  - A is productivity in traditional technology (baseline for adopters as well).
- Profit maximization with capital constraint yields:
  - π_o(z) = max_{0≤k≤̄k, l≥0} { A k^α l^β − r k − w l } =
    - π^u_o := (1−γ) A^{1/(1−γ)} [ α/r ]^{α/(1−γ)} [ β/w ]^{β/(1−γ)}, if z ≥ z^*,
    - π^c_o := (1−β) A^{1/(1−β)} [ β/w ]^{β/(1−β)} ̄k^{α/(1−β)} − r ̄k, if z < z^*,
  - where γ = α + β, and
    - z^* = 1/λ A^{1/(1−γ)} [ α/r ]^{(1−β)/(1−γ)} [ β/w ]^{β/(1−γ)}
      is the minimum collateral needed to operate a financially unconstrained business.

### Adopters
- Adopter production function:
  - y_d = A k^α (η l)^β
  - η is relative difference in labor productivity between adopters and non-adopters.
- Adoption involves per-period total costs of c in capital units (entry cost and variable costs) and risk of output loss with probability p ∈ (0,1).
  - Costs include device purchase/installation, mobile broadband subscription fees, per-period service charges.
  - Probability p captures output loss due to degraded quality/reliability of connectivity and data privacy/cybersecurity threats.
- Maximum profit for adopter:
  - π_d(η,z) = max_{0≤k≤̄k, l≥0} { (1−p) A k^α (η l)^β − w l − r k − r c } =
    - π^u_d := (1−γ) [ (1−p) η^β A ]^{1/(1−γ)} [ α/r ]^{α/(1−γ)} [ β/w ]^{β/(1−γ)} − r c, if z ≥ z^*,
    - π^c_d := (1−β) [ (1−p) η^β A ]^{1/(1−β)} [ β/w ]^{β/(1−β)} ̄k^{α/(1−β)} − r ̄k − r c, if z < z^*.
- Labor demand for adopters, l_d(η,z), is increasing in η. Adopters with sufficiently high η tend to hire more labor than non-adopters for given z.
- Improvement in quality/safety (lower p) narrows labor demand wedge across suppliers and increases proportion adopting.
- Among adopters, financially unconstrained suppliers (z ≥ z^*) exhibit higher labor productivity than constrained suppliers (z < z^*).

### Adoption decision
- Suppliers know η and z, market characteristics r and w, productivity A and its distribution, cost of access c, and probability of output loss p.
- Adoption rule: choose to adopt iff expected profit from adopting ≥ expected profit from not adopting:
  - Π_d(η,z) ≥ Π_o(η,z) in expectations with respect to distribution of A.
- Critical digital skills threshold η^*:
  - Π_d(η,z) increasing in η; Π_o(z) inelastic in η.
  - η^* = { 
    - η^*_u := (1−p)^{-1/β} [ 1 + r c / (1−γ) y^u_o ]^{(1−γ)/β}, if z ≥ z^*;
    - η^*_c(z) := (1−p)^{-1/β} [ 1 + r c / (1−β) y^c_o(z) ]^{(1−β)/β}, if z < z^*.
    }
  - y^u_o and y^c_o(z) are unconstrained and constrained expected outputs had the entrepreneur remained a non-adopter.
- All agents with η ≥ η^* adopt; agents with η < η^* remain non-adopters.
- For financially constrained entrepreneurs, η^*_c(z) is decreasing in z.
- For financially unconstrained entrepreneurs, η^*_u does not depend on personal characteristics but on c and p.

### Selection characterization (Proposition 1)
- For a business owner with characteristics (η,z):
  - If η < η^*_u, then the business owner is a non-adopter.
  - If η ≥ η^*_u, then the business owner is an adopter only if z ≥ z^*. However, if z < z^*, then the business owner is an adopter only if η > η^*_c(z).
- Relationship between thresholds:
  - η^*_c(z) > η^*_u, implying financially constrained suppliers (z < z^*) need higher digital skills than unconstrained counterparts for adoption.

