個別報告論文
バングラデシュ北部におけるモバイルバンキングの加入状況と決定要因
2018 年 54 巻 3 号 p. 133-140
詳細
2018 年 54 巻 3 号 p. 133-140
近年，携帯電話を利用した銀行取引，いわゆるモバイルバンキング（Mバンク）が，より多数の人々の銀行サービス利用を企図して低所得途上国で導入されている．しかし，Mバンク利用に関する実証研究はバングラデシュなどの途上国では限定的である．本研究の目的は，バングラデシュ農村世帯がMバンクに加入する要因を明らかにすることである．計量モデルでは，Mバンク加入の有無（または時期）を被説明変数，Mバンクがバングラデシュに導入される直前の世帯主及び世帯の属性を説明変数として設定した．多項ロジット・モデルの推定から，教育水準，性別等の世帯主属性やMバンクが導入される前の携帯電話の保有がMバンク加入に有意な影響を与えていたことこが見出された．
Financial inclusion contributes to empowering the poor in rural areas (Inoue and Hamori, 2012; Odhiambo, 2009). Despite this fact, banking services have not reached the vast segment of total population worldwide, especially rural underprivileged poor people (Bangladesh Bank, 2017; Beck et al., 2007; World Bank, 2014). One reason for this is low profitability; that is, the maintenance and opportunity cost of banking services is higher relative to the number of users in rural remote areas than in urban areas (Johnson and Nino-Zarazua, 2011). Johnson and Nino-Zarazua (2011) also found that financial exclusion is associated with sociocultural characteristics of the region rather than with the mere urban-rural status in developing countries such as Kenya.
Meanwhile, advanced information technology has extended to rural areas at a rapid pace (International Telecommunication Union, 2017). In line with this issue, recently low income developing countries have embarked mobile-banking (M-bank) to provide banking services to larger populations (Aker and Mbiti, 2010; Yu, 2012). M-bank, a new way of banking services allows users to deposit, withdraw and transfer funds as well as purchase goods and services using mobile phone without using internet (Bangladesh Bank, 2017; Munyegera and Matsumoto, 2016).
Previous studies address that M-bank can reduce the costs and time for financial transaction, increase savings and transfers, and enhance income generating activities (Aker et al., 2016; Azad, 2016; Blumenstock et al., 2016; Jack and Suri, 2014). However, most of these studies on M-bank have focused on problems and prospects, and services and strategies on customer behavior (Ahad et al., 2012; Azad, 2016; Laukkanen and Cruz, 2012). Some studies have investigated socioeconomic determinants of the use of M-bank such as use of mobile phone, age, gender and education level (Ahad et al., 2012; Azad, 2016; Laukkanen and Cruz, 2012; Munyegera and Matsumoto, 2016). However, empirical evidence underlying the use of M-bank is still limited, especially that for the poor households.
Despite the recent economic development, Bangladesh is still classified as one of the least developed countries. About 24% of the nation’s population live below the poverty line (Bangladesh Bureau of Statistics (BBS), 2016). Financial exclusion among poor populations has long prevailed. Only 31% of the total population (about 160 million) have bank accounts. However, thanks to the development of information technology, presently 84% (135 million) are mobile phone subscribers (Bangladesh Bank, 2017; World Bank, 2014).
In this context, the central bank of Bangladesh decided to strategically promote M-bank in January 2011 to include the rural people under financial services. At present, 18 commercial banks have operated M-bank services such as purchasing airtime, transferring money, and paying bills through about 569,000 agents and about 3,900 ATMs. The number of M-bank subscribers has steadily increased and recently reached 57 million (Bangladesh Bank, 2017).
Although there are some studies on the factors of the diffusion of M-bank in Bangladesh (Ahad et al., 2012; Azad, 2016), no study has empirically analyzed the contributing factors to the use of M-bank in Bangladesh using rigid econometric tools. Moreover, no information on financial inclusion is available in the northern part of Bangladesh (BBS, 2016). The northern region has the highest incidence of poverty due to frequent droughts and floods causing lack of income generating activities (BBS, 2016; Shonchoy, 2011).
