ICRFS-ELRF™ 12 Training Videos
The Training Videos give you a guided tour of the product's functionality that includes modeling a number of triangles.
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.
ICRFS-ELRF™ Databases
In this video, the database functionality and navigation is studied. Data, models, and links to reports all reside in one relational database.
The objects in the database are Triangle Groups (TG) and Composite Triangle Groups (CTG). TGs (and CTGs) also contain objects, namely, triangles, exposures, premiums, data sets, models and links to reports.
ICRFS-ELRF™ Professional only: A database can either be remote (on a server that is shareable) or local. Communication between two databases has the same intuitive feel as using Windows Explorer for communicating between two sub-folders.
This video demonstrates:
- Navigation within the database
- Manipulation of triangle group values
- Using variables and values to filter triangle groups
- Database structure
- Triangle group structure
- New triangle groups
This database video is approximately 20 minutes long
ICRFS-ELRF: Introduction to LRT
The Link Ratio Techniques (LRT) module of ICRFS-ELRF™ is introduced.
Navigation of the LRT module is presented.
The methods contained in LRT are covered along with the method for selecting final ratios (including two and three parameter smoothing). User defined ratios can also be created.
Forecast tables, BF and ELR forecasts, and the summaries are also described.
The introduction to LRT video is approximately 10 minutes long
ICRFS-ELRF: Introduction to the Extended Link Ratio Family (ELRF) modelling framework. Mack, Murphy, and extensions
In this video we discuss how link ratios can be reformulated in a regression framework. The Mack method is the regression formulation of the chain ladder or, equivalently, volume weighted average link ratios. The fundamentals of link ratios and regression formulations are developed in detail.
The importance of the residuals (difference between the trends in the data and the trends in the method) is emphasized.
Note the system navigation is covered in detail as the data are being examined in the modelling frameworks.
The Mack method and Murphy methods are introduced briefly along with various variance assumptions, that is, reciprocal of weights in the weighted least squares optimisation.
To view this video, click here. This video is approximately 30 minutes long.
ICRFS-ELRF: The Extended Link Ratio Family (ELRF) modelling framework (cont)
We continue the discussion from the previous video looking at the incurred loss development array that we call "Mack".
The importance of the regression formulation due to Venter (1998): y-x = a + (b-1)x + e is discussed. The x is the cumulative in the current development period, and y is the cumulative in the next period. Link ratios applied to cumulative data actually predict the incrementals in the next development period.
If the incrementals (y-x) versus the previous cumulatives (x) are not correlated, equivalently b-1 is not significant, then the link ratios do not have any predictive power. Several examples are considered.
The assumptions of the Mack method are not supported by the data - we know we need an intercept (Murphy).
The forecast tables and forecast summaries are considered. Along with the full forecast table, results are summarised by accident year and calendar year. Bornhuetter-Ferguson and Expected Loss Ratio Forecast methods are also available when premiums are associated with the dataset (see first video).
It is shown how to easily compare multiple models within the ICRFS-ELRF™ modelling framework.
To view this video, click here. This video is around 25 minutes long.
ICRFS-ELRF: Link ratios, Mack and the bootstrap technique
The bootstrap technique is introduced. The rationale behind the bootstrap technique is explained and related to the link ratio regression formulation.
The bootstrap technique functionality is only available for link ratio-only models.
The output from the bootstrap simulations are discussed in detail including distribution summaries by accident years, calendar years, VaRs and T-VaRs.
To view this video, click here. This video is about 17 minutes long.
ICRFS-ELRF: Worker's Compensation Portfolio
This video illustrates the importance of considering all the diagnostics available in the Extended Link Ratio Family modelling framework.
The Mack method clearly under fits the most recent calendar years. Even when the first eight years are excluded (the link ratios are calculated based on the last three calendar years only), the Mack Method and any other link ratio method do not capture the (recently) high calendar year trend.
The Mack method and link ratio methods in general capture an average calendar year trend. However, there are no descriptors of it. Further, there is no control over the trends in the future.
The optimal model in the ELRF modelling framework is identified.
Bootstrap samples of the real data (triangle) can also be used to assess the quality of the model. The bootstrap samples from the Mack method do not have the same statistical characteristics as the real data, demonstrating significant model specification error.
Briefly, the results in the ELRF modelling framework are contrasted with the results based on the optimal model in the Probabilistic Trend Family (PTF) modelling framework (only available in ICRFS-Plus™).
To view this video, click here. This video is about 17 minutes long.
ICRFS-ELRF: Triangle Group LR High - Link ratio methods including Mack over estimate reserves significantly
This video illustrates a second case study where the traditional link ratio methods (Mack and the like) give results that are far too high - perhaps by as much as a factor of two.
We compare the results from the optimal trend and ratio model for the last six years versus the Mack method on the whole paid loss triangle. We find that the projection using only the last six years is substantially lower (trend in model is more in line with the trend in the data). However, we still don't know what the trend is.
We show the bootstrap technique applied to the Mack method results in distributions with sample means higher than the Mack means. The bootstrap sample means can be shifted to the Mack method means as illustrated in the video.
Simulated data are compared with true parameters to demonstrate the effect of not estimating trends.
To view this video, click here. This video is about 24 minutes long.
ICRFS-ELRF: Triangle Group LR High - Comparison with PTF
We start with the estimates of the Mack method, arithmetic average, and best model from the previous video so all forecasts can be seen.
We then open ICRFS-Plus™ and open optimal models in the Probabilistic Trend Family (PTF) for the Number of Claims Closed (NCC), Case Reserve Estimates (CRE), and Paid Losses (PL). The calendar trends identified for the three are compared and their impact on the future paid losses is considered. The 8.71% +_ calendar year trend in the paid losses is seen to be conservative and projects 594M.
A very conservative assumption of 18.3% +_ is considered and shown to project a mean of 752M - still substantially less than the projections from the Mack method and the arithmetic average.
We conclude by illustrating that the PTF modelling framework gives you control over the future trends and assumptions whereas in the Extended Link Ratio Family no such control is available.
To view this video, click here. This video is 12 minutes long.
ICRFS-ELRF: Worker's Compensation Portfolio, Company APS - major changes in calendar year trends
In this video we consider another worker's compensation portfolio where there are major calendar year trend shifts.
The diagnostics of the Extended Link Ratio Family show that the calendar year trend has not been described by the Mack method. The most recent calendar year trend is consistently underfitted - and also underprojects the future losses by calendar year.
The Mack method is calculated on all the data, for the data following calendar years: 1997, 2001, 2004, and 2006.
The forecasts for each method are shown to increase as less years are included in the modelling method. This means that the calendar year trend described by the method is increasing (but we still don't know what the trend is).
An optimal model (intercept, trend and link ratios) in the ELRF modelling framework is created for the data from 2000 onward however it is still evident that the more recent calendar year trend (2002 onward) is higher than that captured (but not described) by the method.
In contrast, the PTF modelling framework estimates the trend from 2002 as 10%+_; an increase from a previous trend of 0+_. Again, in the PTF modelling framework we can explicitly measure calendar trends in the past and gain full control over future calendar trend assumptions (along with the other trend or variability assumptions) in the future.
To view this video, click here. This video is about 11 minutes long.
For any questions, help in modelling your data, or ways of improving the product, please email our global support team at support@insureware.com.