Will The Widespread Use of Cashless Payments Reduce the Frequency of the Use of Cash Payments?

Fujiki, Hiroshi

doi:https://doi.org/10.15353/rea.v13i3.4522

1 Introduction

Reducing the costs of cash use is often cited as one of the benefits of the widespread use of cashless payment methods in addition to the other important benefit that consumers can enjoy a variety of payment methods, such as contactless payments or payments based on applications on a mobile phone. The private costs spent to use cash are estimated to be 0.12% of the GDP in Germany (Cabinakova et al., 2019), 0.45% of the GDP in Canada (Kosse et al., 2017), 0.60% of the GDP in Uruguay (Álvez et al., 2019), and 0.29% of the GDP, or 1.6 Trillion JPY, in Japan (Nomura Research Institute, Ltd., 2018). In the case of Japan, among the 1.6 Trillion JPY, the largest breakdown of costs is 500 billion JPY in personnel costs for cash management at stores, 412 billion JPY in ATM equipment and installation costs, 146 billion JPY in ATM business operating expenses, 140 billion JPY in ATM security company outsourcing costs, and 100 billion JPY in counter personnel costs for cash-related operations.

Policies that promote cashless payments in Japan that would be beneficial in reducing the costs of cash use focus on subsidizing the use of cashless payment methods. For example, the Japanese government’s Point Reward Project for Consumers Using Cashless Payment subsidized cashless payments in some registered retail shops from October 1, 2019, to June 30, 2020.[1] Increasing the cost of using cash would also promote cashless payments. For example, Japanese commercial banks are reducing the number of ATMs; however, the total number of ATMs increased from 2009 to 2019 because the increase in the number of ATMs in convenience stores exceeded the decrease in the number of ATMs in Japanese commercial banks. Therefore, the increase in the cost of using ATMs seems to have a limited impact on cash users.

To help implement policies to promote cashless payments and to reap the benefit of reducing the cost of cash use, many researchers have examined the question: Will the spread of cashless payments reduce the frequency of cash payments? Such studies have included data from Canada (Payments Canada, 2018), the Netherlands (Jonker et al., 2018), and Switzerland (Brown et al., 2020). Note that the costs of installing ATMs and keeping cash registers in stores, which explain the largest and the second-largest breakdown of the cost of cash use, respectively, are fixed costs that can only be saved if the amount of cash in circulation becomes close to zero. Thus, answering this question is important to determine the extent to which a nation could reap the benefit of reducing the costs of cash use; however, few studies have examined the question in Japan due to the limitation of data. An exception is Wakamori (2020), who used the data from Macro Mill for November 2018 and October 2019 to show a decrease in the frequency of cash use (from 60% to 50%) and an increase in the frequency of Quick Response (QR)-cord payment use (from 0% to 7%) in the choice of household payment methods.

To fill this research gap, this paper explores the question: Will the widespread use of cashless payments, such as credit cards, contactless prepaid cards, and mobile payments via smartphone applications, reduce the frequency of the use of cash payments? To this end, data were collected from the Financial Literacy Survey (the FLS 2019) that was administered from March 1, 2019, to March 20, 2019, to 25,000 individuals aged 18–79 years in Japan, which contains a question regarding the frequency of the use of payment methods (almost every day, once a week, once a month, rarely or never, no adoption). The FLS data show that the frequency of cash use for those who use both cash and noncash payment methods is about once in 2.3 days and about once in two days for those who use cash exclusively. Therefore, there is only a slight difference between the frequency of cash use for those who use both cash and noncash payment methods and that of those who exclusively use cash. The result did not change even if a model that considers the endogenous choice of payment methods was used for the analysis. Two counterfactual simulations were also conducted to examine the potential factors that could promote cashless payments, the decrease in the consumers’ willingness to use cash and the increase in the cost of ATM usage based on the unique data from the FLS 2019; however, the effect of these changes on the frequency of cash usage is also slight. Overall, the results show that the benefit of reducing the cost of cash use due to the widespread use of cashless payment methods is overestimated because the frequency of cash payments is unlikely to decrease despite the use of cashless payment methods.

This paper relates to the estimation of the social costs of using cash and other payment methods. Recent examples in Germany (Cabinakova et al., 2019) and Canada (Kosse et al., 2017) and summaries of the methodology and literature are provided by Hayashi and Keeton (2012) and Krüger and Seitz (2014). This paper focuses on the possible changes in consumers’ choices of payment methods due to the decrease in the willingness to use cash, including the possible increase in the cost of using ATMs and its effect on the frequency of cash use. This paper also relates to the literature on cash demand and the choice of payment methods in Japan and abroad.[2] This study is unique in that the new data on the frequency of cash use obtained from the FLS 2019 were used.

The limitations of the study are as follows. First, the results are based on the FLS 2019 conducted in March 2019, which does not cover the effects of the spread of COVID-19 on cash usage or the Japanese government’s Point Reward Project for Consumers Using Cashless Payments, which subsidized cashless payments from October 1, 2019, to June 30, 2020, and increased the usage of QR-code payments. Second, we could not estimate the cash demand function due to a lack of data on the outstanding amount of cash holdings. Finally, we could not distinguish between cash usage based on transaction values or payment contexts (e.g., day-to-day payments, regular payments, or hoarding) due to a lack of data.

The remainder of the paper is organized as follows. Section 2 discusses the data, and Section 3 explains the methodology. Section 4 reports the results of the estimation. Section 5 concludes.

2 Data

The FLS 2019 is a web survey that was administered from March 1, 2019, to March 20, 2019, to 25,000 individuals aged 18–79 years in Japan. Variables on the use of payment methods, financial literacy, financial behavior, and other demographic variables from the FLS were constructed as follows.

