RDP 2012-06: The Impact of Payment System Design on Tiering Incentives 4. Methodology

4.1 The Simulator

The Bank of Finland has developed a versatile Payment and Settlement System Simulator (BoF-PSS2) for modelling the complex interactions that occur in payment and settlement systems. Simulations can be used for analysing the implications for liquidity and risk of changes in system functionality, market structure (such as increased tiering), and settlement rules or conventions, as well as the effect of specific events. Broadly speaking, the Bank of Finland simulator mimics the functionality of RTGS systems; it requires the user to input transaction, liquidity and other data, which are then processed according to specified algorithms that simulate the workings of an actual RTGS system. The simulations generate a wide range of transaction-level and aggregated data, such as the settlement profile of payments and measures of the liquidity used by participants in the system.

4.2 Simulating Tiering

Our methodology is adapted from Lasaosa and Tudela (2008), who study the benefits and costs of tiering in CHAPS using the Bank of Finland's payment system simulator. Lasaosa and Tudela create tiering scenarios for simulation by amending raw transaction data from CHAPS. For example, to model Bank A settling indirectly through Bank B, they create an amended transaction dataset in which payments originally to or from Bank A become payments to or from Bank B. Payments originally between Bank A and Bank B are deleted from the data, as these are now settled across Bank B's books rather than submitted to the system. These ‘internalised’ payments are an immediate source of liquidity savings.

We create tiering scenarios by amending transaction data from RITS in the same way. The sample period is the month of January 2008, covering 21 business days over which 623860 individual transactions took place with a total value of around $4.04 trillion. Excluding a number of participants for which indirect settlement would be unrealistic (such as the 4 largest participants, CLS Bank and the RBA), there are 49 participants altogether that are considered candidates for tiering in this experiment. Note that only the smallest 25 of these 49 candidates were under the 0.25 per cent threshold in 2008 and therefore eligible to tier.^[10] Notwithstanding this, we model both the cumulative effect of all participants below a given size settling indirectly (the ‘cumulative scenarios’), and each of the 49 candidates individually electing to tier (the ‘individual scenarios’), resulting in 97 unique sets of transaction data representing 97 unique tiering scenarios.^[11]

As Lasaosa and Tudela are primarily interested in the effect of a decrease in tiering in a highly tiered system, they used the results of their simulations to forecast this effect. Given the high level of direct participation in RITS, such forecasting was unnecessary in the context of this paper.

It should be noted that the analysis here is necessarily limited by the fact that it ignores the potential for payments behaviour of participants to change in response to different tiering arrangements. Because the transaction and credit limit inputs to the simulator specify, inter alia, payment submission times, payment status (e.g. priority or active) and maximum liquidity accessible, none of these can be altered in response to different levels of tiering.

4.3 Tiering Order

Although there are a number of ways to select client institutions and their respective settlement banks (see Lasaosa and Tudela (2008) for examples), we allocate institutions based on the value of payments sent and received. In the cumulative scenarios, the 49 candidates are tiered from smallest to largest in order of their share of all payments. Our reasoning is that larger institutions will generally have a lower opportunity cost of collateral as their banking operations naturally result in their holding more eligible securities on their balance sheet, which in turn gives them a competitive advantage in the market for providing payment services. This approach is also consistent with the current formulation of RBA policy, whereby only participants whose share of RTGS payments comprise less than 0.25 per cent of the total value of RTGS transactions are eligible to tier for RTGS transactions.

The settlement bank for each individual tiering candidate is chosen as the institution with which the candidate conducts the largest share of its payments. This approach is likely to maximise the value of payments that are internalised, although this is not a mathematical certainty.^[12]

In practice, decisions about tiering would be interdependent. That is, each institution's choice of settlement bank could change depending on the choices of other institutions and the subsequent sizes of different tiered networks (Adams et al (2010) provide an interesting model of participant tiering choice). However, preliminary work suggested that attempting to account for this would have minimal effect; for instance, when each client institution was assigned to its largest payments partner and the choices of all smaller institutions were taken as given, the choice of settlement bank differed only in four cases.

4.4 System Design

To test the hypothesis that the liquidity-saving features of RITS decrease participants' incentives to tier, we simulate tiering under four RTGS system designs (Table 1). Details of how the bilateral-offset and sub-limit features of RITS have been incorporated in the simulations are contained in Appendix A.

Regardless of our tiering-order methodology, we expect liquidity use to fall as we increase the number of liquidity-saving mechanisms in the system. That is, we expect liquidity use to be the greatest under the pure RTGS system, followed by the central-queue-only system, then the bilateral-offset system.^[13] The RITS replica is expected to require the lowest level of liquidity.

4.5 Liquidity

The liquidity available to participants is modelled in the simulations using limits on credit extended by the system operator to each direct participant. Each participant begins each simulated day with an account balance of zero and, as payments settle, is able to accrue a negative account balance up to that participant's credit limit. Credit limits are set exogenously and may vary throughout the day. In general, the credit limit profile for each participant on each simulated day is modelled on the actual liquidity that was available to that participant at each point in time on the corresponding day of our sample period. This actual liquidity is measured as the sum of the participant's opening settlement account balance and the value of intraday repos it had outstanding at each point in time during the day.^[14]

An exception is made for our simulation of the pure RTGS system. To prevent payments that do not settle immediately from being rejected and remaining unsettled at the end of the day, all participants are assumed to have access to unlimited liquidity. In addition, to ensure that all payments settle in our simulations, we give all participants unlimited access to liquidity under all system designs at the end of each day.^[15]

In the tiering scenarios, we reason that a settlement bank does not have access to collateral on its clients' balance sheets and it will not necessarily commit more of its own collateral to access additional liquidity. Alternatively, we could have assumed that the settlement bank increases the liquidity it accesses (e.g. by the value of the liquidity accessed by its clients when they were direct participants). Indeed, preliminary simulations were run where the credit limits of the settlement bank and its clients were summed, but this resulted in quite substantial and unrealistic increases in liquidity usage under tiering. Therefore, our preference has been to remain with fixed, non-additive access to liquidity.

We measure system liquidity usage as the sum of individual participants' peak intraday liquidity requirements. For an individual participant, this peak intraday liquidity requirement is equal to the absolute value of the participant's minimum account balance. While this liquidity may only have been used for a very brief period during the day, this measure is consistent with the view that the main driver of the cost of liquidity is the maximum value of collateral used, rather than the length of time during the day that the securities are used.

Footnotes

The 28 direct participants eligible for tiering over the whole of 2008 include 1 participant not considered for tiering in our simulations (as it only settles low-value payments on a deferred net basis) and 2 participants that joined RITS after January 2008. [10]

In the cumulative scenarios in which the fifth-largest institution is tiered, the 49 smallest institutions are all settling indirectly via the 4 largest participants. As only 1 institution is tiered in the first cumulative scenario, it is identical to the first individual scenario. [11]

A further scenario, based on the order of the share of total volumes, was not materially different to the one based on values, and so it was not pursued further. [12]

For the purposes of this paper the term ‘pure’ RTGS system is used to refer to an RTGS system that does not have a central queue. [13]

In our simulations, the RBA, CLS Bank and the settlement accounts of the equity and futures clearing and settlement systems are provided with unlimited credit in all system designs. [14]

In the absence of this we find that the simulations result in a small proportion (less than 1 per cent) of payments remaining unsettled at the end of most days. This failure to settle all payments occurs because settlement times differ across the different RTGS systems, while available liquidity is set exogenously. [15]