

In [ ]:

    
%matplotlib inline

Stochastic Bornhuetter Ferguson

There are several aspects of the chainladder module that are demonstrated with this example.

We see how to use the BootstrapODPSample and BornhuetterFerguson to come up with a stochastic view of the Bornhuetter-Ferguson method.
We use the broadcast_axis method of the triangle class (new in 0.4.7)



In [ ]:

    
import chainladder as cl

# Simulation parameters
random_state = 42
n_sims = 1000

# Get data
loss = cl.load_dataset('genins')
premium = loss.latest_diagonal*0+8e6

# Simulate loss triangles
sim = cl.BootstrapODPSample(random_state=random_state, n_sims=n_sims)
sim.fit(loss, sample_weight=premium)


# Fit Bornhuetter-Ferguson to stochastically generated data
model = cl.BornhuetterFerguson(0.65, apriori_sigma=0.10).fit(sim.resampled_triangles_, sample_weight=premium)

# Grab completed triangle replacing simulated known data with actual known data
full_triangle = model.full_triangle_ - model.X_ + loss.broadcast_axis('index', sim.resampled_triangles_.index)

# Limiting to the current year for plotting
current_year = full_triangle[full_triangle.origin==full_triangle.origin.max()].to_frame().T

# Plot the data
current_year.reset_index(drop=True).plot(
    color='orange', legend=False, alpha=0.1,
    title='Current Accident Year BornFerg Distribution', grid=True);