Quadratic Variation: Brownian Motion & Poisson Process

Hey guys! Ever found yourself staring at a complex stochastic process and wondering how to measure its 'wiggliness' or volatility over time? Well, you're in the right place! Today, we're diving deep into the fascinating world of quadratic variation, specifically focusing on a killer combo: the mixed Brownian Motion and Poisson Process. If you're grappling with calculating this, especially if you're thinking Ito's formula is the key (spoiler: it often is!), stick around. We're going to break it down, make it understandable, and hopefully, get you that 'aha!' moment.

Understanding the Building Blocks: Brownian Motion and Poisson Process

Before we get our hands dirty with the mixed process, let's quickly recap what these individual components bring to the table. You've probably heard of Brownian Motion, often denoted as $W_t$ or $B_t$. Think of it as the mathematical model for random walks, the erratic movement of particles suspended in a fluid. Its key characteristic is that its increments are independent and normally distributed. But for quadratic variation, the real magic lies in its path: it's continuous and nowhere differentiable. This 'jaggedness' is precisely what quadratic variation aims to quantify. The quadratic variation of a standard Brownian Motion over an interval $[0, T]$ is simply $T$. Pretty neat, right? It tells us that the total 'squared jump' or 'squared wiggle' accumulates linearly with time. This simple yet profound result is fundamental in many areas, from finance to physics.

Now, let's switch gears to the Poisson Process, often denoted as $N_t$. This guy is all about counting events that happen randomly over time. Think of customer arrivals at a store, calls to a call center, or even radioactive decay. The Poisson process is characterized by its jumps. Each event causes an instantaneous increase (usually by 1) in the process's value. Unlike Brownian Motion, the Poisson process is discontinuous. The quadratic variation of a standard Poisson Process $N_t$ with jump size 1 over an interval $[0, T]$ is also $T$. This might seem counter-intuitive at first glance – how can a process with discrete jumps have the same quadratic variation as a continuous one? The trick here is that while there are only a finite number of jumps in any finite interval, each jump contributes $1^2 = 1$ to the sum of squared increments. Over a long period, the total count of jumps accumulates, and the sum of these squared jumps averages out to be proportional to time. So, both processes, despite their vastly different natures, share this linear growth in quadratic variation.

The Intriguing Mix: Combining Brownian Motion and Poisson Process

So, what happens when we mix these two distinct processes? We're essentially creating a new stochastic process, let's call it $X_t$, which is a combination of both Brownian motion and a Poisson process. A common way to define such a mixed process is as follows:

$X_t = \sigma W_t + J N_t$

Here, $W_t$ is a standard Brownian Motion, $N_t$ is a standard Poisson Process, and $\sigma$ and $J$ are constants. $\sigma$ scales the volatility of the Brownian motion part, and $J$ represents the size of the jump in the Poisson process part. This kind of process is super useful in modeling phenomena where you have a continuous background fluctuation (like stock prices) overlaid with discrete, sudden shocks (like major news events or option exercises). The challenge, and the core of our problem, is to figure out the quadratic variation of this combined process, $X_t$. How does the presence of both continuous randomness and discrete jumps affect the total 'squared fluctuation' over time?

Calculating the quadratic variation of $X_t$ over an interval, say $[0, T]$, involves summing the squares of the increments of $X_t$ over small partitions of the interval. Let $\Pi = \{0 = t_0 < t_1 < \dots < t_n = T\}$ be a partition of $[0, T]$. The quadratic variation is formally defined as the limit (in probability) of the sum of squared increments as the mesh of the partition goes to zero:

$Q V(X)_T = \lim_{\text{mesh}(\Pi) \to 0} \sum_{i=1}^n (X_{t_i} - X_{t_{i-1}})^2$

When we substitute our definition of $X_t$, the increment $(X_{t_i} - X_{t_{i-1}})$ becomes:

$(X_{t_i} - X_{t_{i-1}}) = \sigma (W_{t_i} - W_{t_{i-1}}) + J (N_{t_i} - N_{t_{i-1}})$

Now, squaring this increment gives:

$(X_{t_i} - X_{t_{i-1}})^2 = \left[ \sigma (W_{t_i} - W_{t_{i-1}}) + J (N_{t_i} - N_{t_{i-1}}) \right]^2$

Expanding this, we get:

$(X_{t_i} - X_{t_{i-1}})^2 = \sigma^2 (W_{t_i} - W_{t_{i-1}})^2 + 2 \sigma J (W_{t_i} - W_{t_{i-1}})(N_{t_i} - N_{t_{i-1}}) + J^2 (N_{t_i} - N_{t_{i-1}})^2$

To find the quadratic variation of $X_t$, we need to sum these squared increments and take the limit. This is where the properties of Brownian Motion and Poisson Process become crucial. We need to consider the limit of the sum of each of these three terms separately.

