Artificial Intelligence and Firm Resilience: Evidence from Firm Performance under Disaster Shocks ###

Artificial Intelligence and Firm Resilience: Evidence from Firm Performance under Disaster Shocks

(with Hongchuan Shen, Jing Wu, and Xiaoquan Michael Zhang)

Conditionally accept at Information Systems Research

Conferences: SCECR (2023/6); CIST (2022/10); INFORMS (2022/10); CSWIM (2022/8, Best Paper Award)

Abstract & Keywords

Artificial intelligence (AI) has been increasingly implemented in business operations over the past decade. While AI value in normal times is extensively studied, direct evidence of its effectiveness in uncertain contexts is limited. Our work fills this gap by focusing on the contribution of AI to corporate resilience. Specifically, we measure firm AI investment with the cumulative demands of AI-relevant skills extracted from a comprehensive job posting database. We gauge firm resilience with the unexpected changes in corporate valuation induced by environmental shocks from a universe of natural disasters. Using a pooled event study approach, we find that AI generates resilience: an average firm that demands 2.4\% of total jobs to be AI-related could approximately recover the full damage of disasters reflected in corporate valuation over a short event window. Then, we discuss mechanisms under the framework of an adapted production function. Combined with an instrumental variable that integrates the baseline firm-specific task structure and over time industry-level task-embedded AI growth, we find consistent evidence that, during turbulent periods, AI deployment moderates the decreased responsiveness of firm performance to both labor and capital inputs in the production process. An array of sub-sample analyses reveal a pressing phenomenon that although currently under-performing firms could \textit{potentially} benefit more from an additional unit of AI investment, the \textit{realized} productivity of which is notably restrained due to a lack of complementary organizational designs. Overall, our study makes a distinct contribution relative to prior literature that has focused on AI productivity while assuming constant certainty and not allowing for heterogeneity in firm-level factor elasticity. Our findings provide managerial implications regarding the interplay between environmental uncertainty and firm investment in both AI technology and complementary infrastructures.
**Keywords**: economics of AI, firm production, event study, production function, instrumental variable

An Empirical Study of Algorithm-Induced Online Information Misallocation

(with Hongchuan Shen, Sihan Zhai, and Xiaoquan Michael Zhang)

Under review at Management Science

Conferences: Workshop at HKU (2023/7); CoDE at MIT (2022/10); AFE at U. of Chicago (2022/6)

Abstract & Keywords

In this project, we investigate algorithm-induced information misallocation in the online context. We empirically examine archival data and conduct a large-scale field experiment on Facebook. We find that instead of informing the poorly informed, information is systematically more likely to be distributed to those who are already well informed. With a set of laboratory tests, we identify the cause of this misallocation: algorithms tend to maximize user online engagement, while overlooking the inherent discrepancy between user interests and user needs. Specifically, our results consistently show that individuals with the highest informational needs are not covered by the current algorithms, leading to their exclusion from the allocation pool. Our welfare analysis suggests that allocating more information to the poorly informed can potentially improve efficiency through increased user willingness-to-pay. Overall, our study suggests that it would be socially desirable if content providers were allowed to reveal whether their content pertains to entertainment or information.
**Keywords**: recommendation algorithm, information misallocation, field test, difference-in-differences, lab experiment

Trade and Foreign Economic Policy Uncertainty in Supply Chain Networks: Who Comes Home

(with Ben Charoenwong and Jing Wu)

Manufacturing & Service Operations Management, 25 (1)

Media: Yahoo Finance (2020/2); Reuters (2020/2); The Economist (2020/12)

Conferences: M&SOM SIG (2021/6); AEA Annual Meeting (2021/1)

