RSK4803 · Risk financing · Worked example
The maximum probable yearly aggregate loss (MPY) is the figure a board funds a retained risk against. It answers one question: in a bad year, how much could this exposure cost us in total? In risk financing examinations it is almost always asked through a table of possible losses and their probabilities, and the table is the easy part. The marks go to the method.
01Read the question for the estimator
The study guide gives a family of estimators: the normal approximation, Chebyshev and the normal power approximation. You never have to derive the multiplier. When the question gives you a k value, it has chosen Chebyshev for you:
MPY = E(X) + k · S(X)
E(X) is the mean of the loss distribution, S(X) its standard deviation, and k the multiplier for the confidence level. Chebyshev makes no assumption about the shape of the distribution, which is why it is more conservative than the normal approximation when losses are skewed to the right, as they usually are.
02The Lucky Strike question
In UNISA’s May/June 2018 RSK4803 paper, a diamond mine has bought no theft insurance, and the chief risk officer must size the retained exposure for 2016. The paper gives a monthly theft-loss table with probabilities, sets k = 1,25, and asks for the answer to be multiplied by 10 to arrive at the maximum possible loss (MPL).
Write the formulae down before the table. The marker awards the method before the arithmetic:
03Six steps
-
Check the probabilities. They must add to 1,00. Here 4 × 0,08 + 6 × 0,09 + 2 × 0,07 = 1,00. If they do not, nothing after this can be right.
-
The mean. Multiply each month’s loss by its probability and add the column: μ = R168 075.
-
The deviations. Subtract the mean from each month’s loss. January: 135 000 − 168 075 = −R33 075. The deviation is from the mean of the whole distribution, not from the previous month. That is the most common error in this question.
-
Square and weight. January: (−33 075)² = 1 093 955 625, and × 0,08 = 87 516 450. Squaring is why the sign never matters, and why one extreme month dominates: June’s deviation of R79 425 squares to 6 308 330 625.
-
The variance and standard deviation. The weighted column adds to 1 837 231 875, so σ = √1 837 231 875 = R42 862,94. Carry the cents through each month’s figure; rounding each line to the rand makes the total look R1 out.
-
Apply Chebyshev and scale.
| Line | Working | Result |
|---|---|---|
| k · σ | 1,25 × 42 862,94 | R53 578,68 |
| MPY | 168 075 + 53 578,68 | R221 653,68 |
| MPL | 221 653,68 × 10 | R2 216 537 |
Finish with a sentence that answers the question: the estimated maximum probable yearly loss from diamond theft in 2016 is R221 653,68, and scaled to MPL as instructed it is R2 216 537.
04What the table is really for
The paper also says to use R450 as the price per carat “as the numbers may become difficult to manage”. The calculation never needs it, because the table is already in rand. Stay in rand throughout.
The MPY is not a statistics exercise for its own sake. In the same paper it becomes an entry in the cost of risk, because a loss that is not insured is retained whether or not anyone decided to retain it. Put a number on the retention and the board can decide whether to keep carrying it.