---
title: "Data Strategy and Costs"
book: "The Desk and the Firm"
subject: quant
language: en
chapter: 22
exercises: 8
source: https://one-course.com/books/quant/16/en/chapter/22-data-strategy-and-costs
---

# Chapter 22 — Data Strategy and Costs

In February 2024 the UK’s Financial Conduct Authority published the findings of its study of the wholesale data market. It found no evidence that firms could not get the data they needed, and evidence of market power in all three markets it studied: benchmarks, credit ratings data and the market data vendors that redistribute trading data. Users, it concluded, may be paying higher prices than competition would allow; licensing was complex, many users had to hold licences both from a trading venue and from the vendor that delivered its data, and the complexity itself cost money, down to the compliance teams firms employ to follow their licences. A trading firm’s data bill is set less by how much data it uses than by how it is counted.

## 22.1 What a firm pays for data

**Definition 22.1 (Data budget).**

A *data budget* is a firm’s planned annual spending on data, by product and use: exchange market data (display, non-display and feeds), vendor terminals and platforms, reference and corporate-action data, news, and alternative datasets, with the licences, counts and allocations behind each line.

The chapter’s firm ([Table 22.1](#tab-fm-data-strategy-and-costs-inventory)) spends $2.07 million a year on data: $0.56 million on exchange data, $0.75 million on thirty vendor terminals, $0.46 million on reference data and news, and $0.30 million on an alternative dataset. Terminals are the largest line, 36% of the total, and the one most directly driven by headcount. The fee levels are illustrative inputs; their structure is what every data contract shares: something charged per user, per device, per use, or as a flat fee, with a cap or an enterprise alternative.

| product | charged | usage | $ thousand a year |
| --- | --- | --- | --- |
| exchange real-time display | 0.6 per user | 40 users | 24 |
| exchange non-display (automated) | flat | – | 180 |
| direct feeds | flat, six feeds | – | 360 |
| vendor terminals | 25 per user | 30 users | 750 |
| reference data | 8 per user or 400 | 60 users | 400 |
| news analytics | 15 per device | 4 devices | 60 |
| alternative dataset | flat | – | 300 |
| total |  |  | 2 074 |

***Table 22.1.** The chapter’s data inventory, each product at its cheaper licence (illustrative fee levels). Data: `fm_data`.*

![The data budget by category, $2.07 million a year: terminals, the headcount-driven line, are the largest. Data: fm_data.the_budget.](https://one-course.com/images/onecourse/chapters/quant-16/fm-data-strategy-and-costs/fig-5a7e2599d13d.svg)

***Figure 22.1.** The [data budget](#def-fm-data-strategy-and-costs-budget) by category, $2.07 million a year: terminals, the headcount-driven line, are the largest. Data: `fm_data.the_budget`.*

## 22.2 Licences and their units of count

Book 15 defined the unit of count, the entitlement system and the usage report: the machinery by which a firm knows who and what consumes each dataset. The licences they serve come in a few shapes.

**Definition 22.2 (Enterprise, derived-data and redistribution licences).**

An *enterprise licence* permits unlimited use of a product within the firm (or a defined part of it) for a fixed fee, replacing per-user or per-device counts. A *derived-data licence* permits the firm to create and use, and sometimes distribute, data calculated from the licensed data, such as prices, indices or signals. A *redistribution licence* permits the firm to pass the data, or data derived from it, to third parties such as clients, under conditions and for fees.

The enterprise decision is a break-even. The reference data costs $8 000 per user a year or $400 000 for the enterprise: the [enterprise licence](#def-fm-data-strategy-and-costs-licences) wins from 50 users ([Figure 22.2](#fig-fm-data-strategy-and-costs-enterprise)). At the firm’s sixty users it saves $80 000 a year, and it removes the counting, and with it the audit exposure, for that product. Below fifty users per-user fees are cheaper, unless headcount is expected to grow past the break-even within the contract.

