Conference Agenda

This paper introduces a novel firm-level green innovation measure utilizing ClimateBERT and GPT-3 language models, capturing a broader range of innovative activities than green patents and categorizing firms into inventors and adopters. Green innovating firms, including many from carbon-intensive sectors, experienced lower expected returns than their less innovative industry peers in both groups. These firms exhibit reduced carbon emissions and fewer climate incidents. They demonstrate a notable value increase in response to more stringent environmental regulations and recent heightened attention to green innovation. Climate policies effectively incentivize green innovation but predominantly among financially unconstrained companies within the green inventors.

Our study presents a revolutionary method called Variational Recurrent Neural

Networks (VRNNs) that utilizes a series of graphs to predict future stock price trends.

It works like an animated movie about price trends. We analyze data from the S&P500

index constituents, which are known to be less predictable than other traded stocks,

between 1993 and 2021. Our model generates a Sharpe ratio of 2.94 for equally weighted portfolios and 2.47 for value-weighted portfolios. Even after taking into account a 10 basis points transaction cost, our Sharpe ratio remains approximately twice as high as that of Jiang, Kelly, and Xiu, 2020. By adopting our graph-based approach, we achieve a substantial alpha of 55 basis points per week after controlling for established risk factors (FF3, FF5, FF6, Q5, and DHS). Furthermore, our prediction of price changes, after adjusting for various price trend strategies and firm traits, strongly forecasts the weekly returns of firms.

A key development in social media has been the remarkable growth of influencers and their increasing use by firms to manage their online presence and image, and to promote their products. Despite the huge growth of influencers and their use by firms, there is a lack of analysis of social media influencers and their impact on the financial market performance of firms. Anecdotal evidence suggests mega influencers are able to affect the stock prices of firms via social media. We ask whether the effect on stock prices identified in anecdotal evidence is generalizable to all mega influencers and other financial market characteristics of firms. After developing hypotheses from the Noise Trader model and using a hand collected dataset of more than 11,000,000 mega influencer posts on Instagram (2012–2019), we find that mega influencers affect investors’ attention, volatility, trading volume and, through extreme sentiment posts, stock returns. The effect on returns is, however, very short lived. Companies need to be aware of these stock market consequences if they intend to use influencers for external image purposes and/or product promotion.

We analyze the information content of a variance risk premia extracted from the

weather derivatives contracts written on the local temperature of individual U.S. cities.

We term this the Weather Variance Risk Premia (WVRP). By constructing the WVRP

measure from the CME’s weather futures and options contracts, we examine the role of

weather variance risk on bond credit spreads of local corporations and municipalities.

Our results indicate informativeness of weather derivatives market as a local risk factor

priced in the bond returns of local corporations and municipalities. Our results are

robust to controlling state level economic uncertainty measures.

The impact of economic news varies widely among firms, industries, and countries. Conventional sentiment analysis methods often fail to capture this heterogeneity, potentially leading to misjudged economic consequences for various stakeholders. This study explores the capabilities of large language models (LLMs) in employing economic reasoning to predict industry-specific news impacts.

Our experiments use commodity price news headlines to reveal the capability of LLMs to derive industry-specific sentiments. This capability positively correlates with context granularity, with more detailed information on the industry resulting in more accurate estimates. Furthermore, our experiments show significant variations of this capability along topic areas, language models, and prompting strategies.

We hypothesize that a surge in availability of information coupled with investors’ confirmation bias could aggravate retail investors’ behavioral biases due to their cherry-picking of information that only confirms their priors. We use the staggered EDGAR implementation to provide causal inference for our hypothesis. We find that retail investors’ anchoring tendency is amplified in the post-EDGAR period. The effect is most evident shortly after the release of corporate 10K/Q reports, for Internet users and sophisticated investors, and among stocks with higher information uncertainty. The exacerbated anchoring bias in the post-EDGAR period also makes retail investors lose more money in trading.

We train deep Bayesian neural networks to mimic the trading activity of a large sample of institutional investors. Our methodology allows us to evaluate the predictive power of hundreds of public information signals with potentially complex non-linear effects on trading, and aggregate them into interpretable categories. Deep learning models predict trading decisions with up to 86% accuracy out-ofsample, with macroeconomic data and market liquidity together accounting for

most (66−91%) of the explained variance. Stock fundamentals, corporate news, and analyst forecasts have comparatively low explanatory power. Our results suggest that differences of opinion about macroeconomic conditions or heterogeneous aggregate hedging needs explain most of the observed institutional trading activity, while stock-specific factors other than liquidity are comparatively unimportant.

