Welcome to our Winter 2022 Newsletter

LRI Faculty Seed Grants Reminder

Proposals for the FSG are due March 1, 2022. Visit our application guidelines page for more details. 

NSF I-Corps Northeast Hub Update

Join the first cohort of the new regional I-Corps Hub!

Deadline to apply: 3/22

For full application guidelines, visit the I-Corps site. 

Research Spotlights

Dantong Yu

Business Data Science:  Using Graph Neural Network to process spatial-temporal data from the global climate model, power grid, traffic prediction, and distributed sensors. We first apply the attention mechanism to connect the “dots” (sensors, monitoring devices) and learn dynamic network structures among the system components over time. Next, the end-to-end graph neural networks pipeline diffuses and propagates the sensors’ reading and configuration settings into the learned networks and ultimately predicts the future status. We will also design supervised ML models to uncover the intrinsic relationship between the overall network and the system performance, events, and anomalies. Many recent machine learning algorithms were applied to financial data and gained traction. In my research, we model finance data with spatial-temporal signal, i.e., assets are similar to other assets (spatial) and have cyclic patterns (temporal). Graph neural network, combined with recurrent neural networks, provides effective solutions on asset prices, market prediction, and factor discovery.  

Uddin A, Tao S., Chou, C.,Yu D., Are missing values important for earnings forecast? A machine learning perspective, Quantitative Finance, January 2022.

Uddin A, Yu D. Latent factor model for asset pricing. Journal of Behavioral and Experimental Finance.

Complex Network Models for Asset pricing and Financial Decision Making: FinTech Via A Network Lens: Asset pricing plays a vital role in financial investment decisions and still relies on the linear Fama-French models to identify risk factors and estimate excessive returns. AI and machine learning-based asset pricing models are still in their early stage, but demonstrates their effectiveness in improving decision-making. Recent studies suggest that networks among firms (sectors) play an essential role in asset valuation. It is extremely challenging to capture and investigate the implications incurred by those networks because of the continuous evolution of networks in response to market micro and macro changes. To address this challenge, we propose to develop an end-to-end graph neural network model and show its applicability in asset pricing. First, we apply the attention mechanism to learn dynamic network structures of the equity market over time and then use a recurrent convolutional neural network to diffuse and propagate firms' features in the learned networks.

Uddin, A, Tao S., Yu D., Attention Based Dynamic Graph Learning Framework for Asset Pricing, CIKM '21: Proceedings of the 30th ACM International Conference on Information & Knowledge Management, November, 2021, Queensland, Australia.

Heterogenous Network Analysis: Multiple networks (causality, supply-chain, co-investment, and board member networks) exist among firms and provide a rich set of contexts for detecting global and local trends. The asset pricing might be affected by multiple types of networks simultaneously. To cope with these challenges, we apply the self-supervised approach to learn a consensus graph by exploring the graph embeddings of node features concerning multiple graphs (multi-view). We adopt state-of-the-art graph neural networks (Eigenlearn, APPNP) to generate graph embeddings from individual graph views. The learning objective has two terms: the universal smoothness term in the learned graph and a contrastive loss that regularizes the learned graph to distinguish the nearest neighbors (positive samples) in each graph context from those distant (negative samples). It is necessary to perform graph augmentation to ensure model stability, such as add/drop edges and mask node features.

Yao, S, Yu, D, Jiao, X. Perturbing Eigenvalues with Residual Learning in Graph Convolutional Neural Networks, Asian Conference on Machine Learning 2021, ACML 2021.

Yao S, Yu D, Xiao K. Enhancing Domain Word Embedding via Latent Semantic Imputation. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining; 2019 August; Anchorage, AK, United States. c2019.

Ming Taylor (Fang)

How does the top executive social networks influence firms’ financial reporting?

