Ruin Probability: Adjustment Coefficient and Lundberg's Inequality

The Adjustment Coefficient and Lundberg's Inequality

The core idea is to distill the entire risk profile of the insurer—the rate and severity of claims versus the premium income—into a single number. This number is the adjustment coefficient, often denoted by $\gamma$ (or R, or $\kappa$). It serves as an inverse measure of risk: a larger adjustment coefficient implies a safer portfolio.

Once we have this coefficient, we can use it in a famous result called Lundberg's inequality to get our bound. The following video provides a concise overview of this concept.

As the video explained, Lundberg's inequality states:

This elegant formula tells us that the probability of ruin decays exponentially as the initial surplus $u$ increases. The rate of this decay is governed by the adjustment coefficient $\gamma$.

A More Formal Definition

Given your background in probability theory, we can define the adjustment coefficient more formally. Recall the claim surplus process, $W_t=S(t)-ct$, which represents the insurer's net outflow. The adjustment coefficient $\gamma >0$ is the unique positive value that satisfies the following equation involving the moment generating function (MGF) of the claim surplus increment over any time interval $\Delta$:

Essentially, $\gamma$ is the parameter that makes the related process $Y_t=e^{\gamma W_t}$ a martingale. For our purposes, we don't need the martingale theory, but this connection is what makes the coefficient so fundamental.

Let's look at the formal statement in the following resource.

The key takeaway is that if we can find this special number $\gamma$, we immediately have an upper bound on our ruin probability. The rest of our lesson will focus on exactly how to find it.

Calculating the Adjustment Coefficient

For the standard model in ruin theory—the compound Poisson process for claims—the general condition $M_{W_{\Delta }}(\gamma )=1$ simplifies to a very useful equation. In this model, the premium rate $c$ is set with a premium loading factor $\theta >0$, such that $c=(1+\theta )\lambda \mu$, where $\lambda$ is the claim frequency and $\mu$ is the mean claim severity.

The adjustment coefficient $\gamma$ is the smallest positive root of the equation:

Here, $M_X(\gamma )$ is the MGF of a single claim amount, $X$. This equation provides a practical path to calculation. We are looking for the point where the MGF of the claim size distribution intersects the straight line $1+(1+\theta )\mu \gamma$.

The adjustment coefficient $\gamma$ (labeled $\kappa$ here) is the unique positive value where the convex MGF curve intersects the premium-dependent straight line.

Let's see this in practice with a few examples.

Example 1: Exponentially Distributed Claims

This is a classic textbook case because it yields a neat, analytical solution. Suppose the individual claim amounts $X$ follow an exponential distribution with parameter $a$, so $X\sim \text{Exp}(a)$.

• The mean claim size is $\mu =\mathbb{E}[X]=1/a$.
• The MGF of the claim amount is $M_X(\gamma )=\frac{a}{a-\gamma }$, for $\gamma <a$.

Substituting these into our fundamental equation gives:

Solving this equation for $\gamma$ might look a bit messy, but it simplifies nicely. The resource below provides the final result.

As you saw, the adjustment coefficient is:

So, for an insurer with exponentially distributed claims, Lundberg's bound is $\psi (u)\le \exp \left(-\frac{a\theta }{1+\theta }u\right)$.

Example 2: Discrete Claims and Numerical Methods

In most practical scenarios, the claim distribution is not as convenient as the exponential, and the fundamental equation cannot be solved analytically. This is where your preference for numerical application comes into play. We can find $\gamma$ using iterative methods, such as the Newton-Raphson method.

Let's work through a concrete example.

This example is a perfect illustration of the practical workflow:

1. Define the function: $H(\gamma )=1+(1+\theta )\mu \gamma -M_X(\gamma )$.
2. Find its derivative: $H^{\prime }(\gamma )=(1+\theta )\mu -M_X^{\prime }(\gamma )$.
3. Iterate: Use an initial guess and the Newton-Raphson formula until the value for $\gamma$ converges.

The resulting value, $\gamma \approx 0.3212$, can then be plugged into Lundberg's inequality. For this insurer, $\psi (u)\le e^{-0.3212u}$.

Example 3: Continuous Claims and Root-Finding

Let's consider one more case with a slightly more complex continuous distribution, where the MGF leads to a polynomial equation.

In this case, the equation $M_X(\gamma )=1+(1+\theta )\mu \gamma$ was rearranged into a cubic polynomial with three roots: 0, 1, and 8/3. By definition, the adjustment coefficient is the smallest positive root, so $\gamma =1$. This is a crucial point: the equation may have multiple solutions, but we are always interested in the smallest strictly positive one.

Conclusion

In this lesson, we have built a practical tool to manage and quantify solvency risk. While the exact probability of ruin remains elusive in many cases, Lundberg's inequality provides a robust and easily interpretable upper bound.

Key Takeaways:

• Lundberg's inequality, $\psi (u)\le e^{-\gamma u}$, gives an exponential upper bound for the probability of ultimate ruin.
• The adjustment coefficient, $\gamma$, is a single parameter that summarizes the insurer's riskiness. It is defined as the smallest positive root of the equation $M_X(\gamma )=1+(1+\theta )\mu \gamma$.
• For simple claim distributions like the exponential, $\gamma$ can be found analytically.
• For more complex or discrete distributions, $\gamma$ is typically found numerically, for instance, using the Newton-Raphson method to find the root of the characteristic equation.

We have now seen how to calculate the adjustment coefficient and use it to bound the ruin probability. In our next lesson, we will return to the special case of exponentially distributed claims. Building on today's analytical example, we will see that for this specific distribution, we can go beyond the upper bound and calculate the exact probability of ruin.