T2S workshop on Settlement Optimisation Objectives

Transcription

1 T2S workshop on Settlement Optimisation Objectives Part 3: PRESENTATION ON SETTLEMENT OPTIMISATION OBJECTIVES Paris Sept 16th 2011

2 AGENDA part 3 1. Optimisation features of T2S 1. What is optimisation? 2. T2S optimisation features overview 3. Optimisation strategies and framework 4. Optimisation algorithms 2. Optimisation and evaluation of settlement efficiency 1. Definitions 2. Settlement optimisation objectives 3. Metric 4. Impacting parameters 2

3 AGENDA part 3 1. Optimisation features of T2S 1. What is optimisation? 2. T2S optimisation features overview 3. Optimisation strategies and framework 4. Optimisation algorithms 2. Optimisation and evaluation of settlement efficiency 1. Definitions and Settlement optimisation objectives 2. Metrics 3. Impacting parameters 3

4 On optimisation What does optimisation mean? Formally, finding a maximum/minimum of a function while respecting given constraints. In our context : Function: some quantity involving volumes, values, transaction priority, settlement delay etc Constraints : Functional: respect of business rules (priority, partial settlement thresholds, transactions links, cash balances/cmb/positions 0..) Production: so much CPU/time/memory is allotted to optimisation Best effort optimisation Function: no function is per se better than another but an optimum for a given function is not an optimum for another high sensitivity of the settlement to the function Functional constraints are dictated by the URDs Production constraints are described in the settlement day document (UDFS). Highly sensitive to data: T2S can optimise but cannot make up for participants lacking the resources necessary to the settlement of their transaction. 4

5 Importance of the objective function Simplistic optimisation problem: 100 of available resource (cash, security..) 3 transactions debiting that resource 1. One debits One debits One debits 45 Available resource 100 Debiting 30 Debiting 45 Available resource 20 Available resource 25 Debiting 80 Debiting 80 Debiting 30 Debiting 45 Which solution is the best? 5

6 AGENDA part 3 1. Optimisation features of T2S 1. What is optimisation? 2. T2S optimisation features overview 3. Optimisation strategies and framework 4. Optimisation algorithms 2. Optimisation and evaluation of settlement efficiency 1. Definitions and Settlement optimisation objectives 2. Metrics 3. Impacting parameters 6

7 Optimisation features Optimisation procedures rely on: technical netting; auto-collateralisation; partial settlement. Technical netting is used to assess the settlability of a given set of transactions and performs lack detection, It is a straightforward procedure ( sum/group by ) but nonetheless heavily used in the optimisation. 7

8 Auto-collateralisation feature If a transaction cannot settle due to lacks of cash on T2S dedicated cash accounts or insufficient external guarantee headroom on credit memorandum balances, T2S offers central banks and settlement banks with the capacity to provide intraday credit to credit consumers via an automated process secured with eligible collateral. In addition, the auto-collateralisation process allows automatic release of collateral and automatic substitution to cover lack of securities. The auto-collateralisation is an important feature of T2S optimisation processes as it allows participants to reduce their initial cash balance and limits the risk of failed settlement due to lack of cash. 8

9 Partial settlement Unless explicitly specified at instruction level, all transactions are eligible to partial settlement. Resort to partial settlement is only allowed in: Certain periods; If the resource in lack is a security. Allowing transactions to partial settlement greatly helps the settlement of all transactions in a partial settlement window. 9

10 AGENDA part 3 1. Optimisation features of T2S 1. What is optimisation? 2. T2S optimisation features overview 3. Optimisation strategies and framework 4. Optimisation algorithms 2. Optimisation and evaluation of settlement efficiency 1. Definitions and Settlement optimisation objectives 2. Metrics 3. Impacting parameters 10

11 Optimisation framework : overview An event happens in T2S. If this event is related to an optimisation series, it triggers the launch of an optimisation series. Event Series Feature 1 Feature Feature n Algorithm 1 Algorithm Algorithm n 11

