Public Use and Distribution of Retail CBDC: An Evidence from Thailand’s Retail CBDC Pilot Program

3.1.1 VAR Setup for all Types of Transactions Involving rCBDC

To analyze the intertemporal relationships among variables related to rCBDC transactions, we assess the connections among different types of rCBDC transactions – specifically issuance, redemption, top-up, withdrawal, transfer, and payment – by their value in Thai Baht (THB) on a weekly basis. This analysis is conducted within a vector autoregression (VAR) framework, where.

With the transactions aggregated by type, the VAR specification can be expressed as follows:

We estimate the VAR model, which includes all rCBDC transactions represented in y _t , a 6 × 1 vector at week t, where y_t-j are lagged variables up to p lags.

The VAR estimations are summarized in Table 6, while the results of the Granger causality tests are presented in Table 7. For selecting the lags used in the VAR regression, we employ the Bayesian Information Criterion (BIC) as the determining criterion.

Table 6 presents the VAR estimations of the intertemporal relationships among various types of rCBDC transactions, shedding light on the behavior of financial service providers (FSPs) in managing rCBDC holdings and liquidity. Specifically, there is limited evidence that rCBDC issuance is significantly influenced by other transactions at any of the included lags (such as redemption, top-up, withdrawal, payment, and transfer). In contrast, rCBDC redemption is significantly influenced by other retail transactions across different lags, indicating that FSPs adjust their redemptions based on prior retail transactions. These findings suggest that while FSPs do not adjust their issuance of rCBDC according to past retail transactions, they do gradually adjust their rCBDC holdings through redemption in response to retail transactions. Specifically, the redemption of rCBDC is positively predicted by lagged top-up transactions, indicating that an increase in demand for top-ups leads FSPs to reduce their redemption amounts. Conversely, if the demand for withdrawals increases, FSPs respond by increasing the amount of redemption to convert rCBDC into deposits for their users. Additionally, we find that redemption is significantly and negatively affected by rCBDC transfers. When FSPs observe a high value of retail transactions, they tend to reduce the amount of redemption to accommodate the demand for transfers. A possible explanation for this behaviour is that FSPs may issue rCBDC upfront with a high holding balance to cushion against future demand uncertainty. They then gradually adjust their rCBDC holdings through redemptions based on previous retail transaction activities. Lastly, we observe no strong evidence regarding the relationship between redemption transactions conducted by FSPs and payments made by individual users.

Table 7 presents the VAR-based Granger causality results, highlighting the significance of intertemporal relationships among various types of rCBDC transactions. The findings support the VAR estimations, showing that redemption is significantly predicted by the lags of all retail transactions, particularly withdrawals and transfers. In contrast, there is no significant relationship between issuance and the lags of any retail transactions.

3.1.2 VAR Setup for Retail Transactions Involving rCBDC

We also examine the intertemporal relationships among various rCBDC transactions – including top-ups, withdrawals, transfers, and payments – by their value in THB on a daily basis within a VAR framework, where.

By aggregating transactions by type, the VAR specification can be expressed as follows:

We estimate a VAR model that includes all types of rCBDC transactions. Here y _t is a 4 × 1 vector at day t, and y_t-j represents the lagged variables up to p lags.

The VAR estimations are summarized in Table 8, while the results of the Granger causality tests are presented in Table 9. The selection of lags used in the VAR regression is determined based on the Bayesian Information Criterion (BIC).

Table 8 presents the VAR estimations of the intertemporal relationships among all rCBDC transactions, providing insights into the behavior of individual users and merchants in utilizing rCBDC. The findings align with the hypothesis that rCBDC is primarily used as a means of payment rather than a store of value, as holding rCBDC does not offer incentives (e.g. interest rates or other financial services) compared to traditional deposits. Specifically, individuals and merchants are likely to increase the amount of rCBDC they top up in their wallets when there is a higher demand for payments from the previous day. Conversely, the amount of rCBDC used for payments tends to rise in response to increased top-ups in prior days. Additionally, we observe that users are inclined to withdraw more from their wallets following payments received the previous day, indicating a tendency not to store rCBDC but rather to withdraw it for deposits in interest-earning accounts.