Applying Ito's Formula and the Magic of Orthogonality

Alright guys, this is where the heavy lifting happens, and Ito's formula is our trusty sidekick. If you're feeling a bit lost, remember that Ito's formula helps us deal with the calculus of stochastic processes, especially when they involve things like Brownian Motion. For a process $f(t, Y_t)$, where $Y_t$ is a stochastic process, Ito's formula gives us a way to find its differential $df$. In our case, we are interested in the quadratic variation, which is intrinsically linked to the 'squared increments'.

Let's consider the terms we got from squaring the increment of $X_t$: $\sigma^2 (W_{t_i} - W_{t_{i-1}})^2$, $2 \sigma J (W_{t_i} - W_{t_{i-1}})(N_{t_i} - N_{t_{i-1}})$, and $J^2 (N_{t_i} - N_{t_{i-1}})^2$. We need to find the limit of the sum of each term as the partition gets finer.

1. The Brownian Motion Term: $\sum \sigma^2 (W_{t_i} - W_{t_{i-1}})^2$. As we discussed earlier, the quadratic variation of Brownian Motion is $T$. So, the limit of this sum is $\sigma^2 T$. This part is straightforward and directly comes from the definition of quadratic variation for $W_t$.

2. The Poisson Process Term: $\sum J^2 (N_{t_i} - N_{t_{i-1}})^2$. For a standard Poisson process with jumps of size 1, the increment $(N_{t_i} - N_{t_{i-1}})$ is either 0 or 1. If it's 1, it means an event occurred within that small interval. If it's 0, no event occurred. So, $(N_{t_i} - N_{t_{i-1}})^2$ is also either 0 or 1 (since $1^2=1$ and $0^2=0$). The sum $\sum (N_{t_i} - N_{t_{i-1}})^2$ is essentially counting the number of jumps in the interval $[0, T]$. For a Poisson process with intensity $\lambda$, the expected number of jumps in $[0, T]$ is $\lambda T$. The quadratic variation, in the limit, converges to $\lambda T$. In our case, assuming a standard Poisson process with intensity $\lambda=1$, this term converges to $J^2 T$. If the jump size was different, say $J$, then $(N_{t_i} - N_{t_{i-1}})$ would be $J$ if an event occurs and 0 otherwise. Then $(N_{t_i} - N_{t_{i-1}})^2$ would be $J^2$ if an event occurs and 0 otherwise. The sum of these squared jumps would then converge to $J^2 imes ( ext{number of jumps})$, leading to $J^2 imes ext{intensity} imes T$. Assuming a standard Poisson process with intensity $\lambda$, this term yields $J^2 \lambda T$. If we assume a standard Poisson process where each jump is of size 1, the intensity is usually denoted by $\lambda$. So the quadratic variation of $J N_t$ is $J^2 \lambda T$. If we assume a standard Poisson process with intensity $\lambda=1$ and jump size 1, this gives $J^2 T$.

3. The Mixed Term: $\sum 2 \sigma J (W_{t_i} - W_{t_{i-1}})(N_{t_i} - N_{t_{i-1}})$. This is the most interesting part. Remember that a jump in the Poisson process is an instantaneous event. In any sufficiently small interval $(t_{i-1}, t_i)$, there can be at most one jump in the Poisson process. If a jump occurs, $(N_{t_i} - N_{t_{i-1}})$ is either 0 or 1 (assuming unit jumps). If no jump occurs, $(N_{t_i} - N_{t_{i-1}})$ is 0. Crucially, if a jump occurs in $(t_{i-1}, t_i)$, then $(W_{t_i} - W_{t_{i-1}})$ is a random variable, but it's independent of the event of the jump itself. However, when we consider the limit of the sum, the term $(W_{t_i} - W_{t_{i-1}})$ is multiplied by $(N_{t_i} - N_{t_{i-1}})$. If $(N_{t_i} - N_{t_{i-1}})$ is 0 (no jump), the whole product is 0. If $(N_{t_i} - N_{t_{i-1}})$ is 1 (jump occurs), the product is $(W_{t_i} - W_{t_{i-1}})$. So, the sum effectively becomes $2 \sigma J \sum_{\text{intervals with jumps}} (W_{t_i} - W_{t_{i-1}})$.

