[Paper Review] Determinants of Mobile Money Adoption in Pakistan

Muhammad Raza Khan, Joshua Blumenstock|arXiv (Cornell University)|Nov 13, 2017

ICT Impact and Policies8 references3 citations

TL;DR

This study analyzes mobile money adoption in Pakistan using call detail records (CDR) and mobile money transaction data from a major telecom operator. By applying deterministic finite automata (DFA)-based feature engineering and gradient boosting, the authors identify that mobility-related features are the strongest predictors of adoption across all demographics, while usage and network features show gender- and geography-specific importance, revealing distinct behavioral patterns in men, women, urban, rural, rich, and poor users.

ABSTRACT

In this work, we analyze the problem of adoption of mobile money in Pakistan by using the call detail records of a major telecom company as our input. Our results highlight the fact that different sections of the society have different patterns of adoption of digital financial services but user mobility related features are the most important one when it comes to adopting and using mobile money services.

Motivation & Objective

• To understand how mobile money adoption varies across gender, urban/rural residence, and socioeconomic status in Pakistan.
• To identify the most predictive behavioral features—usage, mobility, and network—using anonymized call detail records (CDR) and mobile money transaction data.
• To evaluate the effectiveness of DFA-based feature engineering in capturing adoption patterns across diverse demographic groups.
• To inform targeted policy and marketing strategies by revealing demographic-specific determinants of mobile money adoption.

Proposed method

• The study uses anonymized call detail records (CDR) and mobile money transaction records (MMTR) from a major Pakistani telecom provider.
• Users are classified into three categories: voice-only users, registered mobile money users, and peer-to-peer (P2P) mobile money users.
• A deterministic finite automaton (DFA) framework is applied to generate a comprehensive set of features across three categories: usage, mobility, and network structural properties.
• Features are derived through recursive filtering, grouping, and aggregation rules (e.g., variance in daily call duration, number of unique contact locations per day).
• Stratified, balanced sampling is used across six demographic groups (male/female, urban/rural, rich/poor), and gradient boosting is applied to classify adoption and P2P usage.
• Feature importance is evaluated via cross-validated model performance, with results visualized per demographic subgroup.

Experimental results

Research questions

• RQ1 How do patterns of mobile money adoption differ between men and women in Pakistan?
• RQ2 What role do mobility patterns play in predicting mobile money adoption across urban and rural populations?
• RQ3 How do usage, network, and mobility features compare in predicting adoption among rich and poor districts?
• RQ4 To what extent do network effects and social endorsement influence adoption in rural versus urban areas?
• RQ5 Which behavioral features are most predictive of peer-to-peer mobile money usage across demographic groups?

Key findings

• Mobility-related features are the most important predictors of mobile money adoption for all demographic groups, especially for males and urban/rural populations.
• For females, usage-related features (e.g., variance in active days with incoming SMS) are the top predictors, indicating higher engagement in communication patterns.
• In rural districts, network-related features such as the size of the incoming SMS contact network are more important, suggesting social influence and awareness play a key role.
• For peer-to-peer usage, mobility features remain dominant in rich districts, while usage features are most predictive in poor districts.
• The model achieves high cross-validated accuracy in predicting both adoption and P2P usage, with top features varying significantly by gender and geography.
• The study reveals that men’s adoption is driven by mobility (e.g., number of unique contact locations on weekdays), while women’s adoption is linked to communication activity and network engagement.