 are inviting me to present this paper. It's a paper, co-written with Pettis de Vendée, who is here in the room. And title, Financial Behavior and Mobile Banking in Madagascar, Learning to Work Before Europe. So what are our purposes of this paper? We have two purposes, establishing and understanding the impacts of the use of mobile banking services on clients' behavior. And our main question is, does the use of mobile banking services, I will explain you later in my presentation, what are these mobile banking services? So does the use of these services have any influence? You can client savings and clients monet transfers. Until recently, only a limited number of studies have been dedicated to the analysis of the impact of mobile banking services on users' behavior. And these studies have mainly been conducted in Africa, in Ghana, in South Africa, in Uganda, and mostly in Kenya, with paper published in 29, 2011, 2012. This literature suggests that the use of mobile banking services may have a positive impact on individual savings, may affect money transfer behavior, and may encourage poor people's access to finance. But such analysis are relevant only if the two groups of the population are quite similar from a statistical standpoint. So keeping this goal in mind, we undertook in Madagascar our own survey in 2012. Why Madagascar? Why is it an interesting ground for our analysis? Because in Madagascar, there is a strong need for financial inclusion. According to the 2012 fine access survey, only 6% of the population hold bank accounts. And in Madagascar, Orange Monet created a mobile banking service called Orange Monet in September 2010. We arrived to the mobile banking services. Initially, at the time of our survey and two years before, since 2010, the Orange Monet services were the deposit cash-in service using mobile phone, the withdrawal cash-out service, the domestic money transfer service, and the bill-pay service. The bill-pay service is the possibility for a client to pay in a shop with his mobile phone if the shop is equipped with an orange e-payment terminal. What about our survey? Our survey was conducted in all districts of the city of Tananarive in March 2012. We surveyed 598 randomly selected orange customers. So we have a population composed of orange customers only. These customers are divided in 196 regular users of Orange Monet, that is to say, using at least one Orange Monet service per month, and 402 orange clients, that is to say, non-regular Orange Monet users, not using Orange Monet services or using them less than once a month. In our study, Orange Monet regular users are our treatment group, and our orange clients, non-users, are the control group. In the following table, we will call OM Orange Monet users and OB Orange Basic Orange Clients non-users. We implement the matching methodology to assess the effect of using Orange Monet services on users' financial behavior. The matching methodology enables a comparison of outcomes among a set of users and non-users statistically comparable. First, we are going to present a social demographic characteristic of our two populations, Orange Monet users and OB clients, Orange Basic clients. And you can see in this table that compared to OB clients, Orange Monet users are more likely to be men, to be younger, to be unmarried, and to earn a low income. But you see in the last column of this table that all these differences are not significant. The significant differences are precisely age, marital status, a number of young people in the household, religion, and some incomes. But we are going to come back on these characteristics after. About financial behavior, we focus the analysis on five individual financial variables, which are the sum of formal savings, the number of remittances sent, and the number of remittances received, and the sum of remittances sent and received. Each one of these variables concerns the free last month before the survey. A few words about the definitions concerning savings. We consider savings in formal financial institutions. And it appears that more than half of orange customers have at least one formal savings account. Concerning domestic remittances, we consider only remittances inside Madagascar because at the time of our survey, international remittances with Orange Monet did not exist. And among our 598 orange customers, 40% sent remittances and 37% received money. This table presents the main differences between Orange Monet clients and Orange Basic clients. And you can see that Orange Monet clients send and receive remittances more frequently than OBI clients. And the difference is significant. But the amount transferred are smaller for Orange Monet users compared to Orange Basic clients. And you see that the sum of remittances sent, the difference is significant, but not concerning the sum of remittances received. Well, how can we explain this phenomenon? Why can we explain that Orange Monet users send and receive more frequently remittances? By safety reasons, as well as lower cost associated with Orange Monet services, we will come back on this point later. Both reasons may lead to transfer more often, but to transfer smaller amounts. But at this stage, our question is the following. Did the ability to make transfers using Orange Monet encourage users to transfer more? Or did they decide to subscribe to this service? Precisely because they already transferred frequently. To answer this question, we have implemented an impact study, which impacts study, a matching study. The goal of a matching process is to find for each treated unit one non-treated unit with similar individual observable characteristics. So we use the available information on untreated units. Our untreated units are Orange Basic clients, non-user of Orange Monet. And we use this information to build