Picture
War Discourse and the Cross-Section of Expected Stock Returns with David Hirshleifer and Dat Mai​


We test whether a war-related factor model derived from textual analysis of media news explains the cross-section of expected asset returns. The war risk factor is motivated by, and builds on a semi-supervised topic model to extract discourse topicsfrom 7,000,000 New York Times stories spanning 160 years, which has been shown to be powerful in predicting aggregate market returns. We find that war risk factors help predict the cross section of returns across a diverse range of testing assets, deriving from both traditional and machine learning construction techniques, encompassing both public and own-constructed sources, and spanning a wide range of 138 anomalies. These findings are consistent with assets that have poor returns during periods of heightened war risk earning higher risk premia, or alternatively, that a factor based upon war sensitivity captures investor mispricing of war risk. The return premium associated with the war factor is incremental to factors from prominent factor models and other measures of news-based uncertainty. Our results are further buttressed through the factor mimicking portfolio of war risk. War risk passes the protocol of factor identification and is shown to be a genuine risk factor.
Picture
War Discourse and Disaster Premia: 160 Years of Evidence from Stock and Bond Markets with David Hirshleifer and Dat Mai​
​  
  • Revise and Resubmit, Review of Financial Studies
  • The War, PLS, and Economics indices are here. Codes and raw data are available upon request. Email me.
  • Chicago Quantitative Alliance (2nd prize winner, 2021)
  • EFA 2021 (PhD symposium), Texas Finance PhD symposium (2021), NFA (2021), FMA (2021), SFA (2021),      FRA (2021),  AFA PhD poster session (2022), MFA (2022), Texas PhD Symposium (2022), CFDM Conference at Boulder (2022), Bank of Portugal (2022), Center for Financial Research Cologne (CFR 2022)
  • Long Tail Alpha presentation (2021)
Using a semi-supervised topic model to extract discourse topics from 7,000,000 New York Times stories spanning 160 years, we test whether topics of media discussion are predictors of future market returns. This provides a test of theories of rare disaster risk premia, and of behavioral hypotheses about the valuation of the prospect of rare disasters. Using data on investors’ perception of rare disaster risks derived from news allows us to circumvent the empirical problem that major disasters are rare. Our methodology also avoids look-ahead bias and addresses the semantic shifts that occur over long time periods. War discourse positively predicts market returns both in-sample and out-of-sample, with an out-of-sample R-squared of 1.35%, and beats all other discourse topics as return predictors. War’s predictive power increases through the sample period and is robust when constructed from the Wall Street Journal. These findings are consistent with a focus by investors on the risk of war in media discourse being associated with low market valuations and high subsequent returns.
Slope Factors Outperforms? Evidence from a Large Comparative Study with Siddhartha Chib, Yi Chun Li and Xiaming Zeng
  • Reject and Resubmit, Management Science
  • Washington University (2021); CICF (2022); SOFIE- New York University Shanghai (2022)
  • This paper compare slope, rank and differential factors altogether. ​
Does the method used to construct long-short factors from firm-level characteristics change the assessment of the factor risks (risk-factors) that are incorporated in the cross-section of expected returns? Is the method of construction important for the pricing performance of those risk- factors? We compare factors constructed by three methods: pure-play slope factors (estimated OLS slopes from cross-sectional regressions), differential factors (difference in returns of 2 by K sorts of stocks) and rank factors (normalized rank weighted returns). Starting from 62 firm level characteristics, we construct 62 such factors from each method. For each set of 62 factors plus the market factor, we examine the posterior distributions of the SDF coefficients to infer risk-factors, and apply a Bayesian pricing criteria to determine the evidence in favor of pricing for a large collection of portfolios, ETFs and stocks. The evidence shows that the slope risk-factors strongly dominate, with important implications for empirical finance.
VC ownership post-IPO: When, why, and how do VCs exit? with Anup Basnet, Harry Turtle and Thomas WalkerWe examine how the ownership of lead venture capital firms (VCs) evolves after their portfolio companies (PCs) are publicly listed. The VC investment period before the IPO, the VC age, the PC age, and the percentage change in the post-IPO stock price all incentivize earlier VC exit. Lead VCs remain invested for a longer period when the PC is of better quality, when the VC has more experience in taking companies public, and when the VC holds positions in the company’s compensation committee. VCs with longer pre-IPO investment periods prefer share distributions (SDs) to continuous sales (C sales), perhaps because SD provides a more expeditious exit compared to C Sales and M&A. Older lead VCs prefer C Sales to M&A, VCs with prior M&A exit experiences prefer M&A exits, and VCs investing in younger PCs prefer SDs. ​
Picture
Climate Risk Preparedness and the Cross Section with Sid Chib and Leifei Lyu


