These types of questions and many more were recently included in the academic research of the three 2021 Nobel Prize winners in Economics, David Card, Guido Imbens, and Joshua Angrist.
Many of the more difficult questions in the social sciences – as well as those researched by the Nobel team – dealt with cause and effect based on observational data relying on the existence of “natural experiments.” This meant using real-world situations where “chance and randomization” “naturally” occur, as opposed to designed and controlled randomized experiments, which are often not possible due to various regulations or business reasons.
A closer look at natural experiments
In their seminal and somewhat controversial paper, “Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania,” Card and Krueger framed the foundations for using natural experiments to address extremely important policy questions, specifically the impact the increase of minimum wage had on employment levels. The paper got credit among liberal economists for challenging the previous consensus that minimum wage increases lead to reduced employment and is cited often by progressives who support raising the minimum wage. In a 2015 opinion column in the New York Times, Prof. Paul Krugman even claimed that the study has set off an “intellectual revolution.”
As it was not possible to randomly increase wages for one set of workers within the same store, or even across different stores, Card and Krueger had to rely on a natural experiment of raising minimum wages. The state of New Jersey enacted a law that raised the minimum wage while the neighbouring state with a similar demography – Pennsylvania – did not.
They used the “difference in differences” approach by looking at the changes in unemployment levels in New Jersey around the period the law was changed, relative to the same changes in the state of Pennsylvania. This enabled them to estimate the causal effect of the changes in the minimum wage on employment levels.
Natural experiment use cases
Today, the work of the three Nobel Prize winners is both far-reaching and highly relevant to many other disciplines where similar questions may arise. Their work demonstrated what cause-and-effect conclusions can be drawn from natural experiments and helped increased the credibility of empirical research
Imagine the case where we would want to understand the impact of increasing the minimum age of driving on fatal accidents as well as insurance premiums. How would we go about addressing this question? The traditional approach of clinical trials or randomized experiments cannot work here. We cannot design a random trial where we force randomly different minimum ages of driving on different people.
Instead, we could use a natural experiment, where some states or regions enacted different minimum-age requirements compared to other states/regions and then use some of the methods developed by these distinguished economists to analyze the impact and draw causal conclusions.
Applying natural experiments to improve pricing strategies
These novel methodologies and approaches are not only relevant for questions related to potential social impact, but they are also as applicable to a vast array of actions or propositions various financial institutions take on a near-daily basis.
When I was a Ph.D. student of Economics at the University of Pennsylvania, I reviewed Card and Kruger’s methodology and seminal paper for one of my classes. I was initially very skeptical about it. However, with time – and after dealing with similar questions in the many insurance and banking engagements I led – I began to rely heavily on the “likes” of natural experiments, often, looking for (or designing) alternative “natural” or pseudo-random experiments to help draw causal conclusions.
For example, consider an insurance company that developed a new pricing strategy that is intended to increase the number of new policyholders, but with minimal to no impact on overall profitability. In some cases, the company can apply the classic A/B testing methodology to test the effectiveness of this new pricing strategy. However, regulatory, business, or functional reasons may prevent the possibility to implement the classic A/B testing methodology within the same region. Instead, a common practice to test the effectiveness of the new pricing strategy and draw causal conclusions is to initially deploy it in a specific region and use other regions, which “naturally” look like the test regions, as the control group.
Questions related to cause and effect with respect to pricing, underwriting, and product offers are popping up daily and could have a tremendous impact on business results and meeting consumers’ expectations. Considering the key lessons learned from natural experiments will help insurance companies estimate causal relationships from observational data and improve business results.
Written by Reuven Shnaps, PhD., Chief Analytics Officer at Earnix