Absolute Continuity: The Foundation of Densities

Hello! Welcome to the first lesson of our course. Our goal is to equip you with the measure-theoretic tools needed to rigorously justify key statistical concepts, specifically for the essay you're planning on conditioning on zero-probability events.

Today, we'll start by tackling a concept that lies at the very heart of this problem: absolute continuity. You're familiar with probability density functions (PDFs) from your statistics and econometrics background, where you integrate a function to find a probability. But what, fundamentally, allows such a density function to exist in the first place? This lesson will answer that question. We will define what it means for one measure to be "absolutely continuous" with respect to another and explore why this is the crucial prerequisite for defining a generalized notion of density. This is the first building block we'll need to construct a formal solution to the conditioning paradox.

What is a "Density," Really?

In your studies, you've frequently worked with probability density functions. For a continuous random variable $X$, its PDF, $f_X(x)$, gives you the probability of $X$ falling into a set $A$ via the integral:

$$P(X\in A)={\int }_A{f}_X(x)dx$$

Notice the $dx$ at the end. This integral is taken with respect to the Lebesgue measure, which is our standard notion of length on the real line. The PDF, $f_X(x)$, can be thought of as telling you the "density" of probability relative to length at the point $x$.

But what if we want to define a density of one probability measure with respect to another, different measure? This happens frequently in finance and econometrics. For example, when moving from the real-world probability measure ($P$) to the risk-neutral measure ($Q$) for pricing derivatives, we are implicitly defining one measure in terms of the other. For this to be possible, we need a more general framework. The starting point for that framework is absolute continuity.

Defining Absolute Continuity

Let's establish a formal definition. The concept is straightforward: a measure $\nu$ is absolutely continuous with respect to another measure $\mu$ if any set that is "invisible" to $\mu$ is also "invisible" to $\nu$.

To see the precise mathematical language, please read the short definition from the Math3ma blog post "Absolute Continuity (Part One)".

As you just read, we write $\nu \ll \mu$ if for every measurable set $A$, the condition $\mu (A)=0$ implies that $\nu (A)=0$.

Think of $\mu$ as a "base" or "reference" measure. Absolute continuity ($\nu \ll \mu$) is a guarantee that the measure $\nu$ doesn't assign positive mass to any set that the reference measure $\mu$ considers to be of size zero. The notation $\ll$ is suggestive—in a way, $\nu$ is "dominated" by $\mu$ because it cannot be positive on sets where $\mu$ is zero.

The Role of Absolute Continuity: A Prerequisite for Density

So, why is this property so important? It turns out to be the essential precondition for a density to exist.

Let's imagine for a moment that a measure $\nu$ does have a density function, let's call it $f$, with respect to a measure $\mu$. This would mean we could write:

$$\nu (A)={\int }_{A}fd\mu$$

Now, consider any set $A$ for which our reference measure is zero, i.e., $\mu (A)=0$. What would the value of $\nu (A)$ be? A fundamental property of the Lebesgue integral (which this generalized integral shares) is that the integral over a set of measure zero is always zero. Therefore, if $\mu (A)=0$, it must be that:

$$\nu (A)={\int }_{A}fd\mu =0$$

This line of reasoning reveals something critical: the very existence of a density $f$ that relates $\nu$ and $\mu$ forces the condition if μ(A)=0, then ν(A)=0. This is exactly the definition of absolute continuity. Therefore, $\nu \ll \mu$ is a necessary condition for $\nu$ to have a density with respect to $\mu$. If a measure $\nu$ is not absolutely continuous with respect to $\mu$, it's simply impossible to define a density for it.

The following video makes this point very concisely.

An Example: When Absolute Continuity Fails

Let's make this concrete with an example where absolute continuity fails, which should be familiar from your work on mixed distributions.

• Let $\mu$ be the Lebesgue measure on the real line $\mathbb{R}$. This is our usual notion of "length."
• Let $\nu$ be the Dirac measure centered at zero, denoted ${\delta }_0$. This measure represents a point mass; it assigns a measure of 1 to any set containing 0, and 0 to any set not containing 0.

Now, consider the set $A=\{0\}$, which contains only the number zero.

• What is the Lebesgue measure of this set? $\mu (A)=\mu (\{0\})=0$. The "length" of a single point is zero.
• What is the Dirac measure of this set? $\nu (A)={\delta }_0(\{0\})=1$.

Here we have a set $A$ where $\mu (A)=0$ but $\nu (A)=1\ne 0$. This violates the definition of absolute continuity, so $\nu$ is not absolutely continuous with respect to $\mu$.

Does it make intuitive sense that we can't define a density for the Dirac measure relative to the Lebesgue measure? Yes. A density function $f(x)$ would need to satisfy ${\int }_{\mathbb{R}}f(x)dx=1$, but $f(x)$ would have to be zero for all $x\ne 0$. No such function exists in the standard sense; you would need an "infinite" spike at zero, which is what the Dirac delta distribution represents, but it is not a function and does not give rise to a density in the measure-theoretic sense we are discussing here.

A Financial Application: Real-World vs. Risk-Neutral Measures

Your background in finance provides a perfect setting to appreciate the importance of this concept. In quantitative finance, asset prices are often modelled under two different probability measures:

1. The real-world measure ($P$): This describes the actual probabilities of future events.
2. The risk-neutral measure ($Q$): This is an artificial construct used for pricing derivatives, where the expected return on all assets is the risk-free rate.

The relationship between these two measures is fundamental. For the standard asset pricing models to work, we require that $P$ and $Q$ are equivalent, which means that $Q\ll P$ and $P\ll Q$. In simple terms, this ensures that an event is impossible in the real world if and only if it is also impossible in the risk-neutral world. This "no-arbitrage" assumption prevents states of the world from being impossible under $P$ but possible under $Q$, which would allow for risk-free profits.

In the discrete case shown in the image, you can see that for any outcome $x$, if its real-world probability $p(x)$ were zero, its risk-neutral probability $q(x)$ would also have to be zero (and vice-versa) for the measures to be equivalent. Since all outcomes have non-zero probability under both measures, they are trivially equivalent.

When $Q\ll P$ holds in a continuous setting, we are guaranteed the existence of a density of $Q$ with respect to $P$. This density, which we will later identify as the Radon-Nikodym derivative $\frac{dQ}{dP}$, is precisely the stochastic discount factor or pricing kernel that you've encountered in financial economics. Absolute continuity is the theoretical bedrock that ensures such a pricing kernel can be formally defined.

For a deeper dive into the formal statement connecting absolute continuity to the existence of a density, the Lebesgue Decomposition Theorem is key.

Conclusion

In this lesson, we have established the first crucial concept for our journey. You should now understand the following key points:

• Absolute continuity ($\nu \ll \mu$) is the formal condition stating that if a set $A$ has zero measure under $\mu$, it must also have zero measure under $\nu$.
• This condition is a necessary prerequisite for being able to define a density of $\nu$ with respect to $\mu$. If it is not met, no such density function can exist.
• This concept generalizes the familiar idea of a PDF (which is a density with respect to Lebesgue measure) and is fundamental in areas like financial economics for relating different probability measures.

We've seen that absolute continuity is necessary for a density to exist. But is it sufficient? The next lesson will answer this question by introducing the Radon-Nikodym theorem. This powerful result will tell us precisely when we can be certain that such a generalized density exists, and it will provide us with the primary tool we need to finally resolve the paradox of conditioning on events of probability zero.