The Art of Term Structure Models: Drift

Welcome to the World of Interest Rate Modeling!

In FRM Part I, you learned how to calculate the price of a bond. In Part II, we take a giant leap forward. We aren't just looking at where interest rates are today; we are trying to model where they might go in the future. This chapter, "The Art of Term Structure Models: Drift," focuses on the "trend" or the "average direction" of interest rates.

Understanding these models is crucial for managing market risk because if you can't model how the yield curve moves, you can't accurately price complex derivatives or manage the risk of a bond portfolio. Don't worry if the math looks intimidating at first—we will break it down piece by piece!

1. The Foundation: What is a Term Structure Model?

At its heart, a short-rate model tries to describe how the instantaneous risk-free rate (\( r_t \)) changes over a tiny sliver of time (\( dt \)). Most of these models follow a standard "recipe":

Change in Rate = (Drift Component) + (Stochastic/Random Component)

Mathematically, it looks like this:
\( dr_t = \mu(r_t, t) dt + \sigma(r_t, t) dW_t \)

• \( dr_t \): The change in the interest rate.
• \( \mu(r_t, t) \): The Drift. This is the "expected" path. If there were no randomness, this is where the rate would go.
• \( \sigma(r_t, t) \): The Volatility. This scales how much "noise" or "uncertainty" there is.
• \( dW_t \): The Wiener Process. Think of this as a random shock (like a surprise news event).

Analogy: Driving a Car
Imagine you are driving a car on a windy day.
- The Drift is your steering wheel and cruise control—it's where you intend to go.
- The Stochastic Component is the wind pushing you side-to-side. You know the wind will blow, but you don't know exactly when or how hard.

Quick Review: Drift tells us the "average" direction, while volatility tells us how much the rate "wiggles" around that average.

2. Model 1: The Ho-Lee Model (Constant Drift)

The Ho-Lee Model was the first "arbitrage-free" model. It assumes that the drift is time-dependent but does not depend on the current level of the interest rate.

The Formula:
\( dr_t = \theta(t) dt + \sigma dW_t \)

Key Features:
1. Time-Dependent Drift (\( \theta(t) \)): This allows the model to be "calibrated" so that it perfectly matches the current market yield curve.
2. Constant Volatility: The "wiggles" are the same regardless of whether rates are 1% or 10%.
3. Normal Distribution: Rates are normally distributed in this model.

The Downside (Common Mistake):
Because the drift is constant and doesn't "pull" the rate back to a center, the Ho-Lee model allows interest rates to become negative quite easily. While negative rates have happened in the real world, this model lacks a "memory" of where rates should normally be.

3. Model 2: The Vasicek Model (Mean Reversion)

The Vasicek Model introduced a game-changing concept: Mean Reversion. In the real world, if interest rates get extremely high, they eventually tend to come back down. If they are extremely low, they eventually drift back up.

The Formula:
\( dr_t = k(\theta - r_t) dt + \sigma dW_t \)

Breaking down the Drift term \( k(\theta - r_t) \):
- \( \theta \): The Long-run Mean. This is the "target" level where rates want to settle.
- \( r_t \): The Current Rate.
- \( k \): The Speed of Reversion. This determines how fast the rate is pulled back toward \( \theta \).

Memory Aid: The Rubber Band
Think of \( \theta \) as a pole and the interest rate as a ball attached to it by a rubber band.
- If the ball (\( r_t \)) moves far away from the pole (\( \theta \)), the rubber band pulls it back.
- The stiffness of the rubber band is \( k \). If \( k \) is large, the rate snaps back quickly!

Key Takeaway: Vasicek is better than Ho-Lee because it reflects the economic reality that rates don't wander off to infinity. However, like Ho-Lee, it still uses a normal distribution, meaning negative rates are still possible.

4. Model 3: Cox-Ingersoll-Ross (CIR) Model

The CIR Model looks very similar to Vasicek, but it fixes one major problem: the possibility of negative rates (mostly).

The Formula:
\( dr_t = k(\theta - r_t) dt + \sigma \sqrt{r_t} dW_t \)

What changed?
Notice the volatility term: \( \sigma \sqrt{r_t} \). In CIR, the volatility depends on the level of the interest rate.
- If the interest rate (\( r_t \)) approaches zero, the volatility (\( \sigma \sqrt{r_t} \)) also approaches zero.
- This makes it very difficult (and impossible, depending on parameters) for the rate to drop below zero.

Did you know?
The CIR model is often called a "Square Root Model" because of that \( \sqrt{r_t} \) term. It produces a non-central chi-squared distribution rather than a normal distribution.

5. Model 4: The Lognormal Model (Black-Karasinski)

Some practitioners prefer to model the natural log of the interest rate rather than the rate itself. This is the Black-Karasinski Model.

The Concept:
By modeling \( d \ln(r_t) \), we ensure that the interest rate \( r_t \) can never be negative. Mathematically, the exponential of any real number is always positive (\( e^x > 0 \)).

Why use it?
It prevents negative rates and allows the volatility to be proportional to the rate level (similar to how stock prices are modeled). It is very popular in building "trees" for American-style options.

Summary Table of the 4 Key Models

1. Ho-Lee: Simple drift, constant vol, rates can be negative.
2. Vasicek: Mean reversion, constant vol, rates can be negative.
3. CIR: Mean reversion, vol depends on \(\sqrt{r}\), rates stay positive.
4. Lognormal: Models \(\ln(r)\), rates stay positive, no closed-form solutions for bond prices.

6. Risk-Neutral vs. Real-World Drift

This is a point where many students get tripped up. There are two "worlds" in finance:

1. The Real World (P-Measure):
This is what we actually expect to happen. It includes a risk premium. Investors demand a higher return for taking on the risk that rates might change.

2. The Risk-Neutral World (Q-Measure):
This is a mathematical construct used for pricing. In this world, we assume investors don't care about risk, so all assets earn the risk-free rate. We adjust the "drift" of our models to account for this.

The Connection: The Market Price of Risk (\( \lambda \))
To move from the Real World to the Risk-Neutral World, we subtract the risk premium from the drift:
\( \text{Risk-Neutral Drift} = \text{Real-World Drift} - (\text{Market Price of Risk} \times \text{Volatility}) \)

Key Takeaway: For the FRM exam, remember that we use Risk-Neutral Drift when we want to price a derivative, but we use Real-World Drift if we are trying to forecast or calculate Value-at-Risk (VaR).

7. Final Tips for Success

Don't let the notation scare you. When you see a formula in this chapter, ask yourself three questions:
1. Does it have mean reversion? (Look for \( k(\theta - r) \))
2. Can rates go negative? (If the vol term doesn't have an \( r \) or \( \sqrt{r} \) in it, they probably can).
3. Is the drift constant or time-varying? (Time-varying drift allows the model to fit today's yield curve perfectly).

Common Pitfall: Students often confuse "Mean Reversion" with "Drift." Remember: Mean reversion is a type of drift. Not all drift models have mean reversion (like Ho-Lee), but all mean-reverting models have a drift component!

Keep practicing the qualitative differences between these models, as the FRM exam often tests your understanding of which model is appropriate for a specific scenario rather than just asking you to crunch numbers!