Session 18: Climate Risk in Insurance

1. Climate Change and Insurance — The Connection

Climate change affects insurance in ways that are simultaneously straightforward and profound. The straightforward part: extreme weather events are becoming more frequent and severe, which increases claims costs for property, crop, motor, and health insurance. The profound part: climate change calls into question the fundamental insurability of certain risks — if an insurer cannot predict the frequency and severity of future losses because the climate is changing faster than the actuarial models, the mechanism of insurance itself is stressed.

1.1 Physical Risks vs. Transition Risks

Climate risks are classified into two categories that affect insurance differently:

1.2 The Challenge to Insurability

For insurance to work — as established in Session 01 — losses must be predictable at the portfolio level. Climate change undermines this predictability in two ways. First, non-stationarity: the statistical distribution of weather events is shifting. The 100-year flood of 1980 may be the 25-year flood of 2025. Actuarial models calibrated on historical data are systematically underestimating current risk. Second, correlated risk escalation: climate change is not increasing all risks equally; it is increasing them in a correlated way. A region that becomes more prone to cyclones also becomes more prone to flooding, and the two perils often coincide in a single event (cyclone storm surge), creating losses that exceed what either single-peril model would predict.

When an insurer's ability to predict and price risk breaks down, the product ceases to function. Some regions — coastal Florida, parts of California, specific cyclone-prone zones in India — are approaching or past the point where private insurers can profitably offer coverage. When insurers exit, the state often becomes the "insurer of last resort," meaning taxpayers bear the growing cost of climate-related disasters. This is not a future scenario — it is already happening.

3. Portfolio Climate Risk Exposure Analysis

The first step in managing climate risk in an insurance portfolio is understanding the exposure: how many policies are in climate-vulnerable zones, what is the total sum insured at risk, and how does the risk concentration align with the insurer's risk appetite and reinsurance programme?

3.1 The Exposure Analysis Framework

A portfolio-level climate risk assessment follows a four-step process:

1. Geographic mapping: Geocode all property, motor, and crop policies to location (address, zip code, district, state). This step is the foundation — without geographic data, climate risk analysis is impossible.
2. Peril overlay: Overlay each policy's location with climate peril maps — flood zones, cyclone wind-speed zones, earthquake zones, heatwave frequency maps, sea-level rise projections. Each policy is assigned a peril score for each relevant peril.
3. Exposure aggregation: Aggregate the total sum insured by peril × region × line of business. Identify concentrations: do any single postal code, district, or region account for more than 5% of the total portfolio exposure?
4. Capital impact: Estimate the probable maximum loss (PML) for each peril at standard return periods (1-in-100 year, 1-in-250 year). Compare to the insurer's risk appetite and available reinsurance capacity. If the PML exceeds the risk appetite, the insurer must: reduce exposure (non-renew high-risk policies), increase premiums for high-risk zones, purchase additional reinsurance, or a combination of all three.

3.2 Exposure Report in Python

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

# Load the policy dataset with location data
policies = pd.read_csv('data/insurance_cleaned.csv')

# For this analysis, we need location information
# If not available, we use policy_type as a proxy for climate exposure
# (Property insurance is most exposed, followed by crop and motor)

# Create climate-exposure-relevant features
# Assign regional exposure scores based on policy characteristics

# 1. Simple exposure proxy: policy type × region vulnerability
region_vulnerability = {
    'Coastal': {'cyclone': 0.8, 'flood': 0.7, 'heatwave': 0.3},
    'Inland': {'cyclone': 0.1, 'flood': 0.4, 'heatwave': 0.6},
    'Mountain': {'cyclone': 0.0, 'flood': 0.5, 'heatwave': 0.1},
    'Urban': {'cyclone': 0.3, 'flood': 0.6, 'heatwave': 0.8}
}

# Assign random regions for analysis (in real life these would come from the data)
np.random.seed(42)
locations = ['Coastal', 'Inland', 'Mountain', 'Urban']
if 'location' not in policies.columns:
    policies['climate_region'] = np.random.choice(locations, size=len(policies), p=[0.3, 0.4, 0.1, 0.2])
else:
    policies['climate_region'] = policies['location']

