Catastrophe-linked bonds (cat bonds) represent the transfer of insured catastrophe risk to capital market investors via a structured bond issuance. Fitch's cat bond rating methodology focuses on the loss probability of a transaction's underlying reinsurance arrangement. The starting point for this approach centres on the base results from sophisticated models developed by independent modelling firms. Fitch then stresses these modelled outputs to account for the uncertainties and risks associated with relying on the models to determine financial loss. These stressed loss outputs, which include both the probability of attachment and the expected loss, are then compared to Fitch's cat bond rating benchmarks to determine the appropriate rating. Such benchmarks are developed from corporate default studies.

Model review
Modelling plays an integral role in the rating of cat bonds. The sophistication of the modelling used in this asset class far exceeds that of many other asset classes. Because of this, cat bond ratings rely extensively on third party expert modelling firms to provide the lion's share of the quantitative risk analysis. As such, cat bond ratings acknowledge the expertise of the modelling firms, as well as the assumption that cat events and their related financial losses can be appropriately modelled.

Fitch performs an in-depth review of the cat models. This includes a step-by-step analysis of the methodology, an evaluation of the underlying data inputs and assumptions, and stress testing.

Risk adjustment factors
Fitch recognises that cat bond modelling is an imperfect process. In an attempt to capture these imperfections, risk adjustment factors are applied to the models' base outputs. The five primary areas that Fitch considers when applying the risk adjustment factors are:

  • event modelling uncertainty, which addresses event predictability, both in terms of frequency of occurrence and severity. With any stochastic model, there is some uncertainty related to generating simulated events. To counteract this risk, Fitch applies a proprietary risk adjustment factor based on the type of peril. It is Fitch's conviction that certain cat perils are more easily modelled than others. For example, some events are well-defined occurrences that, relatively speaking, can be fairly easily parameterized and mathematically described;
  • loss modelling uncertainty, which addresses fundamental loss assessment measures. There is a fundamental uncertainty surrounding the process of modelling losses associated with specific cat events. Accordingly, Fitch expects the modelling of these factors to include conservative assumptions. If these issues are not considered in the model but could have an effect on the ultimate loss payout calculation, Fitch will make adjustments to the model outputs based on Fitch's conservative assumptions;
  • data resolution, which addresses data quality issues and how these affect overall loss assessment measures. The lower the levels of data availability and accuracy, the more uncertainty in the model. In addressing this uncertainty, Fitch examines the level of exposure data resolution by both property and insurance policy. Fitch takes a conservative stance and stresses the model outputs to account for the potential difference between the model inputs and actual policy and exposure data;
  • portfolio growth, which addresses the (potential) dynamic nature of the underlying risk portfolio. Loss estimates are modelled for a specific risk portfolio. However, an insurer's portfolio is not static. It can – and typically does – grow over time due to the addition of new risks and inflation. This serves as an additional risk for bondholders, and Fitch will make adjustments when it is not captured in the model's loss estimates; and
  • other, which addresses other factors that could present additional uncertainties. The most common are currency risk and unmodelled risk.
    Using a cat bond's payout method as a means of categorization, the asset class can be broadly segmented by the following four deal types:
  • indemnity. Under an indemnity deal, bondholder losses are determined based on the actual loss performance experienced by the underlying insurer. This deal type most closely resembles traditional reinsurance;
  • parametric. Under a parametric deal, bondholder losses are determined based on a certain parametric measurement of a defined event occurring in a defined region. With hurricanes, this could be central pressure and landfall location. With earthquakes it could be magnitude, depth, and epicenter location;
  • industry index-based. Under an industry index-based deal, bondholder losses are determined based on the reported value of an index typically defined and used to estimate total industry insured losses associated with certain cat events; and
  • modelled loss. Under a modelled loss deal, bondholder losses are determined based on a model's estimate of the actual loss experienced by the underlying deal sponsor. This estimation is determined based on an event's effect on a notional risk portfolio. Once a defined event occurs, it is parameterised and run through the model to determine the loss payout.

    Loss probability statistics
    The primary output from a cat bond model is the loss exceedance curve (see graph below left). This curve is a representation of potential financial losses to a risk portfolio related to a cat peril, along with the probabilities of those losses.

    By applying the specifics of a cat bond's underlying risk transfer agreement to the loss exceedance curve, a transaction's probability of attachment and probability of depletion can be determined. For example, given the loss exceedance curve as indicated, if it is assumed that the reinsurance cover attaches at $400m of losses (below that level the bondholders are not at risk) and exhausts at $800m, the probability of attachment (frequency) would be 1.0% and the probability of depletion (exhaustion) would be 0.5%. Note that in this case, the reinsurance layer (or protection layer) would be $400m (the $800m exhaustion minus the $400m attachment).

    By considering the severity of loss for each of the events in the stochastic dataset, a transaction's expected loss can be determined. This number is an estimate of the average annual loss as it relates to the protection layer.

    The two statistics that drive Fitch's rating decision are the probability of attachment and the expected loss. Whereas a cat bond's probability of attachment is a long-term estimation of the likelihood that investors will lose at least $1 of investment, its expected loss is a long-term estimation of what investors will lose in total. In determining a cat bond rating, Fitch will apply its risk adjustment factors to these two base model statistics.

    It is important to recognise that of the five risk adjustment factors, not all need to be applied to all deal types. For example, under a parametric deal, only an event modelling uncertainty factor would be applied, whereas under an indemnity deal all five might be applied. The following table highlights Fitch's risk adjustment factor application by deal type.

    Rating benchmarks
    Once all of the risk adjustment factors are applied, Fitch arrives at a transaction's annualised risk-adjusted loss probability assessments.

    These stressed assessments are then compared to Fitch's cat bond rating curves, which categorises certain probabilities of attachment (frequency) and expected loss with certain rating levels.

    Both frequency and expected loss results are considered in assigning ratings. Fitch's primary focus is on expected loss in the non-investment-grade portion of the rating scale. However, ratings assigned based on expected loss must be within two rating notches of the rating that would be assigned based on a frequency focus. Fitch's focus is on frequency alone in the investment-grade portion of the rating scale.