A GLANCE AT LARGE VALUE PAYMENT SYSTEM LITERATURE

Endogeneity, complex interactions, agent multi­plicity, and heterogeneity have not always been a matter of concern in the analysis of large value payment systems.

In fact, the literature on large value payments systems has naturally changed its focus in accor­dance with the innovations undertaken by central banks aimed at promoting the usage of central bank money as a safe settlement asset and limit­ing the size of systemic risk stemming from the settlement of interbank obligations.

Until the eighties, Deferred Netting Settle­ment (DNS) systems represented the standard for large value payment systems in all but one the advanced financial systems. In these systems, usually operating on a multilateral netting fashion, banks grant each other intraday implicit credit by exchanging payment instructions throughout the operational day and paying (or receiving) at the end of the day an amount of central bank money equal to the net debit (credit) stemming from the payments due to, and to be received from, all the other banks participating in the system.

In a DNS system, the liquidity management problem of a bank treasurer is almost simple: it has to borrow or lend in the money market the amount of deposits needed to meet its desired end-of-day level of balances, taking into account the daily cumulative net inflows stemming from its payment obligations. Angelini (2000) builds a model calibrated on the Italian interbank market and on the SIPS, the netting large-value operating payment system operating at the time the paper was written, where the optimal the intraday behavior on the money market depends on the information available from two key stochastic variables: the desired end-of-day clearing balance and a short­term interest rate, typically the overnight rate.

The model by Angelini confirms how the intraday liquidity management is a minor issue in a DNS system: banks rely on the screen-based information on incoming payments provided by the operator of the netting system only to have an informed guess on their end-of-day balances.

On the other hand, the issue of endogeneity is not problematic neither when investigating the systemic risk typically associated to large value DNS systems (Humphrey, 1996). In fact, in such systems, the failure of one participant to meet its net obligations at the end of the day is routinely resolved through a procedure called unwinding. In an unwinding, all the transactions of the defaulter are excluded from the calculus of the net balances, which are recalculated taking into account only the transactions of the non-defaulting bank. The revised balances may differ greatly from those expected by banks in absence of defaults and may cause other banks failing to meet their new payment obligations. Should this be the case, the unwinding procedure is reiterated until all the survivor banks are able to pay their net balances. Multiple defaults may emerge before this occurs: the domino effect triggered by the first default materializes systemic risk into systemic crisis.

Analyzing systemic risk in netting systems does not require making behavioral assumptions and therefore, is an easier task than dealing with Real Time Gross Settlement (RTGS) systems: analytical solutions are easy to derive, and relatively simple models are able to give insight on the impact of a participant default by simply removing from the system all the payment instructions involving that agent as sender or receiver. Subsequent defaults take places for those participants whose revised multilateral net debit position exceeds or equals their available capital³.

When in the late eighties’ majority of central banks started to promote the migration of the large value payment system from DNS systems towards the alternative settlement modality, the RTGS system, with a view to reducing the high level of systemic risk entailed in the DNS systems, the task of policy makers and academics interested in the large value payment system has become more and more complex.

Systems operating in an RTGS fashion settle each payment continuously in real-time, individu­ally, without netting debits against credits (i.e.

Compared to DNS systems, the RTGS mode eliminates virtually the credit risk by providing payment finality throughout the course of the day, but makes the system liquidity greedy, as banks need to maintain sufficient liquidity throughout the operational day in order to settle immediately their payment obligations.

This statement holds particularly true if we take in mind that the intraday liquidity needs, which arise from the time mismatching between incoming and outgoing payments, are usually a multiple of the end-of-day liquidity needs, which in turn are typically limited in size, as banks tend to smooth their end-of-day balances throughout a reserve maintenance period.

To meet their intraday settlement obligations, participants in an RTGS system have at their dis­posal four possible alternative sources of funds: (a) cash balances maintained on account with the central bank, (b) credit facilities granted by the central bank, (c) incoming transfers from other banks, and (d) deposits borrowed from other banks through the money markets.

While cash balances are usually related to re­serve requirements that are imposed for monetary policy purposes and do not entail any additional cost directly addressable to the functioning of the payment systems, the other three sources entail different costs which banks try to minimize by choosing the best mix between the cost of settling customer obligations and their own proprietary operations.

Daylight credit lines by central banks are granted on either an uncollateralized or a fully collateralized basis, being the cost an explicit fee or the opportunity cost of conferring collateral, respectively.

Finally, the cost of borrowing from other banks through the interbank money markets is usually the overnight rate traded in the unsecured market, as there are no intraday money markets in place.

The reliance of RTGS systems on money mar­ket as a funding source creates a close relationship between the monetary policy key variables and the liquidity management carried out by banks in an RTGS environment, as also confirmed by a number of empirical contributions pointing out how the behavior of commercial banks in the money market depends, inter alia, on the intra­day liquidity needs they face to settle payment in RTGS system.

Among these contributions, Ashcraft and Duffie (2007) find a significant relationship be­tween the probability to act in the overnight money market and the liquidity balances held by banks, showing that that participants whose balances at a certain point in time of a particular business day are higher than those held at the same time in the past are more likely to be a lender to their counterparties; conversely, lower than average balances increases the individual probability of being a borrower.

