---
title: "Operations"
book: "The Desk and the Firm"
subject: quant
language: en
chapter: 13
exercises: 8
source: https://one-course.com/books/quant/16/en/chapter/13-operations
---

# Chapter 13 — Operations

On 28 May 2024 most US securities trades moved from settlement two business days after the trade to one. A firm that had had a day to find a missing settlement instruction now had the evening of the trade date: a new rule required allocations, confirmations and affirmations to be completed by the end of the trade date, and the industry’s affirmation cut-off was 9 p.m. Eastern time. The industry report that followed found that nearly 95% of transactions met that cut-off, against 73% in January 2024, and that fail rates stayed where they had been under the two-day cycle. The work had not disappeared. It had moved into fewer hours, and every operations team that did not want more fails had to be sized for the shorter window.

## 13.1 Front, middle and back office

A trade is a promise to exchange securities for cash on a later date. Between the promise and the exchange lie the steps of Book 1’s settlement chapter and of Book 15’s post-trade platform: trade capture, enrichment, confirmation, clearing, settlement instruction, settlement, and the reconciliation of every record with every counterparty’s. A trading firm divides the work into three groups.

**Definition 13.1 (Front, middle and back office).**

The *front office* makes and executes the trading decisions: traders, researchers and sales. The *middle office* controls and reports on what the front office did, independently of it: trade validation, P&L and position reporting, risk control and product control. The *back office* processes the trades to completion: confirmation, clearing, settlement, cash and collateral movements, corporate actions and reconciliation.

The division is also a control. Segregation of duties (Book 6) requires that the person who makes a trade does not book, confirm or settle it, and that the person who reports its P&L does not trade it: the [middle office](#def-fm-operations-offices) exists so that the [front office](#def-fm-operations-offices)’s numbers are checked by someone who is not paid on them. Operational risk (Book 6) lives in the hand-offs between the three.

![The three offices along a trade’s life, and the window in which a break must be fixed before settlement under a two-day and a one-day cycle. Schematic.](https://one-course.com/images/onecourse/chapters/quant-16/fm-operations/fig-4db78383e376.svg)

***Figure 13.1.** The three offices along a trade’s life, and the window in which a break must be fixed before settlement under a two-day and a one-day cycle. Schematic.*

**As of September 2026 — Settlement cycles.**

**United States**: SEC Release 34-96930 (February 2023) amended Rule 15c6-1 so that a broker-dealer may not, for most securities, agree to settle later than the first business day after the trade unless the parties expressly agree otherwise at the time of the trade; compliance began on 28 May 2024. New Rule 15c6-2 requires allocations, confirmations and affirmations to be completed as soon as technologically practicable and no later than the end of the trade date. SIFMA, ICI and DTCC reported in September 2024 that nearly 95% of transactions met the 9 p.m. ET affirmation cut-off (73% in January 2024) and that the July 2024 fail rates, 2.12% in the continuous net settlement system and 3.31% outside it, were consistent with those under T+2. **European Union**: Regulation (EU) 2025/2075 moves the settlement cycle of the Central Securities Depositories Regulation from two business days to one from 11 October 2027.

## 13.2 Reconciliation and breaks

A firm’s records of a trade, a position or a cash balance must agree with its counterparty’s, its clearing house’s, its [custodian](https://one-course.com/books/quant/16/en/chapter/4-the-asset-manager-and-the-fund#def-fm-the-asset-manager-and-the-fund-admin)’s and its own other systems. Reconciliation compares them; a difference is a reconciliation break (Book 15), and its age since it was found is the break ageing that the operations team reports every day. Breaks have causes: a wrong price or quantity at capture, a missing or stale settlement instruction, an allocation not sent, a corporate action booked differently. Most are resolved in minutes; a few take days; a break still open when the securities are due to move becomes a settlement fail.

The team that resolves them is a queue (Book 4): breaks arrive, wait for someone free and are worked. Let breaks arrive at rate $\lambda$ an hour, let each take an exponential time with mean $1/\mu$ to resolve, and let $c$ people work them first come, first served. The queue is stable if $\lambda<c\mu$. A break is late if its time in the system, wait plus work, exceeds the time $D$ left before settlement.

