10:30 19/3/2019 RFS-OP-REVF190015.tex Page: 1983 19832020 The Promises and Pitfalls of Robo-Advising Francesco DAcunto Carroll School of Management, Boston College Nagpurnanand Prabhala Carey Business School, Johns Hopkins University Alberto G. Rossi McDonough School of Business, Georgetown University We study the introduction of a wealth-management robo-adviser that constructs portfolios tailored to investors holdings and preferences. Adopters are similar to non-adopters in terms of demographics and prior interactions with human advisers but tend to be more active and have greater assets under management. Investors adopting robo-advising experience diversification benefits. Ex ante undiversified investors increase stock holdings and hold portfolios with less volatility and better returns. Already well-diversified investors hold fewer stocks, yet see some reduction in volatility, and trade more after adoption. All investors increase attention based on online account logins. We find that adopters exhibit declines in prominent behavioral biases, including the disposition, trend chasing, and rank effect. Our results emphasize the promises and pitfalls of robo-advising tools, which are becoming ubiquitous all over the world. (JEL D14, G11, O33) Received May 31, 2017; editorial decision August 21, 2018 by Editor Wei Jiang. Most individual investors could benefit from stock market participation. However, the benefits of participation depend on whether investors hold appro- priately diversified portfolios (Campbell and Viceira 2002; Campbell 2006). 1 In practice, investors do not diversify (Badarinza, Campbell, and Ramadorai 2016). Financial advising can potentially mitigate underdiversification by We are grateful to Wei Jiang (editor), Cam Harvey (RFS-FinTech discussant), and three anonymous referees for very helpful comments on substance and exposition. We also thank Sumit Agarwal, Li An, Brad Barber, Nick Barberis, Jules van Binsbergen, Kent Daniel, Ken French, Cary Frydman, Theresa Kuchler, Cami Kuhnen, Juhani Linnainmaa, Marina Niessner, Wenlan Qian, Nick Roussanov, Felipe Severino, Kelly Shue, David Solomon, Matthew Spiegel, Geoff Tate, Paul Tetlock, David Yermack, and Stephen Zeldes, as well as participants at the 2018 SFS Finance Cavalcade, the 2018 ABFER meetings, the 2018 Cornell Tech FinTech Workshop, the 2017 RFS FinTech Initiative Workshop, the 2017 NBER Behavioral Finance Fall meeting, the 2017 CEPR Household Finance Conference, the 2017 Miami Behavioral Finance Conference, and seminars at SEC and UNLV. Any remaining errors are our own. Send correspondence to Alberto G. Rossi, Georgetown University, McDonough School of Business, 37th and O Streets, NW, Washington, DC 20057; telephone: (301) 405-0703. E-mail: agr60georgetown.edu 1 Diversification is helpful for retail investors who tend to be uninformed, but for informed investors, portfolio choice depends on the nature of private information. See, for example, Kacperczyk, Sialm, and Zheng 2005 and Ivkovic, Sialm, and Weisbenner 2008. The Author(s) 2019. Published by Oxford University Press on behalf of The Society for Financial Studies. All rights reserved. For permissions, please e-mail: . doi:10.1093/rfs/hhz014 Downloaded from by Renmin University user on 05 December 2019 10:30 19/3/2019 RFS-OP-REVF190015.tex Page: 1984 19832020 The Review of Financial Studies / v 32 n 5 2019 helping investors move towards more diversified portfolios, but financial advisers are prone to behavioral biases and display cognitive limitations (Linnainmaa, Melzer, and Previtero 2017). Our study focuses on a financial technology (FinTech) robo-advising tool that delivers diversification advice to individual investors and does not require the intervention of human advisers. We examine the uptake of the tool and assess its impact on investors financial decision-making. We find that adopting robo- advising has different effects on the investment behavior and performance of investors based on their ex ante level of sophistication. We also find that adopting robo-advising reduces a set of well-known behavioral biases for all investors. The robo-advising tool we examine is an automated portfolio optimizer introduced by a brokerage firm to its clients in India. The tool uses Markowitz mean-variance optimization to construct optimal portfolio weights based on historical data and modern techniques such as shrinkage and short-selling constraints to achieve reasonable portfolio weights. The tool is flexible. It allows investors to rebalance current portfolios or add extra stocks, and it includes a tool to let investors visualize the portfolio choices in meanvariance space. Importantly, the tool incorporates simplified trade execution. Investors merely need to click a button to execute in one batch all the trades to get to their target portfolio. We interpret the robo-adviser as a way to simplify the set of decisions investors have to make to rebalance their portfolios. When