On the other hand, we find no evidence supporting a significant relationship between all rCBDC transactions and lagged transfers, as the transfer of rCBDC does not impact the total amount of rCBDC in circulation. Consequently, the aggregated demand for top-ups and withdrawals remains unaffected.

Table 9 shows the results of the VAR-based Granger causality tests, highlighting the significance of intertemporal relationships among all types of rCBDC transactions. The results confirm that top-ups and withdrawals of rCBDC are significantly influenced by lagged payments, while they are not significantly influenced by lagged transfers.

3.2.1 User Behavior Regarding rCBDC

Understanding user behavior and preferences regarding rCBDC is crucial for formulating effective policies and designing the rCBDC for public use. Therefore, we empirically investigate spending and saving behaviors related to rCBDC, focusing on usage by different transaction types and the value of transactions across various transactions. This analysis is detailed in Tables 10 and 11.

In this section, we summarize the usage of rCBDC by users, categorized by transaction types, to provide insights into user behavior regarding rCBDC (see Table 10). Our findings indicate that the majority of rCBDC transactions in the pilot program are business-to-consumer (B2C) transactions. In other words, rCBDC is primarily used as a means of payment for individuals purchasing goods from merchants, with the highest number of transactions at 42,802 and a total value of 2.72 million THB (78,901 USD). The average transaction value is approximately 63.56 THB (1.84 USD), which aligns with the price of a meal or drink. In contrast, business-to-business (B2B) transactions using rCBDC are relatively low, with only 1,522 transactions totaling 812,302 THB (23,559 USD). However, these transactions have a higher average value of 533.71 THB (15.48 USD) each, suggesting that merchants may use rCBDC to purchase raw materials and supplies for production. Peer-to-peer (P2P) transactions with rCBDC show moderate activity in terms of the number of transactions, total value, and average value. This indicates that users may be transferring value to one another via rCBDC to settle debts.

Additionally, we summarize the time duration between transactions across different types of transactions in Table 11. Our analysis reveals that, among retail transactions, individuals and merchants tend to withdraw funds more frequently than they top up their rCBDC balances. Specifically, individuals and merchants typically top up their rCBDC every 5 days and 9 days, respectively, while withdrawals occur with a 1-day gap. Notably, individual users make payments and transfers more frequently, also with a 1-day gap between these transactions. This consistent pattern indicates that users primarily utilize rCBDC as a means of payment rather than as a store of value. Users maintain sufficient rCBDC balances in their wallets to facilitate daily payments and transfers, topping up their accounts weekly to ensure they have enough funds for their transactions. The frequent top-ups suggest that users may be reluctant to hold large amounts of rCBDC in their wallets, likely due to the absence of interest on rCBDC balances.

3.2.2 User Demographic Attributes and Their Impact on rCBDC Usage

In this section, we investigate the impact of demographic attributes on individual usage of rCBDC. This analysis aims to provide insights that can inform policy design and mechanisms to enhance user adoption of rCBDC. Specifically, we assess the relationship between transfers and payments made using rCBDC by individual users, considering demographic factors such as gender and age. To evaluate these relationships, we employ cross-sectional regression using ordinary least squares (OLS) estimation with heteroskedasticity-robust standard errors. We estimate the relationship between the monetary value of transfers and payments and the demographic attributes of age and gender. The OLS specifications are organized separately for transfers and payment in a cross-sectional setting, where the values for each individual user are aggregated over the entire sample period. In the equation, age _i represents each individual user’s age, while gender _i is a binary variable that takes a value of 1 if the individual user is female, and i denotes each individual user. The sample is limited to those users who have wallets with commercial banks, as demographic attributes are only available for these users and not for those with non-bank FSP wallets. The equation is expressed as follows:

Table 12 displays the OLS estimations of the value of transfers and payments made using rCBDC, considering age and gender as factors for individual users. The results provide important empirical evidence regarding the influence of demographic characteristics on the use of rCBDC. The findings suggest that the age of individual users is significantly and positively associated with the value of rCBDC usage. In other words, older individuals tend to use rCBDC more frequently for payments and transfers than younger users. However, there is no evidence to suggest that gender significantly affects rCBDC usage. These results highlight the need for policy measures promoting rCBDC adoption to target specific demographic groups, particularly by age. Additionally, in Table 13, the results consistently show that individuals aged between 55 and 64 had the highest average value of payments, at 1,603.60 THB (46.51 USD) and 1,004.49 THB (29.13 USD). Similarly, users aged 45–64 demonstrated the highest value of transfers, at 610.02 THB (17.69 USD) and 645.23 THB (18.71 USD).

3.2.3 User’s Adoption of rCBDC

In this section, we aim to provide empirical insights into the factors affecting the adoption of rCBDC, particularly focusing on the effect of reward or incentive schemes on rCBDC usage. This analysis is crucial for understanding how to formulate measures that can increase the adoption of rCBDC. To do so, we conduct statistical tests to determine whether the incentive schemes significantly influence the adoption of rCBDC by using Welch’s t-test. The Welch’s t-test provides a robust assessment when there are unequal variances between two sub-sample. We divide the sample into two sub-samples: one covering the period when rewards were offered (from 6 February to 21 May 2023) and the other for the period when no rewards were offered (from 22 May to 31 August 2023). The rewards were provided to individual users based on the frequency and value of their rCBDC transactions. Specifically, at the end of each month during the reward period, rCBDC was topped up to the wallets of the three individual users with the highest frequency of payment and transfer transactions, as well as those with the highest value of transactions for that month.

Table 14 reports the results of the Welch t-test, which examines the difference in the average daily transaction frequency and value between the period when rewards were offered and the period when they were not. The study finds a positive effect of rewards on the adoption of rCBDC in terms of both transaction frequency and value. Specifically, the average number of daily rCBDC transactions during the reward-offered period was 1.64, compared to 1.33 during the non-reward period. Similarly, the average value of daily rCBDC transactions was higher during the reward period, at 133.15 THB (3.28 USD), compared to 76.45 THB (2.22 USD) during the non-reward period. These results suggest that incentive mechanisms can increase the adoption of rCBDC, both in terms of transaction frequency and value, during the pilot period. This finding provides useful insights for policymakers in designing future rCBDC adoption strategies. However, due to the limited duration of the pilot, the long-term effects of such reward mechanisms were not examined and remain an open question.

4.1.1 Two-Tiered Structure

In payment system networks, a tiered structure of wholesale and retail payments is commonly referenced. Similarly, the rCBDC system exhibits a two-tier structure, often referred to as a hybrid rCBDC architecture, which consists of a wholesale layer and a retail layer, as illustrated in Figure 2.

Figure 2:

Network topology of rCBDC pilot in Thailand (categorized transaction: wholesale and retail transactions). (A) Wholesale Transactions, (B) Retail Transactions. This figure displays network topology of rCBDC pilot in Thailand, broadly categorized of transactions: wholesale and retail transactions. Each node represents a unique wallet and different color corresponding to node indicates different FSPs and Central bank who open and operate that wallet (3 FSPs and 1 Central banks, represented by dark blue). Node size is proportional to degree measures compared among involved wallets (i.e. the number of edges incident on that node). Each edge represents a bilateral relationship between two wallets, with edge width proportional to the number of transactions compared among involved edges. The sample period is 6 February 2023 to 8 September 2023.

For wholesale payments (Panel A), transactions occur among financial service providers (FSPs) and between FSPs and the central bank, which operates the high-value payment infrastructure. In contrast, retail payments (Panel B) involve transactions between individual users and merchants. In this two-tier system, FSPs play a crucial role by bridging the wholesale and retail layers. They facilitate the issuance and redemption of rCBDC from the central bank, distributing it to retail users, and provide wallet services for making payments and transfers. Empirically, the significance of FSPs in the rCBDC system is highlighted by their larger size (indicated by a greater number of edges connected to FSP nodes) and their central positioning within both the wholesale and retail layers.