2.1 Use of payment methods

For the variables regarding the use of payment methods, Question 45 on the FLS for 25,000 individuals was used: “How often do you use the following payment methods: credit cards, debit cards, electronic money, mobile payments using smartphones, or cash? Choose only one answer from the following options: 1. Almost every day, 2. About once a week, 3. About once a month, 4. Scarcely or never, 5. Do not adopt it.” For this question, mobile payments using smartphones could be prepaid or post-paid, QR-code based, or mobile wallets for credit cards, debit cards, and electronic money. Cash includes checks. Note that electronic money refers to Japanese prepaid cards.

Regarding the willingness to use noncash payment methods, Question 46 on the FLS was used for the 24,516 sub-sample for those who responded “Scarcely or never” or “Do not adopt it” for a credit card, and/or debit card, and/or electronic money, and/or smartphone payments: “Under what conditions will you use those payment methods? Choose up to three answers from the following options: 1. Generous rewards programs, 2. Increase in the cost of using cash (Increase in the ATM charge, Decrease in the number of ATMs), 3. Wide acceptance by merchants, 4. Shorter time for settlement, 5. Flexible reloading of electronic money, 6. Safer information security, 7. Some tools that reduce the possibility of overuse, 8. Universal payment methods among merchants that do not require consumers to select particular payment methods for each merchant, 9. Other, and 10. I am satisfied with cash payments and do not plan to use other payment methods.”

Based on Question 46, the FLS regards those who responded “Scarcely or never” or “Do not adopt it” in Question 45 as nonusers of noncash payment methods. Hence, it is assumed that the users of each payment method include those who replied, “Almost every day,” “About once a week,” and “About once a month” to Question 45. Moreover, we are interested in the substitution from cash to noncash payment methods due to changes in the opportunity costs of using cash, and we focus on cash users. Unfortunately, about 30% of respondents did not reply regarding their financial asset holdings, and about 20% did not reply regarding their annual pretax incomes. Thus, we dropped these households from the analysis. Among the 24,516 respondents, we focused on 14,977 cash users.

2.2.1. Combination of the use of payment methods

In this subsection, we explain the combination of the use of noncash payment methods. The third to fifth columns of Table 1 show the combination of the use of payment methods by 14,977 cash users shown in the first column who replied to Question 46. It was found that these cash users chose 15 combinations of four noncash payment methods, as shown in the first column. The second column shows the payment methods that the users of the payment methods in the first column do not use. Specifically, the third column shows that 36% of respondents are credit card (hereafter C) and electronic money (hereafter E) users, 23% uses C, 12% uses C, E, and mobile payments using smartphones (hereafter S), 10% does not use four noncash payment methods (namely, they use cash only, hereafter None), and 7% uses E. The remaining 10 combinations of choices consisted of below 7% of total observations (or below 1,000 observations) and did not provide enough degrees of freedom for the later regression analyses that contained about 50 independent variables. The use of debit cards is rare, as only 3% uses C, E, and debit cards (hereafter D), and we do not analyze the choice of D. The focus is the 13,218 sample of the top five combinations of users of four noncash payment methods listed in the sixth to eighth columns in Table 1: CE, C, CES, None (cash only), and E.

Table 1

**Use of payment methods by cash users**

2.1.2. Number of transactions made using a specific payment method per day

Second, the number of transactions made by a specific payment method per day was measured using Question 45 as follows. We assigned the value of (1/2) to 1 time, (1/14) to (1/7) times, (1/60) to (1/30) times, 0 to (1/60) times, and 0 times for those who replied, “Almost every day,” “About once a week,” “About once a month,” “Scarcely or never,” and “Do not adopt it,” respectively. We assigned these values because if a respondent stated that s/he uses a payment instrument almost every day (or every week or every month) and does not use the payment instrument on a given day (week or month), it was assumed that s/he is likely to use it on the following day (week or month), and thus the expected number of transactions made by this payment instrument within a day (week or month) is from 1/2 time to 1 time (1/14 to 1/7 times a day or 1/60 to 1/30 times a day). We constructed the variables representing the number of transactions per day made by the payment instrument y, where y = Cash, Credit, Emoney, Debit, and Smartphone. We assumed that the values of y take 3/4 times, 3/28 times, 1/40 times, 1/120 times, and 0 times for those who replied, “Almost every day,” “About once a week,” “About once a month,” “Scarcely or never,” and “Do not adopt it,” respectively. Note that in the analysis of cash users, we focus on the respondents who replied “Almost every day,” “About once a week,” and “About once a month” for the use of cash. The mean values of y are reported in the third to seventh rows of Table 2. The first row shows the choice of the top five preferred payment methods, and the second row shows the payment methods not used for each choice of payment method for the sake of reference. The second column reports the average number of transactions per day of the overall sample, and it shows that Cash = 0.44, Credit = 0.13, Debit = 0.00, Emoney = 0.11, and Smartphone = 0.02.

Depending on the use of payment methods, the average number of cash transactions per day differs. Cash for None and E take values of 0.52 and 0.51, respectively, which are higher than the overall average value of 0.44; however, based on the ratio of cash users who chose “Almost every day,” “About once a week,” and “About once a month” for the choice of the top five preferred payment methods in Figure 1, even for those who chose CE, C, or CES, about 90% chose either “Almost every day” or “About once a week,” which is similar to those who chose None and E. Thus, the data show that regardless of the use of noncash payment methods, the majority of cash users uses cash at least once a week. This means that the use of noncash payment methods does not necessarily indicate a less frequent use of cash by cash users. Sections 3 and 4 will examine whether the same tendency would be observed even if we adjusted the demographic background of respondents and possible self-selection bias arising from the endogenous choice of payment methods.

