RDP 2012-05: Payment System Design and Participant Operational Disruptions 5. Results

5.1 System Design and Reaction Times

5.1.1 Actual liquidity

Our results confirm Ledrut's (2007) finding that slower participant reaction times increase the effect of operational disruptions. In other words, the longer the time taken to stop payments to the participant with the disruption (i.e. the reaction time), the higher the average value of unsettled payments between non-stricken participants (Figure 3). If participants do not react, the daily average proportion of unsettled payments is over 40 per cent of the total value of payments between non-stricken participants in systems without bilateral offset. In contrast, the daily average proportion of unsettled payments when participants react 10 minutes after the disruption ranges from 8 per cent to 12 per cent, depending on the system design.

While the results suggest that introducing a central queue, in and of itself, does not mitigate the systemic effect of participant operational disruptions – in fact, the systemic impact is slightly larger moving from the pure RTGS to the central-queue-only design – this is probably due to methodological issues.^[17]

Not surprisingly, use of the bilateral-offset algorithm reduces the systemic impact of participant operational disruptions. Compared to the central-queue-only system, the daily average proportion of unsettled payments in the bilateral-offset system decreases by between 2 and 14 percentage points, depending on the participant reaction time. The results also suggest that a quick reaction by other participants is less important in systems with bilateral offset. However, bilateral offset does not have a noticeable impact on the size of the liquidity sink (Figure 4). The caveat to these results, which is investigated further in the following sub-section, is that participants may respond to the inclusion of a bilateral-offset algorithm by decreasing the amount of liquidity they hold, which may overstate the benefit of having such an algorithm.

As noted in the introduction, the sub-limit feature slows liquidity recycling, which could increase or decrease the systemic impact of a participant operational disruption. The net effect of sub-limits depends on participants' reaction times; as long as participants stop payments to the stricken participant and lower their sub-limits within 2 hours, sub-limits mitigate the effect of a longer reaction time. For example, for a 2-hour reaction time, the daily average proportion of unsettled payments in the sub-limits system is 6 percentage points lower and the liquidity sink is $2.0 billion smaller than in the central-queue-only system. If participants only react at the end of the day, sub-limits slow the development of the liquidity sink without increasing the liquidity available to settle payments between non-stricken participants. As a result, when sub-limits are dropped at the end of the day, the remaining queued payments to the stricken participant settle and the size of the liquidity sink is much the same as it is for systems in which there are no sub-limits.

In our simulations, the systemic consequences of an operational disruption, across all the reaction times, are minimised by the combination of sub-limits and bilateral offset in the RITS replica system. Specifically, if participants react after 2 hours, the daily average proportion of unsettled payments in the RITS replica system are reduced by a further 5 percentage points, on top of the 7 percentage point reduction from introducing the bilateral-offset algorithm. Similarly, for a 10-minute reaction time, the average proportion of unsettled payments is reduced by a further 2 percentage points on top of the 2 percentage point reduction when bilateral offset is introduced.

Inter-day variation in the proportion of unsettled payments is also lower in systems with bilateral offset and sub-limits (Figure 5). On days such as day seven of our data sample, when participants committed a relatively large amount of liquidity to the system, the system design has minimal effect on the impact of the systemic disruption. However, the proportion of unsettled payments is more stable for the RITS replica system across all ten days in the sample, ranging between 6 and 14 per cent.

5.1.2 Scaled liquidity

The results from the simulations in which the liquidity available in the systems that include a bilateral-offset algorithm is reduced by 30 per cent are as follows. In this case, the daily average proportion of unsettled payments in the bilateral-offset system increase by between 2.9 percentage points and 4.5 percentage points (the lighter shaded segments in Figure 6). With a 10-minute reaction time, the decrease in liquidity negates the liquidity-saving benefit of the bilateral-offset algorithm when compared with the central-queue-only system. While the inclusion of a bilateral-offset algorithm does mitigate the systemic impact of a disruption when participants react after 2 hours, the reduction in liquidity means that the bilateral-offset algorithm, by itself, is less effective than sub-limits on their own (as long as participants do react).

Even when participants hold less liquidity, the RITS replica system remains the most effective system for minimising the systemic impact of a participant's operational disruption. A 30 per cent reduction in liquidity causes the average proportion of unsettled payments in the RITS replica system to increase by between 3 percentage points (for the 10-minute reaction time) and 5 percentage points (for the no reaction scenario).

5.1.3 Sub-limits maintained

If non-stricken participants choose to maintain their sub-limits when they react to the operational disruption, the daily average proportion of unsettled payments in the sub-limit-only system increases by between 2 and 7 percentage points (relative to the equivalent scenario in which participants chose to lower their sub-limits when they react to the operational disruption) (Figure 7). This is because liquidity trapped by the sub-limits is not recycled. Similarly, as a result of maintaining sub-limits the daily average proportion of unsettled payments in the RITS replica system also increases by a couple of percentage points. Nevertheless, for a given reaction time and level of liquidity, unsettled payments are still generally lowest in the RITS replica system.

5.2 System Design and Participant Size

It seems likely that the larger the participant (measured in terms of the value of a participant's payments and receipts) experiencing the operational disruption, the larger the systemic effects of that disruption. However, this may oversimplify the issue of size, since the timing of the disruption and the stricken participant's liquidity- and queue-management behaviour also affect the systemic impact of the disruption.

To examine this, operational disruptions at the largest 15 participants are simulated assuming a 2-hour reaction time. The results show that the impact of an operational disruption varies depending on when the disruption is assumed to have occurred. Figure 8 shows, for the RITS replica system, the relationship between the size of the stricken participant (measured as the total value of payments submitted to the system on the specific day to which it was a counterparty) and the systemic impact (measured as the value of unsettled payments) for operational disruption occurring at 9.15 am, 12.00 pm or 3.00 pm. As indicated by the line of best fit, the value of unsettled payments tends to be greatest when the disruption starts at the beginning of the day. The midday disruptions generally have a similar impact to the afternoon disruptions. This ordering broadly holds across all system designs simulated. This is not too surprising since there are more payments yet-to-be settled when a disruption occurs earlier in the day.

Our results also show that the inclusion of hybrid features reduces the systemic impact of an operational disruption for a participant of a given size. For example, as indicated by the line of best fit in Figure 9, for a given participant size, the value of unsettled payments is lower in the RITS replica system than the pure RTGS system.

Some more detailed analysis of individual results underscores the importance of participants' liquidity- and queue-management strategies. In general, when the stricken participant tends to submit payments earlier than its peers, and the operational disruption occurs later in the day, the systemic impact of a disruption is smaller.

Footnote

A central queue is expected to mitigate the systemic effect of a participant's operational disruption because the queued transactions from the stricken participant can continue to settle after the operational disruption occurs, thus reducing the size of the liquidity sink. However, the method used in this paper to select the disruption is likely to understate the benefit of a central queue. Since queued payments from a participant decrease the size of the theoretical liquidity sink, the largest theoretical liquidity sink is likely to occur when there are minimal queued payments from the stricken participant. Thus is it not unexpected that there is very little difference between our results for the central queue and pure RTGS systems. [17]