Using a sample of U.S. listed firms for the 2000–2017 period, we examine how external social networks of top executives and directors affect earnings management in their firms. We find that well-connected firms are more aggressive in managing earnings through both accruals and real activities and that the results are robust after controlling for internal executive social ties. Using a difference-in-differences approach, we find that earnings management decreases after a socially connected executive or director dies. Additional analysis shows that connections forged by past professional working experiences have a greater impact on earnings management than connections forged by education and other social activities. Moreover, CFO social networks have a greater influence on earnings management than CEO social networks. Finally, we explore the underlying mechanisms, finding that 1) firms that are socially connected to each other show more similarities in their earnings management than firms that do not share a connection, and 2) more connected firms are less likely to incur accounting restatements. Collectively, our findings indicate that the external social networks of top executives and directors are important determinants of both their accrual- and real activity-based earnings management.

Fang, M., Francis, B., Hasan, I., & Wu, Q. (2021). External social networks and earnings management. The British Accounting Review, forthcoming.

*The British Accounting Review's 2020 CiteScore (7) and Impact Factor (5.577) rank the journal top 2 and 3 respectively among accounting journals as well as 7 in accounting and finance. It's rated A* in ABDC Journal Quality Guide.

Applying machine learning techniques to asset pricing factor models to improve predictive performance.

We frame asset pricing linear factor models in a machine learning context and consider related comparisons of their predictive performance against ordinary least squares linear regression over a dataset of anomaly portfolios. Specific regression models involved in the comparison include regularized linear, support vector machines, neural networks, and tree based models among others. Performance metrics are presented on a model, portfolio group, and sequential basis, and the strongest predictors are recommended as alternative techniques for the problem of excess return forecasting.

Fang, M., & Taylor, S. (2021). A machine learning based asset pricing factor model comparison on anomaly portfolios. Economics Letters, 204.

Do social networks help firms gain trust and issue financial securities?

We observe that public firms are more likely to issue seasoned stocks rather than bonds when their boards are more socially-connected. These connected issuers experience better announcement-period stock returns and attract more institutional investors. This social-connection effect is stronger for firms with severe information asymmetry, higher risk of being undersubscribed, and more visible to investors. Our conjecture is this social-network effect is driven by trust in issuing firms. Given stocks are more sensitive to trust, these trusted firms are more likely to issue stocks than bonds. Trustworthiness plays an important role in firms’ security issuances in capital markets.

Fang, M., Hasan, I., Sharma, Z., & Yan, A. (2020). Firm social networks, trust, and security issuances. The European Journal of Finance, forthcoming.

A new lens into market responses to earnings announcements.

We studied the relation between immediate market response to corporate earnings announcements and subsequent stock price movement. By adapting an information signal model from Holthausen and Verrecchia (1988), we develop a new measure — the immediate earnings response coefficient (IERC) — to capture immediate market response. We find that a smaller immediate market reaction to earnings surprise, or a lower IERC, leads to a larger subsequent market response. A trading strategy based on our findings can generate an average abnormal return of 5.21% per quarter.

Yan, Z., Zhao, Y., & Fang, M. (2020). Immediate and Subsequent Market Responses to Earnings Announcements. China Accounting and Finance Review, 22(2), 35-53.

Do socially connected firms evade more taxes?

We constructed social networks of CFOs of U.S. companies based on their employment history, education, and non-professional activities. We find that firms with more socially connected CFOs have lower effective tax rates (ETR) than firms with less socially connected CFOs. Moreover, the ETR of a firm decreases if the CFO centrality increases. We do not find similar results for the connectedness of the boards of directors. Further, firm pairs have similar ETRs if their CFOs are socially connected, suggesting tax related information exchange among CFOs through their social networks. We also find that past ETRs of firms with central CFOs predict the ETRs of firms with non-central CFOs. This suggests that less socially connected CFOs follow more socially connected CFOs for their tax planning. Overall, our findings suggest that more socially connected CFOs have more relevant information and resources with regard to tax planning which lead to the adoption of more aggressive tax strategies than less socially connected CFOs.

Fang, M., Francis, B., Hasan, I., & Wu, Q. CFO Social Network and Tax Avoidance. Working paper. Presented at the plenary session of the 7th Journal of International Accounting Research Annual Conference.