An Arbitrage Guide to Financial Markets · Chapter 1
An Arbitrage Guide to Financial Markets — Robert Dubil

Chapter 1: The Purpose and Structure of Financial Markets

This book's opening move is to throw out the "who buys what and why" tour of financial markets and replace it with one idea: every security is a bundle of state-contingent claims, and almost everything a trading desk actually does is either pure arbitrage or — far more often — relative value arbitrage. That single distinction is the closest thing in either reference book to a direct definition of what a quantitative trading strategy is.

Source
Extracted directly from the chapter text in /references
Book structure
Part I (Spot) starts next — this chapter is the conceptual foundation for all four parts
Prerequisite math
None yet — the book deliberately avoids calculus, using numerical examples instead
01

Markets are more than a Saturday bazaar

The standard tour of financial markets — businesses issue stock and bonds, investors buy them, prices go up or down — is real but incomplete. The book's specific complaint: this view focuses exclusively on spot markets, where cash and securities exchange hands immediately. The missing dimension is time of delivery. Once you introduce trades that settle later (forwards), and trades that settle later and only if something happens (options and other contingent claims), the whole structure of financial markets becomes visible.

02

The chessboard — every security is a bundle of claims

The book's central mental model: any financial contract is a package of tiny state-contingent claims, each one a bet on a specific price level at a specific future date. A share of stock trading at €45, held for 2 years, tradeable in €0.05 ticks from €0.00 to €500.00, is literally a bundle of 10,001 price levels × 730 days = 7,300,730 individual claims — a "chessboard" with time on one axis and price on the other. The stock's price today is the sum of the value of every square.

A forward contract to buy the same stock at €60 in one year is a subset of that same chessboard — the first 365 days' worth of claims removed, and the payoff on each remaining claim equal to (stock price − €60). An American call option struck at €60 is an even smaller subset: only the claims where the stock price exceeds €60 have any value at all.

This is also why options are structurally identical to insurance: a $1,000,000 life insurance policy is a bundle of 365 daily claims, each paying out only if the insured dies that day. A put option on the S&P 100 index works the same way — it pays out only if the index falls below the strike, and the further it falls, the more it pays, exactly like a fire-insurance policy paying more for worse fire damage.

Proposition (the book's own): all financial markets evolve to have three structural components — a spot market, a forwards/futures market, and a contingent-claims (options/derivatives) market — regardless of what underlying asset class you're looking at.

⚙ Algo-trading angle

The book's own worked example: an investor holding a corporate bond worried about credit-spread risk shorts a duration-matched government bond, eliminating the interest-rate dimension entirely and leaving only the credit-spread dimension exposed — reducing a 3D risk "cube" (time × rate × spread) to a 2D plane. This is precisely how a systematic strategy isolates the one risk factor it actually wants to trade, hedging out everything else it doesn't have a view on. A strategy that doesn't do this dimension-reduction explicitly is implicitly making bets on risk factors it never intended to take.

03

Pure arbitrage vs. relative value arbitrage — the core distinction

Pure arbitrage (the book's definition): generating riskless profit today by statically or dynamically matching current and future obligations to exactly offset each other, including known financing costs. Classic example: if Czech korunas cost more bought directly with pounds than bought via dollars, buy dollars, buy korunas with the dollars, and you've captured a risk-free profit with zero directional exposure.

In practice, pure arbitrage opportunities are rare and fleeting. Almost all real institutional money-making — hedge funds, prop/arb desks — runs on relative value arbitrage instead: find a close substitute for the primary risk in a position, take an offsetting position in that substitute, and deliberately leave the remaining secondary risk unhedged (but controlled).

⚙ Algo-trading angle
Context
The book's own worked example: buy $100M of a 30-year US government bond, short $102M of a 26-year government bond — the $100M/$102M ratio chosen via duration-matching.
Algo-relevant?
Yes — this is the textbook definition of a systematic relative-value / statistical arbitrage strategy.
What's applied
Duration-matching sizes the two legs so their dollar sensitivity to a parallel rate move is equal and opposite — the position is neutral to the overall level of rates, only exposed to whether the 30y and 26y rates move together or diverge.
Action
The position only makes or loses money when the yield curve changes shape (30y and 26y rates move by different amounts) — never from rates simply going up or down together.
Why
Investors barely distinguish 30y from 26y risk — the two rates move closely together most of the time. The relative arbitrageur is explicitly betting they'll diverge, while staying indifferent to the much larger, much more common parallel moves.
Worked example — the book's structure, with illustrative dollar-duration (DV01) figures to make the P&L concrete
Position: long $100M 30-year bond, short $102M 26-year bond
Duration-matched so DV01 (value of 1bp) ≈ $180,000/bp on both legs