​
​Our paper presents a theoretical model to analyze the relationship between a company’s expected returns and its preparation for climate risk. The model is based on a two-date investment framework and considers the impact of a climate change-induced event (CCIE) that may occur on the second date. The extent of damage caused by this event is dependent on the amount spent on climate preparedness on the first date. The company has an initial capital stock, profitability, and readiness. Our analysis demonstrates that the expected return on equity is directly proportional to the company’s preparedness for climate change, profitability, and size, but inversely proportional to investment. In addition, we have developed an empirical model based on five factors that can be used to price stocks, ETFs, anomalies, and characteristics sorted portfolios.
Picture
Animating Stock Markets by Tomasz Kaczmarek and Kuntara Pukthuanthong
Even wonder if we can use graphs to predict graphs in a form of animation. You got it. We can! Coming soon. 
  • ​MFA 2023
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
Picture
No One-Size-Fits-All Tale: The Diversity and Complexity in Asset Pricing across Global Markets with Zhuo Qiao and Yan Wang


We utilize a Bayesian approach developed by Chib, Zeng, and Zhao (2020) and Chib and Zeng (2020) to identify the best asset pricing models for international stock markets. Our findings reveal that models with Student-t distribution outperform models with Gaussian distribution in all markets. However, the best models vary depending on the specific market, and the factor strength fluctuates over time, highlighting the diversity and complexity of asset pricing in different markets. Although currency risk factors are crucial in international asset pricing, they were weakened during the global financial crisis in 2008. Additionally, we evaluate the efficiency of our local, regional, and global models in explaining various cross-sectional anomalies. Our research indicates that local factor models are the most successful, while regional and global models have similar outcomes.
Picture

Cracking the Code: The Networking Matrix of Finance Academia with  Sujiao (Emma) Zhao

​
  • Bank of Portugal (July 2023)​
  • FMA (2023)
This research delves into social networks among finance academics from 1980 to the present. Engagement in social networks positively correlates with productivity, with co-authors and student-advisor networks having the most significant impact, followed by post-Ph.D. and Ph.D. networks. Focusing on existing relationships rather than expanding connections and connecting with prominent researchers has proven more effective. Scholars with strong networks produce higher-quality research and receive better compensation. Our findings on editorial favoritism are complex. Co-authors of editors and Ph.D. colleagues are less likely to have their papers accepted, while student advisors are more likely. There is no evidence to support bias toward female scholars, but discrimination against Asian and Hispanic females and favoritism towards White females and Hispanic males is evident.
Picture

​
  • University of Missouri Columbia, Missouri State University, University of Missouri St. Louis, the Fields Institute for Research in Mathematical Sciences at University of Toronto 2022, Poznan University, Blackrock 2023, FinTech City University of Hong Kong 2023, and Northern Finance Association meeting 2023
​We present a novel approach to assess firm similarity by analyzing two million images. We leverage machine learning techniques to identify graphical objects that best represent companies’ operations, forming Image Industry Classifications (IIC). IIC mirrors investor-defined peer groups and, akin to the brain’s visual processing superiority, outperforms SIC, GICS, NAICS, and text-based similarity in delivering the greatest diversification benefits and industry momentum profits. This success is attributed to high investor agreement within IIC, leading to substantial aggregated demand and supply effects on stock prices. IIC excels in industries characterized by expectation-driven value, growth, and intangibility.
Picture
AI Narrative and Stock Mispricing with  Arka Bandyopadhyay and Dat Mai


​​
  • https://alphaarchitect.com/2023/06/ai-exposures/
  • https://www.barrons.com/articles/ai-nvidia-tech-winners-has-beens-e0dec10
We apply advanced natural language processing to develop a dynamic dictionary of artificial intelligence (AI). Using this dictionary, we construct a real-time index of AI attention from more than 3,000,000 New York Times articles. Firms having high exposures to AI have higher returns one month ahead and lower returns five to seven months ahead, suggesting initial overreactions to AI news and subsequent corrections. The connection between AI exposures and future returns is concentrated among non-big stocks, indicating that small AI stocks are more difficult to value. A long-short AI exposure portfolio among non-big stocks generates significant annual alphas of 3% against benchmark multifactor models.
Picture
Diversity Narrative and Equity in Firm Leadership with Arka Bandyopadhyay and Dat Mai