Thus, the purpose of this study is to empirically investigate the determinants of the use of M-bank in rural areas of the northern Bangladesh.
As the case of our study, we chose Thakurgaon district situated in the northern part of Bangladesh. Thakurgaon district is composed of 5 upazilas (sub-districts) with 641 villages. The district’s population is about 1,390,000 with about 321,000 households for an average of 4.3 persons per household. About 23% of population live below the poverty line, the literacy rate is 48%, and the male headed households account for 91% of the total households in this district (BBS, 2011).
During December 2016 through January 2017, we conducted the face-to-face questionnaire survey in 2 villages (Gilabari and Akhanagar) and 1 district town/city village (Collegepara). In total, 153 households were selected as our sample from 2,816 households with an average household size 4.4 and sex ratio 105 (BBS, 2011). Since a list of villagers was not available, the households were arbitrarily selected. For these households, average household size, sex ratio, and mobile phone users are 4.5, 105, and 93% respectively.
Because M-bank was initiated in Bangladesh in January 2011, household characteristics as of the end of 2010 are considered initial conditions at the introduction of M-bank. Data for such characteristics as educational attainment and occupation of the household head, and use of M-phone were obtained as the status at the end of 2010. However, some variables (e.g., number of family member and number of children, land holdings and location of the households) were obtained only at the time of the survey period. We treated these variables as proxy for the status as of the end of 2010. This is because such variables, especially, the status of land holdings and location of the households have been infrequently changed over the period 2011–2016/17 in the survey area.
(2) MethodsPast studies (Ahad et al., 2012; Azad, 2016; Laukkanen and Cruz, 2012; Munyegera and Matsumoto, 2016) have reported the determinants of the use of M-bank in developing countries (Uganda, Kenya, and Rwanda). It was reported that individual/household characteristics have influenced the use of M-bank. For example, males are more likely to use M-bank (Aker and Mbiti, 2010). Also, the more educated or the more affluent a person is, the more likely she/he is to use M-bank (Munyegera and Matsumoto, 2016). Thus, we hypothesize that the same behavioral patterns underlying the use of M-bank could apply to the case of rural Bangladesh.
Given this hypothesis, our analysis has proceeded in the following two steps. First, we explore a difference in each household characteristic by M-bank users’ status to capture a rough picture about the characteristics for users and non-users in the study area.
Second, we estimate the econometric model specified by the multinomial logit to identify the determinants of the use of M-bank in a more rigid manner. Three categories of using status are considered: early-users (2011–2013), late-users (2014–2016/17), and non-users. Because 6 years have passed since the introduction of M-bank, the determinants may depend on the start time of usage. The multinomial logit model allows us to examine the possibility as the different effects on each category can be estimated.
The brief definitions of explanatory variables in our econometric analyses are described in Table 2. As shown in this table, the M-phone use represents the situation of 2010: it can be considered as a predetermined variable. Because of some omitted variables such as migrant family members in 2010, however, the M-phone use might be an endogenous variable. We alternatively estimate a reduced form model by omitting M-phone use because of the difficulty of finding adequate instrumental variables. Also, having M-phone might be a prerequisite for using M-bank, while the M-phone use in our analyses is not for the case. This is because the households may have started to use M-phone after 2010. Nevertheless, we estimate the model separately for the users and non-users of M-phone as of 2010 to examine the possible different effects of explanatory variables between these two groups.
In addition to household characteristics, 2 village dummies are specified for 3 sub-samples: village-1 (Gilabari), village-2 (Akhanagar), and district town/city village (Collegepara) to capture the possible existence of regional disparities in use of M-bank which cannot be captured by the differences in characteristics across households.