# Calculate exposure by region and product
exposure = policies.groupby(['climate_region', 'policy_type']).agg(
    policy_count=('policy_id', 'nunique') if 'policy_id' in policies.columns else ('claim_id', 'count'),
    total_premium=('premium', 'sum'),
    total_sum_assured=('sum_assured', 'sum') if 'sum_assured' in policies.columns else ('claim_amount', 'sum')
).reset_index()

print("=" * 75)
print("PORTFOLIO CLIMATE EXPOSURE — Geographic × Product")
print("=" * 75)
for region in locations:
    region_data = exposure[exposure['climate_region'] == region]
    total_prem = region_data['total_premium'].sum()
    print(f"\n{region.upper()} (Total Premium: ₹{total_prem/1e7:.1f} Cr)")
    print(f"{'Product':15s} {'Policies':>10s} {'Premium (₹ Cr)':>15s}")
    for _, row in region_data.iterrows():
        print(f"  {row['policy_type']:15s} {row['policy_count']:>6,.0f}       ₹{row['total_premium']/1e7:>5.1f}")

# Climate risk concentration — % of total portfolio in high-risk zones
# Assume Coastal = high cyclone/flood, Mountain = high flood/GLOF
high_risk_zones = ['Coastal', 'Mountain']
high_risk_exposure = policies[policies['climate_region'].isin(high_risk_zones)]
high_risk_pct = len(high_risk_exposure) / len(policies) * 100

print(f"\n{'=' * 75}")
print(f"CLIMATE RISK CONCENTRATION")
print(f"{'=' * 75}")
print(f"Portfolio in high-risk zones: {high_risk_pct:.1f}%")
print(f"  Coastal (cyclone + flood): {len(policies[policies['climate_region']=='Coastal'])/len(policies)*100:.1f}%")
print(f"  Mountain (flood + GLOF):   {len(policies[policies['climate_region']=='Mountain'])/len(policies)*100:.1f}%")
print(f"Remaining portfolio:        {100-high_risk_pct:.1f}%")
print(f"\nIf concentration exceeds 30% in a single high-risk region,")
print(f"the insurer has a significant accumulation risk that must be")
print(f"addressed through reinsurance or exposure reduction.")

4. Climate Data Analysis in Python

The core analytical task in climate risk insurance is understanding the relationship between weather events and claims. This requires merging weather data (from the India Meteorological Department, satellite data, or global climate datasets) with claims data and analysing the statistical relationship.

4.1 Loading and Merging Weather and Claims Data

# Load weather data (synthetic)
weather = pd.read_csv('data/weather_data.csv')
print(f"Weather data loaded: {len(weather):,} records")
print(f"Columns: {weather.columns.tolist()}")
print(f"Date range: {weather['date'].min()} to {weather['date'].max()}")
print(f"Regions: {weather['region'].unique()}")
print(f"Weather data summary:\n{weather.describe()}")

# Load claims data and merge with weather
claims = pd.read_csv('data/insurance_cleaned.csv')
claims['claim_date'] = pd.to_datetime(claims['claim_date'])
weather['date'] = pd.to_datetime(weather['date'])

# Merge claims with weather by date and region (simplified)
# In production, this would use a spatial-temporal join on coordinates and date
if 'region' in claims.columns:
    merged = claims.merge(weather, left_on=['claim_date', 'region'],
                          right_on=['date', 'region'], how='left')
else:
    # Create a proxy region column from location or distribution
    np.random.seed(42)
    regions_available = weather['region'].unique()
    claims['region'] = np.random.choice(regions_available, size=len(claims))
    merged = claims.merge(weather, left_on=['claim_date', 'region'],
                          right_on=['date', 'region'], how='left')

print(f"\nMerged dataset: {len(merged):,} claims with weather data")
print(f"Match rate: {merged['rainfall_mm'].notna().mean()*100:.1f}%")

4.2 Extreme Weather Event Analysis

# Define extreme weather thresholds (top 10% of each metric)
rain_threshold = weather['rainfall_mm'].quantile(0.90)
temp_threshold = weather['max_temp'].quantile(0.95)
wind_threshold = weather['wind_speed_kmph'].quantile(0.90) if 'wind_speed_kmph' in weather.columns else 0

print(f"Extreme Weather Thresholds:")
print(f"  Heavy rainfall: > {rain_threshold:.1f} mm/day (90th percentile)")
print(f"  Extreme heat:   > {temp_threshold:.1f}°C (95th percentile)")
print(f"  High wind:      > {wind_threshold:.0f} km/h (90th percentile)")

# Flag extreme weather days
merged['extreme_rain'] = (merged['rainfall_mm'] > rain_threshold).astype(int)
merged['extreme_heat'] = (merged['max_temp'] > temp_threshold).astype(int)
merged['extreme_wind'] = (merged['wind_speed_kmph'] > wind_threshold).astype(int) if 'wind_speed_kmph' in merged.columns else 0