Scenario A — parallel move: both yields +10bp
  long leg loses:   10 × $180,000 = −$1,800,000
  short leg gains:  10 × $180,000 = +$1,800,000
  ─────────────────────────────
  Net P&L                            ≈ $0  (exactly the point of duration-matching)

Scenario B — curve steepens: 30y +12bp, 26y +8bp (non-parallel)
  long leg loses:   12 × $180,000 = −$2,160,000
  short leg gains:   8 × $180,000 = +$1,440,000
  ─────────────────────────────
  Net P&L                            −$720,000 — the position's ONLY real exposure
04

Telling arbitrage from speculation — the pairs-trading test

The book draws a sharp, practical line most people miss: buying Pfizer and selling GlaxoSmithKline — both large pharma companies with similar R&D budgets — is called out explicitly as pure speculation, not relative value arbitrage, because the two companies' specific risks are too different to call them close substitutes. Buying Polish zlotys against Czech korunas is the same mistake in FX. An airline hedging jet fuel exposure with crude oil futures (instead of a direct jet-fuel contract) is "clearly on the speculative side" too — it carries real basis risk, since a refinery-level supply shock can move jet fuel prices faster than crude.

The test that actually matters: is the substitute genuinely close to the primary risk, or merely correlated with it? Genuine relative value arbitrage requires the former.

⚙ Algo-trading angle

This is exactly the failure mode of naive statistical arbitrage / pairs-trading algorithms: screening for high historical correlation and calling the resulting pair "market neutral" without asking whether the two assets share the same primary risk driver. Two pharma names with different pipelines, different patent cliffs, and different regulatory exposure are not a hedge of each other just because their prices historically moved together — a backtest built on that correlation can look market-neutral in-sample and still take a large, undiversified directional bet the moment the correlation breaks (exactly what happens around company-specific news, which a pairs algorithm has no mechanism to distinguish from noise).

05

How trading floors are actually organized

Institutional trading floors are built around this same logic: desks likely to trade with each other sit next to each other. Customer desks serve outside clients; proprietary (prop/arb) desks focus specifically on relative-value trades or outright speculation across markets, for the firm's own account. Desks routinely collaborate — the book's example: a money-market desk arranges a short-term note with a coupon tied to a stock index, then loops in the swap desk (to reshape the interest-rate exposure) and the equity derivatives desk (to strip out the equity risk), leaving the customer with cheap financing and no equity exposure, while the dealer lays off the swap and equity risk elsewhere.

Dynamic relative-value arbitrage complicates this further: a seller of a 3-year OTC equity call typically hedges with 3- and 6-month exchange-listed calls plus a short stock position, rebalanced daily as the stock price moves (delta hedging) — leaving the dealer exposed only to differences in implied volatility between the options bought and sold, not to the stock's direction at all.

⚙ Algo-trading angle

A delta-hedging market maker is, functionally, an algorithm that runs continuously: recompute the hedge ratio, trade the underlying to match it, repeat — every single trading day for as long as the option position is open, whether or not new options are sold. This daily-rebalance loop is itself the direct precursor to the market-making bot's re-quoting loop covered elsewhere — the "inventory" being managed here is an options book instead of a single stock position, but the mechanic (measure exposure, trade to neutralize it, repeat) is identical.

06

Asset transformers vs. broker-dealers

Financial institutions split into two types by balance-sheet structure. An asset transformer holds assets with different legal character than its liabilities — a bank holds mortgages and business loans (assets) but issues checking accounts and CDs (liabilities) that are small, liquid, and insured (US FDIC covers up to $100,000/customer/bank in this book's era). A broker-dealer holds the same type of security on both sides — it buys stocks and sells the same stocks, exposed only to the temporary risk of holding them. As dealers, they own inventory temporarily; as brokers, they simply match buyers and sellers without taking a position.

The purest broker-dealer model exists historically in the US and Japan (regulatory separation from banking); continental Europe favors universal banks combining both functions. A visible long-run trend: disintermediation — securitizing previously "transformed" assets (credit card receivables, mortgages) into standardized tradeable packages, moving activity from the transformer model toward the broker-dealer model.

07

Primary vs. secondary markets — and why secondary comes first

Primary markets move funds directly from investors to issuers (IPOs, seasoned offerings, private placements under Rule 144-A in the US). Secondary markets trade only between existing investors, with no involvement from the original issuer — organized as exchanges (NYSE, with human specialists maintaining continuity; Tokyo, fully electronic with no dealer able to earn monopoly rents from order-flow information), OTC dealer networks (corporate and government bonds), or hybrids (Nasdaq — a virtual, members-only dealer network).