![Per-user fees against an enterprise licence for the reference data: the lines cross at 50 users; the firm has 60. Data: fm_data.reference_curve.](https://one-course.com/images/onecourse/chapters/quant-16/fm-data-strategy-and-costs/fig-2d4b3e569b90.svg)

***Figure 22.2.** Per-user fees against an [enterprise licence](#def-fm-data-strategy-and-costs-licences) for the reference data: the lines cross at 50 users; the firm has 60. Data: `fm_data.reference_curve`.*

```python
def annual_cost(p, u, enterprise=False):
    if enterprise:
        if p.enterprise is None:
            raise ValueError(f"{p.name} has no enterprise licence")
        return p.enterprise + p.flat
    return p.per_user * u.users + p.per_device * u.devices + p.flat


def budget(products, usage):
    """Each product at its cheaper licence (per unit or enterprise); totals by category."""
    out = {}
    for p in products:
        c = annual_cost(p, usage[p.name])
        if p.enterprise is not None:
            c = min(c, annual_cost(p, usage[p.name], True))
        out[p.category] = out.get(p.category, 0.0) + c
    return out


def enterprise_breakeven(p):
    """The smallest number of users at which the enterprise licence costs no more than per-user fees."""
    return math.ceil(p.enterprise / p.per_user)


def audit_exposure(annual_fees, under_share, years, monthly_rate):
    """Back-billing when a share `under_share` of true usage went unreported for `years`: each month's unbilled fees
    (the reported fees scaled up to true usage) with interest from that month to the audit."""
    months = int(round(12 * years))
    unbilled = annual_fees / 12 * under_share / (1 - under_share)
    principal = unbilled * months
    total = sum(unbilled * (1 + monthly_rate) ** (months - m) for m in range(1, months + 1))
    return {"principal": principal, "interest": total - principal, "total": total}
```

***Listing 22.1.** A product’s annual cost under its fee rules, the budget at the cheaper licence, the enterprise break-even and the exposure of an audit. code/firm/databudget/firm_databudget.py*

## 22.3 Audits and back-billing

**Definition 22.3 (Back-billing).**

*Back-billing* is the charge a data provider raises after an audit for usage that was not reported or paid for in past periods, usually with interest and sometimes with a penalty, under the audit clause of its licence.

**Proposition 22.4 (What an audit finds).**

If a firm reported fees $F$ a year while a share $s$ of its true usage went unreported for $M$ months, the unbilled fees are $u=Fs/(12(1-s))$ a month and the [back-billing](#def-fm-data-strategy-and-costs-backbill) with monthly interest $i$ from each month to the audit is

$$
u\sum_{k=0}^{M-1}(1+i)^k=u\,\frac{(1+i)^M-1}{i}.
$$

**Proof.** True usage is reported usage divided by $1-s$, so the unreported fees are $F\,s/(1-s)$ a year. The month $m$’s unbilled amount accrues interest for $M-m$ months; summing the geometric series gives the formula. ∎

The firm’s per-user fees, display and terminals, are $774 000 a year. An audit that finds 15% of true users unreported over three years bills $410 000 of fees and, at 1% a month, $81 000 of interest: $490 000, 63% of a year’s per-user fees, due at once ([Figure 22.3](#fig-fm-data-strategy-and-costs-audit)). The exposure grows faster than the share: 5% under-reporting costs $146 000, 30% costs $1.19 million. The defence is the entitlement system of Book 15: counts taken from the systems that grant access, not from what users declare.

![Back-billing after an audit of three years of per-user fees ($774 000 a year reported), by the share of true users that went unreported. Data: fm_data.audit.](https://one-course.com/images/onecourse/chapters/quant-16/fm-data-strategy-and-costs/fig-310fb7a28155.svg)

***Figure 22.3.** [Back-billing](#def-fm-data-strategy-and-costs-backbill) after an audit of three years of per-user fees ($774 000 a year reported), by the share of true users that went unreported. Data: `fm_data.audit`.*

## 22.4 Tutorial: the audit letter

**Goal.** Price a firm’s data inventory, find where an [enterprise licence](#def-fm-data-strategy-and-costs-licences) wins, and estimate the exposure of an audit that finds users unreported. **End state:** the budget by category ([Figure 22.1](#fig-fm-data-strategy-and-costs-budget)), the enterprise break-even ([Figure 22.2](#fig-fm-data-strategy-and-costs-enterprise)) and the audit exposure ([Figure 22.3](#fig-fm-data-strategy-and-costs-audit)).