**Proposition 13.2 (Late breaks in a team of ccc).**

In the stationary M/M/c queue with $a=c\mu-\lambda>0$, let $C$ be the probability that an arriving break waits (Erlang’s C formula). The probability that a break is not resolved within $D$ is

$$
P(T>D)=(1-C)\,e^{-\mu D}+C\,\frac{a\,e^{-\mu D}-\mu\,e^{-aD}}{a-\mu}\qquad(a\neq\mu),
$$

and $P(T>D)\ge e^{-\mu D}$ for every $c$: if $e^{-\mu D}>1-s$, no team is large enough to resolve a share $s$ of breaks in time.

**Partial proof.** In the FCFS M/M/c queue an arriving break waits with probability $C$, and given that it waits, its wait is exponential with rate $a$ (a standard result,

**Proof.** *Admitted here.* ∎

). Its work $S$ is exponential with rate $\mu$ and independent of the wait. With no wait, $P(S>D)=e^{-\mu D}$; with a wait $W$, $P(W+S>D)$ is the tail of the sum of two independent exponentials, which is the fraction shown. Both terms are at least $e^{-\mu D}$, since $W\ge0$. ∎

```python
def p_wait(lam, mu, c):
    """Probability that an arriving break waits (PASTA: the stationary probability of c or more in the system)."""
    return float(mmc_stationary(lam, mu, c)[c:].sum())


def erlang_c(lam, mu, c):
    a = lam / mu
    rho = a / c
    top = a ** c / math.factorial(c) / (1 - rho)
    return top / (sum(a ** k / math.factorial(k) for k in range(c)) + top)


def p_late(lam, mu, c, D):
    """P(wait + work > D) for FCFS M/M/c: the wait is 0 with prob 1 - C, else exponential at rate c mu - lam."""
    C = p_wait(lam, mu, c)
    a = c * mu - lam
    tail_s = math.exp(-mu * D)
    tail_sum = math.exp(-mu * D) * (1 + mu * D) if abs(a - mu) < 1e-12 else \
        (a * math.exp(-mu * D) - mu * math.exp(-a * D)) / (a - mu)
    return (1 - C) * tail_s + C * tail_sum


def staff_for(lam, mu, D, target):
    """The smallest team that resolves a share `target` of breaks within D; none can if the work alone is too long."""
    if math.exp(-mu * D) > 1 - target:
        raise ValueError("target unreachable: the work time alone exceeds D too often")
    c = int(math.floor(lam / mu)) + 1
    while 1 - p_late(lam, mu, c, D) < target:
        c += 1
    return c
```

***Listing 13.1.** The waiting probability from firm.queues’ stationary law, the late share of [Proposition 13.2](#prop-fm-operations-late), and the smallest team for a service level. code/firm/opsmetrics/firm_opsmetrics.py*

The floor is the point of the proposition. Staff shorten the wait, not the work: once no break waits, the late share is the share of breaks whose work alone takes longer than the window. Under a two-day cycle the window is long and the floor invisible; under a one-day cycle it can be the binding constraint.

## 13.3 Tutorial: one day less

**Goal.** Size a breaks team for a two-day and a one-day cycle, and price the fails it lets through. **End state:** [Figure 13.2](#fig-fm-operations-late), [Figure 13.3](#fig-fm-operations-cost) and the staff needed for a 95% service level.

1. **The flow.** 120 breaks a day arrive over an eight-hour day ( $\lambda=15$ an hour) and take 40 minutes of work on average ( $\mu=1.5$ ); the offered load is $\lambda/\mu=10$ people (illustrative).
2. **The windows.** A break must be resolved within 10 working hours under T+2 and within 2.5 under T+1, before the evening cut-off.
3. **The fails.** Half the late breaks become fails, on trades of $5 million, failing for two days at a penalty of 1 basis point a day (the EU rate for liquid shares, an input) plus $1 500 of internal cost: $2 500 a fail. A person costs $120 000 a year.
4. **The sizing.** `fm_ops.staff95(cycle)` applies `firm.opsmetrics.staff_for` ; `fm_ops.table(cycle)` prices teams by staff and fail cost.

