Time-SeriesForecastingScenario Analysis

InsureCast

Workers' compensation forecasting dashboard with SARIMAX time-series models, scenario analysis, and multi-dimensional segmentation across states, industries, and claim types.

Try Live Demo

Overview

The Problem

Insurance companies need accurate claims forecasts to set reserves, price policies, and plan resources. Manual forecasting is slow, doesn't account for seasonality, and can't easily model 'what-if' scenarios across multiple business dimensions.

Our Solution

InsureCast applies SARIMAX time-series models to historical OSHA injury data, producing 12-month claims and cost forecasts with confidence intervals. Users can segment by state, industry, and claim type, and run scenario analyses with frequency and severity adjustments.

Key Outcomes

12-month claims count and cost forecasts with widening confidence intervals
Multi-dimensional filtering: 50+ states, 3 industries, 3 claim types
Interactive scenario analysis with frequency shocks (±10-25%) and severity inflation (0-20%)
Dynamic cost modeling using lognormal severity distributions with state/industry/type adjustments

Models & Tech Stack

AI/ML Models

SARIMAX

Seasonal time-series forecasting of monthly claims counts

Order (1,1,1) with seasonal order (1,1,1,12) capturing 12-month seasonality. Trained on historical data from 2015 onwards. Achieves MAE: 8.4, RMSE: 10.7, MAPE: 6.2%.

Lognormal Severity Model

Cost estimation per claim based on segment characteristics

Lognormal distribution (μ=8.55, σ=0.72) with base cost $7,200, adjusted by state factor (e.g., CA=1.15, NY=1.2), industry factor (Construction=1.25), and claim type factor (LostTime=1.25). Includes 3% annual inflation and ±2% monthly seasonality.

Tech Stack

Backend

Python 3.13FastAPIUvicornPydantic

Frontend

Next.js 16React 19TypeScriptRechartsTailwind CSS

ML/AI

SARIMAXLognormal distributions

Data Processing

PandasNumPy

Data & Methodology

Data Sources

OSHA Severe Injury Reports (SIR), public occupational injury incident reports from the Bureau of Labor Statistics. Data normalized into monthly claim counts by state, industry (Manufacturing, Construction, Healthcare), and claim type (LostTime, MedicalOnly, Indemnity).

Methodology

OSHA data ingested and aggregated by month/state/industry/claim_type. NAICS codes mapped to industry categories, event types mapped to claim types. Missing month-segment combinations filled with synthetic data sampled from segment-specific distributions. Severity parameters computed per segment with lognormal fitting. SARIMAX models fit with 12-month seasonal period.

Evaluation Metrics

SARIMAX performance: MAE 8.4, RMSE 10.7, MAPE 6.2%. Forecast horizon: 12 months with confidence intervals widening over time.