Figure 1

**Frequency of cash usage based on the choice of the top five preferred payment methods**

Regarding the average number of transactions per day for noncash payment methods, Credit for those who use CES, CE, and CE takes the values of 0.21, 0.16, and 0.15, respectively, which are higher than the overall average value of 0.13. Emoney for those who use CES, E, and CE takes the values of 0.25, 0.20, and 0.18, respectively, which are higher than the overall average value of 0.12. Smartphone for those who use CES takes the value of 0.15, which is higher than the overall average value of 0.02.

One might be interested in the respondents who replied “Scarcely or never” or “Do not adopt it” for cash, although they are not included in the analysis thus far. If these respondents use noncash payment methods very frequently, it would be expected that the prevalence of noncash payment methods would indicate very little cash usage. An examination of the data showed that 607 respondents chose “Do not adopt it” for cash. Among them, 381 (or 63%) responded “Do not adopt it” for all other payment instruments, and only 82 (or 14%) responded “Almost every day” or “About once a week” for a credit card. These 82 respondents tended to have lower incomes, lower financial asset holdings, a lower age, and a lower level of financial literacy and tended to be male compared with the overall average reported in Table 2. A total of 416 respondents chose “Scarcely or never” for cash, and among them, 199 (or 48%) responded “Almost every day” or “About once a week” for a credit card. These 199 respondents might serve as a good approximation of cash usage under the frequent use of noncash payment methods. These respondents tended to have higher incomes, higher financial asset holdings, and a lower age and were more likely to be male; however, due to the small number of respondents in this group, we could not include them in the later regression analyses that contained about 50 independent variables. Thus, we approximated the impact of the widespread use of noncash payment methods on the frequency of cash usage by comparing the exclusive cash users and cash users who also use noncash payment methods in the analysis.

2.1.3. Willingness to use cash and conditions for using noncash payment methods

Third, we measured the willingness to use cash based on Question 46. First, among the 13,218 samples, 2,701 (or 20% of the sample) respondents exclusively chose 10 (“I am satisfied with cash payments and do not plan to use the other payment methods”). We created a dummy variable, Satisfied with cash, that takes a value of 1 for these respondents and otherwise zero. If no one replied to this question affirmatively, the satisfaction regarding cash usage is so low that everyone would adopt noncash payment methods. Second, we grouped 1,679 (13%) respondents who chose 2 (Increase in the cost of using cash [Increase in the ATM charge, decrease in the number of ATMs]) because these respondents are willing to use cash if the costs of using ATMs are sufficiently low compared with the cost of using other noncash payment methods. We created a dummy variable, ATM, which takes a value of one for these respondents and otherwise zero to determine the willingness to use cash given the current costs of using ATMs. If no one replied to this question affirmatively, the costs of using ATMs are too high, and everyone would use noncash payment methods.

We developed eight dummy variables to determine the conditions under which a household would be willing to use other noncash payment methods that had not been adopted. Note that these questions ask about the future use of noncash payment methods conditional on the current use of noncash payment methods and are not directly related to the current frequency of cash usage. The following eight dummy variables take a value of 1 for the respondents who chose the eight options in Question 46 and otherwise zero: Reward for those who selected “Generous rewards programs,” Acceptance for those who selected “Wide acceptance by merchants,” Speed for those who selected “Shorter time for settlement,” Reload for those who selected “Flexible reloading to electronic money,” Security for those who selected “Safer information security,” Restraint for those who selected “Some tools that reduce the possibility of overuse,” Standard for those who selected “Universal payment methods among merchants that do not require consumers to select particular payment methods for each merchant,” and Other_factors_only for those who selected “Other.” Note that the variable names Acceptance, Security, and Restraint follow the names of the dummy variables that represent the underlying needs that are satisfied by the consumers’ chosen payment methods, as demonstrated by Borzekowski et al. (2008). Note also that Satisfied with cash and Other factors only are exclusively chosen; however, the respondents who selected ATM could choose other variables—Reward, Acceptance, Speed, Reload, Security, Restraint, and Standard—because the respondents could choose up to three answers.

The average values of Satisfied with cash, ATM, Reward, Acceptance, Speed, Reload, Security, Restraint, Standard, and Other factors only according to the use of payment methods are reported in the eighth to seventeenth rows in Table 2. Observe that as many as 45% of respondents chose Reward, which reflects tough competition between payment methods through a discount program or a reward program. Twenty-five percent of respondents choose Acceptance, which suggests that the wide acceptance by merchants would also increase the use of noncash payment methods. The average values differ substantially across the use of payment methods. For example, the average value of Satisfied with cash for None takes a value of 0.43, which is substantially higher than the average value of 0.20. The average value of Satisfied with cash for C and E takes values of 0.22 and 0.23, respectively. This means that 22% of C do not plan to use electronic money, debit card, or mobile payments via smartphone applications and that 23% of E do not plan to use credit cards, electronic money, debit card, or mobile payments via smartphone applications because they are satisfied with cash payments. The average value of Satisfied with cash for the users of CE and CES takes values of 0.14 and 0.15, respectively, which are lower than the values for None, C, and E. The results suggest that satisfaction with cash payments tends to be one of the reasons for the non-adoption of noncash payment methods when the person adopts fewer kinds of payment methods. Note also that the values of None (exclusively cash users) for Reward, Acceptance, Speed, Reload, Security, and Standard are also smaller than the average values, which suggests that it is difficult to encourage the use of cashless payments for the users of None.