We use machine learning to assess the predictive ability of over a hundred corporate governance features for firm outcomes, including financial-statement restatements, class-action lawsuits, business failures, operating performance, firm value, stock returns, and credit ratings. We discover that adding corporate governance features does not improve the predictive accuracy of models over that of models constructed using only firm characteristics. Our results confirm the challenges in constructing measures of corporate governance with predictive value suggested in prior research. These results also raise doubts about the existence of strong causal effects of corporate governance on firm outcomes studied in prior research.

We study the environmental sustainability of individuals’ consumption choices using unique data from a FinTech App that tracks users’ spending and emissions at the transaction level. Using a randomized encouragement design, we show that individuals are likely to purchase carbon calculator services that provide them with detailed transaction-level information about their emissions. However, such a tool does not cause significant changes in their consumption and emissions. On the other hand, services that offset individuals’ emissions by planting trees are less likely to be adopted but prove effective in reducing users’ net emissions. Conditioning on age, gender, and income does not alter our findings. Our results show the challenges and opportunities associated with the automated tools promoting sustainable behavior that were initially confined to specialized FinTech Apps and are now becoming widespread across large financial institutions

Crowdfunding is an increasingly popular way to raise emergency funding after disasters. However, for victims of a major Colorado wildfire, we find that crowdfunding raised more support for wealthier beneficiaries rather than helping the most vulnerable. Specifically, beneficiaries with income above $150,000 receive 28% more support on GoFundMe than beneficiaries with income below $75,000. High-income households are also 14 percentage points more likely to have a crowdfunding campaign at all. These findings hold conditional on the amount of property value destroyed by the fire. The regressive allocation of disaster crowdfunding relates to several network advantages possessed by high-income households, including more connections outside the disaster area. Our findings highlight substantial disparities in social network insurance, which, as we show, likely exacerbate income inequalities in the recovery process.

We use empirical Bayes (EB) to mine for out-of-sample returns among 73,108 long-short strategies constructed from accounting ratios, past returns, and ticker symbols. EB predicts returns are concentrated in accounting and past return strategies, small stocks, and pre-2004 samples. The cross-section of out-of-sample return lines up closely with EB predictions. Data-mined portfolios have mean returns comparable with published portfolios, but the data-mined returns are arguably free of data mining bias. In contrast, controlling for multiple testing following Harvey, Liu, and Zhu (2016) misses the vast majority of returns. This "high-throughput asset pricing" provides an evidence-based solution for data mining bias.

The relevance of asset return predictability is routinely assessed by the economic value

that it produces in asset allocation exercises. Specifically, out-of-sample return forecasts are generated based on a set of predictors, increasingly via “black box” machine learning models. The return forecasts then serve as inputs for constructing a portfolio, and portfolio performance metrics are computed over the forecast evaluation period. To shed light on the sources of the economic value generated by return predictability in fitted machine learning models, we develop a methodology based on Shapley values—the Shapley-based portfolio performance contribution (SPPC)—to directly estimate the contributions of individual or groups of predictors to portfolio performance. We illustrate the use of the SPPC in an empirical application measuring the economic value of cross-sectional stock return predictability based on a large number of firm characteristics and machine learning.

We create a natural language processing (NLP) machine learning model to identify covenant violations using 10K and 10Q filings provided by the Securities and Exchange Commission (SEC). Our sample includes more than 580,000 filings comprising the universe of all publicly listed U.S. firms and spanning the entire 1996 to 2022 period. We use the MPNET Sentence Transformer as a classification algorithm and obtain a model accuracy of 94.4%, considerably higher than conventional manual approaches. Importantly, we are able to differentiate between firms that obtain an amendment after or before a covenant violation occurs and those that remain in technical default. Covenant violations have significantly declined over the past two decades. During this time, the percentage of firms that received an amendment increased while those in technical default decreased. The decline in violations was driven by non-investment-grade rated and unrated firms and intensified during the COVID-19 pandemic. Firms in technical default perform considerably worse on a number of dimensions, such as leverage or liquidity, and they draw down a larger percentage of their credit lines in the eight quarters before a covenant violation. They recover quickly after a violation, specifically in terms of leverage ratio and operating income. Moreover, they make larger changes in their investment and financial policies as compared to firms that obtain amendments. Again, the effects are mainly driven by non-investment-grade rated and unrated firms. Not surprisingly, non-investment-grade rated firms are about 17% more likely to declare bankruptcy in the eight quarters following a covenant violation. This effect is substantially muted in the sample of firms that obtain loan amendments.