### Taxonomy and inefficiencies
- Four categories emerge: unconstrained adopters, unconstrained non-adopters, constrained adopters, constrained non-adopters.
- In the frictionless benchmark (λ = ∞, c = 0, p = 0), any supplier with η > 1 would adopt; selection driven solely by η.
- With real-world frictions, two inefficiency sources:
  - Inefficiencies due to credit constraints.
  - Inefficiencies due to cost and quality of digital technologies.
- Assumption: labor flows freely between business types; employees have no intrinsic employer-type preference and employers do not select workers by initial digital skills. Authors note this is consistent with the grains and legumes context in the data.

*Source: IMF Working Paper — 3.1 Model set up.*

### 4.2  Impact of Technology Adoption

### 4.2  Impact of Technology Adoption

### Model and Estimation Approach
- Estimated model: Outcome_i = β0 + β1 Adopt_i + X'_i γ + u_i (Equation (6)).
- Main outcome (productivity): number of major monthly business transactions (a “major” transaction involves 5,000 CFA or more in sales).
- Adoption potentially endogenous → instrument: NearUsers (number of nearby users).
  - Instrument rationale: NearUsers significantly influences adoption probability but is unlikely to affect monthly transactions except through adoption.
  - Potential weakness noted: nearby adopters could convey market information enabling additional transactions without adoption; exogeneity tests and structural estimation help address endogeneity concerns.
- Controls included in X_i: Wealth, Age, Education, Experience, Female, Married, Native, Member, and the constant.
- Because dependent variable is a count, both OLS/2SLS and Poisson/IV Poisson specifications are reported.

### Reduced-form: Effect of Adoption on Transactions (Table 3)
- Estimated coefficients on Adopt:
  - OLS: 2.2236 (standard error 0.8117) — significance: ** (p<0.01)
  - Poisson: 2.2507 (standard error 0.3652) — significance: *** (p<0.001)
  - 2SLS IV: 4.3454 (standard error 2.1323) — significance: ** (p<0.01)
  - IV Poisson: 4.5681 (standard error 0.9649) — significance: *** (p<0.001)
- Interpretation:
  - Adopters perform on average 2 to 5 transactions more than non-adopters, everything else equal.
  - Given non-adopters perform on average 10 transactions per month, adoption could lift monthly frequency of trades by up to 50%.
- Additional regressors:
  - Including NearUsers among regressors: coefficient not significant while other statistics remain almost unchanged → NearUsers has no effect on the outcome once Adopt is accounted for.
- Model fit metrics and sample:
  - Adj/Pseudo R2: reported as 0.1120 (OLS), 0.1014 (Poisson), 0.0899 (2SLS), – (IV Poisson).
  - # Observations: 445 (OLS/Poisson), 448 (2SLS/IV Poisson).
  - Notes: All estimates are average marginal effects. Standard errors in parentheses. Significance codes: *p<0.05, ** p<0.01, *** p<0.001.

### Heterogeneous Effects by Gender (Table 4)
- Adoption rates in data:
  - Women: 51.1%
  - Men: 48.8%
- Estimated coefficients (average marginal effects; standard errors in parentheses):
  - Adopt:
    - OLS: 5.2736 (2.3161) — ** (p<0.05)
    - Poisson: 5.7015 (0.7914) — *** (p<0.001)
    - 2SLS: 8.3605 (2.9726) — ** (p<0.05)
    - IV Poisson: 8.9158 (4.3049) — ** (p<0.05)
  - Female:
    - OLS: -0.6762 (1.8653)
    - Poisson: -0.0051 (0.6723)
    - 2SLS: 0.0906 (1.6327)
    - IV Poisson: 0.2123 (0.4074)
  - Adopt × Female:
    - OLS: -3.5928 (1.3806) — ** (p<0.05)
    - Poisson: -4.1074 (0.8391) — ** (p<0.05)
    - 2SLS: -5.4415 (2.2457) — ** (p<0.05)
    - IV Poisson: -5.8541 (2.6092) — ** (p<0.05)
- Interpretation:
  - Coefficient on Adopt represents treatment effect for men:
    - Men adopters perform on average 5.3 to 8.9 more transactions than men non-adopters.
  - Female coefficient insignificant → no significant difference between women non-adopters and men non-adopters.
  - Effect for women adopters = Adopt + Adopt×Female:
    - OLS: about 1.7
    - Poisson: about 1.6
    - 2SLS: about 2.9
    - IV Poisson: about 3.0
  - Difference between effects on men and women given by Adopt×Female: ranges from 3.6 to 5.8 across specifications.
  - Implication: Women adopt more, but men tend to benefit more from adoption. Suggested plausible explanation: women may spend disproportionately more online time on non-business activities (household purchases, children educational contents, etc.). More information required to understand heterogeneity.

