Chapter 8 · Stochastic Processes in Finance

Stochastic Arithmetic

The new rules of calculation — why stock prices need their own laws of physics.
Why school maths breaks here

Everything you have learned so far — Brownian Motion, GBM, Quadratic Variation — described what prices do. Now we need the tools to actually calculate with them. The problem is that the calculus you learned in school was designed for smooth, well-behaved functions. A stock price is neither. It is jagged, infinite in variation, and powered by randomness. To work with it mathematically, we need a new arithmetic — a new set of rules for how things multiply and add in a world full of noise.

School calculus vs stochastic calculus — smooth function vs jagged Brownian path Two panels — left shows a smooth differentiable curve labeled school calculus, right shows a jagged Brownian path labeled stochastic calculus School Calculus Stochastic Calculus Smooth — regular calculus works Jagged — needs new rules!

The same calculus toolkit cannot serve both. Stochastic paths break the assumptions smooth calculus relies on.

This chapter introduces the three building blocks of stochastic arithmetic — the Filtration, the Martingale, and the Quadratic Variation rule — and then reveals the Stochastic Multiplication Table that every quant uses every single day.

1 · The filtration (𝔽ₜ): the information constraint

In any trading model, there is a critical rule that sounds obvious but is surprisingly easy to violate: you cannot use tomorrow's data to make today's decision. The mathematical tool that enforces this rule is called the Filtration. Think of it as a one-way valve on information — only the past can flow in, never the future.

Filtration as a one-way information gate — past flows in, future is blocked A timeline with a vertical wall labeled NOW. Past information flows left to right into a trading decision. Future information is blocked by the wall. 9 AM $98 9:30 $101 9:55 $99 NOW — 10:00 AM Bet Size Δ(t) uses only past ✓ 10:30 ??? 11:00 ??? FUTURE — BLOCKED! PAST — ALLOWED ℱ(t) = everything known up to now

The Filtration ℱₜ is the mathematical wall between past and future. Your strategy can only see what is behind it.

In mathematical notation: Δ(t) is ℱ(t)-measurable. This is a compact way of saying your bet size at 10:00 AM can only be a function of information that existed before 10:00 AM. It is obvious in theory. But in backtesting, it is easy to accidentally let future data slip through — and when it does, your results are fiction.

Look-Ahead Bias is the silent killer of backtests. It happens when your model accidentally uses future data — perhaps a daily average calculated at 9 AM using the full day's data — to make a decision. The Filtration is the mathematical guard that prevents this. If your backtest shows suspiciously perfect results, check whether your ℱₜ is truly non-anticipatory.

2 · The martingale: your null hypothesis

We met the Martingale Property in Chapter 5. Now we can write it with the full mathematical precision it deserves — and understand what it means as a trading baseline.

𝔼[ W(t) | ℱ(s) ] = W(s)
𝔼[ … | ℱ(s) ] = "Our best average guess, given everything we know up to time s"
W(s) = the current price right now
W(t) = the price at some future time t

In plain English: "If you know the price right now, your best guess for any future price is… exactly the price right now."
Martingale as the flat baseline — all paths fan out from the current price with no drift Multiple future price paths fanning out symmetrically from a single current price point, with the expected value flat across time W(s) = $100 RIGHT NOW 𝔼[W(t)] = $100 possible paths possible paths Paths spread out — but on average, they stay at $100

The Martingale is the flat line through the centre of the fan. Every path is possible — but the average stays put.

The Null Hypothesis

In a fair market, 𝔼[W(t)] = W(s). Use this as your baseline. Build your model. If the model predicts something higher or lower — you have found a signal that needs explaining.

Where Profit Hides

If your quantitative model shows 𝔼[W(t)] > W(s), you have found a drift — a violation of the Martingale. Something is pulling prices in a direction. That pull is the alpha you are looking for.

3 · The dW² = dt rule: why school calculus fails stocks

This is the single most important rule in all of stochastic calculus. It looks small. It looks almost trivial. But it is the reason options trading exists. Everything — the Black-Scholes formula, every volatility model ever written — depends on this one line.

In School Calculus (smooth functions)

When a function is smooth, its tiny increments df are proportional to dt. So when you square them: (df)² ∝ (dt)². As dt → 0, this squared term vanishes completely. It becomes zero. You can safely ignore it. This is why regular calculus is clean and tidy.

In Stochastic Calculus (Brownian paths)

The wiggles of Brownian Motion are so violent that squaring them does not make them vanish. Instead: (dW)² = dt. Squaring a shock gives you real time, not zero. This tiny term survives — and it carries all the information about volatility and uncertainty.