12 Optimisation framework : cheat sheet Event Series Feature 1 Feature Feature n Algorithm 1 Algorithm Algorithm n Event: Foreseen: Settled transaction Unsettled transaction Beginning/end of sequence/cycle Possible: Time event (5 mins?) Others Series Succession of algorithms sharing common features Features: Resort to auto-collateralisation Resort to partial settlement Algorithms Build up family Deselect family 12

13 Overview: real-time period Real-time settlement and optimisation Instruction Matching Instructions are matched (LCMM) and transformed in transactions (SPS) as they arrive and are transferred over to the settlement engine. Previously unsettled transactions SPS New transaction Previously unsettled transactions The settlement engine attempts to settle them on a unitary basis. SLA commitment is to minimise the delay of first settlement attempt VPB Recycling and optimisation procedures run in parallel with the settlement and combine previously unsettled transactions so as to make them jointly settlable. Unsettled Settled Optimisation Recycling 13

14 Overview: night-time period Night-time settlement and optimisation Instruction Matching SPS Scheduling Transactions selection New and recycled transactions Optimisation Night-time settlement Start Sequence n Instructions are matched (LCMM) and transformed in transactions (SPS) as they arrive. The night-time is organised into cycles, the latter being themselves split in sequences. Each sequence only deals with a pre-determined type of transactions (Liquidity transfers, Corporate actions, DVP trading etc). There is no continuous settlement : at the beginning of each sequence, the transactions corresponding to the sequence are selected for settlement. SLA commitment is to process the whole volume During this period, optimisation procedures aims is to find the largest jointly settlable subset of the selected transactions. Sequence 0 : Liquidity transfers Sequence 1 : Corporate actions Sequence.. :... Sequence 4 : Trading transactions 14

15 Overview: Real-time vs. night-time Real-time optimisation Triggered by the settlement engine Continuously runs during the Real-time In parallel with settlement Priority is given to real-time settlement Light algorithms used Very many settlement/s Very few (<10) transactions jointly settled Night-time optimisation Triggered by the scheduling module Batch processing Manages the whole settlement Heavy algorithms used Very few settlement/s Very many(>100?) transactions jointly settled 15

16 AGENDA part 3 1. Optimisation features of T2S 1. What is optimisation? 2. T2S optimisation features overview 3. Optimisation strategies and framework 4. Optimisation algorithms 2. Optimisation and evaluation of settlement efficiency 1. Definitions and Settlement optimisation objectives 2. Metrics 3. Impacting parameters 16

17 Two main families of algorithms Build Up: Real time : Start with one unsettled transaction; Pick other unsettled transactions (typically from 1 to 3) so as to build a small settlable set of transactions. Night-time: Gross settlement Keep iterating over the stock of pending transactions and try adding them on a unitary basis while maintaining the selection settlable Deselect: Start with a set of many unsettled transactions. How this set is built is irrelevant to the algorithm; Remove from this set as few transactions as possible so as to make the set settlable, i.e. try to find a large settlable subset. 17

18 Build-up algorithms Build-up based algorithms start with one unsettled transaction. They look for patterns such as: Simple Circle; Back to back. 5 sec1 5 sec1 5 sec1 5 sec1 A B + B C = A B C 500 A has enough securities but B has 0 cash. Transaction cannot settle 600 C has enough cash but B doesn t hold the securities Transaction cannot settle Those two transactions are nonetheless jointly settlable If those patterns are not found, a generic chain is built. P2 180 sec P1 20 sec sec sec P3 P sec P5 18

19 Deselects algorithms: foundations Deselect algorithms operate on a sub-selection of the pending stock, i.e. not yet submitted or already unsettled transactions. A netting is computed and the impact on both cash balances and securities positions is determined. 1. If this shows a resource is debited more than is available; the resource is said to be in lack. 2. All lacks are identified and prioritised. The algorithms starts with the first prioritised lack and removes transactions from the current selection until the netting shows the corresponding resource is not in lack anymore. The algorithms stop when there are no more lacks. All transactions in the current selection are then jointly settled. 19