investors have no access to the tool, rebalancing involves a complex set of decisions. Investors face the daunting task of choosing from a large number of securities and allocating their wealth among the chosen stocks. To simplify this set of problems, investors often use suboptimal rules of thumb (e.g., Frydman, Hartzmark, and Solomon 2018). The robo-advising tool we study simplifies the process, and its automated execution lets investors easily implement the advice they receive. The unique data set we assemble includes information on individual access to robo-advice, investor demographic characteristics, all their trades on a daily basis, and their portfolio holdings at the end of each month. The data span the period from April 1, 2015, to January 27, 2017. The first date users could access the robo-advising tool was July 14, 2017. We propose two empirical strategies. We first show single-difference results that control for unobserved time-invariant investor characteristics. Second, we construct an identification strategy based on a difference-in-differences analysis that exploits quasi-random variation in the introduction of the robo-advising tool to reduce the concerns that time-varying investment motives might drive any results. We start by analyzing the adoption of the robo-advising portfolio optimizer to understand the types of investors receptive to technological innovation in the realm of financial advice. We find that users and non-users are indistinguishable along several demographic characteristics, including their gender, age, and trading experience. At the same time, users have a larger amount of wealth 1984 Downloaded from by Renmin University user on 05 December 2019 10:30 19/3/2019 RFS-OP-REVF190015.tex Page: 1985 19832020 The Promises and Pitfalls of Robo-Advising invested with the brokerage house and are more sophisticated, in the sense that they are more involved with the management of their portfolios and have superior risk-adjusted performance, similar to the U.S. findings reported by Gargano and Rossi 2018. We next analyze the effects of robo-advising on portfolio diversification, risk, and investment returns in a within-investor analysis, which separates out all the time-invariant determinants of adoption. Investors holding less than ten stocks before using the optimizer increased the number of stocks they held and experienced sharp declines in portfolio volatility. For investors with ten or more stocks, the number of stocks held decreased after portfolio optimizer usage, suggesting the optimizer recommended closing positions in stocks that would be shorted had the short-sales constraint not been binding. Although these investors held fewer stocks after adoption, portfolio volatility did not increase but decreased less than for undiversified individuals. The evidence that undiversified investors benefit more from robo-advicewhose technology makes implementation of advice simple also for the less savvy investors suggests robo-advice can be an effective tool to help investors diversify their portfolios, compared with other forms of advice (Bhattacharya et al. 2012; Linnainmaa, Melzer, and Previtero 2017). We move on to assess the effects of the usage of the portfolio optimizer on post-adoption trading. Once again, we sort investors based on their levels of diversification before usage. We find that market-adjusted investment performance improved for less diversified investors. The average returns for the ex ante diversified investors were essentially flat. These investors paid more attention to their portfolio and increased their trading volume, which we proxy by the overall amount of trading fees. We thus continue to find greater benefits of the robo-advising tool for undiversified investors. Our third set of tests examines prominent behavioral biases individual investors exhibit when buying and selling stocks. 2 We follow a before-after design in which we compared the biases one month before the adoption robo- advising with those one month after adoption. We focus the analysis on three prominent biases noted in finance. For selling decisions, we examine the disposition effect, whereby investors are more likely to realize gains than losses on their positions (Shefrin and Statman 1985). To assess buying behavior, we examine trend chasing, whereby investors tend to purchase stocks after a set of positive returns (Barber and Odean 2008). 3 Finally, we also examine the rank effect, whereby investors are more likely to trade the best- and worst- performing stocks in their portfolios (Hartzmark 2014). We test the incidence of these three biases before and after investors accessed robo-advice. The biases were substantially less pronounced, although not entirely eliminated, after 2 For surveys of behavioral finance, see Barberis and Thaler 2003 or Shefrin 2009. 3 For sophisticated investors, this behavior could replicate a momentum strategy. 