23.3 Projection Risk

Different ways to project trends

23.3.1 Simple Trend

Forecast one trend using historical average and into the future

Caveat: Additional uncertainty due to estimate of ultimate losses is uncertain

23.3.2 Severity Trend and Inflation

Claim severity \(\neq\) general inflation

\[\begin{equation} [\text{Claim Severity Trend}] \approx [\text{General Inflation}] + [\text{Superimposed Inflation}] \tag{23.2} \end{equation}\]

2 approaches:

  1. Model the severity trend independent of general inflation

  2. Adjust the data for general inflation and model the residual superimposed inflation

For ERM where general inflation is modeled, severity trend should be dependent on the general inflation

23.3.3 Trend as a Time Series

More realistic, allow for trends to change over time

AR(1) with mean reverting form:

\[\begin{array}{ccccc} r_{t+1} &= &m + \alpha_1(r_t - m) &+ &\epsilon_{t+1} \\ &= &(m - \alpha_1 m) + \alpha_1 r_t &+ &\epsilon_{t+1} \\ &= &\alpha_0 + \alpha_1 r_t &+ &\epsilon_{t+1} \\ \end{array}\]

  • \(m\) is the long term mean estimated from data

  • \(\alpha_1\) is the correlation from one period to the next

  • \(\epsilon_t \sim N(0,\sigma)\)

Simple trend understate the projection risk especially for long tail LoBs

  • Simple trend had a 10 year forecast 99th percentile prediction error of +45%