up a counterfactual for each treated unit, for each Orange Monet user. But it isn't easy to find people who have exactly the same characteristics in both subpopulations. So Rosenbaum and Rubin suggest matching treated and non-treated units using a propensity score. I think you know what is a propensity score. But quickly, the propensity score is the individual probability to belong to the treatment according to a vector of individual observable characteristics. In this paper, the matching process requires estimating the individual probability to be an Orange Monet user conditionally to a vector of covariates, which includes a set of socioeconomic variables assumed to be useful to explain why an individual is using Orange Monet services. Here we have the results. Considering the model 1 in the first column, you can see that people are significantly more likely to be an Orange Monet user if they are young or if they are older, because we have a negative sign of a variable age and a positive sign of a variable age squared. People are more likely to be an Orange Monet user if they are born outside from Antananarivo, if they have young people in their household, and if they earn an income rather low. In the model 2, sorry, in the second column, it's the same, sorry, in the model 2, we retain only the significant variables from the model 1. And these variables keep their significance, so we conclude that our estimation is quite robust. Excuse me. So the covariates we are going to keep for realizing the matching are VH squared, the number of young people, and the income, and the place of birth. Once the probability to be an Orange Monet user is estimated, we can compute the individual propensity score, but it is generally impossible to find two individuals with exactly the same propensity score. So we are going to implement two different ways. We are going to choose two different ways to implement the matching process, nearest neighbor matching and kernel matching. With nearest neighbor matching process, each Orange Monet user is matched with one non-user, whose propensity score is the nearest possible, and a common support region can be defined. And with the kernel matching method, every Orange Monet user is matched with the weighted average of all non-users, the weights are inversely proportional to the distance between the traded groups and the control groups propensity score. Finally, we check the quality of our matching process. If our matching process is correct, covariates should be balanced in both groups and no significant differences should be found. To check this, we conduct two balancing tests, the equality of means and the standardized differences test. And these tables shows that our matching is considered as correct. So differences between the two groups, Orange Monet users and non-users, in savings and money transfers, may only be due to the use of Orange Monet. What about our main question? What are the impacts of using Orange Monet? It becomes now possible to assess the average treatment effect on the traded ATT, calculating the difference between the outcomes of traded individuals and untreated ones. Then, delta ATT is only the average of these differences. Here we have the results. Watching delta ATT significant, you can see that delta ATT is only significant for the number of remittances sent and the number of remittances received. And you can see that delta ATT is positive. So Orange Monet users send and receive more frequently than over clients orange, non-user of Orange Monet services, send and receive more frequently remittances. I'm going to comment these results. Whatever the matching method, Orange Monet users significantly send and receive remittances more frequently. Such a positive effect may be explained by the low cost compared to Western Union Orange Monet transfer service is less expensive. I have tables if you want to show you the fees practiced by Orange Monet and by Western Union, for instance. Over explanation of our result, the safety of the Monet transfer by Orange Monet is safer than the informal transfer and informal remittances. And the use of these services. On the contrary, you have seen in the table that we have no impact of Orange Monet on savings and on the sum of remittances sent and received. The absence of effect on other financial behavior may be explained by the short period elapsed since the deployment of Orange Monet and by the fact that until mobile banking services have modified individual economic situations, Orange Monet users have no incentive and no ability to modify their financial behavior. To conclude, we have to note that all these results are in line with what was found in some previous studies. They voted to MPCA in Kenya. And we can also compare our result with the feeling of Orange Monet users because in our survey, we have questions about the perceptions from Orange Monet users. Among the Orange Monet users who use the Monet transfer service, 55% believe it has led them to transfer more frequently. So this perception is consistent with our analysis. But on the opposite, among Orange Monet users who deposit money into their Orange Monet account, 62% considered that due to this service, their savings has increased. This feeling on the contrary is not consistent with our analysis. Should we then conclude that the mobile banking's promises have not been kept? Sorry, certainly not. Because the fact that the deposit money is the most used service allows the assumption that clients use it as a way to increase their precautionary savings. And if they increase their precautionary savings, we assume that they are going to improve risk management, be encouraged to invest, be encouraged to open a bank account, and to ask for credit. So mobile banking may have a positive impact on the economy. Thank you for your adoption.