# Compare claims on extreme weather days vs. normal days
print(f"\n{'=' * 65}")
print(f"CLAIMS ANALYSIS: Extreme Weather vs. Normal Days")
print(f"{'=' * 65}")

extreme_rain_claims = merged[merged['extreme_rain'] == 1]
normal_claims = merged[merged['extreme_rain'] == 0]

if len(extreme_rain_claims) > 50 and len(normal_claims) > 50:
    rain_freq_extreme = len(extreme_rain_claims)
    rain_freq_normal = len(normal_claims)
    print(f"\nHeavy Rain Analysis:")
    print(f"  Claims on extreme rain days:     {rain_freq_extreme:>6,.0f}")
    print(f"  Claims on normal days:           {rain_freq_normal:>6,.0f}")
    print(f"  Ratio:                           {rain_freq_extreme / max(rain_freq_normal, 1):.2f}x")
    print(f"  Avg claim amount (extreme rain):  ₹{extreme_rain_claims['claim_amount'].mean():>8,.0f}")
    print(f"  Avg claim amount (normal):        ₹{normal_claims['claim_amount'].mean():>8,.0f}")

# Monthly aggregation for trend analysis
merged['month'] = merged['claim_date'].dt.month
merged['year'] = merged['claim_date'].dt.year

monthly_claims_weather = merged.groupby(['year', 'month']).agg(
    claim_count=('claim_id', 'count'),
    avg_claim=('claim_amount', 'mean'),
    total_claim_amount=('claim_amount', 'sum'),
    avg_rainfall=('rainfall_mm', 'mean'),
    max_temp=('max_temp', 'max')
).reset_index()

monthly_claims_weather['date'] = pd.to_datetime(
    monthly_claims_weather['year'].astype(str) + '-' +
    monthly_claims_weather['month'].astype(str) + '-01'
)

# Plot: claims vs. rainfall over time
fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(14, 8))

ax1.plot(monthly_claims_weather['date'], monthly_claims_weather['claim_count'],
         color='#6c5ce7', linewidth=2, marker='o', markersize=4)
ax1.set_ylabel('Monthly Claim Count', fontsize=11)
ax1.set_title('Claims Volume vs. Rainfall — Climate Correlation Analysis', fontweight='bold')
ax1.spines['top'].set_visible(False)
ax1.spines['right'].set_visible(False)

ax2.bar(monthly_claims_weather['date'], monthly_claims_weather['avg_rainfall'],
        color='#00d2d3', alpha=0.7, width=15)
ax2.set_ylabel('Average Rainfall (mm)', fontsize=11)
ax2.set_xlabel('Date')
ax2.spines['top'].set_visible(False)
ax2.spines['right'].set_visible(False)

plt.tight_layout()
plt.show()

# Correlation analysis
corr_rain_claims = monthly_claims_weather['avg_rainfall'].corr(monthly_claims_weather['claim_count'])
corr_temp_claims = monthly_claims_weather['max_temp'].corr(monthly_claims_weather['claim_count'])

print(f"\nCorrelation Analysis:")
print(f"  Rainfall vs. Claim Count:     {corr_rain_claims:+.3f}")
print(f"  Max Temperature vs. Claims:    {corr_temp_claims:+.3f}")
print(f"\nInterpretation:")
if corr_rain_claims > 0.3:
    print(f"  Positive correlation between rainfall and claims — consistent with flood damage.")
if corr_rain_claims > 0.5:
    print(f"  STRONG correlation — rainfall is a significant driver of claims in this portfolio.")
if corr_temp_claims > 0.3:
    print(f"  Heat is associated with higher claims — consistent with health and property impacts.")

4.3 Lag Analysis — Do Weather Events Predict Future Claims?

# Lag analysis: does heavy rainfall predict higher claims in the following week/month?
merged['claim_month'] = merged['claim_date'].dt.month
merged['claim_year'] = merged['claim_date'].dt.year

# Aggregate to monthly — current month rainfall vs. next month claims
monthly_rain = merged.groupby(['claim_year', 'claim_month']).agg(
    rainfall=('rainfall_mm', 'mean'),
    claims=('claim_id', 'count')
).reset_index()

# Create lagged variables
monthly_rain['rainfall_lag1'] = monthly_rain['rainfall'].shift(1)
monthly_rain['claims_lead1'] = monthly_rain['claims'].shift(-1)

current_monthly = monthly_rain.dropna()

if len(current_monthly) > 10:
    lag_corr = current_monthly['rainfall'].corr(current_monthly['claims_lead1'])
    print(f"\n{'=' * 55}")
    print(f"LAG ANALYSIS: Does this month's rainfall predict next month's claims?")
    print(f"{'=' * 55}")
    print(f"  Correlation (rainfall_t → claims_t+1): {lag_corr:+.3f}")

    if abs(lag_corr) > 0.2:
        print(f"  {'↗ A lag effect of ~1 month exists — heavy rainfall predicts higher claims in the following month.' if lag_corr > 0 else '↘ Negative lag correlation — unusual pattern, may indicate data issue or seasonal effect.'}")
        print(f"  This is consistent with the delay between a flood event and the filing of")

