1. EXPANDED LESSON OBJECTIVES (10 TARGETS)

By the end of this lesson, you will be able to:

  • Derive the key properties of continuous and discrete probability distributions used in finance.

  • Compute the Moment Generating Function (MGF) for Normal, Log-normal, and Poisson distributions.

  • Implement the Central Limit Theorem (CLT) to approximate the distribution of portfolio returns.

  • Define and validate the Markov property and the Martingale property for financial time series.

  • Construct Brownian Motion (Wiener process) from a random walk and prove its properties.

  • Derive Ito’s Lemma using a second-order Taylor expansion and apply it to transform stochastic differential equations.

  • Solve the Geometric Brownian Motion (GBM) stochastic differential equation using Ito’s formula.

  • Derive the Black-Scholes-Merton PDE from the GBM dynamics and explain its relationship to the heat equation.

  • Simulate the GBM path using Monte Carlo methods and compute the Euler discretization error.

  • Implement a basic option pricing engine using the risk-neutral valuation framework.

2. ESSENTIAL PROBABILITY DISTRIBUTIONS FOR FINANCE

A. THE NORMAL (GAUSSIAN) DISTRIBUTION:
A random variable X follows a normal distribution X ~ N(μ, σ^2) with PDF:

f_X(x) = [1 / (σ * sqrt(2Ï€))] * exp( -(x – μ)^2 / (2 * σ^2) )

Properties:

  • Mean = μ, Variance = σ^2.

  • Standardized variable Z = (X – μ) / σ ~ N(0, 1).

  • The Moment Generating Function (MGF) is:

M_X(t) = E[e^(tX)] = exp( μ * t + (1/2) * σ^2 * t^2 )

Skewness = 0, Kurtosis = 3. Financial returns, however, exhibit “fat tails” (excess kurtosis) and skewness. This is why the normal distribution is insufficient for risk modeling.

B. THE LOG-NORMAL DISTRIBUTION:
If Y = ln(X) ~ N(μ, σ^2), then X follows a log-normal distribution (used for stock prices, asset returns). The PDF is:

f_X(x) = [1 / (x * σ * sqrt(2Ï€))] * exp( -(ln(x) – μ)^2 / (2 * σ^2) ), for x > 0

Properties:

  • Mean = E[X] = exp( μ + σ^2/2 )

  • Variance = Var(X) = exp(2μ + σ^2) * (exp(σ^2) – 1)
    The log-normal distribution is the foundation of stock price modeling because it ensures prices are positive and percentage returns are normally distributed.

C. THE POISSON DISTRIBUTION (FOR JUMP PROCESSES):
Used to model the number of jumps or arrivals (e.g., number of trades in a second, number of defaults). A Poisson random variable K with rate λ has PMF:

P(K = k) = (e^(-λ) * λ^k) / k!, for k = 0, 1, 2, …

Properties:

  • Mean = λ, Variance = λ.

  • MGF: M_K(t) = exp( λ * (e^t – 1) )

D. THE PARETO DISTRIBUTION (FOR EXTREME EVENTS):
Used in risk management for modeling large losses (extreme value theory). CDF:

F_X(x) = 1 – (x_m / x)^α, for x ≥ x_m, α > 0

Properties:

  • Tail index = α. If α < 2, the variance is infinite. Financial loss data often has α between 1.5 and 2.5, explaining the fat-tailed nature of market crashes.

3. MOMENT GENERATING FUNCTIONS AND CUMULANTS

The MGF is a powerful tool for characterizing distributions and summing independent random variables.

If X_1, X_2, …, X_n are independent random variables, the MGF of the sum S = Σ X_i is the product of the individual MGFs:

M_S(t) = Π_{i=1}^n M_{X_i}(t)

Cumulants: The cumulant generating function is K(t) = ln(M(t)). The first cumulant is the mean, the second is the variance, the third is the skewness, and the fourth is the kurtosis. For a normal distribution, K(t) = μ * t + (1/2) * σ^2 * t^2. All higher cumulants are zero.