This paper studies the extent to which financial covenants are an important consideration for firm decisions outside of violation events. Applying textual analysis to earnings call transcripts, I construct a novel measure of covenant concerns by distinguishing between discussions of covenants that relate to the future as opposed to the past or present. The measure predicts future violations, and covaries intuitively with earnings and leverage. Covenant concerns are associated with significant reductions in investment and financing activity. These responses persist even after controlling for standard measures of investment opportunities and are economically large relative to the effects of actual violations.

Younger adults increasingly look to social media for news and investment guidance about cryptocurrencies. In this paper we combine survey responses and transaction- level data to study how individuals respond to mainstream celebrity endorsements of cryptocurrencies on Twitter. We find that individuals appear to treat these celebrity tweets as financial advice: tweets are associated with a 16% higher probability that an individual invests in cryptocurrencies, with stronger effects for men, wealthier, and older investors. We also find that aggregate market trading volume in a given coin increases by 10% on the day of the celebrity tweet and stays elevated for the following two days, while returns exhibit a 3% spike with no reversal over the following week. We conclude by showing that investors would have been better off buying Bitcoin or Ethereum than the coin mentioned in the celebrity tweet.

This paper studies the rationale and effects of buying bots on X (former Twitter). We observe that large amount of attention to corporate X accounts around earnings announcements is driven by bots. Bot activity is a significant predictor of investor disagreement, which is persistent long-term. Moreover, bot activity increases analyst dispersion for the following quarterly earnings announcement. Consistent with managerial short-termism, bot activity often accompanies intense earnings management. Our results are robust to various specifications, including a matching approach indicating causal interpretation.

The recent surge in retail option trading has sparked concerns about gambling and significant losses. We study a novel trader-level dataset of about $20 billion in retail trades between 2020 and 2022 to show that these concerns may be exaggerated. We find that option trades account for nearly half of all trades in 2022, making them a vital part of retail trading. Despite wide bid-ask spreads, retail investors see minimal losses on option trades. Although options theoretically resemble lottery tickets, we find little evidence of positive skewness in realized dollar profits, contradicting gambling-driven trading. Many retail investors trade options because of high leverage and low option prices; indeed, option trades are concentrated in a few high-priced underlyings. A typical retail trade is the purchase of a one-day S&P 500 index call held for an hour. Overall, we provide the first comprehensive account of modern retail trading in the U.S. options market using trader-level data.

This paper examines employees' trading of own-company options. Using data from Finland, I show that employees' direct and indirect purchases of call options represent 4%-14% of aggregate retail option volume. These purchases contain price-relevant information: weekly returns on the underlying stocks are approximately 50 basis points. The informativeness is most evident before earnings announcements, extends to firms in the employer's supply chain, is not driven by industry knowledge, and disappears upon job separation. Consistent with prospect theory, employees who experience recent losses in their stock portfolios are more willing to exploit their information advantage by trading own-company options.

We examine the effect of market efficiency on the efficiency of capital allocation in the setting of decentralized exchanges of crypto assets. Utilizing data on nearly 100 million trades in concentrated liquidity pools on two leading blockchains, we construct a highly granular, capital-market-based measure of capital allocation efficiency. We also design and implement a method of identifying market-efficiency-restoring arbitrage transactions among all blockchain transactions and construct arbitrage-based granular measures of market efficiency. We find that market efficiency has positive, economically and statistically significant, and causal impact on capital allocation efficiency.

Off-exchange trades are often executed by referencing on-exchange prices. In equilibrium, such price referencing softens market makers' on-exchange competition and makes liquidity expensive for investors. Additionally, by equalizing on- and off-exchange prices, price referencing guarantees “best-execution” and makes investors indifferent where to trade. Market makers effectively obtain a license to fragment orders off exchange, raising their profits but reinforcing market-wide illiquidity. This inefficiency remains tenacious even if more market makers enter and if they are forced to compete off exchange, as in the SEC's proposed order-by-order auction. The model yields important implications for regulating various forms of off-exchange trading.