![The share of breaks not resolved within the settlement window, by team size, for 120 breaks a day of 40 minutes’ work. With 2.5 hours no team gets below the floor e- D=2.35\% (lower dotted line) set by the work itself; the upper dotted line is the 5% of a 95% service level. Data: firm.opsmetrics.p_late.](https://one-course.com/images/onecourse/chapters/quant-16/fm-operations/fig-19540f3e9337.svg)

***Figure 13.2.** The share of breaks not resolved within the settlement window, by team size, for 120 breaks a day of 40 minutes’ work. With 2.5 hours no team gets below the floor $e^{-\mu D}=2.35\%$ (lower dotted line) set by the work itself; the upper dotted line is the 5% of a 95% service level. Data: `firm.opsmetrics.p_late`.*

Under T+2 eleven people, the smallest stable team, resolve all but a few breaks in a million within the window: the late share is 0.0003% and the year’s fail cost is nothing. Under T+1 eleven people let 8.37% of breaks run late, 1 255 fails a year costing $3.14 million; twelve are needed for the 95% service level (3.38% late, $1.27 million) and thirteen minimise staff and fail cost together (2.69% late, $2.57 million in all, against $1.32 million under T+2). Beyond thirteen, people buy almost nothing: the late share approaches the 2.35% floor ([Figure 13.3](#fig-fm-operations-cost)). Waiting is the smaller part of the problem: with thirteen people 28.5% of breaks wait at all.

![Staff cost, fail cost and their sum under a one-day cycle. The total is lowest at thirteen people, $2.57 million a year, where the fail cost is still $1.01 million: the floor of cannot be staffed away. Data: fm_ops.table.](https://one-course.com/images/onecourse/chapters/quant-16/fm-operations/fig-01c22e82dc35.svg)

***Figure 13.3.** Staff cost, fail cost and their sum under a one-day cycle. The total is lowest at thirteen people, $2.57 million a year, where the fail cost is still $1.01 million: the floor of [Proposition 13.2](#prop-fm-operations-late) cannot be staffed away. Data: `fm_ops.table`.*

| scenario | team for 95% | cheapest team | staff cost | fail cost | total |
| --- | --- | --- | --- | --- | --- |
| T+2 | 11 | 11 | 1.32 | 0.00 | 1.32 |
| T+1 | 12 | 13 | 1.56 | 1.01 | 2.57 |
| T+1, half the breaks | 7 | 7 | 0.84 | 0.58 | 1.42 |
| T+1, 30-minute work | 9 | 10 | 1.20 | 0.30 | 1.50 |

***Table 13.1.** Staffing a breaks team under each cycle, and two ways of changing the work itself ($\$$ million a year). Data: `fm_ops.scenarios`.*

The remedy is to change the work, not the team ([Table 13.1](#tab-fm-operations-scenarios)). Halving the breaks, by automating the steps that cause them (standing settlement instructions, electronic allocation and matching on the trade date), needs seven people and costs $1.42 million; making the work faster, at 30 minutes a break, needs ten and costs $1.50 million. Either almost undoes the cost of the shorter cycle, and staffing alone cannot.

**Remark 13.3 (What the queue leaves out).**

Breaks do not arrive evenly: many appear together, when the end-of-day reconciliations run, which is the worst time under a one-day cycle. A queue with batch arrivals at the close gives larger late shares than the chapter’s even flow, and makes the case for moving reconciliation to real time (Book 15) rather than for more people in the evening.

![The ageing profile of resolved breaks: how long each took from arrival to resolution, for three team sizes. More people move breaks into the first half hour; the share older than four hours falls from 1.3% to 0.3%, but not to zero. Data: fm_ops.ageing.](https://one-course.com/images/onecourse/chapters/quant-16/fm-operations/fig-4475093322af.svg)

***Figure 13.4.** The ageing profile of resolved breaks: how long each took from arrival to resolution, for three team sizes. More people move breaks into the first half hour; the share older than four hours falls from 1.3% to 0.3%, but not to zero. Data: `fm_ops.ageing`.*

## 13.4 Corporate actions and reference events

A dividend, a split, a merger, a rights issue or a coupon changes what a position is worth or what it is. Book 1 describes the events; for operations they are breaks waiting to happen, since every system that holds the position must apply the event the same way on the same date, and the firm must decide on elective events (take the cash or the shares, tender or not) before a deadline set by the issuer’s agent. The controls are those of reconciliation, applied before the event rather than after it: a golden source of reference data, an event calendar checked against the positions that will be affected, and a named owner for every elective decision.