### Structural Estimation: Likelihood, Parameters, and Identification
- Structural approach:
  - Joint likelihood matches model-predicted adoption and output with observed adoption and transactions (Equations (8), (9), and log-likelihood in Equation (10)).
  - Digital skills distribution specified as log-normal conditional on Educ_i, Age_i, NearUsers_i:
    - ln η_i = κ0 + κ1 Educ_i + κ2 Age_i + κ3 NearUsers_i + ε_i where ε_i ∼ N(0,1) (Equation (7)).
  - Vector of structural parameters ψ = (κ0, κ1, κ2, κ3, α, β, λ).
  - Production elasticities: α (capital share), β (labor share).
  - λ: proportion of wealth that can be invested (credit constraint parameter).
- Institutional parameter calibration:
  - Gross interest rate r fixed at 1.053 (official average monthly market rate).
  - Wage rate w set at 30,000 CFA Francs (official minimum monthly wage in Benin).
  - Discount factor index for reliability of broadband internet: 1−p.
    - p = 1 − Cover × Safe.
    - Broadband penetration (Cover) in Benin: 42% in 2022.
    - Proportion of respondents that did not report privacy concerns (Safe): 95.6%.
    - Implied discount factor 1−p = 40.3%.
- Identification:
  - Education and Age identify log digital skill parameters by shifting its mean.
  - Exclusion restriction driven by NearUsers shifting productivity only if adopter.
  - Variations in labor and capital shares come from shifts in output, wealth and individual costs.
  - Heterogeneity in initial wealth pins down credit constraint parameter.
  - Maximum likelihood identification inferred from non-singularity of information matrix.

### Structural Estimation Results (Table 5)
- Log Digital Skills:
  - Constant κ0 = 3.4419 (Std. Error 0.4551)
  - Education κ1 = 0.1031 (Std. Error 0.0137)
  - Age κ2 = -0.1551 (Std. Error 0.0206)
  - NearUsers κ3 = 0.0034 (Std. Error 0.0012)
- Production and Constraint:
  - Capital share α = 0.1426 (Std. Error 0.0368)
  - Labor share β = 0.8377 (Std. Error 0.3565)
  - Wealth λ = 5.5013 (Std. Error 0.0177)
- Log-Likelihood = -102.42
- Observations = 451
- Notes: Standard errors calculated using bootstrap samples.
- Interpretations emphasized in text:
  - κ1 = 0.103 → any additional unit of education associated with 10.3 percent increase in digital skills.
  - κ2 = -0.155 → any additional year of age associated with 15.5 percent decrease in digital skills.
  - κ3 = 0.0034 → any additional user in 500-meter radius associated with 0.34 percent increase in digital skills (network learning).
  - λ = 5.5013 → total investment bounded up to 5.5 times the value of initial wealth; caveat: in data about 70% of initial business investment came from personal savings, gifts, family transfers and remittances; loans represented remaining 30% → estimated multiplier should be discounted by about 30% to reflect formal credit market, that is, about 1.65.
  - α = 0.1426 and β = 0.8377 → 1 percent increase in capital associated with 0.14 percent increase in output; 1 percent increase in hired labor associated with 0.84 percent increase in output. Sum close to 1 → suggests a very competitive grains and legumes supply market.