5. Climate Risk Visualization in Looker Studio

Climate risk dashboards are essential for both internal risk management (monitoring portfolio exposure) and external reporting (regulatory disclosure, investor communication, and reinsurance submissions). Looker Studio's integration with Google Maps, its free access model, and its ease of sharing make it a practical choice for smaller insurers and InsurTechs that cannot afford enterprise BI tools.

5.1 Dashboard Structure

LOOKER STUDIO DASHBOARD: CLIMATE RISK MONITOR

Page 1 — Portfolio Exposure Overview:
┌─────────────────────────────────────────────────────────────────────┐
│ CLIMATE RISK MONITOR  [Slicers: Time Period ▼] [Product ▼] [Region ▼]
├──────────────────┬──────────────────┬──────────────────┬─────────────┤
│ Total Policies   │ Total Sum Insured│ Highest Risk     │ Protection  │
│ at Risk          │ at Risk (₹ Cr)   │ Region           │ Gap (%)    │
│ XX,XXX           │ ₹X,XXX Cr        │ XXXX             │ XX%        │
├──────────────────┴──────────────────┴──────────────────┴─────────────┤
│ Portfolio Exposure Map (Google Maps / Geo Chart)                     │
│ • Heatmap overlay: Sum insured by district/state                    │
│ • Layer toggle: Cyclone zones, Flood zones, EQ zones                │
│ • Drill-down: Click state → district → policy detail                │
├──────────────────────────┬──────────────────────────────────────────┤
│ Exposure by Peril (Donut)│ Top 10 High-Risk Districts (Table)      │
│ • Flood:        XX%     │ District │ Pols │ SI (Cr) │ PML100 │    │
│ • Cyclone:      XX%     │ ──────────────────────────────────── │    │
│ • Heatwave:     XX%     │ XXXX     │ XXXX  │ XXXX    │ XXXX   │    │
│ • Earthquake:   XX%     │ XXXX     │ XXXX  │ XXXX    │ XXXX   │    │
└──────────────────────────┴──────────────────────────────────────────┘

Page 2 — Claims × Weather Trend Analysis:
┌─────────────────────────────────────────────────────────────────────┐
│ Time series: Claims per Month vs. Rainfall (dual-axis line chart)  │
│ • Claims volume (bar) + Rainfall (line, secondary axis)            │
│ • Reference lines for extreme weather thresholds                   │
│ • Annotations for major weather events (cyclones, floods)          │
├──────────────────────────┬──────────────────────────────────────────┤
│ Correlation Matrix       │ Claims Impact by Event Type             │
│ (Peril × Claims Increase)│ (Bar chart: % change in claims during   │
│                          │  extreme weather events vs. normal)     │
└──────────────────────────┴──────────────────────────────────────────┘

5.2 Building the Dashboard

Step 1 — Data source: Upload the merged weather + claims dataset (from Section 4) to Google Sheets. In Looker Studio, connect to this Google Sheet as the data source. Ensure date fields are recognised as dates and numeric fields as numbers.

Step 2 — Page 1 — Portfolio Exposure: Add a Google Map chart with "region" as the location field and "total_sum_assured" as the metric. Add a second layer showing cyclone zones (if shapefile is available as a data source). Add a donut chart showing sum insured by peril type. Add a table with the top 10 high-risk districts sorted by PML estimate.

Step 3 — Page 2 — Claims × Weather: Add a time series chart with monthly claim count as the metric and month as the date dimension. Add a second series: average rainfall (right Y-axis). Configure these chart settings: legend at bottom, connect null values (on), missing data as line gaps. Add reference lines at the 90th percentile rainfall threshold. Add annotations for known extreme weather events (cyclone dates, flood events).

Step 4 — Interactivity: Add date range control. Add a product-type filter dropdown. Add a region filter. Ensure all charts update when filters change. Test: selecting "Cyclone" in the peril filter should update the map, the donut chart, and the claims trend simultaneously.

Step 5 — Sharing: Share the dashboard with view-only permissions to the underwriting team and the risk committee. Set up scheduled email delivery (PDF snapshot) to the CRO weekly during monsoon season (June–September) and monthly during the rest of the year.