​We provide causal evidence that the narrative of diversity from the New York Times articles has nudged the corporations to choose female CEOs to be equitable in terms of gender of firm leadership. This channel is independent of the board gender diversity, which was mandated in 2017 in California and later repealed in 2022. Surprisingly, the diversity narrative channel fails to explain the election of Indian CEOs in several HiTech firms over the last few years. We argue that the election of Indian CEOs was motivated to create a favorable image of Tech firms to Indian and Chinese labor. We conclude that providing equity in firm leadership in terms of ethnic diversity is harder compared to gender diversity.
Picture
  • FMA (2023)
​This paper investigates the diversification benefits of commodities to a portfolio of traditional assets from the perspective of domestic investors around the world. We develop a model and derive a set of implications leading us to argue that a country’s degree of commodity dependence should impact the benefits of including commodities into domestic investors’ porfolios. Using an international sample of 39 countries, we find support for our hypothesis in that investors located in high-commodity dependence countries usually do not benefit from adding commodities to their portfolio while investors located in low-commodity dependence countries usually do.
Picture
Do robots hurt humans? Evidence from the dark side of workplace automation with Zhihua Wei, Aoran Zhang, and Yunfei Zhao

  • FMA (2022)
We investigate the influence of utilizing automated capital to replace human workers in U.S. public firms. Firms that can replace human capital with automated capital need better records on employee related corporate social responsibility (employee CSR) and workplace safety. Our results remain robust after addressing the endogeneity concern. We further document that the negative link is more powerful for firms adopting more aggressive financial policies, incurring higher labor-related costs, and locating in less religious areas. Our analysis also unveils an ethical conundrum. The compromise on employees’ well-being among firms that are highly subjective to workplace automation comes at the cost of reputation damage; however, such firms are associated with better financial performance. Overall, we document that workplace automation creates a moral hazard: firms expect to increase productivity with automated capital while paying less attention to current employees’ welfare.
Investor Sentiment and Asset Returns: Actions Speak Louder than Words  with Dat Mai and Guofu Zhou
We analyze daily predictability of investor sentiment across four major asset classes and compare sentiment measures based on news and social media with those based on trade information. For the majority of assets, trade-based sentiment measures outperform their text-based equivalents for both in-sample and out-of-sample predictions. This outperformance is particularly noticeable in long-term forecasts. However, real-time mean-variance investors can only achieve economic gains using Bitcoin trade sentiment, suggesting the challenge of transforming sentiment into daily profitable trading strategies.
Human Capital Valuation, Asset Pricing, and Economic Development with Dave Berger and Richard Roll
  • Semi-Finalist of Best Paper Award 2016, Financial Management Association Meeting World Finance Conference 2021
Human capital is the largest component of aggregate wealth, but its relation to other macroeconomic variables is murky due to the lack of market prices. Valuing human capital using historical costs or expected income is characterized by substantial measurement error. We develop a human capital index using slave prices and relate its dynamics to that of other indicators including equities, GDP, real estate and interest rates. The human capital values are extrapolated from the 19th Century to the modern era. Their evolution has substantial implications for our understanding of the human capital dynamics, with applications to growth and portfolio allocation.
Changing Expected Returns can Induce Spurious Serial Correlation with Richard Roll and Avanidhar Subrahmanyam Changing expected returns can induce spurious autocorrelation in returns. We show why this happens with some simple examples and then investigate its prevalence in actual equity data. Using different measures of expected returns based on realized means, options markets, factor models, and analysts’ price targets we find evidence that supports our premise. Specifically, absolute shifts in expected returns are strongly and positively related to measured autocorrelations in the cross-section of individual stocks, as predicted by our analysis. Well-studied risk factors show no evidence of spurious components. We also show how our analysis implies spurious cross-autocorrelation and find supporting evidence for this phenomenon as well.
Interest Rates and Real Estate Values: The Divergence Effects of Real Rates and Expected Inflation with Richard Roll
​A common belief is that higher interest rates reduce real estate values, but corroborating evidence is underwhelming. A sensible explanation suggests heterogeneity across countries is that mortgage interest deductibility for the U.S. personal income tax makes much of the real value of the mortgage principal tax deductible when inflation is high. This implies that nominal mortgage rates could positively affect U.S. house prices. We find evidence consistent with this possibility; the inflation component of nominal interest is associated positively with U.S. real estate prices but negatively with those in Canada, a country without mortgage interest deductibility.
A New Method for Factor-Mimicking Portfolio Construction with Richard Roll, Junbo Wang, and Tengfei Wang
  • Finalist of Crowell Prize 2019, PanAgora Asset Management
  • AFA (Ph.D. Poster), CICF, EFA, FMA, Louisiana State University, NFA, the Society of Financial Econometrics Conference (SoFiE), the World Finance Conference, and PanAgora Asset Management
We relate factor-mimicking portfolios (FMPs) to the beta-pricing model and propose that each FMP should minimize the mispricing component of its underlying factor. We also examine FMP construction when the underlying factor contains noise and offer a new method to resolve this issue. For both classical and our newly proposed methods, we recommend enhanced necessary criteria. FMPs of several macroeconomic factors constructed by our method satisfy these criteria. We find that equities are priced by consumption growth, inflation, and the unemployment rate; and corporate bonds are priced by consumption growth, industrial production, and the default spread.
Agnostic Tests of Stochastic Discount Factor with Richard Roll and Junbo Wang We propose an estimator for the stochastic discount factor (SDF) that does not require macroeconomic proxies or preference assumptions. It depends only on observed asset returns yet is immune to the form of the multivariate return distribution. It does not depend on factor structure, thus making our SDF ideal for asset pricing test. We find that SDF is correlated with a large number of the factor/firm characteristics, which confirms Kozak, Nagel, and Santosh (2018)’s finding. Moreover, assets within the equity market are not fully integrated as SDFs constructed from different asset groups are uncorrelated.
Picture
Asset Prices and Partisanship: Evidence from Daily Shopper Data with Jialu Shen and Ruixiang Wang