## 13.5 Collateral management

Every margined relationship moves collateral daily: variation margin with clearing houses (Book 1) and, under a credit support annex (Book 2), with bilateral counterparties above a threshold and in amounts no smaller than the minimum transfer amount (Book 6).

**Definition 13.4 (Collateral management, collateral dispute).**

*Collateral management* is the daily process of computing, calling, paying, receiving, valuing and substituting the collateral due under a firm’s margin agreements, and of reconciling its records with its counterparties’. A *collateral dispute* arises when the two parties’ calculations of the amount due differ by more than an agreed tolerance; the undisputed amount is usually paid while the difference is investigated.

**Example 13.5 (A disputed call).**

A CSA has a threshold of $10 million and a minimum transfer amount of $0.5 million; the firm holds $14.6 million. The firm values the exposure at $25.3 million and calls $25.3-10-14.6=\$0.7$ million. The counterparty values it at $24.4 million: its figure is a return of $0.2 million, below the minimum transfer amount, so it sees no call. The valuations differ by 3.6%, above a 2% tolerance: the call is disputed, and the cause is almost always a difference in trade population or in the marks of a few trades, which is where the reconciliation starts.

```python
@dataclass(frozen=True)
class Csa:
    threshold: float = 0.0
    mta: float = 0.0


def call(exposure, held, csa):
    """Collateral to call (positive) or return (negative) under a CSA's threshold and minimum transfer amount."""
    need = max(exposure - csa.threshold, 0.0) - held
    return need if abs(need) >= csa.mta else 0.0


def disputed(ours, theirs, tolerance):
    """A call is disputed when the two valuations differ by more than the tolerance (a fraction of the larger)."""
    return abs(ours - theirs) > tolerance * max(abs(ours), abs(theirs), 1e-12)


def kri(value, amber, red):
    """Traffic light of a key risk indicator; red below amber means lower values are worse (e.g. affirmation rate)."""
    if red < amber:
        return "red" if value <= red else ("amber" if value <= amber else "green")
    return "red" if value >= red else ("amber" if value >= amber else "green")


def report(metrics, limits):
    return [(k, v, kri(v, *limits[k])) for k, v in metrics.items()]
```

***Listing 13.2.** A collateral call under a CSA’s threshold and minimum transfer amount, the dispute test, and a key risk indicator’s traffic light. code/firm/opsmetrics/firm_opsmetrics.py*

## 13.6 Settlement discipline

**Definition 13.6 (Settlement discipline regime).**

A *settlement discipline regime* is the set of rules by which a market deters and handles settlement fails: the reporting of fails, cash penalties charged to the failing party, and buy-ins that replace securities not delivered.

**As of September 2026 — The EU settlement discipline regime.**

The Central Securities Depositories Regulation’s settlement discipline rules (Delegated Regulation (EU) 2018/1229) have applied since 1 February 2022. Cash penalties are charged per day of fail on the value of the failed instruction, at rates set in Delegated Regulation (EU) 2017/389: 1.0 basis point for liquid shares, 0.5 for illiquid shares, 0.25 for SME growth market instruments, 0.10 for sovereign debt, 0.20 for other debt, 0.15 for SME growth market debt and 0.5 for other instruments; a lack of cash is charged at the issuing central bank’s overnight credit rate, floored at zero. Regulation (EU) 2023/2845 made mandatory buy-ins a measure of last resort: they apply only if the Commission introduces them by an implementing act, after a cost-benefit analysis by ESMA, where other measures have not reduced fails.

A penalty of 1 basis point a day on a $5 million trade is $500 a day: small against a trading desk’s P&L, but charged on every fail every day, and paid to the counterparty that did not fail. The larger costs of a fail are those the tutorial prices as internal: funding the unsettled position, claims, the time spent, and the damage to the relationship. A [key risk indicator](#def-fm-operations-kri) makes them visible before they happen.