4.1.2 Clusters

Figure 2 also reveals the presence of several clusters within the rCBDC network. These clusters are differentiated by the various financial service providers (FSPs) that offer wallet and payment services, as indicated by the three different colors representing the players’ nodes associated with each FSP. Furthermore, the clusters highlight a key characteristic of payment transactions: the distinction between on-us and off-us transactions. Specifically, clusters located toward the outer edges of the network are more likely to represent on-us transactions, while those situated closer to the center tend to indicate off-us transactions. This is reflected in the density of edges connecting players’ nodes with different FSPs. In essence, the inner clusters consist of players’ nodes that conduct transfers and payments with nodes associated with various service providers. Empirically, we observe that transactions within the rCBDC system are predominantly on-us, contrasting with traditional payment systems where off-us transactions are more prevalent.

4.1.3 Topological Across Different Types of Transactions

To further analyze the rCBDC network by transaction types, we divide it into two subnetworks (Figure 2): the subnetwork for wholesale transactions (Panel A) and the subnetwork for retail transactions (Panel B). The wholesale subnetwork encompasses the issuance and redemption of rCBDC conducted by financial service providers (FSPs), while the retail subnetwork includes top-ups, withdrawals, payments, and transfers performed by retail users. Panel A illustrates the subnetwork of wholesale transactions within the rCBDC system, comprising nine nodes (0.21 % of the total nodes in the aggregated network) and six edges (0.02 % of the total edges). Each FSP node connects to two central bank nodes (represented as dark blue nodes), indicated by two directed edges. This signifies that FSPs engage in the issuance (receiving rCBDC) and redemption (returning rCBDC) of rCBDC with the central bank. Furthermore, the subnetwork displays several separate clusters corresponding to each FSP, suggesting a lack of bilateral relationships among FSPs in wholesale transactions.

On the retail side, Panel B presents the subnetwork of retail transactions within the rCBDC system, which consists of 4,310 nodes (99.88 % of the total nodes in the aggregated network) and 26,280 edges (99.98 % of the total edges). The dominance of transactions and bilateral connections within the retail domain highlights that the rCBDC pilot primarily serves retail transactions. Additionally, the subnetwork reveals several distinct clusters associated with each FSP. A closer examination of these clusters indicates that most retail users tend to conduct transactions primarily with one another within the same FSP, as evidenced by the proximity of their nodes within the same cluster while remaining separated from nodes in other clusters.

4.2.1 Node-level Measures in the rCBDC Network

To empirically analyze the properties of nodes in the rCBDC network and their implications for the role of financial service providers (FSPs) and their interactions with retail users in a hybrid rCBDC setting, we present two key node-level measures: node degree and node centrality (Figure 3).

Figure 3:

Empirical distributions of in-degree and out-degree corresponding to players’ wallets from rCBDC pilot in Thailand. (A) In-degree, (B) Out-degree. This figure displays empirical distributions of in-degree and out-degree corresponding to players’ wallets from rCBDC pilot in Thailand. In-degree is the number of incoming edges toward a particular node. Out-degree is the number of outgoing edges from a particular node. The empirical distributions of in-degree and out-degree is estimated using Epanechnikov kernel density, with bandwidth equal to one, which are estimated on aggregated data covering sample period of 6 February 2023 to 8 September 2023.