1. **The inventory.** `fm_data.PRODUCTS` and `USAGE` hold seven products with their fee rules and counts; `firm.databudget.budget` prices each at its cheaper licence.
2. **The break-even.** `enterprise_breakeven` gives the headcount from which the [enterprise licence](#def-fm-data-strategy-and-costs-licences) is cheaper; `fm_data.reference_curve` draws both costs against users.
3. **The audit.** `fm_data.audit(under, years, monthly_rate)` applies [Proposition 22.4](#prop-fm-data-strategy-and-costs-audit) to the per-user fees.
4. **The allocation.** `fm_data.allocation()` splits the budget over four strategies by their measured use; `fm_data.alt_value()` prices the alternative dataset with Book 7’s vendor-trial arithmetic.

**What to change next.** Let headcount grow 20% a year and find when the terminals would justify an enterprise negotiation; add a redistribution fee for a client-facing quote page.

## 22.5 Negotiating with exchanges and vendors

**As of September 2026 — Rules on market-data pricing.**

**European Union**: MiFIR (Regulation (EU) No 600/2014) article 13, as adopted, requires trading venues to make pre- and post-trade data available on a reasonable commercial basis, with non-discriminatory access, and free of charge fifteen minutes after publication. **United States**: the SEC’s market data infrastructure rules provide for competing consolidators of NMS market data, registered under Rule 614 (17 CFR 242.614, adopted in 2021). **United Kingdom**: the FCA’s wholesale data market study (February 2024) found evidence of market power among benchmark providers, credit rating agencies and market data vendors, and said it would pursue its findings through regulatory reform and its competition powers.

Negotiation starts from the inventory and the counts. A firm that knows exactly who uses what can move whole categories to enterprise terms, drop unused entitlements before renewal, and resist price rises with a credible plan to switch where a substitute exists. It has less leverage where the data comes from one venue or one administrator, which is the market power the FCA described; there the levers are the licence terms rather than the price: definitions of non-display use, derived-data rights, audit periods and interest, and caps on renewal increases.

## 22.6 Data as a strategic asset

An alternative dataset is bought for the return it adds. Book 7’s vendor-trial harness prices it: the annual fee at which the dataset’s expected contribution, decaying as others buy it, just pays for itself. For the chapter’s dataset, feeding a strategy on $50 million that earns 2% a year from it before trading costs of 0.5%, with an edge that halves every two years, the break-even fee over a three-year contract is $369 000 a year (`fm_data.alt_value`): at $300 000 the dataset is worth buying, with little margin for error in the decay.

Data costs are allocated to strategies by use: on the chapter’s weights, market making carries $933 000 of the budget, statistical arbitrage $622 000, event-driven trading $311 000 and macro $207 000. A strategy whose data costs exceed what it earns from them should not be carrying them, which is only visible once they are allocated.

**Method 22.5 (Running a data budget).**

1. Keep an inventory of every data product: provider, licence, unit of count, current counts, renewal date, owner.
2. Take counts from the entitlement system, reconcile them monthly with providers’ invoices, and remove unused access before each renewal.
3. Compare per-unit and enterprise terms at forecast headcount; price the audit exposure of every counted licence.
4. Price alternative data by its expected contribution before and after a trial, and stop what does not pay.
5. Allocate the budget to strategies by use and review it with their results.

## 22.7 Build: the data budget

**Purpose.** Data products and fee rules as data, the budget, the enterprise break-even, audit exposure, allocation to strategies and the value of an alternative dataset.

**Interface.** `firm.databudget`: `Product`, `Usage`, `annual_cost`, `budget`, `enterprise_breakeven`, `audit_exposure`, `allocate`, `alt_breakeven` (wrapping `firm.vendoreval.breakeven`).

**Rules.** Fee levels are inputs; each product is budgeted at its cheaper licence; [back-billing](#def-fm-data-strategy-and-costs-backbill) accrues interest from each month to the audit.

**Acceptance tests.** `code/firm/databudget/tests/`: costs and the budget on hand numbers; the break-even; the audit’s geometric sum; the wrapper.

**Stretch.** Usage records from Book 15’s entitlement system; renewal dates and notice periods; derived-data and redistribution fees.

Sources and further reading

- Financial Conduct Authority, *Wholesale Data Market Study* , Market Study MS23/1.5, February 2024.
- Regulation (EU) No 600/2014, article 13; 17 CFR 242.614.