This property is used in risk aggregation. If a bank has 100,000 retail loans with independent default probabilities, the distribution of total defaults can be approximated using a normal distribution or a Poisson distribution, depending on the parameter regimes.

4. THE LAW OF LARGE NUMBERS AND THE CENTRAL LIMIT THEOREM

A. LAW OF LARGE NUMBERS (LLN):
For an i.i.d. sequence X_1, X_2, … with mean μ, the sample average converges almost surely to μ:

(1/n) * Σ_{i=1}^n X_i → μ as n → ∞

B. CENTRAL LIMIT THEOREM (CLT):
The CLT is the bedrock of parametric risk models. The distribution of the standardized sample mean converges to a standard normal distribution:

sqrt(n) * [ ( (1/n) * Σ X_i – μ ) / σ ] → N(0, 1)

Proof sketch (using characteristic functions):
Let φ(t) be the characteristic function (the Fourier transform of the PDF) of X_i. For the standardized sum S_n, we expand the log-characteristic function using a Taylor series:

ln(φ_{S_n}(t)) ≈ -t^2/2 + o(1)
As n → ∞, the higher-order terms vanish, leaving the standard normal characteristic function e^(-t^2/2).

Financial Implication: For a large, well-diversified portfolio (n > 30), the portfolio return distribution is approximately normal, regardless of the underlying asset distribution. This is why many banks use the Gaussian Copula for correlation modeling (despite its failure during the 2008 crisis when tail dependencies broke the CLT assumption).

5. STOCHASTIC PROCESSES – MARKOV PROPERTY AND MARTINGALES

A. MARKOV PROPERTY:
A stochastic process {X_t, t ≥ 0} satisfies the Markov property if the conditional distribution of X_{t+1} depends only on the current state X_t, and not on the entire history (X_0, …, X_{t-1}):

P(X_{t+1} ∈ A | X_0, …, X_t) = P(X_{t+1} ∈ A | X_t)

In finance, this means that the future price of an asset depends only on its current price, not on the path it took to get there. This is a fundamental assumption for option pricing (Black-Scholes model).

B. MARTINGALE:
A martingale is a stochastic process where the conditional expectation of the future value, given all past information, is equal to the current value:

E[X_{t+1} | F_t] = X_t

Where F_t is the filtration (information set) up to time t. In risk-neutral pricing, the discounted asset price S_t * e^(-r*t) is a martingale under the risk-neutral measure Q. This ensures there are no arbitrage opportunities.

C. THE DOOB-MEYER DECOMPOSITION:
Any martingale can be decomposed into a continuous part and a jump part. The continuous part is a stochastic integral with respect to Brownian motion.

6. BROWNIAN MOTION (WIENER PROCESS) – THE BUILDING BLOCK

A Wiener process W_t is the continuous-time limit of a simple random walk. It has the following properties:

  1. W_0 = 0.

  2. Independent increments: For any 0 ≤ t_1 < t_2 < t_3, the increments W_{t_2} – W_{t_1} and W_{t_3} – W_{t_2} are independent.

  3. Normal increments: W_t – W_s ~ N(0, t – s).

  4. Continuous paths: The function t → W_t is continuous almost surely (no discontinuities).

CONSTRUCTION FROM A RANDOM WALK:
Let S_n = Σ_{i=1}^n ξ_i, where ξ_i = ±1 with equal probability. The displacement after n steps is S_n. We scale time by Δt = 1/n and space by Δx = sqrt(Δt). The scaled process converges in distribution to Brownian motion as n → ∞:

W_t = lim_{n→∞} (1/sqrt(n)) * S_{nt}

DIFFERENTIAL REPRESENTATION:
We write the infinitesimal increment of Brownian motion as dW_t. The properties are:

E[dW_t] = 0
Var(dW_t) = dt
(dW_t)^2 = dt (This is a critical property for Ito calculus).