Table 1 summarizes the data for the status of M-bank usage by 153 households in the study area. About 50% of households are found to have M-bank accounts in 2016/2017. Almost 60% of households in a district town/city village (Collegepara) used M-bank. The rates in 2 villages are lower than this, especially in Akhanagar (40%), suggesting regional disparities in the use of M-bank across 3 locations.
| Location | Users | Non-users | Total | ||
|---|---|---|---|---|---|
| Early | Late | Total | |||
| Village-1: Gilabary |
13 (26.0) |
12 (24.0) |
25 (50) |
25 (50) |
50 |
| Village-2: Akhanagar |
4 (7.7) |
17 (32.7) |
21 (40.4) |
31 (59.6) |
52 |
| District town: Collegepara |
8 (15.7) |
22 (43.1) |
30 (58.8) |
21 (41.2) |
51 |
| Total | 25 (16.3) |
51 (33.3) |
76 (49.7) |
77 (50.3) |
153 |
Source: Author’s household survey in 2016/2017.
1) Percentages (%) are shown in parenthesis.
Table 2 provides the comparison of household characteristics by M-bank adoption status: users and non-users. We tested the hypothesis of no difference between users and non-users. A χ2 test was applied to categorical variables, and a 2-sample t-test was employed to quantitative variables such as age and owned land size. M-bank users are found to significantly differ in the following characteristics from non-users. First, users tend to be older, more educated, mobile phone user with larger household size and greater size of the homestead area. Second, as for main occupation, users are more likely to be involved in self-employed business and office employee, while non-users tend to be engaged in farming.
| Users | Non-users | p-val1) | Total | |||||
|---|---|---|---|---|---|---|---|---|
| Variable | Description | Mean | SD | Mean | SD | Mean | SD | |
| Socio-demographic variables of household (HH) in 2010 | ||||||||
| HHhead age | Age of head | 38.368 | 11.917 | 35.13 | 12.28 | 0.100 | 36.739 | 12.17 |
| HHhead sex | 1 if head is male | 0.921 | 0.271 | 0.961 | 0.195 | 0.293 | 0.941 | 0.236 |
| Household size | Total members in the HH | 4.474 | 1.935 | 3.584 | 1.26 | <0.001 | 4.026 | 1.686 |
| No. of children | Total children in the HH | 1.447 | 1.182 | 1.299 | 1.113 | 0.424 | 1.373 | 1.146 |
| Educational attainment of household head in 2010 (dummy variables) | ||||||||
| No | None | 0.263 | 0.443 | 0.325 | 0.471 | 0.404 | 0.294 | 0.457 |
| Primary | 1–5 years | 0.145 | 0.354 | 0.221 | 0.417 | 0.224 | 0.183 | 0.388 |
| Secondary | 6–10 years | 0.316 | 0.468 | 0.364 | 0.484 | 0.532 | 0.34 | 0.475 |
| Higher | Longer than 10 years | 0.276 | 0.45 | 0.091 | 0.289 | 0.003 | 0.183 | 0.388 |
| Main occupation of household head in 2010 (dummy variables) | ||||||||
| Agriculture | Farmer | 0.289 | 0.457 | 0.416 | 0.496 | 0.103 | 0.353 | 0.479 |
| Labor | Labor | 0.066 | 0.25 | 0.169 | 0.377 | 0.048 | 0.118 | 0.323 |
| Self-employed | Self-employed business | 0.224 | 0.419 | 0.091 | 0.289 | 0.024 | 0.157 | 0.365 |
| Office employee | Office employed | 0.368 | 0.486 | 0.221 | 0.417 | 0.045 | 0.294 | 0.457 |
| Others | Politics, cooperative member | 0.013 | 0.115 | 0.026 | 0.16 | 0.568 | 0.02 | 0.139 |
| Students | Student | 0.013 | 0.115 | 0.026 | 0.16 | 0.568 | 0.02 | 0.139 |
| Unemployed | Unemployed | 0.026 | 0.161 | 0.052 | 0.223 | 0.414 | 0.039 | 0.195 |
| Mobile phone use in the household in 2010 (dummy variables) | ||||||||
| HH M-phone use | HH uses mobile-phone | 0.776 | 0.419 | 0.416 | 0.496 | <0.001 | 0.595 | 0.493 |
| Land holdings of household in 2016 | ||||||||
| Owned land (ha) | Owned land size | 0.683 | 2.017 | 0.414 | 1.393 | 0.338 | 0.548 | 1.731 |
| Homestead (a) | Size of homestead land | 0.078 | 0.276 | 0.021 | 0.034 | 0.074 | 0.05 | 0.198 |
| Location of household in 2016 (dummy variables) | ||||||||
| Village-1 | Gilabary | 0.329 | 0.473 | 0.325 | 0.471 | 0.955 | 0.327 | 0.471 |
| Village-2 | Akhanagar | 0.276 | 0.45 | 0.403 | 0.494 | 0.099 | 0.34 | 0.475 |
| District town | Collegepara | 0.395 | 0.492 | 0.273 | 0.448 | 0.109 | 0.333 | 0.473 |
| Total | Observations | 76 | 77 | 153 | ||||
Source: Author’s household survey in 2016/2017.