The book's sharpest point here: secondary markets often get built first, specifically to make the primary market viable — a "tail wag the dog" pattern. Developing countries limit exchange seats and trading hours deliberately, to funnel scarce buyers and sellers into one venue and manufacture liquidity. Two real historical examples the book cites: Michael Milken's Drexel Burnham Lambert made speculative high-yield bonds sellable in the 1980s primarily by actively market-making a secondary OTC market for them — not by underwriting alone. Enron, before its 2002 collapse, built its energy-derivatives business by making itself a virtual exchange (with Enron as the sole dealer) for energy forwards and contingent contracts. Both firms failed; both markets they built survived them.

08

Hedgers, speculators, buy-side, sell-side

The book's definition of speculation is deliberately uncomfortable: any position whose return isn't contractually guaranteed to equal your cost of capital is speculation — relative to a benchmark, not in absolute terms. Its own example: John buys $1,000 of stock with his own cash, sells for $1,100 a year later — 10% return. Adam borrows the same $1,000 at 5% to buy the same stock, sells for $1,100 — 10% gross, 5% net of interest. Both earned the same 5% excess return over their cost of capital; both speculated equally. Even a fixed-rate CD holder is "speculating" that rollover rates won't rise above the locked-in rate during the CD's term.

Speculators take on explicit market risk for excess return, with no offsetting hedge. Hedgers enter simultaneous offsetting transactions so their net exposure sits above their cost of capital without directional risk. All arbitrageurs — pure or relative — are hedgers by this definition.

The more common split in practice: sell-side (the ~50 largest global financial institutions — manufacture products, hedge immediately, always looking for "the other side of the trade") vs. buy-side (everyone choosing from the sell-side's menu — mutual and pension funds, insurers, individual investors). Hedge funds sit unusually on both sides: capitalized like buy-side speculators, but nearly every strategy they run is relative value arbitrage — a hedge, by the book's definition. In the late 1990s, funds like Tiger, AIM, and LTCM grew large enough in their niches to flip roles entirely, effectively selling hedges to dealers instead of buying them.

09

Where this book is headed

The book is organized around delivery time, not asset class — Part I (Spot), Part II (Forwards/futures — cash-and-carry arbitrage), Part III (Options — delta-hedged dynamic cash-and-carry), plus an Appendix on credit risk cutting across all three. Each part opens with a "Financial Math" primer chapter (2, 6, 9 — deliberately calculus-free, built on numerical examples) before the descriptive market-survey chapters that apply that math to real benchmark trades.

10

Before moving to Chapter 2

  1. 1

    Pick two securities you'd consider a "pairs trade" and apply the chapter's own test: is the substitute genuinely close to the primary risk, or just correlated with it? Use the Pfizer/GSK example as the bar to clear.

  2. 2

    Work through the duration-matched bond example with your own illustrative numbers (pick two bond maturities, assign DV01s) and compute P&L under both a parallel move and a curve-shape change, the way §03's worked example does.

  3. 3

    Classify your own recent trades using the book's strict definition: for each one, is your expected return contractually guaranteed to equal your cost of capital? If not, you were speculating — even if it didn't feel that way.

  4. 4

    Chapter 2 ("Financial Math I — Spot") is next: present value, rates and yields, and the yield curve — the toolkit this chapter's examples borrowed without deriving.

state-contingent claimpure arbitragerelative value arbitrageduration matching basis riskdelta hedgeasset transformerbroker-dealer primary marketsecondary markethedger vs. speculatorbuy-side / sell-side
11

Relative value arbitrage, systematized

This chapter's entire framework maps almost directly onto algorithmic/quant trading strategy design: pure arbitrage bots (fully automatable, fleeting, latency-sensitive — cross-exchange or triangular-FX arbitrage), static relative-value bots (duration-matched pairs, cash-and-carry structures, rebalanced rarely), and dynamic relative-value bots (delta-hedging market makers, rebalancing continuously). Here's the duration-matched pairs trade from §03, systematized:

// static relative-value bot — duration-matched bond pair, §03
long_leg  = bond(maturity=30y, notional=$100M)
short_leg = bond(maturity=26y, notional=duration_match(long_leg))  // §03 — DV01-neutral to parallel moves

// daily P&L attribution — the whole point of the trade:
parallel_pnl    = 0                              // by construction, near zero
curve_shape_pnl = -DV01_long × Δy_30y + DV01_short × Δy_26y   // §03 — the ONLY real exposure

// risk check, not present in a naive pairs bot (§04's failure mode):
if correlation(long_leg, short_leg, lookback) < min_substitute_similarity:
    flag_as_speculation()   // §04 — the Pfizer/GSK test, automated
    reduce_size_or_exit()

The risk check at the bottom is what a naive stat-arb screener skips: it operationalizes §04's core lesson — measuring whether the "hedge" is still a genuinely close substitute, not just historically correlated, before trusting the position's risk to be what the duration-matching math says it is.

Built directly from the chapter's own text, extracted from the PDF in /references — including its real worked examples and numerical structure. Not a summary from the chapter title or abstract alone.