20 Deselect algorithm: lack resolution Currently two strategies have been identified. Others are under study. Greedy : Remove all transactions debiting the resource in lack; Add the biggest while the resource is not in lack. One by One : While the resource is in lack ; if removing one transaction can solve the lack, than the smallest of those is removed. If not, remove the biggest transaction. Example Greedy One By One Available resource 100 Debiting 30 Debiting 45 Available resource 20 Available resource 25 Debiting 80 Debiting 80 Debiting 30 Debiting 45 Value bias Volume bias 20

21 AGENDA part 3 1. Optimisation features of T2S 1. What is optimisation? 2. T2S optimisation features overview 3. Optimisation strategies and framework 4. Optimisation algorithms 2. Optimisation and evaluation of settlement efficiency 1. Definitions and Settlement optimisation objectives 2. Metrics 3. Impacting parameters 21

22 Definitions Effectiveness: Effectiveness is defined as the ratio between the actual output and the objective that the system is designed to reach. For T2S, effectiveness may be defined as the ratio between the number/value of settled transactions and the total number/value of transactions to be settled. Efficiency: Efficiency is the measure of how well the resources (cash, security but also time) are utilised to reach a predetermined objective. For T2S, efficiency can then be defined as the ratio between a predetermined effectiveness indicator and the time taken to achieve it. 22

23 Settlement optimisation objectives URD provides rules to be translated into settlement optimisation objectives (T2S , T2S , T2S , T2S and T2S ): T2S settlement optimisation process shall favour the settlement of transactions with a higher level of priority. When several transactions with the same level of priority compete for settlement, T2S settlement optimisation process shall favour the settlement of transactions with the oldest intended settlement date (ISD). When several pending transactions with the same level of priority and the same intended settlement date compete for settlement, T2S settlement optimisation process shall favour the settlement of transactions in a way that maximises the volume and value of settlement (in an optimum balance). T2S settlement optimisation process shall use resources for oldest transactions first in order to reduce the time during which a transaction remains unsettled beyond the intended settlement date. During the night-time, T2S settlement optimisation process shall minimise the number and value of auto-collateralisation operations. T2S settlement optimisation process shall minimise the number of transactions submitted to partial settlement. 23

24 AGENDA part 3 1. Optimisation features of T2S 1. What is optimisation? 2. T2S optimisation features overview 3. Optimisation strategies and framework 4. Optimisation algorithms 2. Optimisation and evaluation of settlement efficiency 1. Definitions and Settlement optimisation objectives 2. Metrics 3. Impacting parameters 24

25 Terminology For a transaction t, we call : p as t s priority a as t s age defined by : Age(t)=Min(CurrentBusinessDate - ISD(t),3) c as t s currency As shortcuts : transactions(p,a) refers to the set of transactions of age a and priority p transactions(p,a,c) refers to the set of transactions of age a, priority p and currency c 25

26 Effectiveness ratio s building blocks: volume and value For each relevant pair of priority and age (p,a), we compute the Volume Ratio R Vol (p,a) as described above For each relevant tuple of priority, age and currency (p,a,c), we compute the Value Ratio R Val (p,a,c) as described above so as to compute the Value Ratio R Val (p,a) below 26

27 Effectiveness ratio s building blocks : BRI The Balanced Ratio Indicator BRI(p,a) is a synthetic indicator combining both the Volume and the Value ratios through a parameter in ]0;1[ : BRI (p,a) = R Vol (p,a) + (1- ) R Val (p,a) The higher, the more volume is favoured over value 27

28 Effectiveness ratio and objective function For each pair (p,a), a weight is computed. The effectiveness indicator BRI is defined as a weighted average of all BRI (p,a) : BRI can be used as the objective function that we would seek a maximisation. At first, could be set at 0.5 and further fine-tuned. So as to respect the URDs, we propose to use the following weighting function : A G E Priority Reserved Top High Normal 0 0, E-14 1E-22 1E ,0001 1E-12 1E-20 1E ,01 1E-10 1E-18 1E , E-16 1E-24 28