### Model Fit and Validation
- Model predictions versus observed data:
  - Predicted adoption rate: 49.8%; Observed adoption rate: 50.6%.
  - Predicted average monthly transactions: 11.2 overall; 13.6 for adopters; 10.01 for non-adopters.
    - Observed: 11.6 overall; 13.05 for adopters; 10.09 for non-adopters.
  - Structural model predicts a treatment effect of 4.15; reduced-form 2SLS treatment effect = 4.35 (Table 3).
- Goodness of fit:
  - None of the deviations between data and structural model (percent error rate) for the observed statistics exceeds 5% (referenced Table C2, Appendix C).
- Robustness checks:
  - Alternative specifications for distribution of skills considered (including wealth as covariate) gave worse model fits; wealth not a proxy for digital skills (correlation insignificant).

*Source: wpiea2024163-print-pdf - 4.2  Impact of Technology Adoption*

### 5.2  Policy Evaluations

### 5.2  Policy Evaluations

### Overview
- Counterfactual simulations are conducted using the estimated structural model to evaluate the short-term impacts (wage and interest rates and distribution of wealth fixed) of three policy options on digital technology adoption and associated productivity in the grains and legumes markets in Benin:
  - (1) improving quality of connectivity,
  - (2) lowering broadband internet service costs,
  - (3) increasing access to credit.
- Outcomes reported include fraction of adopters (adoption rate), aggregate/average (log) output, and differential responses between constrained and unconstrained adopters.

### Improving the quality of connectivity
- Policy formulation:
  - New probability of digital output loss ̃p = p − τp, 0 < τp < p, where τp is the increment in quality and ̃p is the new probability of output loss.
- Mechanisms for improving broadband quality cited: broadening internet coverage, upgrading safety of broadband network, reducing latency, deploying fiber optic technologies, fostering competition among providers.
- Key quantitative impacts (from Figure 7 descriptions):
  - Adoption rate increases markedly, reaching almost 100 percent for a 35-percentage point decrement in the probability p.
  - Investment cost benchmark: extending 4G internet coverage to the entire population could amount to 2.5% of Benin’s projected average annual GDP between 2021 and 2030 (Bennouna et al., 2024, as cited).
  - Average (log) output response exhibits a two-speed increase:
    - For τp < 0.35, rapid increase in log output mainly via extensive margin (new entrants).
    - For τp ≥ 0.35, adoption ≈ 100 percent and slower increase in log output mainly via intensive margin (improved efficiency among adopters).
  - Constrained vs. unconstrained adopters:
    - Range of log output response for unconstrained adopters is roughly 10 times that of constrained adopters.
    - Constrained adopters display higher elasticity (steeper slope) of output to connectivity improvement than unconstrained adopters.

### Reducing the costs of connectivity
- Policy formulation:
  - New cost ̃ci = (1 − τc) × ci, 0 < τc < 1, where τc is fraction cost reduction.
- Actionable policies cited: subsidizing fiber optic backbone, exploiting internet exchange points (IXPs), other cost-reduction measures.
- Key quantitative impacts (from Figure 8 descriptions):
  - Adoption rate and log output show a step-wise increase as connectivity costs are slashed.
  - Adoption rate remains virtually unchanged for cost reductions < 45 percent, then increases and plateaus around 53 percent when cost reduction reaches 90 percent.
  - Log output follows a similar step-wise pattern on a larger scale.
  - Interpretation: a nonlinear threshold around a 45 percent cost reduction beyond which services become affordable to some non-adopters; substitution with other consumption expenditures may play a role.
  - Elasticities: constrained adopters’ log output elasticity to connectivity cost reduction is higher (steeper slope) than that of unconstrained adopters.
  - Empirical constraint: even reducing internet cost to zero does not induce full adoption because 25% of suppliers reported not being interested in adopting digital technologies.

