Extended Link Ratio Family (ELRF) modeling framework
The Extended Link Ratio Family (ELRF) modeling framework formulates average link ratio methods as regression estimators and extends them. This includes Mack (volume weighted averages), other weighted averages and extensions:
- Murphy (an intercept)
- A constant trend in the incrementals for each development period down the accident years
- Optimal combinations of intercepts, trends, and link ratios
- Control over variance assumptions (delta)
Standard deviations are computed analytically in all cases.
The bootstrap technique
The Bootstrap technique in the ELRF module gives
- Distributions of reserves by accident years, calendar years, and total
- Percentiles, V@Rs and T-V@Rs
- … and more!
Distributions can be translated from the bootstrap sample mean to any mean specified by the user by accident period totals or calendar period totals. This allows the actuarial analyst to fit any link ratio model, obtain the means by accident (or calendar) period and apply the bootstrap using the Mack method residuals with a selected delta.
Diagnostic statistics and graphs
Diagnostic statistics to test assumptions made by any ELRF model
- Residuals versus development year, accident year, calendar year, and fitted
- Y|X plots to test whether link ratios have predictive power in the presence of an intercept
- … and more!
The regression formulation of link ratios (Mack, Murphy, and other extensions), provides the framework for verifying the assumptions made by the link ratio techniques.
Link ratios, Mack, Murphy, Over-Dispersed Poisson, and the bootstrap technique
Do link ratio methods work for your data?
Read more on the ELRF™ modeling framework.
Link Ratio Techniques (LRT) module
The LRT module include a full range of link ratio methods:
- Chain Ladder (Volume weighted averages);
- Arithmetic Average;
- Geometric Average;
- Average without min/max;
- Wtd. Average of last N*;
- Average of Last N*;
- Geom. Average of Last N*;
- Maximum Ratio
- Minimum Ratio
- Weighted Ex high/low;
- Two and three parameter smoothing;
- and more!
*N is configurable by the analyst.