We analyze how early-stage investors evaluate potential investments by using

ChatGPT to analyze 5,143 expert consultation calls. Companies discussed in these

calls are 15 percentage points more likely to receive financing in the following quarter.

Positive signals about technology integration and customer acquisition increase deal

likelihood by 14% and 10.5%, respectively, with their predictive power declining by

over 75% for mature companies. Market analysis and business strategy discussions –

comprising over 40% of call content—show minimal predictive power for investment

outcomes. Our findings document both how investors overcome information asymmetries

in early-stage investing and a misalignment between the information they seek and

the information that predicts investment outcomes. Methodologically, we demonstrate

the potential of LLMs to extract nuanced insights from complex qualitative data.

We model the multitasking nature of managerial incentives when ESG metrics are
introduced jointly with standard financial or accounting metrics in executive
compensation. Building on insights from multitasking theory, we predict that
pay-performance sensitivity or dollar delta of standard metrics should optimally
decrease when value-adding but less measurable ESG goals are introduced in executive
pay. Empirical tests support the existence of a significant opportunity cost for the
effort of executives to improve ESG metrics that firms mitigate by decreasing
incentives to achieve standard metrics. Consistently, the downward adjustment in
dollar delta of standard metrics is shown to be larger when the number of ESG metrics
increases, they are less material to the firm, or less measurable. This adjustment is not
offset by a simultaneous increase in the time vesting delta or the executive’s total
compensation. The tests show differential effect of E, S, and G metrics on the dollar
delta of standard metrics. In sharp contrast, there is no variation in the dollar delta of
standard metrics when a new standard metric (instead of an ESG metric) is
introduced. Overall, the evidence is consistent with efficient contracting in the
presence of multitasking when ESG metrics are introduced in executive compensation.

We investigate the second-by-second impact of Fed Chair Jerome Powell's facial and vocal emotional expressions during FOMC press conferences on the market microstructure of SPY and TLT ETFs. Using unique high-frequency multimodal data and robust wild cluster bootstrap inference, we find complex and often counterintuitive effects. Contrary to standard hypotheses, negative emotional cues (facial or vocal) predict significantly lower subsequent volatility, while positive facial expressions predict higher SPY volatility. Positive vocal cues improve TLT liquidity (narrower spreads), but negative vocal cues also predict narrower spreads at longer horizons. Significant interactions between nonverbal cues and textual sentiment reveal further complexity, such as congruent positive signals predicting lower TLT depth. The Treasury market appears more sensitive than the equity market. Our results challenge simple valence-based theories for high-frequency nonverbal cues, highlighting the importance of modality, asset class, microstructure dimension, and interactions with text in shaping immediate market dynamics.

We provide a comprehensive analysis of AI adoption across various aspects of asset management by examining U.S. registered investment advisers. Since 2017, the investment advisory industry has seen significant AI adoption, primarily for investment strategies and IT infrastructure, while generative AI adoption remained minimal until 2023. AI adoption varies substantially across advisers and is related to the size and composition of advisers’ assets under management, client base, and workforce. Advisers with larger assets, a higher share of hedge funds, and fewer employees directly interacting with individual clients exhibit higher levels of AI adoption. Moreover, we estimate that AI adoption leads to increased growth of the investment advisory business and a shift toward private funds, with spillover effects extending beyond hedge funds. Our findings underscore the heterogeneity of AI applications across the diverse funds and services in asset management.

This study investigates the impact of Regulation 18 on green innovation within Chinese firms. Introduced in 2013 to sever political ties in corporate governance, Regulation 18 mandated the resignation of government officials from business roles. Studying Chinese listed firms from 2010 to 2016, we utilize a difference-in-differences (DiD) model to evaluate its effects. Our findings reveal a significant influence of Regulation 18 on green innovation. Non-State-Owned Enterprises (non-SOEs) in High Energy Consumption, High Pollution, or Overcapacity (HHO) industries experienced reduced green innovation post-regulation, consistent with the legitimacy theory and social perspective, indicating a positive correlation between political connections and green innovation. Conversely, State-Owned Enterprises (SOEs) in HHO industries exhibited increased green innovation, aligning with the resource curse theory and tunneling arguments, suggesting a negative correlation between political connections and green innovation. Furthermore, the analysis extends to corporate financial performance, revealing a decline for politically connected non-SOEs in HHO industries post-Regulation 18, while improvements are observed for SOEs in the same sector.

