ifrs17sim IFRS waterfall chart

The script draws multiple waterfall charts, each of which depict sources of changes in balances related to IFRS17. The script is broken down into pieces of code, and explained in a Jupyter notebook, IFRS17 waterfall chart notebook.

The live version of the notebook is available online.

Launch this notebook online! launch binder

  • ../../_images/sphx_glr_plot_ifrs_waterfall_001.png
  • ../../_images/sphx_glr_plot_ifrs_waterfall_002.png
  • ../../_images/sphx_glr_plot_ifrs_waterfall_003.png
  • ../../_images/sphx_glr_plot_ifrs_waterfall_004.png
import pandas as pd
import collections
import matplotlib.pyplot as plt
from draw_charts import draw_waterfall

try:
    import ifrs17sim.ifrs17sim as ifrs17sim
except ImportError:
    import ifrs17sim

model = ifrs17sim.build(True)
proj = model.OuterProjection[1]

proj.CSM_Unfloored(15)
data = collections.OrderedDict()

for cells in ['CSM_Unfloored',
              'IntAccrCSM',
              'AdjCSM_FulCashFlows',
              'TransServices']:
    data[cells] = [proj.cells[cells](t) for t in range(3)]


df = pd.DataFrame(data)
df['TransServices'] = -1 * df['TransServices']

draw_waterfall(df, title='CSM Amortization')

# %% Expected Cashflow Rollforwad

data.clear()
for cells in ['PV_FutureCashflow',
              'ExpectedPremium',
              'ExpectedInterestCashflow',
              'ExpectedAcqCashflow',
              'ExpectedClaims',
              'ExpectedExps']:
    data[cells] = [proj.cells[cells](t) for t in range(3)]

estcf = pd.DataFrame(data)

for inflow in ['ExpectedPremium']:
    estcf[inflow] = -1 * estcf[inflow]

plt.figure()
draw_waterfall(estcf, title='Expected Cashflows')

# %% Actual Cashflow Rollforward

data.clear()
for cells in ['prj_AccumCashflow',
              'prj_incm_Premium',
              'prj_InterestAccumCashflow',
              'prj_exps_AcqTotal',
              'prj_bnft_Total',
              'prj_exps_MaintTotal']:
    data[cells] = [proj.cells[cells](t) for t in range(3)]

actcf = pd.DataFrame(data)
for outflow in ['prj_exps_AcqTotal',
                'prj_bnft_Total',
                'prj_exps_MaintTotal']:
    actcf[outflow] = -1 * actcf[outflow]


plt.figure()
draw_waterfall(actcf, title='Actual Cashflows')

# %% IFRS17 Financial Performance

data.clear()
for cells in ['NetBalance',
              'InsRevenue',
              'InsServiceExps',
              'InsFinanceIncomeExps']:
    data[cells] = [proj.cells[cells](t) for t in range(5)]


ifrspl = pd.DataFrame(data)
for pl in ['InsServiceExps']:
    ifrspl[pl] = -1 * ifrspl[pl]

plt.figure()
draw_waterfall(ifrspl, title='IFRS17 Profit/Loss')

Total running time of the script: ( 0 minutes 4.297 seconds)

Gallery generated by Sphinx-Gallery