def test_base_minimum_exposure_triangle():
raa = (cl.load_dataset('raa').latest_diagonal * 0 +
50000).to_frame().reset_index()
raa['index'] = raa['index'].astype(str)
cl.Triangle(raa,
origin='index',
columns=list(cl.load_dataset('raa').columns))
def test_malformed_init():
assert cl.Triangle(
data=pd.DataFrame({
'Accident Date': ['2020-07-23', '2019-07-23', '2018-07-23', '2016-07-23', '2020-08-23', '2019-09-23', '2018-10-23'],
'Valuation Date': ['2021-01-01', '2021-01-01', '2021-01-01', '2021-01-01', '2021-01-01', '2021-01-01', '2021-01-01'],
'Loss': [10000, 10000, 10000, 10000, 0, 0, 0]}),
origin='Accident Date', development='Valuation Date', columns='Loss'
).origin_grain == 'M'
def test_init_vector(raa):
a = raa.latest_diagonal
b = pd.DataFrame(
{"AccYear": [item for item in range(1981, 1991)], "premium": [3000000] * 10}
)
b = cl.Triangle(b, origin="AccYear", columns="premium")
assert np.all(a.valuation == b.valuation)
assert a.valuation_date == b.valuation_date
def test_partial_year(prism):
before = prism['Paid'].sum().incr_to_cum()
before=before[before.valuation<='2017-08'].latest_diagonal
after = cl.Triangle(
before.to_frame(keepdims=True).reset_index(),
origin='origin', development='valuation', columns='Paid', index=before.key_labels)
assert after.valuation_date == before.valuation_date
def test_init_vector():
assert raa == raa_gt
a = raa[raa.development == 12]
b = pd.DataFrame(
{"AccYear": [item for item in range(1981, 1991)], "premium": [3000000] * 10}
)
b = cl.Triangle(b, origin="AccYear", columns="premium")
assert np.all(a.valuation == b.valuation)
assert a.valuation_date == b.valuation_date
def test_exposure_tri():
x = cl.load_sample("auto")
x = x[x.development == 12]
x = x["paid"].to_frame().T.unstack().reset_index()
x.columns = ["LOB", "origin", "paid"]
x.origin = x.origin.astype(str)
y = cl.Triangle(x, origin="origin", index="LOB", columns="paid")
x = cl.load_sample("auto")["paid"]
x = x[x.development == 12]
assert x == y
def test_exposure_tri():
x = cl.load_sample('auto')
x = x[x.development == 12]
x = x['paid'].to_frame().T.unstack().reset_index()
x.columns = ['LOB', 'origin', 'paid']
x.origin = x.origin.astype(str)
y = cl.Triangle(x, origin='origin', index='LOB', columns='paid')
x = cl.load_sample('auto')['paid']
x = x[x.development == 12]
assert x == y
def __init__(self):
self.Data_Handler = Data_Handler()
'''
Using the chainladder (cl) package: https://chainladder-python.readthedocs.io/en/latest/index.html
'''
self.Triangles = cl.Triangle(self.Data_Handler.data , origin='ACCIDENT_MONTH_DATE',
development='ACCOUNTING_MONTH_DATE', columns='CLAIM_PAID_EX_GST_D',
index=['PRODUCT_CODE_CLAIM','CLAIM_CATEGORY', 'CLAIM_CATEGORY_RESERVING',
'CLIENT_NAME', 'RISK_GROUP'
#'PRODUCT_GROUP'
],
cumulative=False
)
self.filter_ = filter_
def test_base_minimum_exposure_triangle():
assert raa == raa_gt
d = (raa.latest_diagonal * 0 + 50000).to_frame().reset_index()
d["index"] = d["index"].astype(str)
cl.Triangle(d, origin="index", columns=d.columns[-1])
def test_correct_valutaion(raa):
new = cl.Triangle(
raa.iloc[..., :-3, :].latest_diagonal.to_frame(keepdims=True, implicit_axis=True),
origin='origin', development='valuation', columns='values')
assert new.valuation_date == raa.valuation_date
def test_create_full_triangle(raa):
a = cl.Chainladder().fit(raa).full_triangle_
b = cl.Triangle(
a.to_frame(keepdims=True, implicit_axis=True),
origin='origin', development='valuation', columns='values')
assert a == b
def summary_triangles(data, reporting_date):
'''
Remove all data which is not know at the reporting date and convert data into triangles, for further analysis
Returns 'chainladder' triangles
'''
#
#fileter out all data not known at reporting date
#
# create masks
data_triangles = data.copy()
date_mask_notpaid = (data_triangles['Claim_payment_date'] -
reporting_date) / np.timedelta64(1, 'D') > 0
date_mask_notreported = (data_triangles['Claim_report_date'] -
reporting_date) / np.timedelta64(1, 'D') > 0
date_mask_paid = (data_triangles['Claim_payment_date'] -
reporting_date) / np.timedelta64(1, 'D') <= 0
date_mask_reported = (data_triangles['Claim_report_date'] -
reporting_date) / np.timedelta64(1, 'D') <= 0
date_mask_notwritten = (data_triangles['Start_date'] -
reporting_date) / np.timedelta64(1, 'D') > 0
# clear paid and reported data which is unknown at reporting date
data_triangles.loc[date_mask_notpaid, 'Claim_payment_date'] = np.nan
data_triangles.loc[date_mask_notreported, [
'Claim_report_date', 'Claim_incident_date', 'Claim_value',
'Claim_value_gu'
]] = np.nan
# add counts for paid and reported
data_triangles['Claim_count_reported'] = 0
data_triangles['Claim_count_paid'] = 0
data_triangles.loc[date_mask_reported, 'Claim_count_reported'] = 1
data_triangles.loc[date_mask_paid, 'Claim_count_paid'] = 1