2.2. Financial literacy[3]

We followed Sekita et al. (2018) in using a proxy for objective financial literacy. Objective financial literacy is defined as the number of correct answers to 12 questions on five categories of financial literacy: two questions on the compound interest rate, two questions on the diversification of risk in stock investments, two questions on life insurance products, four questions on mortgage payments and the relationship between interest rates and bond prices, and two questions on inflation. We discuss the details of the 12 questions in the Appendix. We followed Kadoya and Khan (2020) in using the experience of financial troubles, such as bank transfer fraud or multiple debts (Fraud1). We add to the work of Sekita et al. (2018) and Kadoya and Kahn (2020) via the dummy variables of debt holdings (Debt). The means of Objective financial literacy, Fraud1, and Debt based on the use of payment methods are reported in the eighteenth to twentieth rows in Table 2. The means of Objective financial literacy for the users of CE and CES take higher values compared with the overall means, and the means of those for the users of None take lower values compared with overall means.

As a proxy of information sources, the frequency of obtaining information on financial and economic conditions from mass media (News) was examined.[4] The means of News in the twenty-first row for the users of CE and CES take higher values compared with the overall means, and the means of those for the users of None take lower values compared with the overall means.

We also used dummy variables indicating the source for obtaining knowledge and information when selecting financial products from Question 35 of the FLS (See details on Question 35 in the Appendix). We first created a dummy variable that takes a value of 1 for respondents who do not invest in financial products (S_do_not_choose) and a dummy variable that takes a value of 1 for respondents who were not sure what opportunities would allow them to acquire such knowledge or information (S_do_not_know). Second, we created dummy variable S_fin_inst for respondents who chose at least one of the following information sources that are related to financial institutions: sales staff at financial institutions, pamphlets provided at financial institutions, lecture meetings or seminars, financial professionals/professional financial advisors, and schools. Finally, we created a dummy variable S_exclude_fin_inst that takes a value of 1 for respondents who chose at least one information source—mass media, websites, and conversations with family members or friends—but did not choose the information sources included in S_fin_inst. Respondents take a value of 1 for S_exclude_fin_inst who use family and friends, mass media, and websites as their information sources but do not use formal information sources, such as financial institutions or financial experts. In the remaining regression analysis, we use S_do_not_choose as the base case. The mean values of S_fin_inst and S_exclude_fin_inst in the twenty-third and twenty-fourth rows for the users of CES and CE take higher values than the average value. Overall, the users of None and E are less financially literate compared with users of noncash payment methods.

2.3. Financial behavior

We followed Sekita et al. (2018) and used six variables that capture financial behavior from the perspective of behavioral economics (see Beshears et al., 2018, for literature on behavioral household finance). Over-confidence captures one’s over-confidence regarding financial literacy through the difference between one’s subjective financial literacy (self-evaluation of one’s level of financial literacy in comparison to other people) and Objective financial literacy. Impatience captures the present-biased preferences in which one places extra value on more immediate awards. It is based on the following question: “If I had the choice of (1) receiving 100,000 yen now or (2) receiving 110,000 yen in 1 year, I would choose (1), provided that I can definitely receive the money. Choose from a scale of 1 to 5 where 1 means ‘agree’, and 5 means ‘disagree.’” Impatience is defined as the difference between 5 and the answer to this question, so that a higher value corresponds to a respondent with a higher time preference and thus impatience, assuming that the safe interest rate remains about zero in the Japanese economy. Reputation is a proxy variable that shows whether a person considers reputation in making financial decisions. It is based on the following question: “When there are several similar products, I tend to buy what is recommended as the best-selling product rather than what I actually think is a good product.” Self-control is a proxy of the degree to which a person makes deliberate and thoughtful decisions. It is based on the following question: “Before I buy something, I carefully consider whether I can afford it.”

We created two proxy variables for risk aversion. Risk aversion 1 is a dummy variable that takes a value of 1 for a person who says “no” to the question “If you invested 100,000 yen, you would either get a capital gain of 20,000 yen or a capital loss of 10,000 yen at 50% probability.” Risk aversion 2 is a proxy value for the extent to which a person is reluctant to take a risk on an investment. It is based on the following question: “I am prepared to take a risk when saving or making an investment.”

The means of these variables based on the use of payment methods are reported in the twenty-fifth to thirtieth rows in Table 2. The users of None seem to be less overconfident compared with users of noncash payment methods.

2.4. Demographic variables

We constructed the following demographic variables. We first constructed a variable indicating household annual pretax income (Income). Because the FLS asks about household annual pretax income (in units of million yen) by ranges (zero, below 2.5, 2.5–5, 5–7.5, 7.5–10, 10–15, 15 over, no response, or do not know), we assigned the values of 0, 1.25, 3.75, 6.25, 8.75, 12.5, and 15 for those who chose zero, below 2.5, 2.5–5, 5–7.5, 7.5–10, 10–15, and 15 and over, respectively. We constructed a variable indicating the household total financial asset holdings (Asset). Because the FLS asks about household total financial asset holdings (in units of million yen) by ranges (zero, below 2.5, 2.5–5, 5–7.5, 7.5–10, 10–20, 20 over, no response, or do not know), we assigned the values of 0, 1.25, 3.75, 6.25, 8.75, 15, and 20 for those who chose zero, below 2.5, 2.5–5, 5–7.5, 7.5–10, 10–20, and 20 and over, respectively. We constructed a variable indicating the ages of respondents (Age). Because the FLS asks for the ages of respondents by ranges (below 20, 20–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69, 70–74, and 74–79), we assigned the values of 19, 22, 27, 32, 37, 42, 47, 52, 57, 62, 67, 72, and 77 for those who chose below 20, 20–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69, 70–74, and 75–79, respectively. We constructed a variable indicating years of education (Edu). Because the FLS asks about educational attainment via the choice of below senior high, senior high, vocational college, junior college, university, graduate school, and others, we assigned the values 9, 12, 13, 14, 16, and 18 for those who chose below senior high, senior high, vocational college, junior college, university, and graduate school, respectively. We dropped the observations of those who chose “others.”

