Welcome to State-Dependent Probabilities!
Hello there, future Actuary! In previous chapters, you likely looked at simple survival models where someone is either "Alive" or "Dead." In this chapter, we are leveling up. We are exploring Continuous-Time Markov Models where a person can move between several different states (like "Healthy," "Sick," and "Dead") and where the chance of moving depends on their current age and state.
This might sound intimidating, but think of it like a GPS for life insurance. We are trying to map out all the possible roads a person can take and calculate the probability they end up at a specific destination at a specific time. Let’s dive in!
1. The Basics: What is a State-Dependent Model?
In the real world, risks aren't just "on" or "off." A person might be "Active" in a pension plan, then "Disabled," then perhaps "Recovered," and eventually "Deceased."
In a Continuous-Time Markov Model, we assume:
1. The future depends only on the current state (this is the "Markov Property").
2. Transitions can happen at any tiny moment in time.
3. The probability of moving depends on the age of the individual, which we denote as \( x+t \).
Key Term: Transition Intensity
The transition intensity, denoted as \( \mu_{x+t}^{ij} \), is the "instantaneous" rate of moving from state \( i \) to state \( j \) at age \( x+t \).
Analogy: Think of transition intensity like the speedometer on your car. If your speedometer says 60 mph, it doesn't mean you've traveled 60 miles; it means that at this exact moment, that is the rate you are moving. Similarly, \( \mu_{x+t}^{ij} \) is the speed at which people are "flowing" from state \( i \) to state \( j \).
2. Transition Probabilities: \( {}_t p_x^{ij} \)
While intensities are "speeds," transition probabilities are the actual likelihood of being in a certain state after a certain amount of time.
\( {}_t p_x^{ij} \) = The probability that a person who is in state \( i \) at age \( x \) will be in state \( j \) at age \( x+t \).
Important Distinction:
- \( {}_t p_x^{ii} \): The probability of being in state \( i \) at time \( t \), given you started in \( i \). (You could have left and come back!)
- \( {}_t p_x^{\bar{ii}} \): The probability of staying in state \( i \) continuously from time \( 0 \) to \( t \) without ever leaving.
Quick Formula for Staying:
\( {}_t p_x^{\bar{ii}} = \exp\left( -\int_0^t \sum_{k \neq i} \mu_{x+s}^{ik} ds \right) \)
Memory Aid: This looks just like the survival formula \( {}_t p_x = e^{-\int \mu ds} \)! We just sum up all the ways to "leave" state \( i \).
Summary Table: Probabilities vs. Intensities
Intensity (\( \mu \)): The rate of change (the "speed").
Probability (\( p \)): The likelihood of being somewhere (the "location").
3. The Kolmogorov Forward Equations
Don't let the name scare you! These equations are just a way to describe how probabilities change over a tiny sliver of time.
For any state \( j \), the rate of change in the probability \( \frac{d}{dt} {}_t p_x^{ij} \) is calculated by looking at the Inflow minus the Outflow.
The General Equation:
\( \frac{d}{dt} {}_t p_x^{ij} = \sum_{k \neq j} \left( {}_t p_x^{ik} \mu_{x+t}^{kj} \right) - {}_t p_x^{ij} \sum_{k \neq j} \mu_{x+t}^{jk} \)
Let's break that down:
1. Inflow: You sum up the probabilities of being in any other state \( k \) and then transitioning to state \( j \).
2. Outflow: You subtract the probability of already being in state \( j \) and then transitioning away from it to any other state \( k \).
Example: The Healthy-Sick-Dead Model
If state 0 = Healthy, 1 = Sick, and 2 = Dead, the change in the probability of being sick (\( {}_t p_x^{01} \)) is:
\( \frac{d}{dt} {}_t p_x^{01} = \underbrace{{}_t p_x^{00} \mu_{x+t}^{01}}_{\text{Healthy to Sick}} - \underbrace{{}_t p_x^{01} (\mu_{x+t}^{10} + \mu_{x+t}^{12})}_{\text{Sick to Healthy or Dead}} \)
4. Solving the Equations (Euler’s Method)
In the ALTAM exam, you often can't solve these equations perfectly with calculus. Instead, you use Euler's Method to approximate the next step.
Step-by-Step Euler Approximation:
1. Start with your current probability at time \( t \).
2. Calculate the derivative (the rate of change) using the Kolmogorov Forward Equation.
3. Multiply that rate by a small time step \( h \).
4. Add that to your starting probability to get the probability at time \( t+h \).
Formula:
\( {}_t+h p_x^{ij} \approx {}_t p_x^{ij} + h \cdot \left( \frac{d}{dt} {}_t p_x^{ij} \right) \)
Common Mistake: Forgetting that \( h \) (the step size) is usually small (like 1/12 for a month or 1/365 for a day). Don't forget to multiply the derivative by \( h \)!
5. Working with Non-Constant Intensities
Sometimes the exam will give you intensities that are functions of time, like \( \mu_{x+t}^{01} = 0.01 + 0.001(x+t) \).
Did you know? This makes the model "Non-Homogeneous." All this means is that the "speed" of transitioning changes as the person gets older.
To find the probability of staying in a state when intensities are changing, you must integrate.
\( {}_t p_x^{\bar{ii}} = \exp\left( -\int_0^t \text{total exit rates} \right) \)
Quick Review:
- If the exit rate is constant (\( \mu \)), then \( {}_t p_x^{\bar{ii}} = e^{-\mu t} \).
- If the exit rate is linear (like \( At + B \)), you'll need to integrate to get a quadratic in the exponent!
6. Summary and Key Takeaways
Key Points to Remember:
- States and Transitions: We track movement between multiple statuses using intensities (\( \mu \)).
- Staying vs. Being: \( {}_t p_x^{\bar{ii}} \) is "never left," while \( {}_t p_x^{ii} \) is "back home again."
- Forward Equations: The rate of change is simply "Inflow minus Outflow."
- Euler's Method: Use it for numerical steps: \( \text{New} = \text{Old} + (\text{Step Size} \times \text{Rate of Change}) \).
Final Encouragement:
Continuous-time Markov models are just a formal way of tracking movement. If you can visualize the "flow" between boxes (states), the math usually follows. Practice drawing the state diagrams—it helps clarify which intensities belong in which part of the equation!