Utilities

Utilities contains example datasets and extra functionality to facilitate a reserving workflow.

Sample Datasets

A variety of datasets can be loaded using load_sample(). These are sample datasets that are used throughout the examples in this documentation. The full list is available through list_samples(), which returns a table of every bundled dataset along with its index, columns, and grain.

import chainladder as cl

cl.list_samples()

	index	columns	cumulative	origin_grain	development_grain	origin_periods	development_periods
name
abc	None	[values]	True	Annual	Annual	11	11
auto	[lob]	[incurred, paid]	True	Annual	Annual	10	10
berqsherm	[LOB]	[Incurred, Paid, Reported, Closed]	True	Annual	Annual	8	8
cc_sample	None	[loss, exposure]	True	Annual	Annual	5	5
clrd	[GRNAME, LOB]	[IncurLoss, CumPaidLoss, BulkLoss, EarnedPremD...	True	Annual	Annual	10	10
clrd2025	[GRNAME, LOB]	[IncurredLosses, CumPaidLoss, BulkLoss, Earned...	True	Annual	Annual	19	19
friedland_auto_bi_insurer	None	[Paid Claims, Reported Claims, Earned Premium]	True	Annual	Annual	9	9
friedland_auto_freq_sev	None	[Closed Claim Counts, Reported Claim Counts, R...	True	Semiannual	Semiannual	10	10
friedland_auto_salsub	None	[Reported Salvage and Subrogation, Received Sa...	True	Annual	Annual	11	11
friedland_autoprop	None	[Reported ALAE, Paid ALAE, Reported Claims, Pa...	True	Annual	Annual	11	11
friedland_berq_sher_auto	None	[Paid Claims, Closed Claim Counts, Reported Cl...	True	Annual	Annual	8	8
friedland_gl_insurer	None	[Closed Claim Counts, Reported Claim Counts, D...	True	Annual	Annual	8	8
friedland_gl_self_insurer	None	[Reported Claims, Paid Claims]	True	Annual	Month	11	1
friedland_med_mal	None	[Reported Claims, Paid Claims, Case Outstandin...	True	Annual	Annual	8	8
friedland_qs	None	[Gross Reported Claims, Net Reported Claims, N...	True	Annual	Annual	4	4
friedland_us_auto_chg_prod_mix	None	[Paid Claims, Reported Claims]	True	Annual	Annual	10	10
friedland_us_auto_incr_claim	None	[Paid Claims, Reported Claims]	True	Annual	Annual	10	10
friedland_us_auto_steady_state	None	[Paid Claims, Reported Claims]	True	Annual	Annual	10	10
friedland_us_industry_auto	None	[Paid Claims, Reported Claims, Earned Premium]	True	Annual	Annual	10	10
friedland_us_industry_auto_case	None	[Case Outstanding, Paid Claims]	True	Annual	Annual	10	10
friedland_uspp_auto_increasing_case	None	[Reported Claims, Paid Claims, Earned Premium]	True	Annual	Annual	10	10
friedland_uspp_auto_increasing_claim	None	[Reported Claims, Paid Claims, Earned Premium]	True	Annual	Annual	10	10
friedland_uspp_auto_steady_state	None	[Reported Claims, Paid Claims, Earned Premium]	True	Annual	Annual	10	10
friedland_uspp_increasing_claim_case	None	[Reported Claims, Paid Claims, Earned Premium]	True	Annual	Annual	10	10
friedland_wc_self_insurer	None	[Closed Claim Counts, Reported Claim Counts, P...	True	Annual	Annual	8	8
friedland_xol	None	[Gross Reported Claims, Net Reported Claims, C...	True	Annual	Annual	4	4
friedland_xyz_auto_bi	None	[Paid Claims, Reported Claims, Closed Claim Co...	True	Annual	Annual	11	11
friedland_xyz_disp	None	[Disposal Rate, Closed Claim Counts, Paid Claims]	True	Annual	Annual	8	8
genins	None	[values]	True	Annual	Annual	10	10
ia_sample	None	[loss, exposure]	True	Annual	Annual	6	6
liab	[lob]	[values]	True	Annual	Annual	14	14
m3ir5	None	[values]	True	Annual	Annual	14	14
mack_1997	None	[Case Incurred]	True	Annual	Annual	10	10
mcl	None	[incurred, paid]	True	Annual	Annual	7	7
mortgage	None	[values]	True	Annual	Annual	9	9
mw2008	None	[values]	True	Annual	Annual	9	9
mw2014	None	[values]	True	Annual	Annual	17	17
prism	[ClaimNo, Line, Type, ClaimLiability, Limit, D...	[reportedCount, closedPaidCount, Paid, Incurred]	False	Month	Month	120	120
quarterly	None	[incurred, paid]	True	Annual	Quarter	12	45
raa	None	[values]	True	Annual	Annual	10	10
tail_sample	None	[incurred, paid]	True	Annual	Annual	10	10
ukmotor	None	[values]	True	Annual	Annual	7	7
usaa	None	[incurred, paid]	True	Annual	Annual	10	10
usauto	None	[incurred, paid]	True	Annual	Annual	10	10
xyz	None	[Incurred, Paid, Reported, Closed, Premium]	True	Annual	Annual	11	11

Model Diagnostics

model_diagnostics() summarizes a fitted IBNR model into a single Triangle whose columns are the diagnostic vectors of interest, such as the latest diagonal, IBNR, and ultimate. It accepts a fitted estimator or a Pipeline, and an optional groupby to summarize at a coarser index level.

model = cl.Chainladder().fit(cl.load_sample('raa'))
cl.model_diagnostics(model)

	Triangle Summary
Valuation:	1990-12
Grain:	OYDY
Shape:	(1, 8, 10, 1)
Index:	[Measure, Model, Total]
Columns:	[Latest, Quarter Incremental, Year Incremental, LDF, CDF, Ultimate, IBNR, Run Off 1]

Chainladder Persistence

All estimators can be persisted to disk or database using to_json or to_pickle. Restoring the estimator is as simple as cl.read_json or cl.read_pickle.

import chainladder as cl
model_json = cl.Chainladder().fit(cl.load_sample('raa')).to_json()
model_json

'{"params": {}, "__class__": "Chainladder"}'

cl.read_json(model_json)

Chainladder()

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

The saved Estimator does not retain any fitted attributes, nor does it retain the data on which it was fit. It is simply the model definition. However, the Triangle itself can also be saved allowing for a full rehydration of the original model.

# Dumping triangle to JSON
triangle_json = cl.load_sample('raa').to_json()

# Recalling model and Triangle and rehydrating the results
cl.read_json(model_json).fit(cl.read_json(triangle_json)).ibnr_.sum('origin')

np.float64(52135.228261210155)

Warning

Some features of estimators may not be json-serializable, such as a virtual_column or a callable hyperparameter. In these cases, JSON serialization will fail.