**Definition 13.7 (Key risk indicator).**

A *key risk indicator* is a metric that is monitored against thresholds because it moves before losses do: for operations, the number of breaks older than a day, the fail rate, the share of trades affirmed on the trade date, the number of disputed calls.

**Method 13.8 (A daily operations report).**

1. Choose five to ten [key risk indicators](#def-fm-operations-kri) , each with an amber and a red threshold and an owner.
2. Compute them from the systems of record, not from the teams’ own counts.
3. Report the traffic lights daily to the [middle office](#def-fm-operations-offices) and the head of operations, and weekly with trends to the [risk committee](https://one-course.com/books/quant/16/en/chapter/12-the-risk-management-function#def-fm-the-risk-management-function-cro) .
4. For every red, record the cause and the fix; for every recurring amber, change the process, not the threshold.

On the chapter’s report (`fm_ops.daily_report`), fourteen breaks older than a day, a fail rate of 2.4% and 93% of trades affirmed on the trade date are all amber against thresholds of 10, 2% and 95%; one disputed call is green.

## 13.7 Build: operations metrics

**Purpose.** The operations team’s numbers: the break queue and its ageing, the staff for a service level, fail costs, collateral calls and disputes, and [key risk indicators](#def-fm-operations-kri).

**Interface.** `firm.opsmetrics`: `mmc_stationary`, `p_wait`, `erlang_c`, `p_late`, `staff_for`, `ageing`, `fail_cost`; `Csa`, `call`, `disputed`; `kri`, `report`.

**Rules.** The queue must be stable; a service level below the work-time floor is refused; penalty rates are inputs, never constants; a [key risk indicator](#def-fm-operations-kri) whose red lies below its amber is read the other way.

**Acceptance tests.** `code/firm/opsmetrics/tests/`: the chain’s waiting probability against Erlang’s formula; the floor; staffing at the boundary; calls, disputes and traffic lights on hand numbers.

**Stretch.** Batch arrivals at the close; break types with their own work times; a collateral workflow with substitution and ageing of disputes.

Sources and further reading

- SEC Release 34-96930 (2023); 17 CFR 240.15c6-1 and 240.15c6-2.
- SIFMA, ICI and DTCC, T+1 After Action Report, press release of 12 September 2024.
- Delegated Regulations (EU) 2017/389 and 2021/70; Regulations (EU) 2023/2845 and 2025/2075.

## 13.8 Exercises

**Exercise 13.1 ★.**

Which office does each of these belong to: booking a correction to a trade’s price, approving the day’s P&L, choosing the cash option of a dividend, executing an order?

**Solution of Exercise 13.1.**

Correcting a booked price: [middle office](#def-fm-operations-offices) (validation) or [back office](#def-fm-operations-offices), never the trader; approving P&L: [middle office](#def-fm-operations-offices); the dividend election: [back office](#def-fm-operations-offices), on the [front office](#def-fm-operations-offices)’s instruction; executing an order: [front office](#def-fm-operations-offices).

**Exercise 13.2 ★.**

What does one fail cost under the tutorial’s assumptions, and how much of it is the penalty?

**Solution of Exercise 13.2.**

$5\,000\,000\times10^{-4}\times2+1\,500=\$2\,500$, of which the penalty is $1 000.

**Exercise 13.3 ★.**

With breaks taking 40 minutes on average, what is the floor on the late share for a 2.5-hour window and for a 10-hour window?

**Solution of Exercise 13.3.**

$e^{-1.5\times2.5}=2.35\%$; $e^{-15}$, about three in ten million.

**Exercise 13.4 ★★.**

Compute the probability that a break waits with 11, 12 and 14 people ($\lambda=15$, $\mu=1.5$).

**Solution of Exercise 13.4.**

Erlang’s C: 0.682, 0.449 and 0.174.

**Exercise 13.5 ★★.**

In [Example 13.5](#ex-fm-operations-call), what does each party think is due, and why is the call disputed?