1) Column p-val reports the results for testing the hypothesis of no difference between users and non-users where chi-square tests were applied to categorical variable and 2-sample t-tests to quantitative variables.
It is interesting to note that the proportion of male-headed households in total households is found to be greater for non-users than users though the difference is statistically insignificant. This seems inconsistent with the finding available in the previous studies which found that males are more likely to use M-bank than females (Aker and Mbiti, 2010). This issue will be examined in a more rigid setting relying on the estimation of econometric models.
(2) Econometric estimation resultsTable 3 shows the estimation results of the base model that includes the M-phone use as the explanatory variable. The marginal effects are obtained from the multinomial logit model with the base category set to non-users. It is found that M-phone use, secondary educational attainment, homeland holdings have significant effects on the use of M-bank for both early and late users.
| Variables | Early-users1) | Late-users1) | Non-users1) | |||
|---|---|---|---|---|---|---|
| Socio-demographic variables of household (HH) in 2010 | ||||||
| HHhead age | 0.000 (0.003) | –0.002 (0.004) | 0.002 (0.004) | |||
| HHhead sex | –0.192 (0.113) | * | –0.250 (0.180) | 0.442 (0.179) | *** | |
| Household size | 0.019 (0.018) | 0.072 (0.026) | *** | –0.092 (0.029) | *** | |
| No. of children | –0.019 (0.030) | –0.012 (0.037) | 0.032 (0.036) | |||
| Educational attainment of household head in 2010 (dummy variables) | ||||||
| Primary | –0.058 (0.083) | 0.094 (0.113) | –0.035 (0.102) | |||
| Secondary | –0.193 (0.078) | *** | 0.191 (0.094) | ** | 0.002 (0.085) | |
| Higher | 0.073 (0.075) | 0.209 (0.124) | * | –0.282 (0.124) | ** | |
| Main occupation of household head in 2010 (dummy variables) | ||||||
| Labor | –0.137 (0.123) | –0.049 (0.165) | 0.186 (0.141) | |||
| Self-employed | –0.009 (0.082) | 0.156 (0.112) | –0.147 (0.108) | |||
| Office employee | –0.059 (0.093) | 0.157 (0.115) | –0.097 (0.108) | |||
| Others | –1.655 (147.498) | 0.266 (86.524) | 1.389 (60.975) | |||
| Students | –1.265 (154.971) | 0.779 (90.908) | 0.485 (64.064) | |||
| Unemployed | –1.488 (102.883) | 0.810 (60.352) | 0.678 (42.532) | |||
| Mobile phone use in the household in 2010 (dummy variables) | ||||||
| HH M-phone use | 0.144 (0.063) | ** | 0.122 (0.075) | * | –0.266 (0.060) | *** |
| Land holdings of household in 2016 | ||||||
| Owned land (ha) | –0.004 (0.014) | –0.038 (0.030) | 0.041 (0.025) | * | ||
| Homestead (a) | 0.008 (0.004) | ** | 0.022 (0.009) | ** | –0.030 (0.013) | ** |
| Location of household in 2016 | ||||||
| Village-1 | –0.004 (0.090) | –0.153 (0.126) | 0.157 (0.122) | |||
| Village-2 | –0.136 (0.085) | * | –0.051 (0.098) | 0.187 (0.097) | ** | |
| No. of observation | 153 | |||||
Source: Author’s household survey in 2016/2017. Standard errors are reported in parenthesis. Asterisks *, ** and *** represent significance at 10%, 5% and 1%, respectively.