  • The data of mimicking portfolio of our consumption growth is here.
  • Midwest Finance Association meeting 2021; FutFinInfo 2021; World Finance Conference 2021; FMA 2021, CICF 2023​, and AFA 2024
We propose a novel consumption measure that has a daily frequency and is based on real-time shopping data. Our measure explains the joint equity-premium–risk-free rate puzzle with a risk aversion coefficient much lower than any other consumption measures. It encompasses other consumption measures in explaining the cross-sectional variation of expected returns on various portfolio and is the only consumption measure that passes Kleibergen and Zhan (Journal of Finance, 2020) robust tests. Our model decomposes consumption shocks into different frequency of volatility and shows that ignoring short-term dynamics and intra-annual fluctuations explains the much higher risk aversion from low-frequency consumption measures. At state-level daily consumption, (a) consumption in blue states suggests higher risk aversion than that in red states; (b) only Democratic consumption beta explains a variation of cross-sectional returns, and is more sensitive to overall industry performance.
Average and Marginal Tobin’s q as Indicators of Future Growth Opportunities, Expected Return, and Risk  with Richard Roll

Contrary to popular opinion, average Tobin’s q is a better indicator of future growth opportunities than marginal Tobin’s q. We derive a curious relation between average and marginal q: the more profitable a new investment opportunity, the smaller will be the increase in average q when the opportunity is undertaken. Average q is inversely related to the cost of equity capital, so it represents an inverse measure of risk. The closely- related book/market ratio is also a measure of risk in the cross-section.
Picture
Blaze New Trails for Others to Follow: Evidence from Scanner Data with Ruixiang Wang​
  •  
​
  • FMA 2019, SWFA 2020, AFA 2022, NFA 2022
  •  Twinbeech Capital (2021)
Tracking more than 100 billion weekly transactions of two million products at the barcode level from 2007 to 2017, we identify and categorize new products as pioneers, followers and improvers to study corporate exploratory and/or exploitative innovation strategies. Firms introducing “pioneer” products are associated with greater future profitability and stock returns than those introducing “improver” and “follower” products. Price elasticity of demand explains pioneering (exploratory) innovation’s operating success. Meanwhile, limited investor attention accounts for pioneering firms’ superior stock performance. We exploit two exogenous shocks to firms’ new product development decisions to address endogeneity concerns..
Are stock market anomalies anomalous after all with George Chalamandaris, and Nikolaos Topaloglou
  • World Finance Conference 2021
We propose a stochastic spanning to evaluate whether anomalies are genuine under factor-model framework. Our approach is nonparametric and does not rely on any assumption of return distribution and investor risk preferences. It depends on the whole distribution of returns, rather than only on the first two moments. Of the anomalies considered, only a few expand the opportunity set of the risk-averter and have real economic content. Our approach is consistent in identifying genuine anomalies in and out of samples. This is in contrast to mean-variance (MV) spanning tests where anomalies identified in-sample, not out-of-sample.
NEED MAJOR REVISIONS
​A Generalized Machine Learning Framework for Linear Factor Model Test with Christopher Jones, Jinchi Lv and Junbo Wang
We introduce a generalized statistical learning method, sparse orthogonal factor regression (SOFAR), in testing linear factor models with both large numbers of factors and testing assets. Our approach encompasses most of the existing methods in the literature and can be used in many other scenarios with large data sets. Applying SOFAR, we can select the PC factors from the whole swath of 219 candidate factors proposed by the literature simultaneously, identify test assets associated with the selected PC factors, and interpret them. We can also select the PC factors and correlated characteristics in the IPCA framework without bootstrapping. Without firm characteristics instrumenting, we find that four PC factors (market, investment, intangible, and frictions) are relevant to the covariance of asset returns and three types of factors (profitability, asset liquidity, and liquidity bets) price assets in cross-section. With characteristics as instruments, we only identify one factor, and the correlated characteristics are beta, size, momentum, and liquidity.