Klee (2007) analyses the effect of operational outages in the Fedwire system on the deviation of the federal fund rate (i.e. the overnight rate negotiated in the market by banks) from the policy rate, the FOMC target rate; the author finds that operational outages involving large participants in terms of volume submitted to the system and the time of the operational day the outages occur are positively related to deviations from the policy rate. She motivated these findings arguing that an operational outage is a source of uncertainty about end-of-day balances: banks not receiving funds from the counterparty hit by operational strains fear they will not achieve their desired levels of overnight balances at the end of the day, forcing them to buy and lend funds in the market.

This short-perhaps-not exhaustive selection of empirical evidence confirms the central role of the liquidity management analyses for both the smooth functioning of the payment systems-to which the central bank are interested in their oversight role-and the volatility and level of the interest rate negotiated on the money market, which represents the variable directly targeted by central bank when conducting the monetary policy.

Against this background, it is apparent that, with the advent of RTGS era, policy makers and academics start to face a more complex problem, on the one hand they need to study how the system design and the cost of the four liquidity sources affect the behavior of participants and system per­formance; on the other hand, they need to quantify the right level of liquidity a payment system must maintain to allow its smooth functioning and an avoid adverse affect on the short-term interest rate. To tackle these issues, researchers have been relying on using two different approaches: micro­founded models and simulations.

Among microfounded models, Angelini (1998), starting from the literature on precaution­ary demand for reserves, builds a game theoreti­cal model and analyses the incentive mechanism operating in a RTGS environment and its impact on the behavior of key monetary policy variables, such as overnight interest rate and reserves. He shows that if daylight credit is costly, then banks will optimally postpone their outgoing payments until the perceived marginal cost of delaying equals the marginal cost of daylight liquidity. While such a strategy reduces the expected cost of processing payments for the payer bank, it also tends to increase it for the payee banks. As each bank has an incentive to postpone its outgoing payments, the liquidity redistribution mechanism may be hampered and this can potentially cause a system gridlock. In order to avoid ex-ante such an outcome, daylight credit at cheap conditions has to be granted.

Bech and Garrat (2002) confirm Angelini findings relying on a Bayesian game, in which each bank has private knowledge about its own payment requests.

These theoretical findings are empirically confirmed by McAndrews and Rajan (2000), McAndrews and Potter (2002), and Armantier et al. (2008): using data from the US RTGS system FEDWIRE, they all show that the pattern of pay­ments made over that (RTGS) system peaks in the late afternoon as participants condition their outgoing payments to the incoming funds arriving from the other participants.

The microfounded models, a la Angelini, have been proven effective in giving insights on the policies a central bank should follow when granting intraday credit and may suggest how to design the system in order avoid that banks pursue opportunistic behaviors, hampering the liquidity recycling in the system (i.e. setting throughput guidelines, which require participants to submit a certain share of their outgoing payment before pre-specified cut-off times).

However, such theoretical models may not offer an adequate guidance in other real-life prob­lems, e.g. in quantifying the amount of liquidity participants should have at their disposals to sub­mit their outgoing payments without any undue delay or the additional liquidity needs faced by system participants in case a disruption occurs at a participant level as the simplifying assumptions currently imposed⁴ make these models struggle to cope with the complexity stemming from the high number of heterogeneous players (i.e. the number of participants), the huge quantity of payment transactions processed, the plethora of parameters describing the participants’ behavior which makes analytical solutions impracticable when dealing with real data set.

In fact, even structural models describing op­timal behavior can-at the state of the art-hardly be calibrated and adapted to such environments. Moreover, as time goes by, the RTGS systems have evolved from the original plain vanilla RTGS system (McAndrews & Trundles, 2001), having been enriched more and more with liquidity saving features (queuing mechanisms, optimiza­tions’ algorithms, bilateral limits, reservations’ facilities), which add degrees of complexity to the system, influence the participants behavior and make the system more and more difficult to be replicated with stylized neoclassical or game theoretical set-ups.

To overcome these drawbacks, researchers and policymakers have turned their attention to simulation tools able to mimic existing or fictitious payment arrangements and, if fed with real data set, replicate closely the settlement process of an actual business day in the life of a payment system. By modifying the input dataset-e.g. by remov­ing the outgoing transactions from a participant, which is simulated to fail-or the system design, it is possible to build counterfactual scenarios and evaluate policy decisions.

The most popular tool is by far the Bank of Finland Payment System Simulator Bof-PSS2⁵, which can to replicate a wide variety of large value payment systems, functioning either alone or in connections with other (ancillary) systems.

The availability of the Bof-PSS2-which was adopted by 80 users among central banks, academ­ics, and clearing organizations as of August, 2010 (Hellqvist, 2010)-has given rise to a flourishing of studies aimed, inter alia, at quantifying the effect of different liquidity levels on the system performance (Leinonen & Soramaki, 1999; Heji- mans, 2009) or of operational outages occurring at a participants level (Bedford, 2004; Glaser & Haene, 2007).

It is apparent that the main drawback of this strand of literature based on the BoF-PSS2 is that every action the participants undertake in a simula­tion exercise is exogenously determined ex-ante, before the start of the simulated operational day⁶. This means that, when an event, like a default, is simulated to occur, the artificial participants do not adjust their behavior according to the new scenario: in other words, these studies fall gener­ally under the Lucas critique argument⁷.

3.