1) The dependent variable is M-bank user status that classifies the households as follows: early user who started to use M-Bank between 2011 and 2013 (y=2), late user who started to use it between 2014 and 2016/17 (y=1), and non-user (y=0).
The estimation results of the reduced form model are presented in Table 4. The results are quite similar with those for the base model. Although the absolute values of marginal effects are likely to be greater for the reduced form model, the importance of the variables remains almost constant. The variables which are significant in both models have at least indirect effects on the use of M-bank. The estimation with the sub-sample of M-phone users showed similar results with our base model (not presented here because of the space limitation).1
| Variables | Early-users1) | Late-users1) | Non-users1) | |||
|---|---|---|---|---|---|---|
| Socio-demographic variables of household (HH) in 2010 | ||||||
| HHhead age | 0.000 (0.003) | –0.001 (0.004) | 0.002 (0.004) | |||
| HHhead sex | –0.195 (0.112) | * | –0.229 (0.185) | 0.424 (0.184) | ** | |
| Household size | 0.023 (0.019) | 0.072 (0.026) | *** | –0.094 (0.030) | *** | |
| No. of children | –0.026 (0.030) | –0.011 (0.038) | 0.037 (0.039) | |||
| Educational attainment of household head in 2010 (dummy variables) | ||||||
| Primary | –0.047 (0.084) | 0.085 (0.115) | –0.037 (0.109) | |||
| Secondary | –0.172 (0.082) | ** | 0.211 (0.093) | ** | –0.039 (0.089) | |
| Higher | 0.103 (0.075) | 0.198 (0.122) | * | –0.301 (0.124) | ** | |
| Main occupation of household head in 2010 (dummy variables) | ||||||
| Labor | –0.149 (0.127) | –0.043 (0.165) | 0.192 (0.145) | |||
| Self-employed | –0.007 (0.084) | 0.200 (0.112) | * | –0.193 (0.114) | * | |
| Office employee | –0.062 (0.092) | 0.169 (0.115) | –0.107 (0.112) | |||
| Others | –1.784 (129.710) | 0.191 (67.843) | 1.592 (61.868) | |||
| Students | –1.401 (129.228) | 0.783 (67.591) | 0.619 (61.639) | |||
| Unemployed | –1.511 (95.668) | 0.819 (50.037) | 0.692 (45.631) | |||
| Land holdings of household in 2016 | ||||||
| Owned land (ha) | –0.008 (0.013) | –0.032 (0.028) | 0.040 (0.025) | |||
| Homestead (a) | 0.011 (0.004) | *** | 0.027 (0.009) | *** | –0.038 (0.013) | *** |
| Location of household in 2016 | ||||||
| Village-1 | 0.005 (0.089) | –0.139 (0.125) | 0.134 (0.127) | |||
| Village-2 | –0.126 (0.086) | –0.070 (0.098) | 0.196 (0.100) | ** | ||
| No. of observation | 153 | |||||
Source: Author’s household survey in 2016/2017. Standard errors are reported in parenthesis. Asterisks *, ** and *** represent significance at 10%, 5% and 1%, respectively.
1) Refer to note 1) in Table 3.
