Steps of this lesson, in order.
System design deep dive: clearing house architecture
What you will learn
Your objective. By the end of this lesson, you will be able to explain system design deep dive: clearing house architecture. A clearing house processes millions of instructions every day, often 10 to 50 million in a large domestic system, with peak rates of 5,000 to 10,000 instructions per second.
Why it matters. This reduces the liquidity each bank needs by 90 percent or more in a typical cycle.
Chapter objective: Learner explains clearing, bilateral netting, multilateral netting, and the role of a clearing house, and calculates net settlement positions from a set of gross payment obligations. Learner demonstrates this skill using realistic banking examples, explaining each step clearly and using correct terminology throughout.
Key learning points
- Clearing house processes millions of instructions every
- Batch clearing and real time clearing serve
- Netting engine calculates what each participant owes
- Settlement file
- Imbalance stops the file
- Failure containment to one participant
Key terms
Settlement: The final movement of funds between banks to discharge obligations created by cleared payments. Settlement is usually made in central bank money or through correspondent accounts.
Liquidity: Cash or assets that a bank can use immediately to meet payment obligations. Banks manage liquidity by forecasting inflows and outflows, funding nostro and settlement accounts, and using central bank...
Clearing: The process of exchanging payment instructions between banks and calculating what each bank owes the others.
Clearing and settlement timeline showing System design deep dive: clearing house architecture.
Practise this lesson
Exercise
Work out the net settlement position for the clearing cycle described in this lesson.
Use the clearing rules and the netting logic in this lesson.
- What to do. List the payments in the cycle, calculate the net position for each bank, and state what settles.
Correct result. Clearing validates, sorts, matches and nets the instructions, then produces a net position per bank. Only the net position settles, which needs far less liquidity than settling every payment individually.
Why. Netting is what makes high volume payments possible. Without it, every payment would need full liquidity at the moment it is sent.
Check your understanding
Practice question
Which of the following best describes system design deep dive: clearing house architecture?
Takeaway and next step
What to remember
This design contains the failure to one participant instead of letting it cascade across the whole system.
Send a payment through the hub and watch it pass each component named in this lesson.
Open the Payment Hub Architecture Simulator →
Builds on "Why Netting Matters for Risk and Liquidity".
Transcript and full text
Show transcript
A clearing house processes millions of instructions every day, often 10 to 50 million in a large domestic system, with peak rates of 5,000 to 10,000 instructions per second. Instructions arrive through secure file transfers and real time APIs, are validated on receipt, and are stored in a high throughput intake database. The intake layer is designed to accept a burst without blocking, so a bank that sends a file of 500,000 instructions at 9 in the morning does not slow down the banks sending smaller files at the same time. Each instruction is given a unique clearing reference, a timestamp, and a status, and it is acknowledged to the sender within seconds so the sending bank can confirm receipt. Batch clearing and real time clearing serve different needs, and most clearing houses run both. Batch clearing groups instructions into cycles, typically four to six cycles per day, and processes each cycle as a single job, which is efficient for high volume low value payments such as salaries and direct debits. Real time clearing processes each instruction as it arrives, with a target of under five seconds from receipt to confirmation, which suits urgent and high value payments. The batch engine is tuned for throughput, processing up to one million instructions per cycle in under 30 minutes, while the real time engine is tuned for latency, with in memory processing and minimal database writes on the critical path. The netting engine calculates what each participant owes or is owed after offsetting payments in both directions. For each participant and each settlement cycle, it sums all outgoing payments and all incoming payments, and the difference is the net position. A bank that sent 800 million and received 750 million in a cycle has a net debit of 50 million, and only that net amount moves at settlement, not the full 1.55 billion of gross flow. This reduces the liquidity each bank needs by 90 percent or more in a typical cycle. The engine runs in parallel across participants, locks the position once calculated, and produces a net position report that each bank can verify before settlement begins. The settlement file is the output of netting, and it is the instruction that moves money at the central bank. It lists every participant, its net debit or credit, the settlement account, and the total for the cycle, and it is signed and sent to the central bank's real time gross settlement system. Settlement usually happens once or twice per day for batch cycles, and continuously for real time systems. The file is generated only after all participants have confirmed their net positions or the confirmation deadline has passed, and it is checked for balance, so total debits must equal total credits to the cent. Any imbalance stops the file and triggers an immediate investigation. When a participant fails to settle, the clearing house follows a pre agreed default procedure. First it checks whether the shortfall is a timing issue and grants a short grace period, often 30 to 60 minutes, for the bank to fund its account. If the funds do not arrive, the clearing house may use the default fund, a pool of collateral contributed by all participants, to cover the shortfall so other banks still settle on time. The defaulting bank is then suspended from new cycles until it restores its position, and its unsettled instructions are held or returned according to the scheme rules. This design contains the failure to one participant instead of letting it cascade across the whole system.
A clearing house processes millions of instructions every day, often 10 to 50 million in a large domestic system, with peak rates of 5,000 to 10,000 instructions per second. Instructions arrive through secure file transfers and real time APIs, are validated on receipt, and are stored in a high throughput intake database. The intake layer is designed to accept a burst without blocking, so a bank that sends a file of 500,000 instructions at 9 in the morning does not slow down the banks sending smaller files at the same time. Each instruction is given a unique clearing reference, a timestamp, and a status, and it is acknowledged to the sender within seconds so the sending bank can confirm receipt.
Batch clearing and real time clearing serve different needs, and most clearing houses run both. Batch clearing groups instructions into cycles, typically four to six cycles per day, and processes each cycle as a single job, which is efficient for high volume low value payments such as salaries and direct debits. Real time clearing processes each instruction as it arrives, with a target of under five seconds from receipt to confirmation, which suits urgent and high value payments. The batch engine is tuned for throughput, processing up to one million instructions per cycle in under 30 minutes, while the real time engine is tuned for latency, with in memory processing and minimal database writes on the critical path.
The netting engine calculates what each participant owes or is owed after offsetting payments in both directions. For each participant and each settlement cycle, it sums all outgoing payments and all incoming payments, and the difference is the net position. A bank that sent 800 million and received 750 million in a cycle has a net debit of 50 million, and only that net amount moves at settlement, not the full 1.55 billion of gross flow. This reduces the liquidity each bank needs by 90 percent or more in a typical cycle. The engine runs in parallel across participants, locks the position once calculated, and produces a net position report that each bank can verify before settlement begins.
The settlement file is the output of netting, and it is the instruction that moves money at the central bank. It lists every participant, its net debit or credit, the settlement account, and the total for the cycle, and it is signed and sent to the central bank's real time gross settlement system. Settlement usually happens once or twice per day for batch cycles, and continuously for real time systems. The file is generated only after all participants have confirmed their net positions or the confirmation deadline has passed, and it is checked for balance, so total debits must equal total credits to the cent. Any imbalance stops the file and triggers an immediate investigation.
When a participant fails to settle, the clearing house follows a pre agreed default procedure. First it checks whether the shortfall is a timing issue and grants a short grace period, often 30 to 60 minutes, for the bank to fund its account. If the funds do not arrive, the clearing house may use the default fund, a pool of collateral contributed by all participants, to cover the shortfall so other banks still settle on time. The defaulting bank is then suspended from new cycles until it restores its position, and its unsettled instructions are held or returned according to the scheme rules. This design contains the failure to one participant instead of letting it cascade across the whole system.
Full lesson text (all scenes)
Completing a lesson updates your progress dashboard on the homepage.
Scene timestamps are estimated by allocating the narration text proportionally across the real audio length. The audio playback position is the source of truth: pause freezes the visual, seeking restores the correct scene, replay restarts the scene animation.