We also constructed the following dummy variables. They include dummy variables indicating the gender of respondents (Male = 1 for men), the employment status of respondents (Private company, Public company, Teacher, Self-employed, Part-time, House [stay-at-home mum/dad], Student, No Job and Other Job, where the base case is the sum of No Job and Other Job), and nine areas of residence (Hokkaido, Tohoku, Hokuriku, Kanto, Chubu, Kinki, Chugoku, Shikoku, and Kyushu, and the base case is Kanto region). The means of these variables based on the use of payment methods are reported in the thirty-first to fiftieth rows in Table 2. Regarding the differences in the average values among the use of payment methods, users of None and E tended to have a lower income, a lower Asset, and shorter years of education. The users of None and E included many students. The users of CES tended to have a higher income and longer years of education, to be young, and to be male. This suggests that one should consider possible differences in demographic characteristics among users of a particular combination of payment methods.

3 Methodology

In Section 4, we approximate the impact of the widespread use of noncash payment methods on the frequency of cash usage by comparing exclusive cash users and cash users who also use noncash payment methods. In doing so, we note that the choice of payment methods is an endogenous decision made by the respondents. Thus, we began by examining whether the choice of payment methods is affected by the respondents’ demographic characteristics. We then examined cash transactions per day conditional on the use of noncash payment methods using the ordinary least square (OLS) regression. We also used Heckman’s (1979) model when the endogenous choice of payment methods should be considered in estimating cash transactions per day conditional on the use of noncash payment methods. We used the predicted value of cash transactions per day conditional on the use of noncash payment methods to examine whether the frequency of cash use was similar for those who use both cash and noncash payment methods and those who exclusively use cash. Finally, we conducted counterfactual simulations to examine whether the decrease in consumers’ willingness to use cash and the resulting changes in the choice of payment methods would reduce cash usage.

First, to examine whether the choice of payment methods is affected by the respondents’ demographic characteristics, we estimated a multinomial logit model to explain the respondents’ use of the top five preferred payment methods conditional on the vector of conditioning variables X explained in the previous section: financial literacy, financial behavior, demographic variables, two dummy variables for the willingness to use cash—Satisfied with cash and ATM—and a vector Z that contains the variable asking under which conditions the household would be willing to use the other noncash payment methods that they have not adopted: Reward, Acceptance, Speed, Reload, Security, Restraint, Standard, and Other_factor_only. We assumed that the multinomial logit model would approximate a household’s choice of noncash payment methods. Assume that an indirect utility function of a respondent i conditional on the choice of noncash payment methods j = CE, C, CES, None, and E at time t is approximated by the following linear function:

where and are the vector of parameters, and v_ijt are unobservable preferences for payment method j of respondent i. If the respondent i at a time t (in our case, 2019) chooses the noncash payment method k instead of l, it means that . Let v_ijt follow an independent extreme value distribution whose cumulative distribution function is exp(–exp(–v_j)) for each noncash payment choice j. Under these assumptions, the choice of a noncash payment method follows a multinomial logit model, where the probability of the choice of a noncash payment method j for respondent i at time t, P_ijt, depends on X_ijt and Z_ijt in equation (1) after normalizing the parameter value for choice None to zero. Therefore, the estimation of a multinomial logit model of equation (2) is as follows:

where D_ikt is an indicator variable that takes a value of 1, 2, …, 5 if the choice of a noncash payment method k for respondent i is CE, C, CES, None, or E, respectively. The mlogit command of Stata 16 was used to estimate the model. Standard errors in the following analysis were adjusted to an intragroup correlation within the clusters formed by gender, age group, and prefectures because the respondents of the FLS were selected from the people registered with an internet survey company by the weight of gender, six age groups, and 47 prefectures (2*6*47 = 564 clusters) based on the Japanese census.

Second, we estimated the conditional mean value of Cash using two methods. First, we used the ordinary least square (OLS) regression. Because the variables in Z ask about the future use of noncash payment methods and are not directly related to the frequency of current cash usage, we regressed Cash on X but not on Z. We estimated the following equation (3) for each choice of noncash payment method j:

where is a vector of parameters, and are error terms of respondent i who selects payment method j. The reg command of Stata 16 was used to estimate the model.

We also estimated a model by Heckman (1979) to correct possible self-selection bias from the choice of payment method following Schuh and Stavins (2010):

Because the variables in Z ask about the future use of noncash payment methods conditional on the current use of noncash payment methods and are not directly related to the frequency of current cash usage, we use Z only for the second equation; a probit equation was used for self-selection only. The Heckman command of Stata 16 was used to estimate the model using the maximum likelihood method.

Finally, using the estimates of the multinomial logit model, equation (2), we first simulated the effects of the decrease in the values of Satisfied with cash and ATM on the use of noncash payment methods measured by the changes in P_ijt. We then used the estimates of OLS regression, equation (3), to simulate the effects of the decrease in the values of Satisfied with cash and ATM on the conditional mean value of Cash. To cope with the possible self-selection problem, we also used the conditional cash demand function, equation (4), to simulate the changes in the conditional mean value of Cash. This methodology could be criticized because the choice of the values of Satisfied with cash and ATM is endogenous; however, we assumed that these two variables reflect the preference for the willingness to use cash and that the exogenous changes in these two variables would approximate preference shocks for the willingness to use cash.