## 22.8 Exercises

**Exercise 22.1 ★.**

At what headcount does the reference data’s [enterprise licence](#def-fm-data-strategy-and-costs-licences) win, and what does it save at sixty users?

**Solution of Exercise 22.1.**

From $400/8=50$ users; at sixty it saves $480-400=\$80\,000$ a year.

**Exercise 22.2 ★.**

What did the FCA’s study find about access to data and about prices?

**Solution of Exercise 22.2.**

No evidence that firms cannot get the data they need, but evidence of market power in all three markets studied, so that users may pay higher prices than competition would allow.

**Exercise 22.3 ★.**

Without interest, what does an audit bill for 15% of true users unreported over three years on $774 000 a year of per-user fees?

**Solution of Exercise 22.3.**

$774\times0.15/0.85=\$136\,600$ a year of unbilled fees, $410 000 over three years.

**Exercise 22.4 ★★.**

Prove the closed form of [Proposition 22.4](#prop-fm-data-strategy-and-costs-audit) and compute the interest at 1% a month.

**Solution of Exercise 22.4.**

See the proof of [Proposition 22.4](#prop-fm-data-strategy-and-costs-audit); with $u=\$11\,382$ a month, $M=36$ and $i=1\%$, the total is $490 000, of which $81 000 is interest.

**Exercise 22.5 ★★.**

Explain the difference between a [derived-data licence](#def-fm-data-strategy-and-costs-licences) and a [redistribution licence](#def-fm-data-strategy-and-costs-licences) with an example from a trading firm.

**Solution of Exercise 22.5.**

A [derived-data licence](#def-fm-data-strategy-and-costs-licences) lets the firm compute and use its own fair values or signals from an exchange’s prices; a [redistribution licence](#def-fm-data-strategy-and-costs-licences) lets it pass prices or derived values to clients, for example on a quote page or in a research note.

**Exercise 22.6 ★★.**

Why does the audit exposure grow faster than the unreported share?

**Solution of Exercise 22.6.**

The unbilled fees grow as $s/(1-s)$, which is convex in $s$, and interest compounds on each month’s amount.

**Exercise 22.7 ★★★.**

*Coding.* With a half-life of one year instead of two, what is the alternative dataset’s break-even fee? Should the firm buy at $300 000?

**Solution of Exercise 22.7.**

$162 000 a year: below the $300 000 price, so the dataset is not worth buying on that decay.

**Exercise 22.8 ★★★.**

*Find the flaw.* “Our users tell us which terminals they use each quarter, so our counts are right.”

**Solution of Exercise 22.8.**

Self-reported counts are the under-reporting an audit finds; counts must come from the entitlement system that grants access, reconciled with invoices.

## 22.9 Problem: The Audit Letter

**Problem 22.1.**

Weekend problem — the audit letter

A chief operating officer receives an exchange’s notice of audit, and must estimate the exposure and redesign how the firm buys data.

**Part I — The market.**

1. What did the FCA find in 2024, and why does it matter for a buyer?
2. What does MiFIR article 13 require of trading venues?
3. Define a [data budget](#def-fm-data-strategy-and-costs-budget) and list its categories.
4. Give the firm’s budget by category.

**Part II — Licences.**

5. Define enterprise, derived-data and [redistribution licences](#def-fm-data-strategy-and-costs-licences) .
6. Compute the reference data’s enterprise break-even.
7. What else, besides price, does an [enterprise licence](#def-fm-data-strategy-and-costs-licences) change?
8. Which terms would you negotiate where the provider has market power?

**Part III — The audit.**

9. Define [back-billing](#def-fm-data-strategy-and-costs-backbill) .
10. State and prove [Proposition 22.4](#prop-fm-data-strategy-and-costs-audit) .
11. Give the exposure for 15% under-reporting over three years, and for 5% and 30%.
12. How would you respond to the notice?
13. How does an entitlement system reduce the exposure?

**Part IV — Strategy.**

14. How would you price the alternative dataset, and what is its break-even fee?
15. Allocate the budget to strategies.
16. What does the allocation tell the firm?
17. What would you stop buying, and how would you decide?
18. How often should the inventory be reconciled?
19. State the *named result* : the headcount at which an [enterprise licence](#def-fm-data-strategy-and-costs-licences) beats per-user fees, and the [back-billing](#def-fm-data-strategy-and-costs-backbill) exposure of an audit that finds 15 per cent under-reporting over three years.
20. In two sentences, write the data policy.