### Improving access to credit
- Policy formulation:
  - Budget limit adjustment ̃ki = max{k̄i, zi + τz k̄z}, 0 < τz < ∞, where τz is scaling factor and k̄z is average household wealth.
- Expected impact channel: primarily affects financially constrained suppliers.
- Key quantitative impacts (from Figure 9 descriptions):
  - Adoption probability increases gradually, driven by constrained adopters, from 50% up to 56% when 3 < τz ≤ 13; adoption rate plateaus when τz > 13.
  - Maximum adoption rate observed in simulations: 56% irrespective of further top-up funding.
  - Average (log) output grows monotonically, mainly reflecting gains of constrained suppliers able to borrow more.
  - Constrained suppliers’ (log) output rises steadily with improved credit access; unconstrained suppliers’ output remains flat.
  - Among unconstrained suppliers, skilled ones are expected to adopt as credit becomes easier; unskilled ones are not.

### Comparative effectiveness and policy mixes
- Single-policy ranking (based on simulations):
  - Upgrading internet quality could deliver the largest digital adoption rate and the largest output gains among the three individual policies.
- Policy bundling effects:
  - Two-pronged packages:
    - Improving internet quality + substantially reducing service costs → sizeable mutually reinforcing effects on adoption and output.
    - Improving internet quality + facilitating access to credit (especially for financially constrained suppliers) → higher adoption rates and output.
    - Reducing connectivity costs + easier access to credit → better outcomes than either policy alone, though reinforcing effects weaker than those including internet quality improvements.
  - Three-pronged package (better internet quality, lower service costs, better access to credit) is expected to deliver the largest beneficial outcomes.

### Key quantitative findings and context (from the chapter and concluding text)
- Sample and baseline statistics:
  - Suppliers predominately women: 80%.
  - Average age: 45.8.
  - Half employ 6 to 9 people.
  - Digital technology adoption rate among suppliers: about 51%.
  - One third of suppliers use their phones to trade products and/or settle business transactions.
  - Education: 52% of suppliers had no formal education.
  - Non-interest in adoption: 25% of suppliers reported not being interested in using digital technologies.
- Reduced-form and structural estimation highlights:
  - Probability of adoption increases in wealth, the number of nearby digital users, and education; decreases in age.
  - Adopters perform on average 2 to 5 more digital transactions per month than non-adopters, a 50% increase; effect higher for men than women.
  - Structural model reproduces key data moments and supports counterfactual policy experiments.
- Limitations noted affecting interpretation:
  - Counterfactuals are short-term (wage, interest rates, wealth distribution fixed); do not capture general equilibrium or aggregate welfare gains.
  - Digital nomads not included.
  - Demand-side behaviour (consumers/retailers) absent from model; no cost-benefit analysis provided in simulations.

*Source: 5.2 Policy Evaluations, wpiea2024163-print-pdf*

### References

### References

### Major themes in the referenced literature
- Technology adoption and diffusion (theory and empirical evidence), including directed technical change and skill-biased technical change.
- Digital technologies and broadband: determinants of access, adoption, and demand-side programs; quality, costs, and un-adopters.
- Mobile phones, ICTs, and economic development in Africa, with applications to agricultural markets, productivity, innovation, and entrepreneurship.
- Agricultural technology, productivity, risk, credit, insurance, and modernization of agriculture in developing countries.
- Firm-level heterogeneity, misallocation, and productivity differences across firms and countries.
- Labor markets, informality, occupational choice, and workforce age effects on technology adoption.
- Financial frictions, constrained entrepreneurship, and the role of networks in market development.
- Measurement and empirical methods: structural econometrics, identification, segmented labor markets, and experimental evidence.