QUADRATIC VARIATION:
The quadratic variation of Brownian motion over [0, T] is T:

Σ_{i=0}^{n-1} (W_{t_{i+1}} – W_{t_i})^2 → T

This implies that Brownian motion has infinite first variation but finite quadratic variation, which is why it is non-differentiable in the classical sense.

7. ITO’S LEMMA – THE CHAIN RULE FOR STOCHASTIC CALCULUS

Ito’s Lemma is the most important theorem in financial mathematics. It allows us to compute the differential of a function of a stochastic process.

DERIVATION:
Let X_t follow a stochastic differential equation (SDE):

dX_t = μ(X_t, t) * dt + σ(X_t, t) * dW_t

Consider a twice-differentiable function f(X_t, t). We expand f using a Taylor series up to second order:

df = (∂f/∂t) * dt + (∂f/∂x) * dX_t + (1/2) * (∂^2f/∂x^2) * (dX_t)^2 + (∂^2f/∂x∂t) * dX_t * dt + …

Substitute dX_t and the rule (dW_t)^2 = dt, (dW_t * dt) = 0, and (dt)^2 = 0:

(dX_t)^2 = (μ dt + σ dW_t)^2 = μ^2 (dt)^2 + 2 μ σ dt dW_t + σ^2 (dW_t)^2 = σ^2 dt

Therefore, Ito’s Lemma simplifies to:

df = ( ∂f/∂t + μ * ∂f/∂x + (1/2) * σ^2 * ∂^2f/∂x^2 ) * dt + σ * ∂f/∂x * dW_t

APPLICATION TO GEOMETRIC BROWNIAN MOTION (GBM):
Let the underlying S_t follow GBM:

dS_t = μ * S_t * dt + σ * S_t * dW_t

We want to find the process for f(S_t) = ln(S_t).

  • ∂f/∂t = 0

  • ∂f/∂S = 1/S

  • ∂^2f/∂S^2 = -1/S^2

Applying Ito’s Lemma:

d(ln S_t) = ( 0 + μ * S * (1/S) + (1/2) * σ^2 * S^2 * (-1/S^2) ) * dt + σ * S * (1/S) * dW_t
d(ln S_t) = ( μ – (1/2) * σ^2 ) * dt + σ * dW_t

Integrating from 0 to t:

ln(S_t) – ln(S_0) = ( μ – (1/2) * σ^2 ) * t + σ * W_t

Exponentiating:

S_t = S_0 * exp( ( μ – (1/2) * σ^2 ) * t + σ * W_t )

This is the closed-form solution for GBM. Note the drift correction term ( – (1/2) * σ^2 ). This is a direct consequence of the quadratic variation term in the Taylor expansion.

8. BLACK-SCHOLES-MERTON (BSM) PARTIAL DIFFERENTIAL EQUATION

The BSM PDE is derived by setting up a risk-neutral portfolio that hedges out the stochastic dW_t term.

ASSUMPTIONS:

  • The asset price follows GBM: dS = μ S dt + σ S dW.

  • No transaction costs, no dividends.

  • Risk-free rate r is constant.

  • Markets are frictionless and arbitrage-free.

THE HEDGING PORTFOLIO:
We construct a portfolio Π that contains one derivative option V(S, t) and a short position of Δ shares of the underlying stock. The portfolio value is:

Π = V – Δ * S

We choose Δ = ∂V/∂S (the Delta) to eliminate the stochastic (dW) term.

THE DIFFERENTIAL:
Using Ito’s Lemma on V:

dV = ( ∂V/∂t + μ S ∂V/∂S + (1/2) σ^2 S^2 ∂^2V/∂S^2 ) * dt + σ S ∂V/∂S * dW_t

The portfolio differential is:

dΠ = dV – Δ * dS

Substitute Δ = ∂V/∂S:

dΠ = ( ∂V/∂t + μ S ∂V/∂S + (1/2) σ^2 S^2 ∂^2V/∂S^2 – ∂V/∂S * μ S ) * dt + ( σ S ∂V/∂S – ∂V/∂S * σ S ) * dW_t

