# Errata for the 2017 Exam 8 Cookbook

If you spot an error in the Exam 8 Cookbook, please quick send me a message so that I can correct the error and document it here for others. Thanks!

Many thanks to everyone who pointed out the errors corrected below:

**Fisher - Limited Table M (10/13/17): **

- The entry ratio for the policy should be calculated based on the expected aggregate limited loss for the policy, which would be $540k * (1 - .12). This changes the calculation for the Limited Table M charge. See the fixed recipe below.
- A couple of the risks have a higher limited loss than unlimited loss in the problem. That's a mistake on my part when I was creating the problem and making sure some of the calculations worked out nicely. This is corrected in the fixed recipe.

Fisher - Limited Table M_Fixed

**NCCI 2 - NCCI Retro Premium (9/30/17)**: The arrow in step 6 should say "Closest to 0.576."

**Couret & Venter - Multi-Dimensional Credibility (9/24/17)**: In the discussion section, the VHM(W) arrow should be pointing to Cov(W,w), not Cov(V,w).

**Grossi - Occurrence Exceedance Probability (9/10/17)**: The axes in the step 2 description are switched and the graph is correct. The x-axis is the Loss Size and the y-axis is the Occurrence Exceedance Probability (OEP).

**Bahnemann - Risk Load (9/5/17)**: The problem should have the following changes:

- The prompt should state "... including the risk load
"**using the standard deviation method.** - The k' formula should be: k /
**sqrt(E[N])** - ULAE should be loaded as a percentage of
**Loss & ALAE**(not just Loss)

**Clark - Casualty Excess of Loss-Experience Rating (8/27/17)**: The premium in the problem should be labeled "**Subject Premium" **not "On Level Subject Premium."

**Fisher - Table L (8/24/17)**: The prompt *should *say to calculate Table L charges and savings "from 0 to **1.25** in

increments of 0.25" instead of up to 1.5.

**Couret & Venter - Quintiles Test (8/20/17)**: The credibility-weighted ratio for the overall HG in step 2 should be calculated as a weighted average using the even-year TT claims (**.1628**, *not* .1718). Here is the fixed version: **Couret & Venter - Quintiles Test - Fixed**

## The errata below was corrected in the final version

**Mahler - Credibility-Weighted Predictions (8/4/17)**: A subscript on* *the third *Y* in the written-out formula for the "Credibility Weight of the Most Recent Data and Previous Estimate" is incorrect. It should be *Y_i-*2, not *Y_i-*1. The correct formula is below:

**Robertson - Weighted k-means Cluster Analysis (8/4/17):** The subscript on the weight in step 1 is incorrect. The calculation shows *w_A*. It should show *w_1* since this is the weight for class 1.

**Couret & Venter - Multi-Dimensional Credibility (8/4/17)**: The calculation in step 1 is incorrect. I didn't include VHM when calculating Var(*V*i) and Var(*W*i). The rest of the calculation is off because of this. You can download the fixed recipe here: **Couret & Venter - Multi-Dimensional Credibility - Fixed**

**GLM - ROC Curve (8/4/17)**: The denominators in the written-out formulas under step 3 are incorrect. Also, I accidentally wrote sensitivity a second time for the specificity formula. You can download the fixed recipe here: **GLM - ROC Curve - Fixed**