4 Results

4.1 Results of the multinomial regression on the use of noncash payment methods

Table 3 reports the results of the marginal effects computed from the parameter estimates of a multinomial logit model equation (2). *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. The baseline case corresponding to the constant term is a respondent who does not select financial products by themselves, who has no experience with financial trouble, who does not have debt, whose value of Risk aversion1 is zero, whose gender is female, whose occupation is other occupation, and who is living in the Kanto area. In Table 3, the first row shows the top five preferred payment methods, and the second row shows the payment options that were not chosen by the users of the top five preferred payment methods. We focused on the statistically significant estimates of the marginal effects reported in Table 3.

Regarding the variables related to the willingness to use cash, the third row shows that Satisfied with cash is positively associated with the use of None and E and negatively associated with the use of CE and CES, as expected. The fourth row shows that ATM is positively associated with the use of None. ATM is negatively associated with CES because these people would use ATMs rarely and would be insensitive to the increase in the cost of using ATMs. Taken together, we anticipated that the decrease in the willingness to use cash as expressed by the decrease in the value of Satisfied with cash and ATM would reduce the use of None and E (except for ATM) and would increase the number of users of CES and CE (except for ATM).

Regarding the eight variables contained in vector Z reported in the fifth to twelfth rows, the users of None are negatively associated with Reload and Standard and positively associated with Security and Restraint. The users of E are positively associated with Speed, Reload, Security, and Restraint. The users of E are concerned about the Speed and convenience of reloading, while the users of None are not. The users of C are negatively associated with Reload, Standard, and Other factors only and positively associated with Security. These users seem to value the benefit of security but do not value the speed of electronic money or are not interested in widely accepted mobile payments. The users of CE are negatively associated with Restraint and positively associated with Acceptance, Reload, Security, Standard, and Other factors only. These users seem to be prepared to use mobile payments if safe and widely accepted applications are available. The users of CES are negatively associated with Acceptance, Speed, Security, and Restraint. These users seem to be unprepared to use debit cards in the future.

Regarding the variables related to financial literacy, Objective financial literacy is negatively associated with the use of None and E and positively associated with the use of CE and CES. This is a reasonable result because the use of electronic money is easy enough for a less financially literate person or students because it does not involve credit from the providers of the electronic money and has the upper limit of the daily usage; however, the use of credit cards in Japan is restricted to above age 18 with a certain level of annual income. Fraud1 is negatively associated with the use of CE and C and positively associated with the use of None, E, and CES. News is negatively associated with the use of C and None and positively associated with the use of CES and CE. It also shows that users of None tend to be negatively associated with S_fin_inst and S_exclude_fin_inst and that the users of CES tend to be positively associated with these variables. Overall, the users of None and E are less financially literate compared with users of noncash payment methods.

Table 3

**Use of payment methods (Marginal effects, Equation [2])**

The variables related to financial behavior show that the users of CES are positively associated with Overconfidence and Risk aversion2 and negatively associated with Risk aversion1 and Self-control. In contrast, users of None are negatively associated with Overconfidence and Risk aversion2 and positively associated with Self-control. Negative association with Overconfidence suggests that while the users of None are less financially literate, they understand their low level of financial literacy and behave in financially conservative ways. For example, their positive association with Self-control could mean that they avoid using noncash payment methods so they do not overspend The users of C are positively correlated with Impatience, Self-control, and Reputation. In contrast, the users of CE are negatively associated with Impatience, Self-control, and Reputation. Finally, the users of E are negatively associated with Overconfidence.

Regarding the demographic variables Income, Assets, Age, and Edu, the users of CE tend to have a higher income, a higher amount of financial assets, a higher age, and long years of schooling, while the users of C tend to have a lower income, a higher amount of financial assets, a higher age, and shorter years of schooling. The users of CES tend to have a higher income, a lower age, and longer years of schooling, while the users of None tend to have a lower income, a lower amount of financial assets, and shorter years of schooling. The users of E tend to be associated with a lower amount of financial assets and shorter years of schooling.

Regarding the results for gender, the users of C, None, and E are positively associated with Male, and the users of CE are negatively associated with Male.

Regarding the results for the dummy variables for occupations, the users of CE are positively associated with Private, Public, Part-time, and Students, and in contrast, the users of C are negatively associated with Private, Public, Part-time, and Students. The users of CES are positively associated with Private and Self-employed. The users of None are negatively associated with Private, Pubic, and Self-employed and positively associated with Students. The users of E are negatively associated with Private and positively associated with Students.

Finally, the results for regional dummies show that the users of C and None tend to live in the non-Kanto area (the base case, most high-density populated area in Japan), and the users of CE, CES, and E tend to be negatively associated with the non-Kanto area because shops that accept electronic money and mobile payments are located more often in highly populated areas.

The actual and forecasted proportion of the payment method chosen by equation (2) is reported in the forty-ninth and fiftieth rows. Equation (2) was effective in forecasting the proportion of the payment methods chosen by the respondent.

The results of forecasting the proportion of the payment methods chosen by setting the value of Satisfied with cash and ATM to zero to examine the effect of a declining willingness to use cash by respondents are reported in the fifty-first and fifty-second rows. The reduction in the value of Satisfied with cash reduces the proportion of users of C, None, and E and increases the proportion of users of CE and CES, as expected. The reduction in the value of ATM reduces the proportion of the users of C, None, and E and increases the proportion of the users of CES slightly, as expected. Regarding the potential effects on the adoption of noncash payment methods by people exclusively using cash, the proportion of respondents using None (11.4% of the sample) would decrease by 2.3% point and 0.3% point if all households replied negatively to Satisfied with cash and ATM, respectively. These results suggest that the decrease in the willingness to use cash does not lead to a substantial decrease in the proportion of respondents exclusively using cash.

