ICRFS-Plus™ Training videos for new users
If for any reason you are unable to view the training or demonstration videos, please contact our support staff at support@insureware.com and we will arrange to send you a copy of the videos on CD-Rom. You will be able to run the videos from the CD.
The training videos should be used for hands on training.
We suggest you run the videos on a separate computer using a data projector, and train as a group.
The only way you will learn all the new concepts and be able to
exploit all the immense benefits is by using the system. Experiential
learning is imperative.
It is important that you study the videos in sequential order set
out below.
Table of Contents
1. Introduction to ICRFS-Plus™ 10 and modelling modules
- 2.1 PTF (and ELRF) modelling of TG CTP
- 2.2 Heteroscedasticity and Varying Parameter
- 2.3 Compa and SDF
- 2.4 Layers, predictive aggregate loss distributions and lognormals
3. ICRFS-Plus™ 10 modelling wizard and other features
5. ICRFS-Plus™ 10 and NAIC Schedule P
2. PTF Modeling
2.1 PTF (and ELRF) modeling of TG CTP
A model is identified (built or designed) for the real data CTP, first manually, in order to illustrate some intuitive statistical concepts. It is shown how to use residual graphs to add/remove parameters, set a parameter to zero and assign zero weight. The Validation module is introduced and validation analyses are used to confirm the stability of trends, and to extract additional information from the data.
Each diagnostic model included under the model templates (button) is used to answer a specific question about the data.
Modeling of constraints (manually) is also described.
The cumulative data are modeled in ELRF and as a result of condition 1 being satisfied; link ratios have no predictive power whatsoever.
Click here to view.
2.2 Heteroscedasticity and Varying Parameter
Triangle group CS5 is used to model heteroscedasticity by applying the heteroscedasticity module. This is first done manually and subsequently done automatically by pressing one button. The word heteroscedasticity is a big word but the concept is very simple.
Varying parameter modeling is introduced here and its importance is stressed. Three competing models are designed for this study and their relative merits discussed.
The triangle group ABC has major shifts in calendar year trends and very little process variability. It satisfies condition 3 (explained in training video 1.1).
The diagnostic statistical chain ladder model is used to detect major calendar year shifts. These data are used to illustrate that calendar year trends project onto development years.
Analyses of the cumulative data are also conducted in ELRF and the major calendar year trend shifts are also seen in the residual graphs. In ELRF these can not be estimated or controlled for.
The warp speed modeling wizard is introduced and is applied to model the data for triangle group CTP studied in training video 2.1. It identifies the same model in less than one second that was identified in training video 2.1 manually in a few minutes.
By the end of this session the user will be familiar with the various buttons in the PTF module and should have a good idea of how to model the data (manually) relatively quickly.
Click here to view.
2.3 Compa and SDF
The triangle group Compa has extremely volatile paid losses. A good model measures this volatility and projects it into the future based on explicit assumptions. Even though the data are extremely volatile, the reserve distributions (beyond the evaluation period) estimated nine calendar periods earlier are the same as the ones estimated at the current evaluation period, because the calendar year trend is stable. Moreover, the distributions of the (161) observations left out in the last nine calendar periods are forecast accurately nine calendar periods ago using only 91 observations!
Three triangles are simulated from the model and it is shown that it is impossible to distinguish in respect of variability between the real triangle and the simulated triangles, and forecast distributions are statistically the same.
The residuals in ELRF look very bad. Not only do link ratio methods not capture the trend structure, the error structure is not symmetric. That is, the log normal distributions on a dollar scale are very skewed because the corresponding normal distributions on the log scale (in PTF) have a high variance. The data must be modeled on a log scale!
The triangle group SDF contains a paid loss array that is quite volatile and is simulated from a simple trend model with high process variability. The true parameters values are used to illustrate a number of statistically concepts and to debunk a critical loss reserving myth. Another triangle is simulated from the same trends but this time with very low process variability.
Click here to view.
2.4 Layers, predictive aggregate loss distributions and lognormals
For three layers, limited $1M, $1Mxs$1M, limited $2M, the trend structure is the same and residuals are very highly correlated. For the layer $1Mxs$1M the calendar year trend estimate is subject to high uncertainty that is it is non- credible, equivalently, insignificant. This is due to high process variability in the layer. The distributions of aggregate losses for calendar years, accident years and grand total are obtained by simulating from the log normal distributions (using their correlations) projected for each cell. Different prospective and retrospective adverse development cover programs are designed for the layer $1Mxs$1M.
In training video 4.2 we design a composite model in the MPTF module that includes the correlations between the layers.
A model is designed manually for a workers comp portfolio, TG WCOM.The triangle group CS5 is revisited to explain the reason for using log normal distributions on a dollar scale, equivalently, normal distributions with possible changing variances on a log scale – paid losses data are variance invariant on the log scale under a trend transformation.
Click here to view.