The dW_t terms cancel completely. The portfolio is now risk-free, so it must earn the risk-free rate r:

dΠ = r * Π * dt = r * (V – Δ * S) * dt

Equating the deterministic parts:

∂V/∂t + (1/2) * σ^2 * S^2 * ∂^2V/∂S^2 = r * (V – S * ∂V/∂S)

Rearranging gives the Black-Scholes PDE:

∂V/∂t + (1/2) * σ^2 * S^2 * ∂^2V/∂S^2 + r * S * ∂V/∂S – r * V = 0

SOLVING THE PDE (HEAT EQUATION TRANSFORMATION):
The BSM PDE can be transformed into the heat equation (diffusion equation) by a change of variables:

  • Let S = e^x

  • Let V = e^{αx + βt} * u(x, t)

The PDE reduces to:

∂u/∂τ = ∂^2u/∂x^2

Which has the fundamental solution using the Fourier transform. The closed-form solution for a European Call option is the famous Black-Scholes formula:

C = S_0 * N(d_1) – K * e^{-rT} * N(d_2)

Where:

d_1 = [ ln(S_0 / K) + (r + σ^2/2) * T ] / (σ * sqrt(T))
d_2 = d_1 – σ * sqrt(T)

And N(·) is the cumulative distribution function of the standard normal distribution. This formula is implemented in every FinTech trading system.

9. MONTE CARLO SIMULATION OF GBM (DISCRETIZATION)

In FinTech, closed-form solutions are not always available (e.g., for path-dependent options like Asian or Barrier options). We use the Monte Carlo method to simulate thousands of paths.

THE EULER-MARUYAMA DISCRETIZATION:
We discretize time into N steps of size Δt = T / N. The SDE is approximated as:

S_{t+Δt} = S_t + μ * S_t * Δt + σ * S_t * sqrt(Δt) * Z

Where Z ~ N(0, 1). This is the Euler-Maruyama scheme.

CONVERGENCE AND ERROR:
The strong order of convergence is 0.5, meaning the error is O(Δt^(1/2)). The weak order (for expectation) is 1.0, meaning the error is O(Δt).

MILSTEIN SCHEME (IMPROVED ACCURACY):
To improve convergence, we use the Milstein scheme, which adds an extra term from the Ito-Taylor expansion:

S_{t+Δt} = S_t + μ * S_t * Δt + σ * S_t * sqrt(Δt) * Z + (1/2) * σ^2 * S_t * Δt * (Z^2 – 1)

The strong order of Milstein is 1.0, which is superior for path-dependent options.

VARIANCE REDUCTION TECHNIQUES:

  • Antithetic Variates: For each random path, generate a second path using -Z. This reduces variance by exploiting symmetry.

  • Control Variates: Use a known analytical solution (e.g., the Black-Scholes price) as a control variate to reduce the simulation error.

10. RISK-NEUTRAL VALUATION AND ARBITRAGE-FREE PRICING

The fundamental theorem of asset pricing states that in an arbitrage-free market, there exists a risk-neutral measure Q such that the discounted price process is a martingale.

Under Q, the drift of the asset is the risk-free rate r, not the real-world drift μ:

dS_t = r * S_t * dt + σ * S_t * dW_t^Q

The option price is the expected present value of the payoff under Q:

V(S_0, 0) = e^{-rT} * E^Q[ Payoff(S_T) ]

For a European Call, the payoff is max(S_T – K, 0). Substituting the GBM solution and integrating over the log-normal distribution yields the Black-Scholes formula.

NUMERICAL IMPLEMENTATION IN FINTECH:
The systems use the QuantLib library or custom C++/Python engines that implement the Monte Carlo framework with parallel processing (multithreading on GPUs). A single Monte Carlo engine can simulate 1,000,000 paths in under 1 second for a 252-step annual simulation. The standard error of the estimate is:

SE = σ_Payoff / sqrt(N_Paths)

If σ_Payoff is 20 and we run 10,000 paths, the standard error is 0.2 (roughly 20 cents per dollar). For high precision, 500,000 paths are typically used.

 
PreviousNext