### Selected cited works and topical contributions (as listed)
- Acemoglu, D. (1998). Why do new technologies complement skills? directed technical change and wage inequality. The Quarterly Journal of Economics, 113(4):1055–1089.
- Acemoglu, D. (2002). Directed technical change. The Review of Economic Studies, 69(4):781–809.
- Africa Telecom Review (2024). Ugandan government aims to decrease internet bandwidth costs.
- African Union (2020). The digital transformation strategy for Africa (2020-30).
- Aker, J. C. (2010). Information from markets near and far: Mobile phones and agricultural markets in Niger. American Economic Journal: Applied Economics, 2(3):46–59.
- Aker, J. C. and Ksoll, C. (2016). Can mobile phones improve agricultural outcomes? evidence from a randomized experiment in Niger. Food Policy, 60:44–51.
- Aker, J. C. and Mbiti, I. M. (2010). Mobile phones and economic development in Africa. Journal of Economic Perspectives, 24(3):207–232.
- Akerman, A., Gaarder, I., and Mogstad, M. (2015). The skill complementarity of broadband internet. The Quarterly Journal of Economics, 130(4):1781–1824.
- Baily, M. N. and Solow, R. M. (2001). International productivity comparisons built from the firm level. Journal of Economic Perspectives, 15(3):151–172.
- Bartel, A., Ichniowski, C., and Shaw, K. (2007). How does information technology affect productivity? plant-level comparisons of product innovation, process improvement, and worker skills. The Quarterly Journal of Economics, 122(4):1721–1758.
- Beaman, L., Karlan, D., Thuysbaert, B., and Udry, C. (2023). Selection into credit markets: Evidence from agriculture in Mali. Econometrica, 91(5):1595–1627.
- Bennouna, H., Bloch, C., Tawk, N., and Wang, M. (2024). Benin 2024 fourth EFF ECF review, first RSF review, and Article IV consultation. IMF Selected Issues Paper.
- Birke, D. (2009). The economics of networks: A survey of the empirical literature. Journal of Economic Surveys, 23(4):762–793.
- Bloom, N., Sadun, R., and Reenen, J. V. (2012). Americans do it better: US multinationals and the productivity miracle. American Economic Review, 102(1):167–201.
- Brynjolfsson, E. and Hitt, L. (1995). Information technology as a factor of production: The role of differences among firms. Economics of Innovation and New technology, 3(3-4):183–200.
- Buera, F. J. and Shin, Y. (2013). Financial frictions and the persistence of history: A quantitative exploration. Journal of Political Economy, 121(2):221–272.
- Chen, C. (2020). Technology adoption, capital deepening, and international productivity differences. Journal of Development Economics, 143:102388.
- Comin, D. and Hobijn, B. (2004). Cross-country technology adoption: making the theories face the facts. Journal of Monetary Economics, 51(1):39–83.
- Conley, T. G. and Udry, C. R. (2010). Learning about a new technology: Pineapple in Ghana. American Economic Review, 100(1):35–69.
- Costa, V., Piedrahita, N., Mane, E., Davis, B., Slavchevska, V., and Gurbuzer, Y. (2023). Women’s employment in agrifood systems—background paper for the status of women in agrifood systems.
- DePaula, G. (2023). Bundled contracts and technological diffusion: Evidence from the Brazilian soybean boom. Journal of Development Economics, 165:103163.
- D’Andrea, A. and Limodio, N. (2023). High-speed internet, financial technology, and banking. Management Science.
- Elliott, J. T., Houngbonon, G. V., Ivaldi, M., and Scott, P. (2024). Market structure, investment, and technical efficiencies in mobile telecommunications. Journal of Political Economy.
- Emerick, K., De Janvry, A., Sadoulet, E., and Dar, M. H. (2016). Technological innovations, downside risk, and the modernization of agriculture. American Economic Review, 106(6):1537–1561.
- Fafchamps, M. (1993). Sequential labor decisions under uncertainty: An estimable household model of West-African farmers. Econometrica, 61(5):1173–1197.