(dW)² = dt
The squared "shock" of Brownian Motion equals a real slice of time.
It does not vanish. It accumulates. It is the Quadratic Variation.

This is why volatility scales with √T, not T.
Variance (the squared jiggle) grows linearly with time.
Standard deviation (the jiggle itself) therefore grows with the square root of time.
Volatility scales with square root of time — two charts showing variance linear and std dev square root Left panel shows variance growing as a straight line with time. Right panel shows standard deviation growing as a curved square root function. Variance grows with T Std Dev grows with √T σ²T (linear) Time T σ√T (curved) Time T Double the time → double the variance Double the time → √2 × more std dev

Variance (squared uncertainty) grows linearly. But the actual spread you experience — standard deviation — grows much more slowly, as √T.

This is why a 1-year option is not twice as expensive as a 6-month option — it is roughly √2 ≈ 1.41 times more expensive. The market prices time in square-root units because uncertainty itself grows in square-root units. The rule (dW)² = dt is the mathematical reason for this.

4 · The stochastic multiplication table

When you write an equation that mixes random shocks dW and time increments dt, you need to know what their products are worth. This table is the "arithmetic of stochastic calculus" — the rules every quant uses every single day. Memorise it and you can follow any derivation in any textbook.

Multiply Result Layman meaning
dWₜ × dWₜ dt Two random shocks multiplied = a real slice of time. Volatility leaves a real, measurable mark. This is why uncertainty matters.
dWₜ × dt 0 A random shock multiplied by a tiny time slice is so small it vanishes completely. Ignore it.
dt × dt 0 Time squared is infinitesimally tiny. A second-order effect so small it contributes nothing. Always zero.
Visual multiplication table — what survives and what vanishes in stochastic calculus Three boxes showing dW times dW equals dt with a green checkmark, dW times dt equals zero with a red cross, and dt times dt equals zero with a red cross dW × dW = dt SURVIVES ✓ carries volatility info dW × dt = 0 VANISHES ✗ too small to matter dt × dt = 0 VANISHES ✗ infinitesimally tiny

Only one product survives — dW × dW = dt. That single survivor is the engine of all stochastic calculus.

The reason only the first row survives is that dW is of order √dt (recall: variance grows linearly, so std dev grows as √T). So dW × dW ≈ √dt × √dt = dt. But dW × dt ≈ √dt × dt = dt^(3/2), which vanishes. And dt × dt = dt² vanishes even faster. Only the product that lands on exactly order dt survives the limit.

5 · Mapping it to the price landscape

Your Price Landscape has three dimensions — time, price, and volume. Each dimension moves according to its own rule from our multiplication table. Here is how the arithmetic maps onto each axis of the landscape:

dt
X-Axis: Time
Moves smoothly and predictably. Regular calculus works here. dt is the quiet, steady ticking of the clock — no surprises.
dW
Y-Axis: Price
Moves randomly via Brownian shock. dW is the jiggle — unpredictable at every step. This is where all the excitement (and risk) lives.
(dW)²
Surface Texture
Defined by (dW)² = dt. The roughness of the landscape — not zero, not random — perfectly equal to elapsed time. This is what makes options have value.

The fact that (dW)² ≠ 0 is the only reason options trading exists. If (dW)² were zero — if Brownian Motion were smooth — there would be no uncertainty, no Vega, no volatility to hedge, and no reason to buy protection. The entire multi-trillion dollar derivatives market exists because this one squared term refuses to vanish.

Try it — verify dW² = dt live

Run a Brownian path and compute the accumulated sum of squared increments at every step. Watch it track the elapsed time (the diagonal line) with remarkable precision — this is (dW)² = dt converging in real time. Increase the number of steps to see the convergence get tighter.

500
3
Steps N
QV (should → T=1)
Error |QV − T|
Convergence

Purple/teal lines = accumulated QV for each path  |  Gold dashed = the true value T  |  Paths converge to the gold line as N increases

The four rules of stochastic arithmetic:

Filtration ℱₜ — Your model can only use the past. Any future data that leaks in creates look-ahead bias and invalidates your results.

Martingale 𝔼[W(t)|ℱ(s)] = W(s) — Fair markets have no free drift. If your model predicts one, you've found a signal — or a bug.

(dW)² = dt — The squared shock of Brownian Motion is not zero. It equals real time. This is why volatility scales with √T, and why options have value.

The Multiplication Table — Only dW × dW survives. Everything involving dt² or dW·dt vanishes. This single table is the engine behind Itô's Formula — the subject of the next chapter.

Next Chapter: Realized Volatility →