4.2.2 Node Degree

To assess the importance of each node in the rCBDC network, we estimate the node degree, which indicates a particular player’s connectivity to other nodes within the network. Node degree provides insight into the extent of transactions between players using rCBDC. For the directed rCBDC network, we present empirical distributions of players’ node degrees using nonparametric kernel density estimation, as shown in Figure 3. The empirical distributions are divided into two measures: in-degree (Panel A), which represents the number of incoming transactional relationships to a player’s node, and out-degree (Panel B), which reflects the number of outgoing transactional relationships from that node. Our findings indicate that the rCBDC network exhibits characteristics of a scale-free network, as the distributions of in-degree and out-degree follow a power law. This is illustrated by the downward-sloping nature of the distributions, suggesting that most player nodes engage in transactions with only a few other nodes (resulting in low in-degree and low out-degree), while a select few nodes transact with many others (resulting in high in-degree and high out-degree). In essence, retail users tend to engage in limited transactions with others, whereas FSPs conduct transactions with a larger number of retail users due to their broader business scope.

4.2.3 Node Centrality

In addition to node degree, node centrality is a key focus in network literature as it provides valuable insights into various aspects of the network, including transmission capabilities, reliance on specific nodes, and overall robustness. To assess the significance of particular player nodes in the rCBDC network regarding the transmission of rCBDC and reliance on specific players, we estimate two common measures of network centrality: eigenvector centrality (which indicates the extent to which a node is connected to other central nodes) and betweenness centrality (which measures how often a node lies on the shortest path between other nodes), as shown in Table 15. The results consistently reveal that the player nodes of FSPs rank highest in both eigenvector centrality and betweenness centrality. Their eigenvector centrality scores are 1.00 and 0.68, while their betweenness centrality scores are 0.73 and 0.60. This suggests that FSPs play a critical role in managing and distributing rCBDC to retail users within the network, highlighting their systemic importance.

The results from the node-level measures indicate that, within the hybrid rCBDC framework, FSPs play a crucial role in distributing rCBDC to individual users and providing customer service. A failure in this role could have systemic repercussions for the entire rCBDC network. Consequently, implementing policies to increase the number of FSPs in the hybrid rCBDC system could help mitigate concentration risks associated with over-reliance on a small number of FSPs, thereby enhancing the system’s resilience.

4.2.4 Network-Level Measures

To thoroughly evaluate the rCBDC network and assess the effectiveness of the hybrid rCBDC model – including its distribution and the interrelationships among customers – we conduct an empirical analysis of network-level measures. Specifically, we estimate three common metrics that characterize the network as a whole: density, average clustering coefficient, and modularity (Table 16).

4.2.5 Density

To understand the characteristics of a network, density is a crucial metric. In the rCBDC network, the density measure reflects the number of transactional relationships between players’ nodes. This metric provides insight into the distribution of rCBDC within the network. According to Table 16, the density of the rCBDC network is estimated at only 0.001, indicating that it is a sparsely connected network. This suggests that, under the hybrid rCBDC pilot setting, the distribution of rCBDC is limited and not widely utilized within the system. Consequently, there is potential for introducing new intermediaries to facilitate the acquisition and redistribution of rCBDC, as well as for developing new business models or use cases that encourage broader adoption among FSPs and users.

4.2.6 Average Clustering Coefficient

In analyzing the structure of a network, the clustering coefficient is a key measure that indicates the degree to which nodes tend to form clusters. This coefficient helps us understand how players transact with each other using rCBDC during the pilot. Our analysis reveals an average clustering coefficient of 0.242 for the rCBDC network, which is moderately low. This suggests the presence of small clusters surrounding individual nodes, meaning that players connected to a particular node tend to transact with one another but within a limited scope. This finding indicates that the use of rCBDC among users is still restricted, pointing to opportunities for enhancing business use cases related to rCBDC transfers to encourage broader usage. Another important structural property of the network is modularity, which measures the strength of the division of the network into subclusters. This measure provides insight into how players transact with one another in the rCBDC pilot. Our estimation of modularity for the rCBDC network yields a value of 0.529, indicating a moderate level of modularity. This suggests that there are multiple subclusters with dense connections among players within each cluster, but sparse relationships between nodes in different clusters. In other words, retail users frequently conduct transactions with one another within the same clusters of FSPs (referred to as on-us transactions) but seldom engage in transactions with users in different clusters (off-us transactions).