**Solution of Problem 22.1.**

1. Market power in benchmarks, ratings data and data vendors, and complex, multiple licences that add cost; a buyer has limited price leverage and must negotiate terms.
2. Pre- and post-trade data on a reasonable commercial basis, non-discriminatory, free after fifteen minutes.
3. See [Definition 22.1](#def-fm-data-strategy-and-costs-budget) .
4. $564 000 market data, $750 000 terminals, $460 000 reference data and news, $300 000 alternative data: $2.07 million.
5. See [Definition 22.2](#def-fm-data-strategy-and-costs-licences) .
6. 50 users.
7. It removes counting and the audit exposure for that product, and fixes the cost as headcount changes.
8. Definitions of use, derived-data rights, audit periods and interest, renewal caps.
9. See [Definition 22.3](#def-fm-data-strategy-and-costs-backbill) .
10. See [Proposition 22.4](#prop-fm-data-strategy-and-costs-audit) .
11. $490 000; $146 000 at 5%; $1.19 million at 30%.
12. Reconcile the auditor’s counts with the entitlement records, challenge what is not supported, negotiate interest and settlement, and fix the counting.
13. Counts come from access actually granted, so under-reporting cannot arise from users’ declarations.
14. By its expected contribution net of costs, decaying as others buy it; $369 000 a year over three years, against a $300 000 price.
15. $933 000 market making, $622 000 statistical arbitrage, $311 000 event-driven, $207 000 macro.
16. Which strategies’ data costs exceed what the data earns them.
17. Products with unused entitlements and datasets whose trial contribution does not cover the fee.
18. Monthly against invoices and before every renewal.
19. 50 users; $490 000 including $81 000 of interest.
20. Count every licence from the entitlement system, buy enterprise terms where headcount exceeds the break-even and remove unused access before renewals; price every dataset by what it earns and allocate the budget to the strategies that use it.

## 22.10 Interview questions

**Interview question 22.1 ★ developer.**

What is a non-display use of market data, and why does it matter to a trading firm?

**Solution of Interview question 22.1.**

Use by machines rather than people viewing a screen, such as algorithmic trading or risk calculation; exchanges charge for it separately, often by category, and it is a large share of a trading firm’s market-data fees.

*What the interviewer is looking for: machine use and separate fees.*

**Interview question 22.2 ★ trader.**

Why might a firm pay for an [enterprise licence](#def-fm-data-strategy-and-costs-licences) it does not fully use?

**Solution of Interview question 22.2.**

Because above the break-even it is cheaper, and because it removes counting and audit exposure and lets headcount grow without new fees.

*What the interviewer is looking for: break-even and audit risk.*

**Interview question 22.3 ★★ researcher.**

How would you decide whether an alternative dataset is worth its price?

**Solution of Interview question 22.3.**

Trial it for incremental information over what the firm already has, estimate the return it adds net of costs and its decay, and compare the break-even fee with the price.

*What the interviewer is looking for: incremental value and decay.*

**Interview question 22.4 ★★ developer.**

Design a system that keeps market-data counts audit-ready.

**Solution of Interview question 22.4.**

Grant access only through an entitlement system, log every grant and use by the provider’s unit of count, reconcile with invoices monthly and keep the records for the audit period.

*What the interviewer is looking for: counts from access, not declarations.*

**Interview question 22.5 ★★ risk.**

An exchange audit finds under-reporting. What is the firm’s exposure, and what do you do next?

**Solution of Interview question 22.5.**

The unreported fees over the audit period plus interest ([Proposition 22.4](#prop-fm-data-strategy-and-costs-audit)); verify, negotiate, pay, and fix the counting.

*What the interviewer is looking for: the formula and the remediation.*

**Interview question 22.6 ★★★ researcher, trader.**

How should data costs be allocated across strategies, and what behaviour does the allocation create?

**Solution of Interview question 22.6.**

By measured use (entitlements, messages, queries); it makes strategies weigh data costs against returns, and encourages dropping unused data, but can also discourage sharing data that would benefit others.

*What the interviewer is looking for: allocation by use and its incentives.*