Abstract & Keywords

The uncertainty around trade and foreign economic policy contributes to supply chain risk. Whether such policy uncertainty will bring some production back to the United States or only redistribute the global supply chains among foreign sources is theoretically ambiguous and warrants an empirical analysis. In this paper, we study the relationship between trade and foreign economic policy uncertainty and the supply chain networks of American firms. We use firm-level global supply chain data, transaction-level shipping container data, and policy uncertainty indexes constructed from leading media outlets to study how policy uncertainty correlates with changes in supply chain networks. When U.S. trade policy uncertainty rises, firms with majority domestic sales decrease their supplier base abroad, whereas firms with majority foreign sales increase the number of foreign suppliers. Firms also substitute among foreign countries in response to their respective economic policy uncertainty—shifting suppliers from countries with higher uncertainty to ones with lower uncertainty. Firms requiring more specific inputs, producing more differentiated products, having higher market shares, and more central to the production network are more sensitive to policy uncertainty. Supply chain restructuring following higher policy uncertainty puts the market value at risk. Managers should consider customers’ locations when making global supply chain restructuring decisions.
**Keywords**: supply chain network, economic policy uncertainty

The impact of government regulation on sharing platform growth: A channel of supplier behavior change

(with Xiaoquan Michael Zhang)

ICIS 2020 Proceedings

Abstract & Keywords

There has been practical debates over the sharing economy and government intervention. Applying a difference-in-differences approach on Airbnb listing records in 20 U.S. cities from 2015 to 2020, we conclude a negative impact of government regulation on platform growth, through a channel of supplier behavior change. Our empirical analysis finds that implementation of license policy causes a significant decrease in the individual supplier activeness, quantified by 3.04% lower response rate, 3.22-hour longer response time and 4.71% lower acceptance rate. Subsequently, the change in behaviors leads to reduced revenue at host level and depressed booking demand, an estimate of 648 fewer requests each month, at platform level. We provide mechanisms by which the supplier-behavior channel exists, and discuss managerial implications for local policymakers and platform managers.
**Keywords**: digital platform, supplier behavior, policy regulation, difference-in-differences

Passion Matters but not Equally Everywhere: Predicting Achievement from Interest, Enjoyment, and Efficacy in 59 Societies

(with Xingyu Li, Geoffrey L Cohen, and Hazel Rose Markus)

Proceedings of the National Academy of Sciences, 118 (11)

Media: Tweet (by Prof Steven Strogatz); British Psychological Society

Abstract & Keywords

How to identify the students and employees most likely to achieve is a challenge in every field. American academic and lay theories alike highlight the importance of passion for strong achievement. Based on a Western independent model of motivation, passionate individuals—those who have a strong interest, demonstrate deep enjoyment, and express confidence in what they are doing—are considered future achievers. Those with less passion are thought to have less potential and are often passed over for admission or employment. As academic institutions and corporations in the increasingly multicultural world seek to acquire talent from across the globe, can they assume that passion is an equally strong predictor of achievement across cultural contexts? We address this question with three representative samples totaling 1.2 million students in 59 societies and provide empirical evidence of a systematic, cross-cultural variation in the importance of passion in predicting achievement. In individualistic societies where independent models of motivation are prevalent, relative to collectivistic societies where interdependent models of motivation are more common, passion predicts a larger gain (0.32 vs. 0.21 SD) and explains more variance in achievement (37% vs. 16%). In contrast, in collectivistic societies, parental support predicts achievement over and above passion. These findings suggest that in addition to passion, achievement may be fueled by striving to realize connectedness and meet family expectations. Findings highlight the risk of overweighting passion in admission and employment decisions and the need to understand and develop measures for the multiple sources and forms of motivation that support achievement.
**Keywords**: culture, passion, achievement, hierarchical linear model

On the limits of algorithmic prediction across the globe

(with Xingyu Li, Difan Song, Yu Zhang, and Rene F Kizilcec)

arXiv preprint arXiv:2103.15212

Abstract & Keywords

The impact of predictive algorithms on people's lives and livelihoods has been noted in medicine, criminal justice, finance, hiring and admissions. Most of these algorithms are developed using data and human capital from highly developed nations. We tested how well predictive models of human behavior trained in a developed country generalize to people in less developed countries by modeling global variation in 200 predictors of academic achievement on nationally representative student data for 65 countries. Here we show that state-of-the-art machine learning models trained on data from the United States can predict achievement with high accuracy and generalize to other developed countries with comparable accuracy. However, accuracy drops linearly with national development due to global variation in the importance of different achievement predictors, providing a useful heuristic for policymakers. Training the same model on national data yields high accuracy in every country, which highlights the value of local data collection.
**Keywords**: big data, machine learning, algorithmic bias, cultural differences, Human Development Index