学术论文

The mandatory switch from the incurred loss model to the more forward-looking current expected credit loss (CECL) model was originally scheduled to begin in 2020. However, when the COVID-19 pandemic started in early 2020, US regulators made the switch voluntary. Our study investigates how banks' exposure to the pandemic affects their decision to adopt CECL as well as adopting banks' pandemic-era pattern of loan loss provisions. First, consistent with pandemic-driven economic uncertainty reducing banks' willingness to adopt the new model, we find a negative association between banks' pandemic exposure and their CECL adoption. This association is more pronounced for banks with more lending opportunities, more lending competition, and worse loan quality. Second, compared with non-adopters, CECL adopters report more loan loss provisions during the pandemic's early period, and less or even negative loan loss provisions during the late period. The latter scenario reflects a reversal of earlier loan loss reserves and is more pronounced for banks with more exposure to states with a higher level of vaccination, consistent with banks having a more positive economic outlook because of improving pandemic conditions. Overall, our study offers useful insights into the adoption and implementation of accounting standards during periods of economic uncertainty.

We propose a two-step procedure to model and predict high-dimensional functional time series, where the number of function-valued time series p is large in relation to the length of time series n. Our first step performs an eigenanalysis of a positive definite matrix, which leads to a one-to-one linear transformation for the original high-dimensional functional time series, and the transformed curve series can be segmented into several groups such that any two subseries from any two different groups are uncorrelated both contemporaneously and serially. Consequently in our second step those groups are handled separately without the information loss on the overall linear dynamic structure. The second step is devoted to establishing a finite-dimensional dynamical structure for all the transformed functional time series within each group. Furthermore the finite-dimensional structure is represented by that of a vector time series. Modeling and forecasting for the original high-dimensional functional time series are realized via those for the vector time series in all the groups. We investigate the theoretical properties of our proposed methods, and illustrate the finite-sample performance through both extensive simulation and two real datasets. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.

This article provides the first comprehensive evidence that the return extrapolation behavior of investors leads to biases in the expectations of volatility. Lower past returns are associated with higher expectations of volatility when using the physical, risk-neutral, and survey measures to estimate volatility expectations. Consistent with the return extrapolation framework, recent past returns have a larger impact than distant past returns on volatility expectations. Biases in volatility expectations are i) distinct from extrapolating past realized volatility, ii) asymmetrically induced by recent past negative returns, and iii) lead investors to pay more to insure against the perceived higher expected volatility.

Whereas service firms have started to deploy service robots on the service front line, it remains less understood whether all types of service firms gain a similar benefit from this advance in information technologies. With one secondary data analysis of review content from a leading travel website, one field study conducted at a hotel, and three randomized controlled experiments, the current research addresses this important managerial issue. We propose that replacing human service providers with service robots dampens customers’ feelings of superior social status over service providers, which drives a lower relative preference for service robots over human service providers in luxury services than in mainstream services. Supporting this mechanism, we show that the effect of price image (luxury versus mainstream) on customers’ relative preference for robots over humans is mediated by the desire to be treated as a customer of superior status by service providers. Furthermore, the robot disadvantage in luxury services can be mitigated when a luxury service firm adopts a social equality–oriented positioning and when the use of service robots is endowed with superior status.

Firms increasingly use a combination of image and text description when displaying products and engaging consumers. Existing research has examined consumers’ response to text and image stimuli separately but has yet to systematically consider how the semantic relationship between image and text impacts consumer choice. In this research, we conduct a series of multimethod empirical studies to examine the congruence between image- and text-based product representation. First, we propose a deep-learning approach to measure image-text congruence by building a state-of-the-art two-branch neural network model based on wide residual networks and bidirectional encoder representations from transformers. Next, we apply our method to data from an online reading platform and discover a U-shaped effect of image-text congruence: Consumers’ preference toward a product is higher when the congruence between the image and text representation is either high or low than when the congruence is at the medium level. We then conduct experiments to establish the causal effect of this finding and explore the underlying mechanisms. We further explore the generalizability of the proposed deep-learning model and our substantive finding in two additional settings. Our research contributes to the literature on consumer information processing and generates managerial implications for practitioners on how to strategically pair images and text on digital platforms.

Does inward foreign direct investment (FDI) promote or inhibit the technological innovation of local firms? Prior research has generated mixed views and findings on this important question. To address this puzzle, we propose that FDI spillover and crowding-out effects may apply to different phases of innovation (i.e., idea generation and R&D implementation) respectively, which then leads to differential impacts of FDI on incremental and radical innovation. The results from a panel dataset of Chinese listed firms show that industry-level FDI is positively related to incremental innovation but has an inverted U-shaped impact on radical innovation. Furthermore, we find that technology gap reinforces the positive impact of FDI on incremental innovation while making its inverted U-shaped impact on radical innovation more pronounced, and when FDI local market focus is higher, the inverted U-shaped relationship between FDI and radical innovation is steeper. These findings reconcile the inconsistency regarding how FDI may affect local innovation and provide a novel analysis framework for the FDI literature.

While prior research has emphasized the economic threats posed by political risk, it is unclear how geopolitical risk (GPR), a supranational-level risk, affects global supply chain decisions. Drawing on the political economy perspective, we posit that GPR presents both opportunities and threats for multinational corporations (MNCs), depending on the political affinity between MNCs’ home and host countries. We also identify the risk-mitigation roles of MNCs’ political lobbying and market diversification. Using panel data from publicly listed MNCs in the US, we find that host-country GPR increases MNCs’ first-tier supply base when home–host country political affinity is high, but decreases when political affinity is low. Moreover, the positive effect of high-affinity host-country GPR on MNCs’ supply base is stronger, and the negative effect of low-affinity host-country GPR is weaker for MNCs with high levels of political lobbying or market diversification. These findings enrich the international business research and political economy perspective by elucidating both the opportunities and threats of GPR, and highlight the importance of risk-coping capabilities in managing GPR. These findings also provide insights for MNCs to adapt their strategies amid GPR by leveraging home–host political affinity, engaging in political lobbying, and pursuing market diversification to mitigate geopolitical challenges.

Our paper examines analyst reports and online stock opinion articles which recommend buying stocks that, based on the literature, trade at high prices and earn low future returns ("short-leg securities"). Using a textual analysis, we test whether the justifications primarily (1) emphasize safe-haven qualities, (2) indicate exuberance, or (3) highlight lottery-like features. Our results strongly point to (3). We subsequently validate our text-based inferences through a survey of institutional and retail investors with long positions in short-leg securities. Overall, perceived upside potential appears to play a material role in driving investor demand for stocks in the short legs of anomalies.

Hate speech is a major problem on social media platforms. Automatic hate speech detection methods relying on machine learning models, which learn from manually labeled datasets, have been proposed in both academia and industry. However, there is increasing evidence that hate speech detection datasets labeled by general annotators (e.g., amateurs or MTurk workers) contain systematic bias, as they cannot effectively consider language use differences among different speakers. When such biased datasets are used to train machine learning models, the resulting models will also be biased. Unlike general annotators, experts can produce much less biased annotations. However, expert annotations cannot be efficiently obtained in large quantity. This paper bridges the gap by adopting a weakly supervised learning method for hate speech detection using a small number of expert annotations. We propose a novel design that uses contrastive learning and prompt-based learning based on large language models, incorporating a group estimator, a pair generator, and knowledge injection. Using real-world Twitter posts written by African American English speakers and other racial groups as an example, extensive experiments were conducted to demonstrate the superior performance of the proposed method. The proposed approach was also evaluated on data in the LGBTQ+ community and achieved consistent results. The study has important academic and practical implications for hate speech detection and large language models.