Exploiting the adoption of mandatory central clearing by U.S. regulators, we explore the effects of this regulatory reform on banks’ derivative offsetting. Using a triple-difference testing procedure, we find that derivative offsetting increases (decreases) for banks with higher (lower) capital ratios after the adoption of mandatory central clearing, in comparison to the control group. The results are economically significant and robust to a variety of alternative measurements and tests. Our findings suggest banks with different target capital ratios respond to central clearing reform differently.

With the phenomenal rise of generative AI models (e.g., large language models such as GPT or large image models such as Diffusion), there are increasing concerns about human creatives’ futures. Specifically, as generative models’ power further increases, will they eventually replace all human creatives’ jobs? We argue that the answer is “no,” even if existing generative AI models’ capabilities reach their theoretical limit. Our theory has a close analogy to a familiar insight in financial economics on the impossibility of an informationally efficient market [Grossman and Stiglitz (1980)]: If generative AI models can provide all the content humans need at low variable costs, then there is no incentive for humans to spend costly resources on content creation as they cannot profit from it. But if no human creates new content, then generative AI can only learn from stale information and be unable to generate up-to-date content that reflects new happenings in the physical world. This creates a paradox.

We explore the director labor market consequences of social media use. Specifically, we identify directors in S&P 1500 firms who are active on Twitter and examine various director labor market outcomes. We find directors, particularly females and minorities, on Twitter are more likely to gain an additional directorship, a larger directorship and a directorship in a new industry each year than those who are not on Twitter. These results hold when controlling for time invariant unobserved director characteristics and when using an instrumental variable approach to control for endogeneity. They are strongest for directors, primarily females and minorities, who engage more with other directors via social media. Shareholders show more support for social-media-active directors through a greater (lesser) percentage of votes casts “For” (“Against”) their election and through a greater stock price reaction to the announcement of first-time director appointments. These results suggest that social media can play an important role in reducing traditional labor market frictions and facilitating more opportunities for minority and female directors.

Cryptocurrencies exhibit high volatile behavior and observable anomaly effects, yet systematic anomaly detection within these markets remains underexplored. This study introduces a tailored Business Intelligence (BI) system for cryptocurrency anomaly detection. Leveraging advanced analytics and visualization, our novel BI system provides a comprehensive and real-time overview of market behaviors, enabling swift identification of cryptocurrency price anomalies. Our unique system empowers users to navigate the dynamic cryptocurrency trading, enhancing anomaly detection informed decision-making and risk management.

Hash-linked timestamping is the key feature behind blockchain technology. It makes it possible to design financing contracts that are based on reliable and up-to-date records of transactions. For this reason, it is considered to enhance trust. This paper develops a theoretical model that uses dynamic contract theory to derive optimal financing contracts in a blockchain environment. I show that a dynamically adjusting profit sharing rule is optimal and I highlight properties that determine the splitting rule. In contrast to the view that blockchain enhances traditional contracts like debt and equity by bringing efficiency gains, I emphasise that blockchain allows borrowers to learn from data and take effoort decisions more frequently, which make debt and equity contracts costlier.

We use Target Date Funds (TDFs) to study how managers of funds behave when shielded from their investors' short-term needs. We document that asset managers exploit reduced investor attention to deliver lower performance quantifiable in 21% for an average investor holding the fund for 50 years. This underperformance is driven by fund families using TDFs to smooth the flow shocks of affiliated open-end funds and to boost fees by investing in the affiliated expensive share classes. We use the Pension Protection Act of 2006 as an exogenous shock that made TDFs the default investment option within 401(k) retirement plans.

Liquid staking allows agents to sell ownership of an illiquid claim to satisfy a liquidity need. We develop a model of liquid staking and characterize the effect of the secondary market on protocol stability. We establish that the liquid market has two effects: first, it allows agents to redeem illiquid assets and thus reduces run risk on the protocol, but also conveys information and can act as a coordination mechanism and increase market run risk. Using novel data on the Lido protocol, we present stylized facts on the staking market and relate our results to the design of digital deposits.

