Tail risks focus on extreme, low-probability losses. The tail of the loss distribution points to which perils could cause very large losses if they actually occur, even though those events are unlikely. This is exactly what risk managers need to know: what kind of extreme events could blow up losses and by how much. That awareness guides how much capital to hold, what hedges or contingency plans to deploy, and how to stress-test the portfolio. The other ideas don’t capture that core idea as precisely: tail risk isn’t primarily about geographic areas affected, nor about standard deviation alone, and tail losses aren’t inherently more probable simply because more time passes.

Aggregating losses across all events in a year captures the full annual loss distribution—frequency and magnitude of every event—so you can see how likely different loss levels are and how big total losses can become. This is what EP curves and AALs reflect: they summarize the entire yearly risk, not just a single event. If you look at the occurrence view, you’re focusing on the single largest loss driving event in a year. That misses the adds-up effect of multiple smaller events and how they can push total losses higher over the course of a year. In practice, several modest events can sum to a larger annual loss than the biggest single event, which means the risk profile shown by aggregate measures better informs capital needs, pricing, and risk transfer. So, using aggregate EP curves and AALs provides a complete, representative picture of yearly losses, whereas focusing only on the highest loss event gives an incomplete view of annual risk.

In CAT modeling, events are generated through stochastic simulations that build a large catalog of possible catastrophes. The approach starts with real historical events to anchor the model, then adds plausible events that could occur but haven’t happened, creating synthetic scenarios. This Monte Carlo-style process lets you explore a wide range of intensities, frequencies, and locations, and to estimate losses and their distributions, including extreme tail events. Relying solely on historical events misses potential extremes; using a fixed, SAMPLEnon-simulation-based distribution can’t capture the full variety and dependencies of real catastrophes; simply labeling events as random doesn’t describe the structured, calibrated generation that CAT models perform.

A tsunamigenic event is any event that can displace enough seawater to generate a tsunami. This displacement often comes from vertical movement of the seafloor or a sudden mass movement of water, such as submarine earthquakes, volcanic eruptions, large landslides into the ocean, or even a meteor impact. Because of that, the defining idea is the ability to create a tsunami, not just any tsunami wave itself or an event that cannot cause one. So the correct choice describes an event that can produce a tsunami. The other options don’t fit: something that cannot cause a tsunami isn’t tsunamigenic; an earthquake happening anywhere isn’t guaranteed to generate a tsunami, and not all earthquakes do; and a tsunami about to hit land is the wave itself, not the initiating event that could create it.

Claims data provide historical loss information that helps calibrate how catastrophes translate into losses, feeding the model development process by estimating parameters for frequency, severity, and the vulnerability and loss distributions. They’re also essential for model validation, where you compare what the model would have predicted against actual past losses to judge accuracy and make adjustments. In addition, claims data can reveal how vulnerability varies with secondary features—things like construction type, occupancy, age of building, and other exposure characteristics—allowing you to refine vulnerability curves and segmentation to improve predictive performance. Gaining relationships with a client is relevant to business operations, but it doesn’t directly inform the model’s structure or predictive accuracy, so it’s not a core modeling use of claims data.