We construct a heterogeneous-firm growth model of the data economy, where data, crucial for business optimization, is at risk of being damaged and destroyed by cyber criminals. Digitally-savvy firms invest in in-house cybersecurity, which can be used to improve the quality of their other products, and trade cybersecurity protection with non-digitally-savvy firms. We use the model to study the impact of cybercrime risk on firm innovation and aggregate growth. Theoretically, we find that cyber-crime unequivocally leads to reduced knowledge stocks, decreased productivity, and slower overall economic growth for all firms. Cybercrime risk mitigates some of the adverse effects as it ex-ante prompts digitally-savvy firms to pursue digital innovation that enhances productivity in other domains. We then test the theoretical prediction using several unique data sets on firms’ in- vestments in cyber-protection. Empirically, we observe increased innovation rates in response to higher cyber-crime risk, driven primarily by data-intensive firms and by firms which intensively pursue in-house cybersecurity protection rather than third-party cybersecurity delegation.

Open banking can alter the information structure of the loan markets by furnishing fintechs with more financial data on their customers and enhancing their screening capabilities. We investigate the welfare implications of open banking by constructing a model of a loan market with adverse selection. In this market, a fintech, reliant on information technology to assess borrowers' credibility, competes with a traditional bank that employs collateral to differentiate between various borrower types.

Our primary finding is that enhancing the fintech's monitoring capacity through the provision of free access to borrowers' information may not necessarily lead to improved welfare. This suggests that complete data sharing (granting the fintech full access to borrowers' data) is not always the optimal solution and could potentially reduce welfare. Specifically, when the bank is sufficiently proficient in utilizing collateral, partial data sharing might be the preferred option.

We develop a new, topics based method of classifying content by identifying the characteristics of named entities through their wikipedia entry. We apply the method to a new corpora - the closed caption data from three major TV news networks (CNN, FOXNEWS and MSNBC) and three business TV news networks (Bloomberg TV, FOX Business and CNBC). TV news networks have converged around content discussed over the last decade, but diverged in sentiment around the topics of politics, race, religion, gender, and sexual orientation. Polarization in modern news media appears to be more about how topics are presented rather than what topics are discussed. Across business news networks we find starker differences in content: CNBC and Bloomberg appear to be focused more on economics and market related content relative to FOX Business, which maintains the political tilt of its news only affiliate.

The capacity to aggregate information from diverse perspectives has positioned social finance forums as a potent source of signals that shape investors’ beliefs and actions. We investigate how investors react to the information provided by male and female non-professional analysts on social platforms and the financial market consequences. Although male and female contributors exhibit similar informativeness and skills, female-authored perspectives receive significantly lower engagement, lower trust, and higher disagreement from platform users. The higher disagreement is subsequently associated with higher abnormal trading volume and lower price efficiency. Additional analysis indicates that the platform consensus becomes less informative about future cash flows when female contributors quit due to their less favorable engagement.

I study the effects of information technology (IT) progress in a model where stock prices aggregate speculators' information and guide firms' investments. A firm with higher exposure to risky production technology attracts more information from speculators. IT progress (i.e., lowering information costs) improves stock price informativeness and corporate efficiency when information is costly. Yet, when information is inexpensive, speculators use up their limited attention. Then IT improvements can backfire: Firms excessively increase risky technology exposure to engage in zero-sum competition for attention, reducing corporate efficiency and social welfare. Raising firms' growth opportunities can reinforce the adverse effects of IT progress.

This paper demonstrates that liquidity transformation provided by asset managers can boost firm-specific information production. We examine a setting where stocks become perfectly illiquid during trading suspensions: the prices and shares held by mutual funds “freeze.” Consistent with a model of liquidity-driven information acquisition, we find that investors analyze these illiquid holdings and reallocate capital in funds to take advantage of these stale prices. Once trading resumes, stocks exposed to liquidity transformation exhibit informative price movements about future firm fundamentals, reflecting the information produced by investors. Our findings suggest a liquidity channel through which asset management influences information production in capital markets.