**Solution of Exercise 13.5.**

The firm calls $0.7 million; the counterparty computes a return of $0.2 million, below the minimum transfer amount, so nothing. The valuations differ by 3.6% of the larger, more than the 2% tolerance.

**Exercise 13.6 ★★.**

Why does adding people beyond thirteen barely reduce fails under T+1, and what does?

**Solution of Exercise 13.6.**

With thirteen people few breaks wait; the rest of the late share is breaks whose work alone takes more than 2.5 hours, 2.35% of them, which no team size changes. Fewer breaks, or faster work on each, move the floor.

**Exercise 13.7 ★★★.**

*Coding.* Run `fm_ops.scenarios()`. Which change to the work saves more, halving the breaks or cutting the work to 30 minutes, and why does the faster work need more people?

**Solution of Exercise 13.7.**

Halving the breaks: $1.42 million a year against $1.50 million for faster work. Faster work lowers the floor ($e^{-2\times2.5}=0.67\%$) but the offered load stays at 7.5 people, so it needs ten; halving the breaks halves the load, to five, but leaves the floor at 2.35%.

**Exercise 13.8 ★★★.**

*Find the flaw.* “The move to T+1 halves our window, so we will double the operations team.”

**Solution of Exercise 13.8.**

The team’s size is set by the offered load, which the cycle does not change, and by the waiting it can remove; the window’s effect beyond that is the work-time floor, which people cannot remove. On the chapter’s numbers one or two more people (11 to 12 or 13) meet the service level; the rest of the cost must be taken out of the work.

## 13.9 Problem: One Day Less

**Problem 13.1.**

Weekend problem — one day less

A head of operations must present a plan for a firm’s move from a two-day to a one-day settlement cycle.

**Part I — The organisation.**

1. Define the front, middle and [back office](#def-fm-operations-offices) .
2. Why is the [middle office](#def-fm-operations-offices) independent of the [front office](#def-fm-operations-offices) ?
3. State the US rules on the settlement cycle and on same-day affirmation.
4. What did the industry report find after the US move?

**Part II — The queue.**

5. Model the breaks team as a queue and state its stability condition.
6. State [Proposition 13.2](#prop-fm-operations-late) and prove the floor.
7. Give the floor for the tutorial’s T+1 window.
8. Give the probability that a break waits with 11, 12 and 14 people.

**Part III — The sizing.**

9. Give the staff needed for a 95% service level under each cycle.
10. Give the late share, fails and fail cost of eleven, twelve and thirteen people under T+1.
11. Which team minimises the total cost under T+1, and what is that cost against T+2?
12. Give the cost of the two changes to the work in [Table 13.1](#tab-fm-operations-scenarios) .
13. How does the ageing profile change from eleven to fourteen people?
14. What does the model leave out, and in which direction does that bias it?

**Part IV — Discipline and collateral.**

15. Define a [settlement discipline regime](#def-fm-operations-discipline) and give the EU penalty rates for shares and sovereign debt.
16. What changed for buy-ins in 2023?
17. Define [collateral management](#def-fm-operations-collateral) and a [collateral dispute](#def-fm-operations-collateral) , and work through [Example 13.5](#ex-fm-operations-call) .
18. Define a [key risk indicator](#def-fm-operations-kri) and give four for operations.
19. State the *named result* : the staff needed to keep 95% of breaks resolved before settlement under a two-day and a one-day cycle, and the fail cost avoided.
20. In two sentences, write the plan.