7. The Protection Gap

The protection gap — the difference between total economic losses from a disaster and the portion covered by insurance — is the most important climate-finance metric for the insurance industry. It measures the proportion of economic loss that is uninsured — meaning the loss is absorbed by individuals, businesses, or the government (ultimately, taxpayers). In India, the protection gap for natural catastrophes exceeds 90% — meaning less than 10% of climate-related economic losses are insured.

7.1 The Numbers

# Protection gap simulation for Indian natural catastrophes
events = pd.DataFrame({
    'event': ['Cyclone Amphan (2020)', 'Kerala Floods (2018)',
              'Cyclone Fani (2019)', 'Uttarakhand Flood (2021)',
              'Chennai Floods (2021)', 'Heatwave (2024, estimated)'],
    'economic_loss_cr': [97200, 33600, 58560, 3640, 14550, 21800],
    'insured_loss_cr': [9750, 1680, 4380, 240, 900, 520]  # Estimated
})

events['protection_gap_pct'] = ((events['economic_loss_cr'] - events['insured_loss_cr']) /
                                  events['economic_loss_cr'] * 100)
events['gap_amount_cr'] = events['economic_loss_cr'] - events['insured_loss_cr']

print("=" * 80)
print("INDIA'S NATURAL CATASTROPHE PROTECTION GAP")
print("=" * 80)
print(f"{'Event':30s} {'Economic':>10s} {'Insured':>10s} {'Gap %':>8s} {'Uninsured':>12s}")
print("-" * 70)
for _, row in events.iterrows():
    print(f"{row['event']:30s} ₹{row['economic_loss_cr']:>7,.0f}Cr ₹{row['insured_loss_cr']:>5,.0f}Cr {row['protection_gap_pct']:>6.1f}% ₹{row['gap_amount_cr']:>6,.0f}Cr")

total_economic = events['economic_loss_cr'].sum()
total_insured = events['insured_loss_cr'].sum()
total_gap_pct = (total_economic - total_insured) / total_economic * 100

print("-" * 70)
print(f"{'TOTAL':30s} ₹{total_economic:>7,.0f}Cr ₹{total_insured:>5,.0f}Cr {total_gap_pct:>5.1f}%  ₹{total_economic - total_insured:>6,.0f}Cr")
print(f"\nThe protection gap for these major events averages {total_gap_pct:.1f}%.")
print(f"For every ₹100 of economic loss from natural catastrophes in India,")
print(f"approximately ₹{100 - total_gap_pct:.0f} is insured — the rest is absorbed")
print(f"by individuals, businesses, and the government (taxpayers).")

7.2 Why the Protection Gap Matters

The protection gap is not just an insurance metric — it is a development metric. When a flood destroys a village and 90% of the losses are uninsured, the consequences are severe: families fall into poverty (the World Bank estimates that 26 million people globally are pushed into poverty each year by natural disasters), businesses cannot rebuild because they lack the capital, and the government must redirect budget from other priorities (education, health, infrastructure) to disaster relief. Insurance — particularly climate risk insurance — is a climate adaptation tool, not just a financial product. Closing the protection gap is one of the most important contributions the insurance industry can make to sustainable development.

7.3 How to Close the Gap

Closing the Indian climate protection gap requires action on multiple fronts simultaneously:

• Product innovation: Traditional indemnity insurance is too expensive to distribute and administer for low-premium, high-volume climate-exposed populations. Parametric insurance (Session 19), micro-insurance, and mobile-first products can reach customers that traditional products cannot. The goal is not "cheaper insurance" but "different insurance" — products with lower distribution costs, faster claims, and simpler terms that match the needs of climate-exposed populations.
• Data infrastructure: Better weather data, flood mapping, cyclone trajectory models, and exposure data enable better pricing — which enables insurers to offer coverage in areas they currently avoid because the risk is unquantifiable. The India Meteorological Department's expanding weather station network, satellite data from ISRO, and the government's PMFBY data infrastructure are reducing this barrier.
• Public-private partnerships: The protection gap cannot be closed by private insurers alone — the risks are too large and the premiums too low for purely commercial solutions. The government must play a role as: (a) reinsurer of last resort for catastrophic events, (b) premium subsidiser for low-income populations, and (c) data provider and standard-setter for climate risk modelling. PMFBY is India's largest example of this PPP model.
• Financial literacy and trust: Many climate-exposed households in India have never purchased insurance and do not trust insurance companies. Building trust through fast, fair claims (the most important marketing investment for climate risk insurance), simple product terms (one-page policies in the local language), and distribution through trusted local institutions (cooperatives, banks, microfinance institutions) is essential.