- Fafchamps, M. (2001). The role of business networks in market development in Sub-Saharan Africa. Community and Market in Economic Development, 1:186–215.
- Fageda, X., Rubio-Campillo, R., and Termes-Rifé, M. (2014). Determinants of broadband access: Is platform competition always the key variable to success? Information Economics and Policy, 26:58–67.
- FAO, IFAD, UNICEF, WFP, and WHO. (2023). The state of food security and nutrituion in the world. urbanization, agrifood systems transformation and healthy diets across the rural–urban continuum. Technical report, Food and Agriculture Organization of the United Nations.
- Foster, A. D. and Rosenzweig, M. R. (2010). Microeconomics of technology adoption. Annual Review of Economics, 2(1):395–424.
- Galor, O. and Tsiddon, D. (1997). Technological progress, mobility, and economic growth. The American Economic Review, pages 363–382.
- Ginè, X. and Yang, D. (2009). Insurance, credit, and technology adoption: Field experimental evidence from Malawi. Journal of Development Economics, 89(1):1–11.
- Gollin, D. (2010). Agricultural productivity and economic growth. Handbook of Agricultural Economics, 4:3825–3866.
- Gollin, D. and Rogerson, R. (2016). Agriculture, roads, and economic development in Uganda. African Successes, Volume IV: Sustainable Growth, 4:69.
- Gollin, D. and Udry, C. (2021). Heterogeneity, measurement error, and misallocation: Evidence from African agriculture. Journal of Political Economy, 129(1):1–80.
- Hauge, J. A. and Prieger, J. E. (2010). Demand-side programs to stimulate adoption of broadband: What works? Review of Network Economics, 9(3).
- Heckman, J. J. and Honoré, B. E. (1990). The empirical content of the Roy model. Econometrica, 58(5):1121–1149.
- Hjort, J. and Poulsen, J. (2019). The arrival of fast internet and employment in Africa. American Economic Review, 109(3):1032–1079.
- Houngbonon, G. V., Mensah, J. T., and Traore, N. (2022). The impact of internet access on innovation and entrepreneurship in Africa. World Bank Policy Research Working Paper.
- INStaD (2016). Enquête Régionale Intégrée sur l’Emploi et le Secteur Informel (ERIESI) Bénin, 2018. Institut National de la Statistique et de la Démographie, Benin.
- International Telecommunication Union (2022). Measuring digital development: Facts and figures. Technical report, International Telecommunication Union.
- Jackson, M. O., Rogers, B. W., and Zenou, Y. (2017). The economic consequences of social-network structure. Journal of Economic Literature, 55(1):49–95.
- Jelfs, A. and Richardson, J. T. (2013). The use of digital technologies across the adult life span in distance education. British Journal of Educational Technology, 44(2):338–351.
- Karaivanov, A. (2012). Financial constraints and occupational choice in thai villages. Journal of Development Economics, 97(2):201–220.
- Karaivanov, A. and Townsend, R. M. (2014). Dynamic financial constraints: Distinguishing mechanism design from exogenously incomplete regimes. Econometrica, 82(3):887–959.
- Karaivanov, A. and Yindok, T. (2022). Involuntary entrepreneurship–evidence from thai urban data. World Development, 149:105706.
- Katz, M. L. and Shapiro, C. (1994). Systems competition and network effects. Journal of Economic Perspectives, 8(2):93–115.
- Koundouri, P., Nauges, C., and Tzouvelekas, V. (2006). Technology adoption under production uncertainty: theory and application to irrigation technology. American Journal of Agricultural Economics, 88(3):657–670.
- Kumar, U., Amaglobeli, D., and Moszoro, M. (2023). Determinants and social dividends of digital adoption. IMF Working Paper No. 2023/65.
- Lee, C. and Coughlin, J. F. (2015). Perspective: Older adults’ adoption of technology: an integrated approach to identifying determinants and barriers. Journal of Product Innovation Management, 32(5):747–759.