**Solution of Problem 13.1.**

1. See [Definition 13.1](#def-fm-operations-offices) .
2. So that the [front office](#def-fm-operations-offices) ’s trades and P&L are checked by people not paid on them: segregation of duties.
3. Settlement no later than the first business day after the trade unless expressly agreed otherwise (since 28 May 2024); allocation, confirmation and affirmation by the end of the trade date.
4. Nearly 95% of transactions affirmed by the 9 p.m. ET cut-off (73% in January 2024), and fail rates of 2.12% and 3.31% in July 2024, consistent with T+2.
5. M/M/c with $\lambda=15$ , $\mu=1.5$ ; stable if $c\mu>\lambda$ , that is $c\ge11$ .
6. See [Proposition 13.2](#prop-fm-operations-late) ; the floor holds because the wait is never negative.
7. 2.35%.
8. 0.682, 0.449 and 0.174.
9. 11 under T+2 and 12 under T+1.
10. 8.37%, 3.38% and 2.69% late; 1 255, 508 and 403 fails; $3.14, 1.27 and 1.01 million.
11. Thirteen people, $2.57 million a year, against $1.32 million under T+2.
12. $1.42 million with half the breaks (seven people) and $1.50 million with 30-minute work (ten people).
13. Breaks resolved within 30 minutes rise from 28.6% to 50.3%; those older than four hours fall from 1.3% to 0.3%.
14. Batch arrivals at the end-of-day reconciliation; they make late shares larger than the model’s.
15. See [Definition 13.6](#def-fm-operations-discipline) ; 1.0 basis point a day for liquid shares, 0.5 for illiquid ones, 0.10 for sovereign debt.
16. Regulation (EU) 2023/2845 made mandatory buy-ins a last resort, to be introduced only by a Commission implementing act.
17. See [Definition 13.4](#def-fm-operations-collateral) and [Exercise 13.5](#exo-fm-operations-5) .
18. See [Definition 13.7](#def-fm-operations-kri) ; breaks older than a day, the fail rate, the share affirmed on the trade date, disputed calls.
19. Eleven people under T+2 and twelve under T+1 for 95%; going from eleven to thirteen people under T+1 avoids $2.13 million of fail cost a year, and $1.01 million remains.
20. Add two people to cover the waiting, and spend the rest on the work itself: standing settlement instructions, electronic allocation and matching on the trade date, and real-time reconciliation. Track breaks older than a day, fails and trade-date affirmation daily as [key risk indicators](#def-fm-operations-kri) .

## 13.10 Interview questions

**Interview question 13.1 ★ trader, developer.**

What is the difference between the [middle office](#def-fm-operations-offices) and the [back office](#def-fm-operations-offices)?

**Solution of Interview question 13.1.**

The [middle office](#def-fm-operations-offices) controls and reports on the [front office](#def-fm-operations-offices)’s activity (validation, P&L, risk); the [back office](#def-fm-operations-offices) processes trades to settlement (confirmation, clearing, settlement, cash and collateral).

*What the interviewer is looking for: control against processing.*

**Interview question 13.2 ★ developer.**

What is a reconciliation break, and what are its usual causes?

**Solution of Interview question 13.2.**

A difference between two records of the same trade, position or balance; wrong capture, missing or stale settlement instructions, missing allocations, corporate actions applied differently, timing.

*What the interviewer is looking for: records that should agree.*

**Interview question 13.3 ★★ researcher.**

Breaks arrive at 15 an hour and take 40 minutes each. How many people do you need at the least, and why is that not enough?

**Solution of Interview question 13.3.**

At least eleven, since the load is $15/1.5=10$ people; a stable queue still leaves breaks waiting, and a short window adds a floor from the work itself, so the service level needs more, or less work per break.

*What the interviewer is looking for: load, stability and the service level.*

**Interview question 13.4 ★★ risk.**

A counterparty disputes your margin call. What do you do?

**Solution of Interview question 13.4.**

Pay or accept the undisputed amount, reconcile the trade population and the marks of the largest differences, agree or escalate, and track the dispute’s age as a [key risk indicator](#def-fm-operations-kri).

*What the interviewer is looking for: undisputed amount first, then reconciliation.*

**Interview question 13.5 ★★ trader, risk.**

What did the US move to T+1 change for a trading firm’s operations?

**Solution of Interview question 13.5.**

It compressed allocation, confirmation and affirmation into the trade date and moved the breaks work into the evening; it rewarded automation and penalised manual steps.

*What the interviewer is looking for: same-day affirmation.*

**Interview question 13.6 ★★★ developer, researcher.**

Design a system that reduces settlement fails under a one-day cycle.

**Solution of Interview question 13.6.**

Capture trades with complete reference data and standing settlement instructions, match and affirm electronically on the trade date, reconcile in real time, route exceptions to a queue sized for the load, and track ageing and fails as [key risk indicators](#def-fm-operations-kri).

*What the interviewer is looking for: prevention first, then queue sizing.*