# remove unwritten policies
data_policies = data_triangles.loc[~date_mask_notwritten].copy()
tri_paid = cl.Triangle(
data_policies,
origin='Start_date',
index='Class_name',
development='Claim_payment_date',
columns='Claim_value',
cumulative=False).incr_to_cum().to_frame().reset_index().melt(
id_vars="index",
var_name='Development Month',
value_name='Paid Value').rename({'index': 'Origin Month'},
axis='columns')
tri_paid_count = cl.Triangle(
data_policies,
origin='Start_date',
index='Class_name',
development='Claim_payment_date',
columns='Claim_count_paid',
cumulative=False).incr_to_cum().to_frame().reset_index().melt(
id_vars="index",
var_name='Development Month',
value_name='Paid Count').rename({'index': 'Origin Month'},
axis='columns')
tri_reported = cl.Triangle(
data_policies,
origin='Start_date',
index='Class_name',
development='Claim_report_date',
columns='Claim_value',
cumulative=False).incr_to_cum().to_frame().reset_index().melt(
id_vars="index",
var_name='Development Month',
value_name='Reported Value').rename({'index': 'Origin Month'},
axis='columns')
tri_reported_count = cl.Triangle(
data_policies,
origin='Start_date',
index='Class_name',
development='Claim_report_date',
columns='Claim_count_reported',
cumulative=False).incr_to_cum().to_frame().reset_index().melt(
id_vars="index",
var_name='Development Month',
value_name='Reported Count').rename({'index': 'Origin Month'},
axis='columns')
tri_premium = cl.Triangle(
data_policies,
origin='Start_date',
index='Class_name',
development='Policy_premium_date',
columns='Policy_premium',
cumulative=False).incr_to_cum().to_frame().reset_index().melt(
id_vars="index",
var_name='Development Month',
value_name='Premium Value').rename({'index': 'Origin Month'},
axis='columns')
#join all triangles together into a single dataframe
join_cols = ['Origin Month', 'Development Month']
tri_all = tri_paid.merge(tri_paid_count, on=join_cols)
tri_all = tri_paid.merge(tri_reported, on=join_cols)
tri_all = tri_all.merge(tri_reported_count, on=join_cols)
tri_all = tri_all.merge(tri_premium, on=join_cols)
return tri_all, data_policies
=================
Although triangles have both origin and development attributes, it is often
convenient to create premium or exposure vectors that can work with loss
triangles. The `Triangle` class treats the development parameter as
optional. This example instantiates a 'premium' triangle as a single vector.
"""
import chainladder as cl
import pandas as pd
import chainladder as cl
# Raw premium data in pandas
premium_df = pd.DataFrame(
{'AccYear':[item for item in range(1977, 1988)],
'premium': [3000000]*11})
# Create a premium 'triangle' with no development
premium = cl.Triangle(premium_df, origin='AccYear', columns='premium')
# Create some loss triangle
loss = cl.load_sample('abc')
ultimate = cl.Chainladder().fit(loss).ultimate_
# Plot
(ultimate / premium).plot(
kind='area', title='Loss Ratio by Accident Year',
alpha=0.7, color='darkgreen', legend=False, grid=True).set(
xlabel='Accident Year', ylabel='Loss Ratio');
def test_base_minimum_exposure_triangle():
d = (raa.latest_diagonal * 0 + 50000).to_frame().reset_index()
d['index'] = d['index'].astype(str)
cl.Triangle(d, origin='index', columns=d.columns[-1])
def test_df_period_input():
assert raa == raa_gt
d = raa.latest_diagonal
df = d.to_frame().reset_index()
assert cl.Triangle(df, origin="index", columns=df.columns[-1]) == d
import chainladder as cl
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
# Read in the data
lobs = 'wkcomp'
data = pd.read_csv(
r'https://www.casact.org/research/reserve_data/wkcomp_pos.csv')
data = data[data['DevelopmentYear'] <= 1997]
# Create a triangle
triangle = cl.Triangle(
data,
origin='AccidentYear',
development='DevelopmentYear',
index=['GRNAME'],
columns=['IncurLoss_D', 'CumPaidLoss_D', 'EarnedPremDIR_D'])
# Output
print('Raw data:')
print(data.head())
print()
print('Triangle summary:')
print(triangle)
print()
print('Aggregate Paid Triangle:')
print(triangle['CumPaidLoss_D'].sum())
plot_data = triangle['CumPaidLoss_D'].sum().to_frame().unstack().reset_index()
plot_data.columns = [
#http://chainladder-python.readthedocs.io/en/master/quickstart.html
import chainladder as chainl
import numpy as numpi
import pyliferisk as pylife
import matplotlib.pyplot as plt
##%matplotlib inline
###############vars####################
s = "This is a test"
##############objects###############
RAA = chainl.load_dataset('RAA')
RAA_TriangleObj = chainl.Triangle(RAA)
RAA_ChainL = chainl.Chainladder(RAA_TriangleObj)
#############calls##############
RAA.round(0)
############ sys out #####################
def main():
# my code here
print(s + " 1")
print(RAA.round(0))
print("\n\n\n\n")
print(s + " 2")
plt.plot(RAA_TriangleObj.cum_to_incr())
plt.ylabel("Aggregate Claim")
plt.xlabel("Year")
plt.show()
print("\n\n\n\n")
print(s + " 3")