1985 Downloaded from by Renmin University user on 05 December 2019 10:30 19/3/2019 RFS-OP-REVF190015.tex Page: 1986 19832020 The Review of Financial Studies / v 32 n 5 2019 the usage of the portfolio optimizer. The result holds regardless of investors diversification before usage. Biases were not fully eliminated, because investors could place additional trades on top of what the robo-adviser suggested, and investors could decide to not follow the robo-advice in full. The results described are based on single-difference tests, in which we compare diversification, trading behavior, and trading performance within individuals, before and after usage of the portfolio optimizer. The single- difference tests control for systematic, time-invariant investor characteristics but cannot rule out that time-varying shocks to trading motives that simultaneously triggered both the usage of the optimizer and the change in trading behavior after usage. We tackle this issue by exploiting quasi- random variation in adoptions induced by the way the portfolio optimizer was introduced to the market. Our identification strategy builds on the fact that on a set of dates set by the head of the individual investors division of the brokerage firm, human advisers were provided with lists of clients to call to promote usage of the portfolio optimizer and initiate usage of the tool. The brokerage house had no underlying motivations for pushing the portfolio optimizer at any point in time, apart from the fact that their technology team thought the device was ready to use broadly and they wanted to market it as a free service to their clients. Crucially for our purposes, we accessed a unique field in the data set that identifies all the outbound and inbound calls between human advisers and clients at each point in time. Moreover, we know whether calls went through and, for those that did, the length of the call. In this identification strategy, treated clients were the ones the human advisers reached on the days they were promoting the portfolio optimizer, and who used the optimizer. Control clients were those clients the human advisers tried to contact on the same day to promote the optimizer, but did not answer the phone and hence were not exposed to the tool. 4 We provide direct evidence that in the pre-call period, treated and control clients were similar in terms of observable characteristics that might have predicted their likelihood of answering their advisers phone calls, such as investors initiated calls to human advisers, investors access to their online accounts, their volume of trades, as well as the average performance of their trades. We note the potential concern that the subset of clients advisers called might have been selected based on characteristics the econometrician does not observe. Advisers might have called clients whose unobserved characteristics made them more likely to adopt the optimizer, or clients they thought would benefit the most from using the optimizer. This potential selection is barely a problemand perhaps even advantageousfor our difference-in-differences 4 We only consider non-responsive clients who do not use the portfolio optimizer in the 30 days after the attempted call by their human adviser, which serves to exclude confounding effects of investors who decide to use the tool on their own volition. The results are not sensitive to using different horizons for this restriction. 1986 Downloaded from by Renmin University user on 05 December 2019 10:30 19/3/2019 RFS-OP-REVF190015.tex Page: 1987 19832020 The Promises and Pitfalls of Robo-Advising strategy, because, if anything, the clients advisers called would have been selected based on dimensions the econometrician cannot observe and that made the treatment and control groups even more similar to each other. The control sample contains clients who did not answer the phone, despite being as likely to benefit from the optimizer as clients who answered the phone. Overall, the difference-in-differences specifications confirm our results. 1. Related Literature Our work contributes to multiple strands of literature in finance and economics. First, we contribute to the literature on household finance. Campbell 2006 points out in his presidential address that the benefits of financial markets depend on how effectively households use financial products. 5 Participation in the stock market is optimal from a portfolio-allocation viewpoint given the historically high risk premia of stock market investments. However, attaining these high returns depends on whether investors hold appro- priately diversified portfolios, conditional on not having private information. The actual risky holdings of investors deviate considerably from theoretical predictions (Badarinza, Campbell, and Ramadorai 2016). In particular, participants in the stock market tend to be underdiversified. Underdiversified portfolios result in investors bearing idiosyncratic risk. The field of cultural finance has recently r