The exponential growth and variety of studies on returns highlights the lack of comprehensive asset pricing theory for explicitly explaining the empirical data. The paper addresses this challenge by proposing a hierarchical state-based asset pricing model based on two interconnected solutions: the state space of assets and explanatory gain decomposition approach. As a result, the states of assets extend the conventional state of nature for bringing fundamental and macroeconomic characteristics into existing asset pricing models. Then, the decomposition approach tackles the complexity and heterogeneity of asset pricing by advancing all-in analyses with hierarchical piecewise finer-grained regressions. The direction is demonstrated with a multi-step analysis subsequently boosting the explanatory power of regressions between price-to-fundamental ratios and asset quality characteristics and resolving the weak correlation between the HML and RMW factors. Furthermore, the proposed model establishes a direct link between theory and empirics encompassing multi-dimensional data and growing stack of data science techniques.

To support a growing financial system, many recent blockchain innovations (e.g., layer-2 scaling solutions, proof-of-stake layer-1 chains, cross-chain interoperability bridges, and data availability layers, etc.) adopt an “optimistic” design principle, that is, the system proceeds as if all participants are well-behaving, presumably sustained by some “stake-and-slash” mechanisms. We formulate the logic behind such optimistic systems in a simple game and characterize its equilibria. We point out that the “optimistic” assumption under current designs is not compatible with all participants being rational. Therefore, to ensure security, protocol designers have to impose trust assumptions regarding the presence of altruistic participants. We explore alternative design changes that may help alleviate the identified problems, including introducing redistribution across participants or adopting succinct proof-based designs.

It is assumed in the news-based model of stock pricing (NBSPM) that stock prices are determined with macroeconomic news (modeled with the total market return in the spirit of CAPM), industry news (modeled with the relevant industry ETF returns), and the company-specific news and momentum that are described using an optimal ARMA-GARCH model. In this work, the NBSPM accuracy for forecasting stock prices is compared with that of the momentum-enhanced five-factor Fama-French model. The results for a representative list of holdings of nine major US equity sector ETFs demonstrates superiority of the NBSPM in most cases.

This presentation synthesises evidence from several new survey-based studies (supported by Nasdaq N.F.) that cover more than 2,300 individual investors in Denmark, Finland, Sweden and France, in order to explain why, how and for whom cryptocurrencies are attractive. The combined findings reveal multiple behavioural channels, i.e., demographics, memory accuracy, interpersonal and institutional trust, gendered knowledge gaps in risk perception, promotion-versus-prevention motivation, and values-based beliefs, that jointly determine adoption and price expectations in a market where traditional fundamentals are weak, offering academics new micro-foundations for expectation-formation models.

We propose a new analytical model for the valuation of loan commitments and some of their main features including the MAC (Material Adverse Change) clause. A two-period contingent claims approach in continuous time is developed. The advantage of this approach is that it is based on rational economic considerations that are not based on utility functions.

We offer a new metric, "net-zero alignment strength" (NZAS), which can be used by socially responsible investors for portfolio selection. NZAS is based on the current corporate GHG emission intensity (EI) and its reduction rate (ERR) and, hence, can be used for choosing among the companies with high EI and high ERR versus the companies with low EI and low ERR. We have incorporated NZAS into the mean-variance portfolio (MVP) framework, which yields simultaneous optimization upon high returns and NZAS, and low price volatility. The NZAS contribution to the MVP minimization function is controlled by the net-zero commitment parameter, which is investor’s choice. An example for a portfolio with 29 major constituents of the Energy sector illustrates an interplay between the company Sharpe ratios and NZAS, which determines the effects of the net-zero commitment on the major portfolio holdings. We suggest to use this framework for finding the best-in-class companies within the chosen equity sectors.

Public debate about cryptocurrency reveals strong and disparate opinions on potential adoption. The paper argues that this uncertainty is the driving factor of cryptocurrency prices. In the model presented, uncertainty about a cryptocurrency's eventual adoption demand amount leads to a discounted cryptocurrency price. However, over time, investors learn about the likely extent of eventual adoption, and the discount reduces. The model replicates the long-term decline in price growth rates and variance of Bitcoin and other major cryptocurrencies. The model can forecast expected price growth and variance conditional on low interim adoption, providing guidance to cryptocurrency allocation sizing within an investment portfolio. The model also estimates the probability distribution of a cryptocurrency's extent of eventual adoption.

