Facilitating Regulatory Impact Assessments: The Benefits of Machine Learning in Legislation
Sylvana Walprecht* 1, Catharina Lewerenz* 1
Abstract
Laws have consequences. Besides their regulatory nature to govern behaviour, they frequently impose costs on the addressees of a regulation in the form of material expenses and time expenditure. To mitigate the burden on those affected by regulations, the associated costs are estimated beforehand as part of the legislative process. However, ex-ante calculations are very time-intensive and consume a lot of staff capacity in the ministries as well as in the Federal Statistical Office. That is why, we propose a method to catalyse and facilitate regulatory impact assessments by means of computer-based calculations.
Following the extraction of regulations spanning from 2012 to the present, along with their synopses to account for alterations, our initial step connects segments from each regulation with corresponding entries in our compliance costs database (OnDEA). This process is aimed to generate a learning dataset for a machine learning algorithm capable of discerning word combinations signalling either a change or no change in compliance costs. This approach is expected to streamline the identification of passages containing compliance costs in future legal texts.
In the subsequent step, we enhance an existing machine learning model, with the objective of enabling rapid cost estimations with just a few key pieces of information concerning the legal requirement under examination. The concept is as follows: If the anticipated effort is highly likely to be minimal, it can be swiftly computed automatically. Conversely, if there is a likelihood of substantial effort, it is advisable to conduct more precise manual estimates instead. By doing so, staff capacities will be redirected to focus on the truly crucial estimations.