PlumBot: Bot: Creating new article from JSON

2026-03-10T04:54:27Z

Bot: Creating new article from JSON

New page

📊 '''Generalized linear model (GLM)''' is a statistical modeling framework that serves as the workhorse of modern insurance [[Definition:Actuarial science | actuarial]] pricing. GLMs extend classical linear regression by allowing the response variable — such as [[Definition:Claim frequency | claim frequency]] or [[Definition:Claim severity | claim severity]] — to follow non-normal distributions like Poisson, gamma, or binomial, which are far more representative of how insurance losses actually behave. Since the late 1990s, GLMs have become the industry standard for [[Definition:Rate-making | rate-making]] across [[Definition:Property and casualty insurance | property and casualty]] lines, and [[Definition:Regulatory compliance | regulators]] in many jurisdictions explicitly accept GLM-based rate filings.

⚙️ Building a GLM involves selecting an appropriate probability distribution for the target variable, defining a link function that connects the linear predictor to the expected value of that distribution, and fitting the model to historical [[Definition:Loss data | loss data]] using maximum likelihood estimation. In practice, an [[Definition:Actuary | actuary]] might model auto insurance claim counts with a Poisson distribution and a log link, then model average claim costs separately with a gamma distribution. [[Definition:Rating factor | Rating factors]] — driver age, vehicle type, territory, [[Definition:Credit score | credit score]], and others — enter the model as covariates, and the resulting multiplicative [[Definition:Relativities | relativities]] translate directly into the insurer's [[Definition:Rating algorithm | rating algorithm]]. [[Definition:Underwriting | Underwriters]] and product teams can interpret GLM outputs with relative ease compared to black-box [[Definition:Machine learning | machine learning]] methods, which is a significant practical advantage in a regulated industry where rate justifications must be transparent and defensible.

🔍 Despite the rise of more complex [[Definition:Predictive analytics | predictive modeling]] techniques — gradient boosting, neural networks, and ensemble methods — GLMs remain deeply embedded in insurance pricing workflows because of their interpretability, stability, and regulatory acceptance. Many insurers now employ a hybrid approach: using [[Definition:Machine learning | machine learning]] for feature discovery and variable transformation, then feeding those insights into a GLM structure that regulators and business stakeholders can readily examine. This pragmatic blend reflects the insurance industry's dual mandate of analytical sophistication and transparency. For [[Definition:Insurtech | insurtechs]] building pricing engines from scratch, a solid GLM foundation is often the fastest path to a defensible, deployable product — and overlooking it in favor of trendy algorithms can create unnecessary friction with [[Definition:State insurance department | state regulators]] and [[Definition:Rating bureau | rating bureaus]].

'''Related concepts'''
{{Div col|colwidth=20em}}
* [[Definition:Actuarial science]]
* [[Definition:Predictive analytics]]
* [[Definition:Rate-making]]
* [[Definition:Machine learning]]
* [[Definition:Loss ratio (L/R)]]
* [[Definition:Rating factor]]
{{Div col end}}

Definition:Generalized linear model (GLM) - Revision history

PlumBot: Bot: Creating new article from JSON