- Lucas, R. E. (1978). On the size distribution of business firms. The Bell Journal of Economics, pages 508–523.
- Magnac, T. (1991). Segmented or competitive labor markets. Econometrica: journal of the Econometric Society, pages 165–187.
- Magruder, J. R. (2018). An assessment of experimental evidence on agricultural technology adoption in developing countries. Annual Review of Resource Economics, 10:299–316.
- McAnally, A. P. (1963). The measurement of retail productivity. The Journal of Industrial Economics, pages 87–95.
- Meghir, C., Narita, R., and Robin, J.-M. (2015). Wages and informality in developing countries. American Economic Review, 105(4):1509–1546.
- Meyer, J. (2011). Workforce age and technology adoption in small and medium-sized service firms. Small Business Economics, 37:305–324.
- Miassi, Y. E. S., Dossa, F. K., and Şinasi, A. (2018). Women and market garden production in benin republic. Horticulture International Journal, 3(4):191–193.
- Ndubuisi, G., Otioma, C., Owusu, S., and Tetteh, G. K. (2022). ICTs quality and technical efficiency: An empirical analysis. Telecommunications Policy, 46(10):102439.
- Nguimkeu, P. (2014). A structural econometric analysis of the informal sector heterogeneity. Journal of Development Economics, 107:175–191.
- Nguimkeu, P. (2016). Some effects of business environment on retail firms. Applied Economics, 48(18):1647–1654.
- Nguimkeu, P. and Okou, C. (2021). Leveraging digital technologies to boost productivity in the informal sector in Sub-Saharan Africa. Review of Policy Research, 38(6):707–731.
- Okou, C., Spray, J., and Unsal, D. F. (2022). Staple food prices in Sub-Saharan Africa: An empirical assessment. IMF Working Paper No. 2022/135.
- Olabisi, M., Maredia, M., Liu, J., Ajibade, T., and Ajeigbe, H. (2024). In-group competition for incentives. Journal of Development Economics, page 103320.
- Oliva, P., Jack, B. K., Bell, S., Mettetal, E., and Severen, C. (2020). Technology adoption under uncertainty: Take-up and subsequent investment in Zambia. Review of Economics and Statistics, 102(3):617–632.
- Paulson, A. L., Townsend, R. M., and Karaivanov, A. (2006). Distinguishing limited liability from moral hazard in a model of entrepreneurship. Journal of Political Economy, 114(1):100–144.
- Pratap, S. and Quintin, E. (2006). Are labor markets segmented in developing countries? a semiparametric approach. European Economic Review, 50(7):1817–1841.
- Rothenberg, T. (1971). Identification in parametric models. Econometrica, 69(3):577–591.
- Suri, T. and Udry, C. (2022). Agricultural technology in Africa. Journal of Economic Perspectives, 36(1):33–56.
- Urama, N. E. and Ogbu, O. (2018). Evaluating consumer perception and willingness to pay for broadband in Nigeria. Telecommunications Policy, 42(6):421–432.
- Usai, A., Fiano, F., Petruzzelli, A. M., Paoloni, P., Briamonte, M. F., and Orlando, B. (2021). Unveiling the impact of the adoption of digital technologies on firms’ innovation performance. Journal of Business Research, 133:327–336.
- Violante, G. L. (2008). Skill-biased technical change. The New Palgrave Dictionary of Economics, 2:1–6.
- Whitacre, B. and Rhinesmith, C. (2016). Broadband un-adopters. Telecommunications Policy, 40(1):1–13.
- Whitacre, B., Strover, S., and Gallardo, R. (2015). How much does broadband infrastructure matter? decomposing the metro–non-metro adoption gap with the help of the national broadband map. Government Information Quarterly, 32(3):261–269.
- World Bank (2016). World development indicators 2016. The World Bank.

*Can Mobile Technologies Enhance Productivity? A Structural Model and Evidence from Benin Food Suppliers Working Paper No. WP/2024/163*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024163-print-pdf.pdf_