An Arbitrage Guide to Financial Markets — Robert Dubil

Chapter 2: Financial Math I — Spot

This is the book's first "primer" chapter — the toolkit Chapter 1's examples borrowed without deriving. It's also where Chapter 1's illustrative duration-matching example (the $100M/$102M bond pair, with made-up DV01 numbers) gets its real formula: this chapter shows you exactly how duration is computed, with the book's own fully worked table.

Source
Extracted directly from the chapter text in /references
Chapter type
Financial Math primer — deliberately calculus-free, built on numerical examples
Connects to
Formalizes the duration concept used illustratively in Ch. 1
01

Discounting: the one idea everything else builds on

The book's opening claim, stated flatly: cash-flow discounting is the basis of all securities valuation. A stock's value is the present value of its future dividends and capital gains; a bond's value is the present value of its coupons and principal. You'd always pay less than $100 today for a promise of $100 later — not only because the promise carries risk, but because $100 today, invested, would grow to more than $100 by then. Whether you use your own cash or borrowed cash doesn't change this: borrowed cash costs explicit interest, your own cash costs the interest you gave up by not investing it elsewhere (an opportunity cost — just as real a cost as interest paid).

02

Present value, worked from scratch

Invest $500 at 5% for 1 year: $500(1.05) = $525. For 2 years, you earn interest on interest: $500(1.05)² = $551.25. General formula: FV = PV₀(1+r)ⁿ. Reversed: how much would you need today to have $500 in 2 years at 5%? PV₀ = 500/(1.05)² = $453.51 — the two amounts are equivalent at that rate; you'd be indifferent between them.

For a stream of cash flows — say $500 a year for 4 years — you discount and sum each one separately: PV₀ = 500 × [1/(1+r) + 1/(1+r)² + 1/(1+r)³ + 1/(1+r)⁴] = $1,772.98. A constant cash flow like this is an ordinary annuity, and the bracketed term is its present-value annuity factor.

03

Compounding frequency changes the real return — a lot

A quoted rate is always per annum, but how often it compounds changes what you actually earn. The book's own rollover example: invest €1,000 in a 3-month CD at 3.25% (p.a.), rolled over 4 times across a year.

Same quoted rate (3.25% p.a.), different day-count conventions — real book numbers
ConventionFormula1-year result
30/360 (simplified quarters)1,000(1+0.0325/4)⁴€1,032.898
Act/360 (real day counts: 92/91/90/92)1,000·∏(1+0.0325×days/360)€1,033.361

What this means: the exact same 3.25% quoted rate produces two different real amounts — €0.463 apart — purely from which day-count convention is used. That's not rounding noise; the book calls this a real risk: "many unsuspecting investors have been burnt in the past by ruthless dealers playing day-count tricks." The equivalent annual rate (EAR) — 3.2898% for the 30/360 case — is how you make different compounding conventions comparable at all: (1+r/m)^m = 1+EAR, where m is the number of compounding periods per year.

04

Rate vs. yield — a distinction the market blurs but shouldn't

The book's own line: "rate" is the stated percentage used to compute the actual interest paid; "yield" is the percentage actually earned on what you invested. A 5% coupon bond selling at par (100% of face value) yields exactly 5%. The same bond selling below par yields more than 5%, because you paid less than the principal for the same coupon stream. The two numbers coincide only by coincidence, not by definition.

05

Duration — turning bond-market chaos into one comparable number

Where this connects: Chapter 1's duration-matched bond trade (long $100M 30-year, short $102M 26-year) used illustrative DV01 figures to make the point concrete. This section is where that number actually comes from.

Any single issuer can have wildly different bonds outstanding — different maturities, coupons, structures. Macaulay duration is defined as the present-value-weighted average time to a bond's cash flows — it collapses all that heterogeneity into one number, in years, that lets you compare completely different bonds on equal footing. Modified duration reframes the same idea as a price sensitivity: the % change in price per unit change in yield, ModD = −(ΔP/P)/Δy.

The book's own worked example: a 6-year, 7% semi-annual coupon bond yielding 8% — a $100 face bond paying $3.50 every 6 months plus $100 principal at year 6, currently worth $95.3075.

Macaulay duration calculation, abbreviated from the book's Table 2.5 (selected rows)
Time (yrs)Cash flowPV of CF% of total PVt × %PV
0.53.503.3653.53%0.0177
3.03.502.7662.90%0.0871
6.0 (final)103.5064.64667.83%4.0697
Total (all 12 periods)95.3075100.00%4.9720

What this means: Macaulay duration = 4.9720 years — note the final principal repayment alone accounts for 67.83% of the bond's present value, which is why duration (4.97 years) comes out so much shorter than the bond's 6-year maturity: the coupons pull the weighted-average time to cash flows earlier. Modified duration = 4.9720/(1+0.08/2) = 4.7807. If yield rises from 8.00% to 8.15% (+15bp), predicted price change = −4.7807 × 0.15% = −0.7171%, or $95.3075 → $94.6240. The exact recalculated price is $94.6270 — duration is a very close linear approximation for a small yield move like this one.

