Chapter 9 · Grand Finale before Itô's Formula

Realized Volatility & The GBM Masterclass

How to read a price chart and extract the hidden "Volume of Chaos" inside it.
The grand finale

You have learned the physics of price — Brownian Motion, GBM, Quadratic Variation, the Filtration, the Martingale. Now it is time to put it all to work. This chapter is where abstract mathematics meets a real NSE price chart. The goal is one thing: to look at a sequence of prices and extract the single hidden number — σ (sigma), the volatility — that describes how violently the market is truly shaking. Everything else, the drift, the trend, even the direction, drops away. Only σ survives.

From a raw price chart, extract sigma — the hidden volume of chaos Three panels — left shows a raw NSE price chart, middle shows the mathematical extraction process, right shows the single output: sigma Raw Price Chart NSE tick data Math Sieve Σ(log returns)² Drift drops away... Cross-talk drops away... Only σ² survives! Output σ Realized Volatility!

Three steps: observe prices → apply the math sieve → extract σ. Drift and noise vanish. Only volatility remains.

1 · The complete GBM formula: two ingredients

We met GBM in Chapter 3. Now we see its full, explicit form — the exact equation that tells you the price at any future time, given where it started. It has exactly two ingredients, and they work in opposite directions: one is steady and predictable, the other is wild and random.

S(t) = S(0) · exp{ σW(t) + (μ − ½σ²)t }
S(0) — The starting price (what you know for certain today)
(μ − ½σ²)t — The Drift: the steady escalator pulling prices in one direction over time
σW(t) — The Shock: the drunken random walk, scaled by volatility σ
exp{ } — The exponential ensures the price can never go negative (Log-Normal)
GBM decomposed into drift escalator and random shock — two forces acting on price simultaneously A price chart showing the pure drift line as a smooth upward curve, with the actual GBM path zigzagging around it due to the shock term (μ − ½σ²)t — Drift S(t) actual — Drift + Shock σW(t) S(0) Time T →

The drift is the smooth escalator. The shock is the drunken wobble around it. GBM is both, simultaneously.

The −½σ² correction in the drift is subtle but critical. Without it, the average price would grow too fast. The −½σ² term is an "Itô correction" — a penalty that accounts for the fact that the log-normal price path's average is slightly pulled down by the convexity of the exponential function. It keeps the expected return honest.

2 · Log returns: the percentage microscope

Instead of watching raw price changes (from $100 to $101), mathematicians look at log returns: the natural logarithm of the ratio between consecutive prices. This seemingly small change of perspective has a powerful consequence — it turns the compounding multiplication of prices into simple addition, which the mathematics can handle cleanly.

log[ S(t_{j+1}) / S(t_j) ]
Why log? Because prices compound multiplicatively — a 10% gain followed by a 10% loss does not get you back to zero. Log returns convert this multiplication into addition, making the Central Limit Theorem apply cleanly.

Each log return is an independent draw from a Normal distribution — which is exactly what we need to apply all the Brownian Motion machinery we have built.

Raw Price Change (additive)

$100 → $110 → $99
Change: +$10, then −$11
The second drop looks bigger, but percentage-wise it's the same. Dollar changes mislead when the base price shifts.

Log Return (multiplicative)

log(110/100) = +9.5%
log(99/110) = −10.5%
Perfectly captures the asymmetry. Log returns sum cleanly across time and are directly comparable regardless of price level.

3 · The math sieve: three pieces, two disappear

When you square the log returns and sum them up, the algebra breaks the result into exactly three pieces. The remarkable thing is that as you use finer and finer time steps — looking at shorter and shorter intervals — two of those pieces vanish to zero. Only one piece survives. And that survivor is σ².

PURE RANDOMNESS
σ² Σ(ΔW)²
As steps → 0, this converges to σ²(T₂−T₁). This piece SURVIVES — it is the signal we want.
DRIFT SQUARED
(μ−½σ²)² Σ(Δt)²
Contains (Δt)² — time squared. As steps → 0, this vanishes to zero. The drift cannot compete with the randomness.
CROSS-TALK
2σ(μ−½σ²) Σ(ΔW·Δt)
Contains ΔW·Δt — shock times time. As steps → 0, this also vanishes to zero. See Multiplication Table, Ch. 8.
Three pieces of the sum of squared log returns — pieces 2 and 3 shrink to zero as resolution increases, piece 1 converges to sigma squared A chart showing three lines representing the three pieces as the number of steps increases. Pieces 2 and 3 fall to zero while piece 1 converges to a flat non-zero value equal to sigma squared times T. Increasing resolution (smaller steps) → Value ① σ²T — SURVIVES ② Drift² → 0 ③ Cross-talk → 0 0

As resolution increases (steps get smaller), Pieces ② and ③ fall to zero. Only Piece ① — the pure volatility signal — survives.

This is the mathematical proof that Realized Volatility works. You can ignore the drift, the trend, even the direction the market is moving — none of it survives the limit. Sum the squared log-returns and you have extracted σ² from the raw price data. Square-root it and you have σ itself — the true heartbeat of the market.

4 · The reality check: microstructure noise

The math says: the finer your time steps, the purer your σ estimate. Use nanosecond data and you get a perfect reading. But in the real world, zooming in too far hits a wall — not a mathematical wall, but a market structure wall.

