AI reveals hidden labor market attachment in Spain
The Banco de España proposes a novel machine learning methodology to measure labor market attachment for the non-working population in Spain. This data-driven approach aims to overcome biases in traditional official statistics by considering a broader range of individual-level characteristics.
Beyond traditional unemployment metrics
Official statistics often rely on self-reported variables and rule-based procedures to classify individuals' labor market status, leading to potential biases.
This binary distinction, based on willingness, availability, and active job search, fails to capture the true degree of labor market attachment (LMA).
Factors like reservation wages, the quantity and type of job offers received, or job search intensity are often overlooked.
The continuous nature of attachment is evident in transition probabilities: in 2022, 26% of unemployed in Spain transitioned to employment, while 11% of inactive individuals also found jobs, with some inactive subgroups showing even higher transition rates of 33%.
These figures highlight the need for a more nuanced understanding beyond the simple unemployed-inactive dichotomy.
Machine learning's continuous spectrum
To address these limitations, the Banco de España paper introduces a machine learning, data-driven methodology to measure LMA for non-workers.
Using Spanish Labor Force Survey data, researchers train an estimator with a high-dimensional dataset including covariates related to job search, labor market outcomes, and socioeconomic indicators.
The study employs both unsupervised and supervised machine learning algorithms to recover LMA probability distributions.
It finds that the proposed LMA measure presents a continuous, multi-modal probability distribution, allowing for a more granular understanding of individual attachment.
When clustering these distributions into two groups (attached vs. non-attached), significant differences emerge compared to official methodologies.
Crucial for nuanced policy
This research offers a significant methodological advancement in understanding labor market dynamics beyond simplistic binary classifications.
By leveraging machine learning, it provides a more granular and predictive measure of attachment, which is vital for accurate economic forecasting and targeted policy interventions.
The supervised methodology, in particular, demonstrates superior performance in predicting future employment flows and shows a closer relationship with the economic cycle, making it a valuable tool for policymakers.