4.2 Results of the regressions on the frequency of cash usage conditional on the use of noncash payment methods

Table 4 reports the parameter estimates of equation (3) by the OLS for the top five combinations of the use of payment methods.

First, the third and fourth rows show that Satisfied with cash is positively correlated with Cash for the users of CE, C, CES, and None, and ATM is positively correlated with Cash for the users of CE and CES. These users would reduce the frequency of cash use if the willingness of using cash decreased as expressed by the decrease in the values of Satisfied with cash and ATM.

Second, Objective financial literacy is negatively correlated with Cash for the users of CE. News is positively associated with Cash for the users of CE and C. Third, for the users of CE, Cash is negatively associated with Overconfidence. Impatience is positively correlated with Cash for most cases except for the use of CES, which suggests that impatient people tend to use cash more frequently regardless of the choice of noncash payment methods except for the use of CES. Reputation is positively correlated with Cash for the users of None and E. Risk aversion2 is negatively associated with Cash for the users of C and None.

Fourth, regarding the demographic variables, Cash is negatively correlated with Asset for the users of CE. Cash is positively correlated with Income and Age and negatively correlated with Asset for the users of C. Cash is positively correlated with Age and Edu for the users of None. Cash is negatively correlated with Age for the users of E.

Fifth, regarding occupations, Cash is positively correlated with Private for all users, positively correlated with Public for the users of C and CES, positively correlated with Self-employed for the users of CES and None, positively correlated with Part-time and Students for the users of CE, C, CES, and None, and positively correlated with House for the users of CES and None.

Table 4

**Estimates of the conditional frequency of cash usage (Equation [3])**

Finally, regarding regions, people living in the Kinki and Kyusyu areas tend to use Cash more frequently compared with the Kanto area (the base case, most high-density populated area in Japan).

Using the parameter estimates reported in Table 4, we forecasted the average number of cash transactions per day conditional on the choice of the combinations of the use of noncash payment methods, which is reported in the fortieth row labeled Predicted cash. Compared with the actual value of average Cash reported in the forty-fourth row labeled Actual cash, the models predicted the cash transaction per day accurately. This means that even after controlling for the heterogeneity of respondents by regression, the average frequency of the cash use of exclusively cash users (0.517) is close to those of both cash and noncash payment method users (a range from 0.384 to 0.505).

We also calibrated the model to estimate the average number of cash transactions per day under a decrease in the willingness to use cash by setting the values of Satisfied with cash and ATM to 0 for all respondents, which is reported in the forty-second and forty-third rows labeled Cash at satisfied with cash = 0 and Cash at ATM = 0, respectively. The reduction in the value of Satisfied with cash and ATM reduced the average predicted value of Cash only slightly, at most by 0.01 times a day, except for the users of E for Cash at ATM = 0, as anticipated.

4.3 Results of Heckman’s (1979) model on the frequency of cash usage conditional on the use of noncash payment methods

The parameter estimates of equation (4) are reported in Tables 5 and 6. Specifically, Table 5 reports the parameter estimates of the conditional cash demand equation shown in the first line of equation (4), and Table 6 reports the marginal effects computed from the parameter estimates of the probit selection equation shown in the second line of equation (4). The results of the conditional cash demand equation in Table 5 are similar to the results reported in Table 4, except for the users of CES, which makes the parameter estimates of News, Debt, S_fin_inst, Impatience, Self-control, Income, Asset, and Age statistically significant and the parameter estimates of Male not statistically significant.

The forty-first row labeled as Predicted cash shows that the average predicted values of Cash for the users of CE and CES are 0.511 and 0.619, respectively, which are higher than the corresponding average predicted values made by OLS, 0.421 and 0.437, respectively. Moreover, the average predicted value of Cash for the users of CES in the forty-first row takes a small value of 0.091 and is not statistically different from zero based on its standard error (0.063), reported in the forty-second row. It was found that 262 of the 1,737 (or 15%) predicted Cash values for the users of CES took negative values, which is inconsistent with the assumption that Cash must be greater than or equal to zero. Thus, we did not use the results from the Heckman model for the use of CES.

Table 5

**Estimates of the conditional frequency of cash usage (Equation [4])**

Table 6

**Use of payment methods (Marginal effects, Equation [4])**

The results of the selection equation reported in Table 6 are similar to the results in Table 3, and the predicted proportions of the use of payment methods reported in the fifty-second row labeled as Predicted proportion are fairly precise; however, note that the parameter estimates of ρ, the correlation of error terms in the conditional cash demand equation and probit selection equation, reported in the forty-sixth row labeled as rho, are not statistically significantly different from zero at 5% points for the users of None and E, as the forty-seventh row labeled Wald test for rho=0 and chi2(1) and the forty-eighth row labeled P-values of the Wald test show. The results mean that we have no justification to use the Heckman model for the users of None and E. Taken together, we should not use the results from the Heckman model for the uses of CES, None, and E.

4.4 Results of counterfactual simulations for average cash use

Based on the results in Sections 4.2 and 4.3, we used the parameter estimates of the Heckman model for the uses of CE and C and the results from the OLS for the users of CES, None, and E to conduct counterfactual simulations on the changes in the willingness to use cash.

The third and fourth rows of Table 7 show the average predicted probabilities of the use of each payment method obtained from the multinomial logit model or probit model (the estimates of average obtained from equation [2] for the users of CES, None, and E or the Heckman model of equation [4] for the uses of CE and C) and the estimates of Cash conditional on the use of each noncash payment method (the forecast values of the number of cash transactions per day conditional on the use of each noncash payment method obtained from equation [3] for the users of CES, None, and E or equation [4] for the uses of CE and C). The average Cash weighted by the proportion of the choice of payment method is 0.523.