Next, let us offer further discussion on the main estimation results. First, the results from the estimation of the base model suggest that the use of mobile phone has a positive effect on the introduction of M-bank in both early and late stages. The use of M-phone increases the probability of using M-bank by about 14% in the early stage and by about 12% in the late stage, with all other things held constant. In contrast, the use of M-phone reduces the probability of no use of M-bank by 27%, ceteris paribus. The use of M-phone in the household in 2010 is crucial even for the adoption of M-bank in the late stage. This result is consistent with the previous literature which found that M-phone has a significantly positive effect on the use of M-bank (Aker and Mbiti, 2010; Munyegera and Matsumoto, 2016).
Second, male headed-households are less likely to become early users and more likely to become non-users of M-bank. This suggests that the female-headed household is more likely to use M-bank. As mentioned before, this finding is inconsistent with the previous studies. For example, the World Bank (2014) reported that the rural females are the most often excluded from the formal financial sector in developing countries. Moreover, Azad (2016) found no gender difference in use of M-bank in Bangladesh at the individual level. This inconsistency may be attributed to the nature of our sample in which female household heads account for only 5.9%. Another plausible explanation is that currently households headed by females in Bangladesh may use M-bank to receive public welfare benefits (Parvez et al., 2015) the effect of which cannot be captured by variables in our models. Although we could not interview the all respondents in the sample, our follow-up survey conducted in February, 2018 confirmed that some female-household heads started to use M-bank because government or NGO provided stipend for child/female education and allowance for old-age directly via M-bank.
Third, the size of household is one of important factors for the use of M-bank for late users. One household member increase would raise the probability of being a late-users by 7% in the survey area. This seems inconsistent with the previous finding. Munyegera and Matsumoto (2016) found no significant effect of household size on M-bank use in Uganda. This dissimilarity may happen due to the difference in analytical tools. It should be noted that the size of household does not matter in the case of early users. It matters only for late users. This may be because as M-bank is diffused and known to rural households, whether the household decides to use M-bank would rely on expected benefits from use of M-bank rather than household characteristics including well-being conditions. The larger households would be more likely to enjoy benefits from use of M-bank such as reduction in transaction costs for purchasing goods and services and financial transfers that increase with the household size with all other things held constant.
Forth, an effect of educational attainment on M-bank is found to be rather mixed: a negative effect of secondary education on early users but a positive effect on late users. Therefore, the marginal effect on current non-users is not significant. One of plausible reasons for this may be that the household heads with secondary education are more cautious to use M-bank, taking possible risks into account than the one with no education. In contrast, higher education would reduce the probability of not using M-bank, suggesting a positive impact on the use of M-bank.
Fifth, the occupation types are found to have no significant effect on the use of M-bank in the base model. The households may tend to use M-bank as a tool for non-business purposes such as money transactions including the receipt of remittances from family members working in urban areas.
Sixth, the household well-being conditions represented by the size of homestead, has a significantly positive effect on the use of M-bank in all stages although their marginal effects are not large in value. This expected result is consistent with the previous finding (Aker and Mbiti, 2010).
This study has investigated the determinants of M-bank use in northern Bangladesh. There are some interesting results that are inconsistent with the previous literature with respect to i) an effect of the sex of household head, ii) an effect of the household size, and iii) an effect of occupation on the use of M-bank. The possible reasons for them are as follows. As for i), female-headed households are more likely to use M-bank to receive some government income transfer targeted at single-mother families. As for ii), the larger households are more likely to use M-bank because they can enjoy greater benefits from the use of M-bank. When the use of M-bank is diffused to some extent, external factors such as increases in the expected occurrence of use of M-bank may be more important than household internal factors such as wellbeing conditions for the household to decide to introduce M-bank. As for iii), the households in the survey area may not use M-bank for their businesses; rather their main purposes might be to exchange remittance with migrant family members.2 While the use of M-bank has facilitated personal money transactions, it may not play a crucial role in generating household income through enhancing business activities. Nevertheless, further research is required to investigate the mechanism underlying the use of M-bank in rural Bangladesh.
