The Radon-Nikodym Theorem and Its Derivative

Welcome to your second lesson. In our previous discussion, we established that for one measure ν to have a density with respect to another measure μ, it is necessary for ν to be absolutely continuous with respect to μ (written ν << μ). This means that any set deemed to have zero size by μ must also have zero size by ν.

This naturally leads to the next question: is this condition also sufficient? If we know that ν << μ, can we be certain that a density function exists? This lesson introduces the celebrated Radon-Nikodym theorem, which provides a definitive and powerful answer. We will state the theorem, identify its precise conditions, and interpret its result—a generalized density called the Radon-Nikodym derivative. This theorem is the central piece of machinery we need to build a rigorous framework for conditioning.

The Radon-Nikodym Theorem: From Condition to Guarantee

The Radon-Nikodym theorem elevates absolute continuity from a mere necessary condition to a nearly sufficient one. It tells us that, under one additional, very mild assumption, absolute continuity guarantees the existence of the density function we are looking for.

Let's watch a short, clear explanation of the theorem's statement.

As you just saw, the theorem can be stated as follows:

The Radon-Nikodym Theorem: Let μ and ν be two σ-finite measures on a measurable space (X, Σ). If ν is absolutely continuous with respect to μ (ν << μ), then there exists a non-negative, Σ-measurable function f : X → [0, ∞) such that for any measurable set A ∈ Σ:

This function f is called the Radon-Nikodym derivative of ν with respect to μ and is denoted by dν/dμ. Furthermore, this function is unique almost everywhere with respect to μ.

Let's break down the components of this statement.

The Conditions

1. σ-finiteness: This is the new condition. A measure μ is σ-finite if the entire space X can be covered by a countable collection of measurable sets, each of which has a finite measure under μ. Your background in probability and time-series analysis deals almost exclusively with σ-finite measures. For instance, the Lebesgue measure on ℝ is σ-finite because ℝ can be covered by the countable union of intervals [-n, n], each of which has finite length. Probability measures are σ-finite by definition, as the measure of the entire sample space is 1. This is a technical condition that is satisfied in almost all practical applications in statistics and econometrics.

2. Absolute Continuity: This is the ν << μ condition from our last lesson. The theorem confirms its central importance.

The Result: Existence and Uniqueness

The theorem's conclusion is profound. It doesn't just suggest a density might exist; it guarantees it. The function f = dν/dμ acts as a generalized density, allowing us to express the measure ν of any set A by integrating f over that set with respect to the reference measure μ.

The "uniqueness almost everywhere" clause means that if there are two functions, f and g, that both satisfy the theorem, then the set on which they differ, {x ∈ X | f(x) ≠ g(x)}, must have a μ-measure of zero. For all practical purposes in integration, f and g are interchangeable.

For a more formal statement, presented in the direct context of probability theory, please review the following resource.

The Duke notes frame the theorem slightly differently, leading to the conditional expectation E[X|G]. We will explore this connection in great detail in a future lesson. For now, the key takeaway is that given a base measure (P in the notes) and another measure defined from it (λ in the notes), the theorem provides a function (Y) that acts as the density of λ with respect to P.

A Powerful Application: The Pricing Kernel

Your background in corporate finance and derivative securities provides a perfect setting to see the Radon-Nikodym theorem in action. In asset pricing, the concept of the Stochastic Discount Factor (SDF), or pricing kernel, is central. The SDF is the random variable that connects asset prices to their expected future payoffs.

The pricing equation you've seen many times is:
$P_t=E_t^P[m_{t,t+1}{X}_{t+1}]$
where P denotes the expectation is taken under the real-world (or "physical") probability measure.

Alternatively, in a risk-neutral world, prices are simply discounted expected payoffs under a different measure, Q:
$P_t=e^{-r}{E}_t^Q[X_{t+1}]$

How do we formally connect these two pictures? The Radon-Nikodym theorem is the key. The SDF, m, is precisely the Radon-Nikodym derivative that allows us to "change measure" from P to Q. Specifically, m is proportional to dQ/dP.

The following research paper on estimating the SDF relies explicitly on this theoretical foundation.

As you read, the paper makes the relationship explicit:

Here, dF^Q and dF^P represent the measures induced by the risk-neutral and physical distribution functions. The pricing kernel that bridges the real and risk-neutral worlds is a Radon-Nikodym derivative.

For this relationship to hold, we must have Q << P (the risk-neutral measure is absolutely continuous with respect to the physical measure). This is a cornerstone "no-arbitrage" condition in finance: any event that has zero probability of occurring in the real world must also have zero probability in the risk-neutral pricing model. If this were not true, you could construct an arbitrage. The Radon-Nikodym theorem then guarantees that the pricing kernel dQ/dP exists.

This provides a powerful, tangible example of the theorem's importance. It's not just an abstract mathematical curiosity; it's the formal foundation for the change-of-measure techniques that are fundamental to modern financial economics.

Conclusion

In this lesson, we have moved from the necessary condition of absolute continuity to the guarantee provided by the Radon-Nikodym theorem. Here are the key takeaways:

• The Radon-Nikodym Theorem: Given two σ-finite measures ν and μ, if ν << μ, then a generalized density f = dν/dμ is guaranteed to exist.
• The Radon-Nikodym Derivative (dν/dμ): This is the formal name for the function f. It generalizes the concept of a density to arbitrary reference measures. It is unique up to a set of μ-measure zero.
• Application in Finance: The Stochastic Discount Factor (SDF) or pricing kernel is a prime example of a Radon-Nikodym derivative, specifically dQ/dP, which facilitates the change of measure from the physical world to the risk-neutral world for asset pricing.

You now have the two core theoretical components in place: the definition of absolute continuity and the theorem that leverages it to guarantee a density. In our next lesson, we will solidify this understanding by connecting the abstract Radon-Nikodym derivative back to the familiar probability density function (PDF) from your statistics courses. We will show that a standard PDF is simply the Radon-Nikodym derivative of a probability measure with respect to the Lebesgue measure.