Now, here's the key insight: the increments of Brownian Motion $(W_{t_i} - W_{t_{i-1}})$ are normally distributed with mean 0 and variance $(t_i - t_{i-1})$. The expectation of $(W_{t_i} - W_{t_{i-1}})$ is 0. When we sum these terms only over intervals where a Poisson jump occurs, the expectation of this sum tends to zero as the mesh size goes to zero. The crucial part here is that the Brownian motion and the Poisson process are independent. The probability of a jump occurring in a small interval is proportional to the length of the interval, while the Brownian increment is centered around zero. As the interval shrinks, the contribution from this mixed term vanishes in the limit for the quadratic variation. Think about it: the Brownian motion's 'wiggle' is smooth, while the Poisson process's 'wiggle' comes in sharp, instantaneous spikes. When you multiply a smooth random fluctuation by an indicator of a spike (which is rare in small intervals), the product's contribution to the squared variation tends to zero. So, the limit of this sum is 0.

This orthogonality, stemming from the independence and the nature of the processes, is why Ito's formula works so elegantly. Ito's lemma for a process like $X_t = \int_0^t \sigma_s dW_s + \int_0^t J_s dN_s$ (where $\sigma_s$ and $J_s$ could be functions of time or other processes) would yield the quadratic variation as the sum of the quadratic variations of its independent components. In our simplified case, $X_t = \sigma W_t + J N_t$. The differential $dX_t = \sigma dW_t + J dN_t$. Applying the rules of quadratic covariation (which stem from Ito calculus), we have:

$d<X>_t = d<\sigma W_t + J N_t>_t$

Using the properties of quadratic variation and covariation:

$d<X>_t = d<\sigma W_t>_t + d<J N_t>_t + 2 d<\sigma W_t, J N_t>_t$

Since $W_t$ and $N_t$ are independent, their quadratic covariation $d<\sigma W_t, J N_t>_t = \sigma J d<W_t, N_t>_t = 0$.

The quadratic variation of $\sigma W_t$ is $\sigma^2 dt$. The quadratic variation of $J N_t$ is $J^2 d<N_t>$. For a standard Poisson process with intensity $\lambda$, $d<N_t> = \lambda dt$. Thus, $d<J N_t>_t = J^2 \lambda dt$.

Putting it all together, the quadratic variation process $d<X>_t$ is:

$d<X>_t = \sigma^2 dt + J^2 \lambda dt = (\sigma^2 + J^2 \lambda) dt$

Integrating from 0 to $T$, we get the total quadratic variation over the interval $[0, T]$:

$Q V(X)_T = <X>_T = \int_0^T (\sigma^2 + J^2 \lambda) dt = (\sigma^2 + J^2 \lambda) T$

This confirms our earlier intuition! The quadratic variation of the mixed process is simply the sum of the quadratic variations of its independent components, scaled by the constants $\sigma^2$ and $J^2$, and the intensity $\lambda$ of the Poisson process.

Final Result and Implications

So, after all that math, what's the punchline? The quadratic variation of a mixed Brownian Motion and Poisson Process, defined as $X_t = \sigma W_t + J N_t$, over the time interval $[0, T]$ is:

$$Q V(X)_T = (\sigma^2 + J^2 \lambda) T$$

Where:

• $\sigma^2$ is the variance parameter of the Brownian Motion part (it's $\sigma^2$ times the quadratic variation of $W_t$, which is $T$).
• $J^2$ is the square of the jump size of the Poisson Process.
• $\lambda$ is the intensity (or rate) of the Poisson Process.
• $T$ is the length of the time interval.

This result is incredibly insightful. It tells us that the total 'squared roughness' of the process $X_t$ is composed of two parts: one part comes from the continuous fluctuations of the Brownian Motion (proportional to $\sigma^2 T$), and the other part comes from the discrete jumps of the Poisson Process (proportional to $J^2 \lambda T$). The independence of the Brownian Motion and the Poisson Process is key here; it allows us to simply add their contributions. This makes intuitive sense: the 'jumpiness' adds to the overall variability, and the total squared variability is the sum of variability from continuous movements and discrete jumps.

Why is this important, you ask? Well, in fields like quantitative finance, understanding the quadratic variation is crucial for pricing options and managing risk. For instance, in models where asset prices might experience both continuous diffusion (modeled by Brownian Motion) and sudden jumps (due to news or market shocks, modeled by Poisson Processes), this formula helps us capture the total risk profile. The $\sigma^2 T$ term relates to the volatility risk from continuous trading, while the $J^2 \lambda T$ term accounts for the jump risk. Being able to calculate this accurately allows for more robust risk management and better pricing of financial derivatives.

Furthermore, this concept extends to more complex scenarios. If the jump sizes were random, or if the intensity $\lambda$ varied with time, the calculation would become more involved, potentially requiring stochastic calculus on a larger probability space. However, the fundamental principle of decomposing the quadratic variation into contributions from continuous and jump components remains. The application of Ito's formula and the properties of stochastic integrals are central to solving these problems. So, next time you see a process with both smooth drifts and sudden jumps, you'll know how to quantify its overall 'roughness' using quadratic variation. Keep exploring, keep calculating, and happy stochastic modeling, guys!