29 Example 1/2 SA1 SA2 SA3 DCA1 DCA2 DCA3 P11 P12 P13 P21 P22 P23 P31 P32 P33 Trx Trx Trx Trx Trx Trx Trx Trx Trx Trx Trx Trx Available The settlement attempt of all pending transactions is not a valid solution because there are not enough resources to settle all transactions. In this case, the provision check process shows four lacks : PCK

30 Example 2/2 SA1 SA2 SA3 DCA1 DCA2 DCA3 P11 P12 P13 P21 P22 P23 P31 P32 P33 Trx Trx Trx Trx Trx Available PCK Solution 1 indicators : R Vol = 42% R Val =57% ; BRI 5% = 56 % BRI 50% = 49 % BRI 95% = 43 % SA1 SA2 SA3 DCA1 DCA2 DCA3 P11 P12 P13 P21 P22 P23 P31 P32 P33 Trx Trx Trx Trx Available PCK Solution 2 indicators : R Vol = 33% R Val =60% ; BRI 5% = 59 % BRI 50% = 47 % BRI 95% = 34 % 30

31 Efficiency ratio The objective function has just been defined as BRI. The optimisation of this function is computationally expensive. It is computationally cheap to optimise a relaxed problem where we allow all settlement transactions to be partialisable and ignore the various business rules driving transaction partialisation. Of course, the relaxed solution is highly unlikely to be valid from a business standpoint but it is a bound for all achievable solutions. 31

32 Integrality gap and convergence During optimisation, one can measure the distance ( integrality gap ) between the current solution and the relaxed maximum and decide whether to continue optimising. 32

33 Delay indicator Delay indicator is used to assess real-time settlement period s effectiveness. For a single transaction, the delay is defined as : Delay(t) = SettlementDate(t)-max(StartRTSDate,FirstSettlementDate(t)) The Delay Indicator (DI) is defined as : Here again, a weighted average of DI(p,a) could be used to present a synthetic indicator easily computable to monitor in real-time the effectiveness of recycling and optimisation in the real-time settlement period. 33

34 AGENDA part 3 1. Optimisation features of T2S 1. What is optimisation? 2. T2S optimisation features overview 3. Optimisation strategies and framework 4. Optimisation algorithms 2. Optimisation and evaluation of settlement efficiency 1. Definitions and Settlement optimisation objectives 2. Metrics 3. Impacting parameters 34

35 What are the parameters to which optimisation is sensitive? So far, three parameters have been identified : 1. Cash account structure and concentration 2. Resource scarcity 3. Volumes 35

36 Cash account structure and concentration The problem is therefore more coupled in structure 2, which has the following consequences : For the day-time settlement : cash account DCA 1 is more frequently accessed in structure 2, leading to possibly more locking issues in case of simultaneous settlement. For the night-time settlement: more netting opportunities and resource management simplification 36

37 Resource scarcity The 2 graphs above illustrate experimental R Vol and R Val dependency to resource scarcity. The less resource available, the harder the optimisation, and this relationship is not linear ; i.e. 80% of resource does not yield R Vol and R Val in the range of 80%. Cash is the critical resource Independently of optimisation s effectiveness, it is always the participant s resources in T2S that limits the overall optimisation results. 37

38 Volumes All things being equal, in night-time settlement, the number of transactions unique influence is the time to process the data. As always, there is a trade-off to be made between optimisation effectiveness of the optimisation efficiency. We have several algorithms, each one having a different relationship to volumes (and opposite relation to effectiveness) and we can switch from one to another if need be. As we have limited time to process each sequence during night-time, we will need to reduce the quality of the optimisation when a peak volume is announced. 38

39 Conclusion Retained Indicators : Night-time period : Effectiveness indicator : Balanced Ratio Indicator Efficiency indicator : Integrality Gap Real-time period : Effectiveness indicator : Delay Indicator Best effort based optimisation: all things being equal, those indicators are driven by the presence (or absence) of resources. Volumes drive efficiency not effectiveness Optimisation depends on the settlement day 39

40 Thank you! 40