⚠ The Bid-Ask Bounce Trap

Imagine a stock sits at $100.00 bid / $100.01 ask. Every millisecond, trades alternate: buy at $100.01, sell at $100.00, buy at $100.01, sell at $100.00. Your formula sees: log(100.01/100.00) = +0.01%, then log(100.00/100.01) = −0.01%, alternating every millisecond. The sum of squared returns will be enormous. The math screams that volatility is sky-high. But in reality, the price hasn't moved a single penny. You are not measuring market volatility — you are measuring the bid-ask spread.

The volatility smile — realized volatility estimate versus sampling frequency, showing a U-shape A U-shaped curve showing that sampling too slowly misses volatility while sampling too fast picks up microstructure noise — the sweet spot is in the middle Sampling Frequency — Slow (left) to Fast (right) σ est. SWEET SPOT Too slow: miss real moves Too fast: bid-ask noise ~5 min bars

The "Volatility Signature" — the U-shaped curve every quant uses to find the optimal sampling frequency for their market.

In practice, most quant researchers use 5-minute bar data for intraday volatility estimation on liquid markets like NSE Nifty. This sits in the sweet spot: fast enough to capture real price moves during the day, but slow enough to average out the bid-ask bounce. The optimal frequency varies by stock — liquid large-caps can handle finer sampling than illiquid small-caps.

The family tree: who built this?

The mathematics of Brownian Motion was not invented by one person overnight. It was assembled over a century, by people who had no idea their work would one day power a trillion-dollar derivatives industry.

1828
Robert Brown — The Pollen Problem
Botanist, not a mathematician
Looked through a microscope at pollen grains floating in water. Saw them jiggling uncontrollably. Could not explain why. Named the phenomenon — but had no math for it. The observation sat unexplained for 80 years.
1900
Louis Bachelier — The Stock Market Connection
French mathematician, PhD thesis
Wrote his doctoral thesis on the Paris bond market. Noticed that bond prices moved in a pattern that looked exactly like Brown's pollen. Proposed the first ever mathematical model of a financial market. Was largely ignored for 60 years.
1905
Albert Einstein — The Physics Proof
The same year as Special Relativity
Published the mathematical proof that Brownian Motion is caused by millions of invisible water molecules (our "ping-pong balls") bombarding the pollen grain. Gave the first physical derivation of the diffusion equation. Finance borrowed this directly.
1923
Norbert Wiener — The Rigorous Foundation
MIT mathematician
Proved mathematically that the random process Brown observed actually exists — that such a continuous but everywhere-jagged path can be constructed rigorously. Which is why Brownian Motion is also called the Wiener Process. The entire edifice of stochastic calculus rests on his 1923 paper.
The three laws of a "fair market"

You now have the complete mathematical description of a perfectly fair, efficient market. Three axioms summarise everything studied across all nine chapters. These are the laws of Brownian Motion as a market model — and understanding where real markets deviate from them is where every trading edge begins.

The Three Axioms of a Fair Market

1
Martingale — No Free Drift: 𝔼[W(t) | ℱ(s)] = W(s). The current price is the best prediction of all future prices. There is no predictable direction. No free lunch without risk.
2
Markov — No Useful Memory: The future depends only on the present, not the past. Chart patterns, support levels, and historical averages contain no information not already in the current price.
3
Quadratic Variation — Constant Volatility: (dW)² = dt. The "jiggle" accumulates at a perfectly steady rate. Volatility is not random — it is a fixed physical property of the process.

Why models and reality differ: These three axioms describe an idealised, mathematically perfect market. Real markets are more complicated. Empirical research has documented that prices sometimes exhibit momentum (short-term trends that persist), that investors demand risk premiums above the risk-free rate, and that volatility is not constant — it clusters during crises and calms during stable periods. Understanding these three axioms deeply is what allows a researcher to ask the right questions when they observe real data: Is this pattern a Markov violation? Is this return a genuine risk premium or just noise? Is this volatility spike a temporary cluster or a structural change? The axioms are not limitations — they are the measuring stick against which real market behaviour is compared and understood.

Try it — the realized volatility calculator

Generate a GBM price path with a known σ. Then use the math sieve — summing squared log-returns — to recover σ from the path alone. Try different sampling frequencies and watch the microstructure effect: too fine a resolution adds noise, too coarse misses moves. Find the sweet spot.

20%
10%
20 bars/day
True σ (input)
Realized σ (found)
Estimation error
Drift recovered?

Green = price path  |  Gold dashed = true drift (escalator)  |  The stat chips show how well we recovered σ — ignoring the drift entirely

The Grand Finale summary — nine chapters in four lines:

① Brownian Motion is the mathematical model of randomness: independent, normally distributed, continuous but jagged.

② GBM adds a drift and makes moves proportional to price, giving us a log-normal distribution that can never go negative.

③ Realized Volatility is how we extract σ from a price chart: sum the squared log-returns, and everything except σ² vanishes in the limit.

④ The three axioms (Martingale, Markov, constant QV) define a perfect fair market. Your edge is in the violations. You are now ready for Itô's Formula — the bridge that tells you how any function of price changes when the underlying Brownian Motion moves.

Next Chapter: Itô Integral →