​Machine Learning Classification Methods and Portfolio Allocation: An Examination of Market Efficiency with Yang Bai We design a novel framework to examine market efficiency through out-of-sample (OOS) predictability. We frame the asset pricing problem as a machine learning classification problem and construct classification models to predict return states. The prediction-based portfolios beat the market with significant OOS economic gains. We measure prediction accuracies directly. For each model, we introduce a novel application of binomial test to test the accuracy of 3.34 million return state predictions. The tests show that our models can extract useful contents from historical information to predict future return states. We provide unique economic insights about OOS predictability and machine learning models.
Convenience Yields of Collectibles, with Elroy Dimson and Matthew Vorsatz. 
​Please also see the Appendix and additional materials.

​We propose a novel method to estimate convenience yields of collectibles based on factor mimicking portfolios. Using up to 110 years of collectibles returns for 13 distinct asset classes, we apply machine learning techniques to address challenges from non-synchronous trading. We use these estimates to study how convenience yields affect equilibrium pricing. Convenience yield estimates for 24 of our 30 collectibles return series are positive, with an annualized mean (median) of 2.64% (2.53%). Despite various forms of underestimation, these results provide evidence that assets with positive emotional returns have lower equilibrium financial returns. This finding has important implications for ESG investing.
NO BIG DEAL PAPERS
Is Media Tone just a Tone? Time-Series and Cross-Sectional Evidence from the Currency Market with Kari Heimonen and Heikki Lehkonen

  • FMA 2021, NEU (brownbag)
Media tone constructed from 7,000,000 articles from 2,000 global media and 800 social media sites predicts excess US dollar returns up to six months out of sample. Its predicted value contains information beyond those predicted by currency factors and business cycles. Our evidence collaborates with the theory that Media tone increases investment returns, has pronounced predictive power for the currencies associated with hard-to-value characteristics, and its predictive power increases with media sources. Surprisingly, Media tone is not just a tone but a genuine risk factor that cross-sectionally prices currencies. Rational investors, including banks, trade on Media tone.
An Unambiguous Statement of Interest Rate Parity with Lee R. Thomas III
Interest rate parity (IRP) states that the difference between the interest rates of two currencies equals the expected percentage change in the associated exchange rate. Unfortunately, the IRP is ambiguous. There are two, inconsistent values of the expected percentage change in the exchange rate, depending on which currency is treated as the numeraire. We derive a corrected IRP condition that is unambiguous but subtly different. The equilibrium interest rate differential equals the percentage difference between the spot exchange rate and the geometric mean of the future exchange rate distribution, not its expected value.
Foreign Exchange Exposure – No Returns, only risk with Kari Heimonen and Heikki Lehkonen
Returns of portfolios formed based on US dollar foreign exchange exposure are found to have a negative relationship with the absolute value of the foreign exchange exposure. The result is state-dependent and visible during the appreciation periods of the US dollar. We relate this effect to mispricing where investors fail to consider to true impacts of the exchange rate changes on firm values and report several results supporting this claim.