Adaptive AI research engine

Journal Scout

Search journals in any field, blitz any article into key findings, real-life applications, practice translations and protocols, then turn the ideas into 5K–400K-word books, handouts and presentations — exported as print-ready PDF (Amazon KDP interior rules), Word and PowerPoint. Upload your own PDF or Word document to start from there. Every blitz you open is kept in your Blitz library to read, listen to, quiz on and reuse.

Also: Actuarial bibliography (every curated textbook, paper and journal) · Webinar library · Practice dashboard · Publish to Amazon KDP

Real actuarial data

Genuine NAIC Schedule P triangles from the CAS Loss Reserve Database (Meyers & Shi) — 1988–1997 accident years, paid and incurred, with earned premium and the realised ultimates for hindsight testing. Pick a company and line; the analysis below runs live (chain ladder, Bornhuetter–Ferguson, Mack, ODP bootstrap, lognormal severity, VaR/TVaR). Turn on grounding and every new blitz applies its article to these numbers.

Dataset

Basis

Earned premium

2,238,741

Ultimate loss ratio

78.6%

CL IBNR

193,320

Mack CV

30.3%

Bootstrap 95th pct

270,136

CL hindsight error

9.1%

Workers' Compensation

$000

1988–1997

CAS source

Open in Decision Studio

Real data: Allstate Ins Co Grp — Workers' Compensation (paid basis, amounts in thousands of USD)

Source: Meyers, G.G. and Shi, P. (2011). The Retrospective Testing of Stochastic Loss Reserve Models. CAS E-Forum, using data pulled from NAIC Schedule P Parts 1-4, 1988-1997. https://www.casact.org/publications-research/research/research-resources/loss-reserving-data-pulled-naic-schedule-p

Development, reserves and loss ratios by accident year

AY	Earned premium	Latest paid	Age	CDF	CL ultimate	BF ultimate	CL IBNR	Ult. loss ratio	Actual ultimate	CL hindsight error
1988	394,742	325,322	10	1.000	325,322	325,322	0	82.4%	325,322	0.0%
1989	374,252	273,873	9	1.011	276,864	277,149	2,991	74.0%	277,574	-0.3%
1990	280,320	256,788	8	1.047	268,961	267,070	12,173	95.9%	263,000	2.3%
1991	313,982	239,195	7	1.080	258,402	258,110	19,207	82.3%	248,319	4.1%
1992	252,698	159,496	6	1.130	180,151	182,977	20,655	71.3%	168,844	6.7%
1993	201,055	87,215	5	1.196	104,286	113,889	17,071	51.9%	90,686	15.0%
1994	174,381	91,077	4	1.307	119,003	124,243	27,926	68.2%	94,730	25.6%
1995	146,366	87,311	3	1.514	132,157	127,565	44,846	90.3%	91,161	45.0%
1996	93,294	44,916	2	2.025	90,948	83,185	46,032	97.5%	49,255	84.6%
1997	7,651	691	1	4.501	3,110	5,514	2,419	40.7%	2,909	6.9%
Total	2,238,741	1,565,884			1,759,204	1,765,024	193,320	78.6%	1,611,800	9.1%

Selected volume-weighted age-to-age factors: 2.2230, 1.3377, 1.1584, 1.0927, 1.0586, 1.0455, 1.0314, 1.0361, 1.0109 (tail 1.000).

Reserve risk

Measure	Value
Mack total standard error	58,633
Mack CV of total reserve	30.3%
ODP bootstrap mean reserve	195,949
ODP bootstrap s.d.	42,833
75th / 95th / 99th percentile	221,978 / 270,136 / 302,665
VaR 99.5% / TVaR 99%	310,739 / 323,642

Incremental-cell severity (lognormal MLE on 55 positive cells): $\mu = 9.819$ , $\sigma = 1.091$ , mean 33,316, CV 1.51.

Key formulas: $\hat f_k = \sum_i C_{i,k+1} / \sum_i C_{i,k}$ ; $\hat U_i^{CL} = C_{i,n-i}\prod_{k\ge n-i}\hat f_k$ ; $\hat U_i^{BF} = C_{i,n-i} + (1 - 1/\mathrm{CDF}_i)\,\mathrm{ELR}\cdot P_i$ ; $\mathrm{TVaR}_{99\%} = \mathbb E[R \mid R > \mathrm{VaR}_{99\%}]$ .

Observations

Volume-weighted age-to-age factors: 2.223, 1.338, 1.158, 1.093… — slow early development typical of a long-tailed line.
Chain-ladder IBNR on the paid basis is 193,320 (thousands of USD), 12.3% of the latest diagonal.
Premium-weighted ultimate loss ratio 78.6%; by year it ranges 40.7%–97.5%.
Mack standard error of total reserve 58,633 (CV 30.3%); ODP bootstrap 95th percentile 270,136 vs mean 195,949.
Hindsight test: chain ladder over-estimated the realized ultimate by 9.1% (actual data through age 10 is known for every year).
Bornhuetter–Ferguson (ELR 81.0%) is 0.3% above chain ladder in total — the gap sits in the immature years.
Incremental paid amounts fit a lognormal with σ = 1.09 (CV 1.51), i.e. heavy-tailed cell values.