GLM Reserving#
import chainladder as cl
import pandas as pd
This example demonstrates how you can use the TweedieGLM estimator to incorporate
the GLM framework into a chainladder workflow. The specific case of the Over-dispersed
poisson GLM fit to incremental paids. It is further shown to match the basic
chainladder Development estimator.
genins = cl.load_sample('genins')
# Fit an ODP GLM
dev = cl.TweedieGLM(
design_matrix='C(development) + C(origin)',
link='log', power=1).fit(genins)
# Grab LDFs vs traditional approach
glm = dev.ldf_.iloc[..., 0, :].T.iloc[:, 0].rename('GLM')
traditional = cl.Development().fit(genins).ldf_.T.iloc[:, 0].rename('Traditional')
# Plot data
results = pd.concat((glm, traditional), axis=1)
Show code cell source
import matplotlib.pyplot as plt
plt.style.use('ggplot')
%config InlineBackend.figure_format = 'retina'
ax = results.plot(kind='bar', title='LDF: Poisson GLM vs Traditional');
