# “Good and bad investments” in public health stocks amid the COVID-19 shock: evidence from a transformer-based model

## TL;DR

This study uses a Transformer model to analyze how public health stocks performed during the pandemic, showing how markets evolved and how investments can be guided.

## Contribution

The study introduces a Transformer-based model to analyze investment dynamics in public health stocks during the pandemic, revealing market evolution and investment patterns.

## Key findings

- The Transformer model identified excess returns in healthcare and epidemic-prevention stocks during the initial pandemic phase.
- Market volatility and policy uncertainty in the mid-pandemic stage reduced model stability.
- Post-pandemic, stock performance became driven by fundamentals, improving model accuracy.

## Abstract

Major public health emergencies have profoundly reshaped the risk structure and resource allocation logic of capital markets. The market performance of public health-related enterprises has exhibited substantial heterogeneity across different stages of the pandemic, characterized by both considerable risks and emerging opportunities. Understanding this dynamic process is essential for maintaining financial stability and promoting rational investment behavior.

Using the COVID-19 pandemic as the research background, this study selects 55 constituent stocks from the China Securities Index (CSI) Public Health Index as the research sample. A deep learning model based on the Transformer architecture is employed to forecast stock returns and construct long-short investment portfolios. By conducting stage-wise comparisons spanning the pre-pandemic period, the initial outbreak, the normalization phase, and the post-pandemic era, the study reveals the profound temporal evolution and dynamic impacts of public health crises on market investment behavior.

The empirical results reveal that the capital market underwent substantial structural reshaping during the initial phase of the pandemic. The Transformer model effectively identified excess return signals from healthcare and epidemic-prevention enterprises, thereby achieving outstanding investment performance. In the mid-pandemic stage, increased market volatility and policy uncertainty weakened the model's stability. As the market transitioned into the post-pandemic period, rationality gradually returned. Similar to the pre-pandemic stage, firms' performance became increasingly driven by fundamentals rather than policy influences, leading to a marked improvement in the model's predictive accuracy and screening capability.

This study systematically reveals the structural differentiation and dynamic evolution of public health-related enterprises during the pandemic, thereby extending the research frontier at the intersection of public health emergency response, financial risk, and investment portfolio construction. By bridging these domains, it provides both theoretical foundations and empirical evidence to guide investment strategies and policy formulation for industries closely associated with public health, contributing to more resilient and informed financial decision-making amid future crises.