When liabilities are deemed improbable or cannot be reliably estimated by management, they are not recorded on the balance sheet. Instead, they are disclosed qualitatively in the company's filings. Examples include obligations related to pending or future lawsuits, product liability, environmental matters, false advertising, or patent and copyright infringements. We refer to these obligations as intangible liabilities (IL). We construct a firm-level, text-based measure of IL. IL is positively related to firm size, volatility, and share turnover, and is negatively correlated with accounting performance, abnormal returns, and Tobin’s Q. IL also predictably varies across industries. Moreover, IL predicts future lawsuits against firms and the future deterioration of their reputations. Companies with higher IL trade at lower valuation ratios and have significantly higher future crash risks. A portfolio that is long on high IL and short on low IL yields an annual abnormal return of 3% after accounting for common factors. Overall, the results suggest that intangible liabilities are a significant determinant of firm value and stock returns.

Tradefeedr applies the principles of open banking APIs to financial markets, reducing the friction of data sharing between market participants. Through a unified, secure analytics API connected to over 20 major liquidity providers (LPs) and 10 trading platforms, we provide a single data agreement that covers all participants. Our standardized data, metrics, and analysis empower trading teams, while independent client reporting for LPs and bilateral client–LP data dialogue foster transparency and collaboration. Access to this unique dataset enables market participants to unlock the full value of their data and leverage AI and machine learning to make smarter trading decisions. Everything we do is driven by the needs of the trading community.

The quality of cryptocurrency market data is critical for academic research and financial applications, yet the topic remains understudied. We analyze data from leading vendors and document pervasive mislabeling, measurement errors, and discrepancies in reported market metrics. To address these issues, we propose a novel aggregation methodology that achieves asymptotic accuracy by identifying unreliable data instances. We also introduce a data quality grading system, offering practical guidance for data consumers. Our findings underscore the risks of relying on a single provider. They highlight a possible need for regulation in the market for crypto data.

We introduce an out-of-sample neural-network-based measure of firm-level operating leverage, which outperforms existing ones in capturing the elasticity of operating profits to gross profits. Strikingly, our analysis uncovers a non-monotonic—and potentially negative—relationship between operating leverage and the risk premium. This challenges conventional wisdom and contradicts explanations that link operating leverage to the value premium. A production-based asset pricing model incorporating both variable and fixed costs provides a possible rationale for these empirical findings. Furthermore, our analysis offers a fresh perspective on the idiosyncratic volatility premium by emphasizing the interplay between the operating hedge effect induced by variable costs and the operating leverage effect induced by fixed costs.

Japan’s FinTech sector has undergone a rapid transformation in the past five years, driven by progressive open-banking rules, a vibrant Web3 start-up scene, and a public-private push toward cashless payments ahead of Expo 2025 Osaka-Kansai. Representing 500+ member firms, the Fintech Association of Japan (FAJ) will share fresh data on investment flows, consumer adoption, and regulatory sandboxes that enable speedy experimentation without compromising consumer protection.

Building on this evidence, the session explores how Japanese and European innovators can co-create next-generation payment rails, tokenized capital-market pilots, and inclusive digital-finance programmes. By highlighting flexible regulatory pathways, reciprocal sandbox access, and practical partnership models—rather than focusing on any single legislative regime—the talk offers a future-oriented roadmap for policymakers, founders, and investors eager to scale solutions across Asia-Europe corridors.

We document that blocks on the blockchain are rarely filled to capacity, even though
there is excess demand for block space by fee-paying users who want their transactions to
be recorded. In spite of this, higher fee orders are not always prioritized. We show these
patterns are consistent with miners exercising market power: If users believe that only high
fee transactions will be executed expeditiously we show how strategic capacity management
can be used to extract higher fee revenue. Using a novel data set, we provide evidence
consistent with this market power, and estimate that mining pools have extracted least 300
million USD a year in excess fees by making processing capacity artificially scarce.

We investigate open-source innovation by public firms and the private value it generates for these firms. Unlike patents, which grant inventors exclusive rights to their inventions, open-source innovations can be used by anyone. Nevertheless, using an extensive dataset of public-firm activity on GitHub, we find that firms with open-source projects represent 68% of the U.S.~stock market across 86% of industries. We estimate the private value of all projects in our sample to be nearly $25 billion, with the average project generating $842,000. We find that projects with fully permissive licenses are generally less valuable and firms facing higher competition tend to generate less private value from their projects. We also find that the degree to which a project complements commercial products is not a primary driver of private value. Finally, open-source value is associated with a firm's substantial growth in terms of sales, profits, employment, and patenting, yet it also induces creative destruction. These results contribute to our understanding of the private value generated by innovation in the absence of legal excludability.