⚙ Algo-trading angle
Context
The book pushes the same bond to an extreme: what if yield fell all the way to 0%?
Algo-relevant?
Yes — this is exactly where a duration-only risk model breaks, and where a real fixed-income algorithm needs convexity too.
What's applied
Duration predicts: −4.7807 × (−8%) = +38.2456%, i.e. $95.3075 → $131.7584. But the true price at 0% yield is just the sum of undiscounted cash flows: $142 — duration's linear approximation misses by over $10.
Action
An algorithm sizing a hedge using duration alone (as Ch. 1's illustrative DV01 example did, for a small parallel move) is implicitly assuming small moves. For large moves, or for options-embedded bonds, it needs the second-order term (convexity) or full repricing, not just the first derivative.
Why
Duration is a first derivative — the slope of a tangent line to a curved price-yield relationship. A tangent line is a good local approximation and a bad global one. The gap between the linear prediction ($131.76) and the true price ($142) is convexity, quantified.

Portfolio duration has one very convenient property the book highlights: it's simply the weighted average of the individual bonds' durations, weighted by their share of the portfolio — because duration is a first derivative, and first derivatives are additive. This is exactly why duration-matching (Ch. 1) works as a hedge: you don't need to model each bond separately, just match the weighted-average sensitivity.

06

Equity math — why stock valuation can't be as clean as bond valuation

Bonds have contractually guaranteed cash flows (barring default); stocks don't. A company owes shareholders nothing — dividends and capital gains happen only if the company chooses and is able to deliver them. The dividend discount model handles this by discounting all expected future dividends: P₀ = D₁/(1+r) + D₂/(1+r)² + ... to infinity — notably, this holds regardless of your own holding horizon, since a future buyer's price already embeds their own expected future dividends.

Three companies, same $6 dividend, discount rate 15% — the book's own numbers
CompanyDividend patternFormulaFair value
ABC-NoGrowthconstant $6/yr forever6/0.15$40.00
ABC-ConstGrowth$6 next year, +5%/yr forever6/(0.15−0.05)$60.00
ABC-NonConstGrowth$6 flat for 3 yrs, then +5%/yr3-yr PV + terminal value$55.12

What this means: the exact same $6 near-term dividend produces fair values ranging from $40 to $60 — a 50% spread — purely from the assumed long-run growth rate, not the current payout. This is the book's own point: equity valuation is far more sensitive to a subjective growth assumption than bond valuation is to anything comparable, which is why "complicating the math further doesn't make it more accurate" — the uncertainty is in the input, not the formula.

07

Currency quote conventions — a real, recurring source of errors

Unlike a barrel of oil, currencies are quoted both ways — GBP costs USD, and USD costs GBP, and the two are just reciprocals of each other. Convention varies by currency: the euro and former-Commonwealth currencies (AUD, NZD) are typically quoted in American terms — USD per unit of that currency. Most others (CHF, JPY, HKD) are quoted in European terms — units of that currency per USD. The book's specific warning: a "cable rate" is often written GBP/USD followed by a number like 1.65 — which actually means 1.65 USD per GBP, not 1.65 GBP per USD, despite the ordering. Getting this backwards is a real, recurring, entirely avoidable error.

08

Before moving to Chapter 3

  1. 1

    Pick a real bond you can find data for (coupon, maturity, yield) and compute its Macaulay and modified duration by hand, the way the book's Table 2.5 does — then check it against whatever duration figure your broker or a bond screener reports.

  2. 2

    Revisit Ch. 1's duration-matched pairs trade with real durations from step 1 instead of the illustrative $180,000/bp DV01 figure — recompute the parallel-move vs. curve-steepening P&L with real numbers.

  3. 3

    Pick a stock you follow, and compute its fair value under two different long-run growth assumptions (e.g. 3% vs. 6%) using the constant-growth formula — see how sensitive the "fair value" really is.

  4. 4

    Chapter 3 moves to fixed income securities themselves (money markets, government/corporate bonds, mortgages) — the instruments this chapter's math actually gets applied to.

present valueEARday-count conventionrate vs. yield Macaulay durationmodified durationportfolio durationconvexity dividend discount modelGordon growth modelFX quote convention
09

Duration and DDM as the backbone of two algo strategy types

This chapter's two big tools map onto two different systematic-strategy families:

Fixed-income relative value algos§05

Duration is the actual risk metric a rates/bond algorithm hedges on — Ch. 1's duration-matched pairs trade, now with a real formula and a worked table instead of an assumed DV01.

Fair-value / stat-arb equity algos§06

A systematic strategy that estimates "fair value" for a stock and trades deviations from it needs exactly the dividend discount model's structure — and needs to know how sensitive that fair value is to its own growth assumption, since a small assumption change swung fair value from $40 to $60 in the book's own example.

Built directly from the chapter's own text, extracted from the PDF in /references — including its real worked numerical examples (the CD rollover, the duration table, the dividend discount model). Not a summary from the chapter title or abstract alone.