The fifth and sixth rows of Table 7 show the average predicted probabilities of the use of each payment method and the estimates of Cash conditional on the use of each noncash payment method when we set the value of Satisfied with cash to 0. Because the parameter estimates of Satisfied with cash in the conditional cash demand for the users of E are not statistically significantly different from zero in Table 4, we did not use the simulated value of Cash when we set the value of Satisfied with cash to 0 for the users of E. Thus, it was assumed that there is no change in conditional cash demand for the users of E. The average Cash weighted by the proportion of the choice of payment method is 0.507: a decrease of 0.016 from the baseline estimates. The impact on the frequency of cash transactions is small because the changes in the average cash transactions per day when we set the value of Satisfied with cash to 0 is at most -0.02. Note that if we change the proportion of the choice of payment methods alone, average Cash weighted by the proportion of the choice of payment method is 0.521, a decrease by 0.002 from the baseline estimates, which accounts for 13% (i.e., 0.002 out of 0.016) of the decrease from the baseline estimate.

Table 7

**Results of the counterfactual simulations**

The seventh and eighth rows of Table 7 show the average predicted probabilities of the use of each payment method and the estimates of Cash conditional on the use of each noncash payment method when we set the value of ATM to 0. Because the parameter estimates of ATM in the conditional cash demand for the users of C is not statistically significantly different from zero in Table 5 and those for None and E are not statistically significantly different from zero in Table 4, we did not use the simulated value of Cash when we set the value of ATM to 0 for the users of C, None, and E. Thus, it was assumed that there is no change in conditional cash demand for these three cases. The average Cash weighted by the proportion of the choice of payment method is 0.516—a decrease of 0.007 from the baseline estimate. The impact on the frequency of cash transactions is small because the changes in the average cash transactions per day when we set the value of ATM to 0 is at most -0.01. Note that if we change the proportion of the choice of payment methods alone, average Cash weighted by the proportion of the choice of payment method is 0.520—a decrease of 0.003 from the baseline estimate—which accounts for a 43% (i.e., 0.003 of 0.007) decrease from the baseline estimate. The results suggest that the number of cash transactions per day would decrease only slightly even if the consumer’s willingness to use cash decreased.

4.5 Robustness checks

This subsection reports the results of three robustness checks on the analysis presented in the previous sections.

First, regarding the choice of estimation methods, the value of the average Cash weighted by the proportion of the choice of payment method, 0.523, in Table 7 is higher than the sample average, 0.438. This is because we used the predicted value from the Heckman model rather than the OLS for the users of CE and CES. If we used the OLS estimates for the simulation for the users of CE and CES, the value of the baseline average of Cash weighted by the proportion of the choice of payment method would be 0.440, which is close to the sample average; however, if we use the forecast value of Cash by OLS for the users of CE and CES and set the value of Satisfied with cash or ATM to 0, the average Cash weighted by the proportion of the choice of payment method is 0.431 and 0.438, respectively, a decrease of 0.01 and 0.003 from the baseline estimate, respectively. Therefore, the choice of estimation methods affects the baseline average of Cash weighted by the proportion of the choice of payment method; however, the results of the counterfactual simulation measured by the changes in the average of Cash remain unchanged.

Second, we used interval regressions instead of OLS regressions for the conditional cash demand, but the overall results were unchanged. Again, the Wald test for the parameter of ρ justified the use of the interval regression with a sample selection for the users of CE and C. The average of Cash weighted by the proportion of the choice of payment method when we set Satisfied with cash and ATM to 0 fell from the benchmark by 0.014 and 0.006, respectively, which is similar to the results shown in Table 7.

Third, regarding the effects of the benefit of using noncash payment methods, the variable Security is statistically significant for all five top choices of payment methods shown in Table 3. Thus, we set the value of Security to 0 for all respondents to approximate a situation in which there is no concern regarding the security of noncash payment methods and thus a higher willingness to use noncash payment methods. As anticipated based on the marginal effects reported in Table 3, this change would increase the proportion of CES users from 0.131 to 0.151 and would decrease the proportion of other users; however, the average Cash weighted by the proportion of the choice of payment method would be 0.520, a decrease of 0.003 from the baseline average of Cash. We could perform the same exercise for the other variables contained in Z; however, the decreases from the baseline average of Cash are at most 0.001. Therefore, in our model, the effects on the use of Cash through the increase in the benefit of using noncash payment methods are smaller than the effects of the decrease in the consumer’s willingness to use cash.

5 Conclusions

Will the widespread use of cashless payments reduce the frequency of cash payments? Using Japanese microdata, this paper examines how much the frequency of cash use for those who use both cash and noncash payment methods differs from those who exclusively use cash. The data show that the frequency of cash use for those who use noncash payment methods is about once in 2.3 days and that of exclusively cash users is about once in two days, and thus they are only slightly different. The result did not change even if we used a regression model that considers the endogenous choice of payment methods. Two counterfactual simulations were also conducted to examine potential factors that could promote cashless payments, the decrease in the consumers’ satisfaction with cash, and the increase in the costs of ATM usage; however, the effect of these changes on the frequency of cash usage is also small. The results show that the often stated benefit of the spread of cashless payments to reduce the cost of cash use is likely to be overestimated because the frequency of cash payments is unlikely to decrease despite the use of cashless payment methods. Note that our results are based on data from March 2019, and thus recent data, if available, might suggest that the frequency of cash usage by exclusive cash users and both cash and noncash payment method users differ more than our results suggest. This limitation will be resolved as new data on cash usage become available in the future.

Will The Widespread Use of Cashless Payments Reduce the Frequency of the Use of Cash Payments?

Abstract

1 Introduction