How borrowers respond to future changes in the interest rate on their debt is of crucial importance for the transmission of monetary policy and for financial stability. Combining data from a large bank, a letter RCT, and an online survey, we study this question in the context of the German mortgage market, where borrowers face high interest rates since 2022 when their rate fixation period ends. We find that borrowers take various actions to reduce the impact of higher rates on interest payments. Survey responses indicate high awareness of the evolution of interest rates and corroborate a strong propensity to prepare for the rate reset, which we show experimentally is sensitive to the size of the rate increase and to the distance from reset. Our letter intervention does not affect rate beliefs, consistent with high ex-ante knowledge, but increases awareness of available options and the desire to prepare. Ongoing tracking will reveal whether this awareness translates into actual behavior.

This paper considers introducing asymmetric privacy in the design of central bank digital currencies (CBDC) and digital currencies more generally, to preserve the privacy of money spent while keeping the benefits of digital records for money received. It is shown that this feature would help minimize real distortions between consumers, firms, and financiers, while enabling tax optimization and better access to external financing. Protecting the privacy of consumers is desirable from a welfare and efficiency standpoint as long as there exist noticeable privacy concerns. Implementing asymmetric privacy is technologically feasible, using for instance Zero-Knowledge proofs or other privacy tools.

The interconnectedness of firms through various networks, such as production, credit, and competition, plays a critical role in determining firm-level and aggregate outcomes. However, data on these connections are often limited. This paper introduces a novel artificial intelligence methodology that extracts explicit firm relationship networks from financial news articles, providing comprehensive and interpretable data across multiple dimensions. Applying this methodology to New York Times articles since 1981, we generate extensive networks that predict key macroeconomic indicators. Our publicly accessible dataset offers valuable insights for future research on firm networks and aggregate fluctuations.

What is an effective signal in crowdfunding? We asked this question to 83 expert researchers who have published the top-notch articles in this field. They stated that, in theory, strong signals include past crowdfunding success, business experience, patent ownership, and the equity share offered. Examining 145 articles published in leading business and economics journals, we find that the empirical evidence from a meta-analysis does not accord with this perception among expert researchers. Signals that expert researchers consider to be theoretically less strong are more often statistically significant predictors of crowdfunding success and have neither larger nor smaller standardized effect sizes than strong signals. A meta-regression suggests that domain-specific signals play the most important role in crowdfunding. The findings of our literature review provide important insights for investors, platform managers, and the academic review process.

This paper analyzes trading patterns and investor behavior in the market for tokenized
financial assets. Even though investors can trade tokenized assets around the clock, trading
activity is substantially higher when the primary listing exchange is open, in particular during
the regular trading session. Likewise, price differences between the tokenized and underlying
asset are wider during the extended trading hours than during the regular trading session.
The majority of trades result from fractional order sizes, but orders still cluster at relatively
round sizes, values, and prices. Additionally, a disproportionate number of trades, especially
buy trades, use the minimum permissible order size, suggesting the presence of retail traders
interested in owning an asset for purposes other than financial gain.

This paper analyzes investor search behavior across financial information providers by modeling demand within a multi-class classification framework. Examining searches on Bloomberg Terminals and EDGAR, I explore how subscription prices, expected stock returns, and investor income influence the selection of financial data sources. The findings offer insights into information-seeking behavior, retrieval patterns, and access dynamics, highlighting the economic factors that drive the use of proprietary and open-access financial platforms.

This study exploits the non-fungible token (NFT) setting to estimate the value of non-pecuniary benefits, a long- standing empirical challenge in private-value markets such as art, antiques, and other collectibles. After developing and validating our emotional dividend proxy (LOVE), we apply deep learning algorithms and discover that contemporaneous price fluctuations, certain collection features, and ownership wealth significantly contribute to its formation. Leveraging these parameters, we employ AI-powered models to estimate NFT prices with high accuracy, but find their predictive ability is decreasing in LOVE. Additionally, we demonstrate that LOVE-driven trading results in long-term financial losses for the average investor, highlighting a trade-off between wealth and emotional utility. Our study pro- vides novel economic insights into the factors shaping emotional dividends and their role in the pricing of private-value assets. It also surfaces the limitations AI faces in emotionally charged markets, revealing new challenges for algorithmic trading when assets carry significant emotional utility.