Loss Reserving (cont...)

Deciding on a future calendar period trend and its uncertainty is a crucial step in developing reserve distributions. To help you make this decision, the validation process allows you to assess how stable the trends have been in the past. The most recent calendar years of data are removed in turn and the forecast is done again on the reduced data. In the example on the right, the forecast hardly changes even when four years (45% of the data!) have been removed. So it is very likely the same trend will continue into the future. When you forecast, you get a wealth of information. The table to the right shows the mean (in black) and standard deviation (in red) of the forecast probability distribution in each accident and development period. Totals for calendar periods give you the mean and standard deviation of the distributions for the liability stream in future years. The aggregate of these distributions gives you the grand distribution for the reserve. The forecast summary gives you a number of useful tables and graphs. For example, the mean and standard deviation of the total reserve for each accident year can be compared with the Case Reserve Estimates. Distributions of totals for accident periods allow you to answer questions like “What is the probability that the future losses in 1992 will exceed $3M?”. You can also calculate percentiles of any distribution – the graph to the right shows the 75th percentile and Value-at-Risk. Correlations between future periods are also calculated and discount factors applied as required, for asset liability matching.
	How skewed is this distribution?

In this flexible framework, the model can be as simple (just one parameter!) or as complex as your data requires. One test of a good model is to simulate data from the model and see if you can distinguish the original data from the simulated data. In ICRFS-Plus™, you can do this check in just a few clicks. For good forecasting performance, the model should not contain more parameters than are needed (the principle of parsimony). The automatic optimisation option ensures this by removing all unnecessary parameters.