The insurance business relies on the classification of insured individuals into risky and non-risky categories, with actuarial fairness dictating that those who are deemed riskier should pay higher premiums than those who are less risky. Actuaries use econometric or machine learning techniques to classify individuals, but there is a fine line between fair actuarial classification and discrimination. As such, there is growing interest in fairness and discrimination within the actuarial community. Non-sensitive characteristics can serve as substitutes or proxies for protected attributes, but using them alone may not be enough to maintain adequate accuracy. Traditional pricing models are built in a two-stage structure that considers many potentially biased components such as car or geographic risks, which has significant limitations in achieving fairness. To address this issue, the authors have developed a novel pricing model approach that utilizes autoencoders to generate geographic and car pricing components needed to predict pure premiums while mitigating unwanted bias according to desired metrics. This approach can be generalized to multiple pricing factors and is perfectly adapted for a fairness context since it allows for debiasing of the set of pricing components. The authors' findings suggest that debiasing the predictor alone may not be sufficient to achieve adequate accuracy in predicting insurance prices, highlighting the importance of developing fairer pricing models in the insurance industry.
- Insurance companies divide people into groups based on how risky they are to insure.
- People who are more likely to have accidents or get sick pay higher insurance prices.
- Experts called actuaries use math and computers to figure out which group someone belongs in.
- It's important to be fair when deciding how much someone should pay for insurance, but it can be hard to do without being unfair to some people.
- Some researchers made a new way of deciding insurance prices that tries to be fairer by using special computer programs.
Definitions
- Insurance: a system where people pay money (called premiums) so that if something bad happens (like getting sick or having a car accident), the insurance company will help them pay for it
- Actuary: an expert who uses math and statistics to figure out how much risk there is in different situations
- Premium: the amount of money someone pays for their insurance policy
- Fairness: treating everyone equally and not unfairly favoring one group over another
- Discrimination: treating someone unfairly because of things like their race, gender, or age
Fairness and Discrimination in the Insurance Industry: A Novel Pricing Model Approach
The insurance business relies on the classification of insured individuals into risky and non-risky categories, with actuarial fairness dictating that those who are deemed riskier should pay higher premiums than those who are less risky. Actuaries use econometric or machine learning techniques to classify individuals, but there is a fine line between fair actuarial classification and discrimination. As such, there is growing interest in fairness and discrimination within the actuarial community.
Non-sensitive characteristics can serve as substitutes or proxies for protected attributes, but using them alone may not be enough to maintain adequate accuracy. Traditional pricing models are built in a two-stage structure that considers many potentially biased components such as car or geographic risks, which has significant limitations in achieving fairness. To address this issue, researchers have developed a novel pricing model approach that utilizes autoencoders to generate geographic and car pricing components needed to predict pure premiums while mitigating unwanted bias according to desired metrics.
Autoencoders for Generating Pricing Components
Autoencoders are artificial neural networks used for unsupervised learning tasks such as dimensionality reduction or feature extraction from raw data inputs without any labels or supervision signals. Autoencoders consist of an encoder component which compresses input data into latent representations (or embeddings) and a decoder component which reconstructs the original input from these embeddings. The authors' proposed approach uses autoencoders to generate geographic and car pricing components needed for predicting pure premiums while mitigating unwanted bias according to desired metrics. This approach can be generalized to multiple pricing factors and is perfectly adapted for a fairness context since it allows for debiasing of the set of pricing components by removing any correlations between sensitive attributes (such as gender or race) and other non-sensitive characteristics (such as age).
Findings Suggest Debiasing Alone May Not Be Sufficient
The authors' findings suggest that debiasing the predictor alone may not be sufficient to achieve adequate accuracy in predicting insurance prices when compared with traditional two-stage approaches that consider many potentially biased components such as car or geographic risks. This highlights the importance of developing fairer pricing models in the insurance industry if we want accurate predictions without introducing unfair biases based on certain protected attributes like gender or race.
Conclusion
This research paper presents a novel approach utilizing autoencoders for generating geographic and car pricing components needed to predict pure premiums while mitigating unwanted bias according to desired metrics. The authors' findings suggest that debiasing alone may not be sufficient when compared with traditional two-stage approaches considering many potentially biased components like car or geographic risks, highlighting the importance of developing fairer pricing models in order to accurately predict insurance prices without introducing unfair biases based on certain protected attributes like gender or race.
At the core of insurance business lies classification between risky and non-risky insureds, actuarial fairness meaning that risky insureds should contribute more and pay a higher premium than non-risky or less-risky ones. Actuaries, therefore, use econometric or machine learning techniques to classify, but the distinction between a fair actuarial classification and "discrimination" is subtle. For this reason, there is a growing interest about fairness and discrimination in the actuarial community Lindholm, Richman, Tsanakas, and Wuthrich (2022). Presumably, non-sensitive characteristics can serve as substitutes or proxies for protected attributes. For example, the color and model of a car, combined with the driver's occupation, may lead to an undesirable gender bias in the prediction of car insurance prices. Surprisingly, we will show that debiasing the predictor alone may be insufficient to maintain adequate accuracy (1). Indeed, the traditional pricing model is currently built in a two-stage structure that considers many potentially biased components such as car or geographic risks. We will show that this traditional structure has significant limitations in achieving fairness. For this reason, we have developed a novel pricing model approach. Recently some approaches have Blier-Wong, Cossette, Lamontagne, and Marceau (2021); Wuthrich and Merz (2021) shown the value of autoencoders in pricing. In this paper, we will show that (2) this can be generalized to multiple pricing factors (geographic, car type), (3) it perfectly adapted for a fairness context (since it allows to debias the set of pricing components): We extend this main idea to a general framework in which a single whole pricing model is trained by generating the geographic and car pricing components needed to predict the pure premium while mitigating the unwanted bias according to the desired metric.
