Big Data: Long-Term Implications for Financial Markets and Firms

Big Data: Long-Term Implications for Financial Markets and Firms

An NBER conference on Big Data: Long-Term Implications for Financial Markets and Firms took place March 8 in Cambridge. Itay Goldstein of University of Pennsylvania, Research Associate Chester S. Spatt of Carnegie Mellon University, and Faculty Research Fellow Mao Ye of University of Illinois at Urbana-Champaign organized the meeting, supported by the National Science Foundation, in conjunction with the Review of Financial Studies. These researchers' papers were presented and discussed:


Zheng Tracy Ke, Harvard University; Bryan T. Kelly, Yale University and NBER; and Dacheng Xiu, University of Chicago

Predicting Returns with Text Data


Amber Anand, Syracuse University; Mehrdad Samadi, Southern Methodist University; Jonathan Sokobin, Financial Industry Regulatory Authority; and Kumar Venkataraman, Southern Methodist University

Institutional Order Handling and Broker-Affiliated Trading Venues

Using detailed order handling data over the life of 330 million institutional orders, Anand, Samadi, Sokobin, and Venkataraman study whether order routing by brokers to Alternative Trading Systems (ATSs) that they own affects execution quality. In a multivariate regression specification that controls for stock attributes, order characteristics and market conditions, orders handled by brokers with high affiliated ATS routing are associated with lower fill rates. Trading costs based on the implementation shortfall approach are higher when clients select a broker with high affiliated ATS routing. Broker outcomes are highly persistent suggesting that improved disclosures on order handling could help institutional clients with broker selection.


Michael Gofman, University of Rochester; Sajjad Jafri, Queen's University; and James T. Chapman, Bank of Canada

High-Frequency Analysis of Financial Stability

Gofman, Jafri, and Chapman study empirically efficiency and stability trade off in the design of large value payment systems using $500 trillion CAD of intraday transaction level data from Canadian Large Value Transfer System (LVTS). They develop measures of systemic risk and apply these measures to millions of LVTS payments during 2001-2014. LVTS showed stress during 2007-2009. The main source of fragility of the system are binding collateral and credit constraints that cause delays and rejections of payments. Unprecedented injection of liquidity by the Bank of Canada prevented a spillover of systemic risk to global systemically important payment and settlement systems.


David Easley, Cornell University; Marcos Lopez de Prado, AQR; Maureen O'Hara, Cornell University; and Zhibai Zhang, NYU Tandon

Microstructure in the Machine Age

Understanding modern market microstructure phenomena requires large amounts of data and advanced mathematical tools. Easley, Lopez de Prado, O'Hara, and Zhang demonstrate how a machine learning algorithm can be applied to microstructural research. They find that simple microstructure measures designed to reflect frictions in a simpler market continue to provide insights into the process of price adjustment. The researchers find that some of these microstructure features with apparent high explanatory power can exhibit low predictive power, and vice versa. They also find that some microstructurebased measures are useful for out-of-sample prediction of various market statistics, leading to questions about the efficiency of markets. The results are derived using 87 of the most liquid futures contracts across all asset classes.


Jura Liaukonyte, Cornell University, and Alminas Zaldokas, Hong Kong University of Science and Technology

Background Noise? TV Advertising Affects Real Time Investor Behavior

Using minute-by-minute television advertising data covering approximately 326, 000 ads, 301 firms, and $20 billion in ad spending, Liaukonyte and Zaldokas study the real-time effects of TV advertising on investor search for online financial information and subsequent trading activity. The researchers identification strategy exploits the fact that viewers in different U.S. time zones are exposed to the same programming and national advertising at different times, allowing to control for contemporaneous confounding events. The researchers find that an average TV ad leads to a 3% increase in SEC EDGAR queries and an 8% increase in Google searches for financial information within 15 minutes of the airing of that ad. Such advertising effects spill over through horizontal and vertical product market links to financial information searches on closest rivals and suppliers. The ad-induced queries on the advertiser and its key rival lead to higher trading volumes of their respective stocks. For large advertisers, around 0.8% of daily trading volume can directly be attributed to advertising. This suggests that advertising, originally intended for consumers, has a sizable effect on financial markets.


Hedi Benamar and Clara Vega, Federal Reserve Board, and Thierry Foucault, HEC School of Management

Demand for Information, Uncertainty, and the Response of U.S. Treasury Securities to News

Benamar, Foucault, and Vega conjecture that an increase in investors' information demand about an asset signals that their perceived uncertainty about the value of this asset has increased. One implication is that an increase in investors' demand for information should be predictive of a stronger role of news in price discovery. Consistent with this hypothesis, the researchers find that the impact of non-farm payroll news on U.S. Treasury note futures more than doubles when information demand (measured by the number of people reading related news) is high before the release of the announcement.


Robert P. Bartlett III, Richard Stanton, and Nancy Wallace, University of California, Berkeley,and Adair Morse, University of California, Berkeley and NBER

Consumer-Lending Discrimination in the FinTech Era

Ethnic discrimination in lending can occur in face-to-face decisions or in algorithmic scoring. The GSEs' model for pricing credit risk provides us with an identified setting to estimate discrimination for FinTech and face-to-face lenders, as well as to offer a workable enforcement interpretation of U.S. fair -lending laws using the court's justification of legitimate business necessity. Bartlett, Morse, Stanton, and Wallace find that face-to-face and FinTech lenders charge Latinx/African-American borrowers 6-9 basis points higher interest rates for purchase mortgages, consistent with the extraction of monopoly rents in weaker competitive environments and from profiling borrowers on shopping behavior. In aggregate, Latinx/African-American pay $750M per year in extra mortgage interest. FinTech algorithms have not removed discrimination, but two silver linings emerge. Algorithmic lending seems to have increased competition or encouraged more shopping with the ease of applications. Also, while face-to-face lenders discriminate against minorities in application rejection, there are reasons to believe FinTechs may discriminate less.


Isil Erel, Ohio State University; Léa H. Stern, University of Washington; Chenhao Tan, University of Colorado, Boulder; and Michael S. Weisbach, Ohio State University and NBER

Selecting Directors Using Machine Learning (NBER Working Paper No. 24435)

Can an algorithm assist firms in their nominating decisions of corporate directors? Erel, Stern, Tan, and Weisbach construct algorithms tasked with making out-of-sample predictions of director performance. They run tests of the quality of these predictions and show that directors predicted to do poorly indeed do poorly compared to a realistic pool of candidates. Predictably unpopular directors are more likely to be male, have held more directorships, have fewer qualifications, and larger networks than the directors the algorithm recommends. Machine learning holds promise for understanding the process by which governance structures are chosen, and has potential to help firms improve their governance.


Bo Cowgill, Columbia University, and Eric Zitzewitz, Dartmouth College and NBER

Stock Compensation and Employee Attention

Cowgill and Zitzewitz show that daily stock price movements affect the mood, effort level, and decision making of employees. Positive current-day stock returns are accompanied by greater reported economic confidence and job satisfaction, shorter working hours, more optimistically biased beliefs about firm performance, tougher grading of innovative ideas, and tougher evaluation of interviewees. These effects are very short lived, lasting one or two business days. The effects on mood and many types of behavior are larger for employees with larger prior stock and option grants. The researchers show that the short-term effects of (plausibly exogenous) shock to moods is the opposite sign of cross-sectional correlations. Whereas happier employees in the cross section perform better and are more lenient evaluators, shocks that increase happiness longitudinally are accompanied by lower work effort and tougher evaluation.