Behavioral Finance Working Group Meeting, Spring 2020

Using a machine-learning algorithm that can predict visually salient portions of images, Bose, Cordes, Nolte, Schneider, and Camerer construct decision weights based on salient parts of a stock price chart. They analyze these weights in three experimental studies that vary in the realism of the price path images and task complexity. The researchers find that these decision weights are predictive of future investments. They conclude that visual salience captures attention paid to historical returns. Visual salience goes beyond overweighting returns at the tails of the historical distribution or with respect to their difference to a reference return as in the models of Barberis et al. (2016) and Bordalo et al. (2013). Moreover, Bose, Cordes, Nolte, Schneider, and Camerer find that visual salience affects investment decisions independently from recency effects.

Many economically important settings, from financial markets to consumer choice, involve sequential decisions under risk. Data from these dynamic settings runs counter from findings in one-shot settings: people are anomalously risk-averse in the latter, while taking on even negative expected-value risk in the former. Heimer, Iliewa, Imas, and Weber use two pre-registered experiments (N = 940) and a unique brokerage dataset of traders' investment plans and subsequent decisions (N = 190, 000) to shed light on this discrepancy. Participants in their experiments made a sequence of choices over whether or not to accept a symmetric fair gamble with feedback provided after every decision. Heimer, Iliewa, Imas, and Weber use both between-subject and within-subject design to compare participants' ex-ante strategies conditional on future outcomes with their actual behavior. A large majority of participants plan to follow 'loss-exit' strategies - to continue taking risk after gains and to stop after losses. Actual behavior exhibited the reverse pattern: participants cut their gains early and chased their losses. The researchers find an analogous dynamic inconsistency in the investment plans and subsequent decisions of traders in their unique brokerage dataset. They formally demonstrate that this behavioral pattern identifies the dynamic predictions of Cumulative Prospect Theory. A significant demand for commitment devices points to at least partial sophistication about the dynamic inconsistency. Heimer, Iliewa, Imas, and Weber use simulations to quantify that the welfare costs of naı̈veté for a representative agent are over one hundred and ten percent of the stakes in a one-round investment. Moreover, participants' wide-spread demand for non-binding commitment, which is ineffective in mitigating dynamic inconsistency, highlights a second form of naı̈veté with respect to the effectiveness of such 'soft' commitment. Their results have implications for evaluating unintended effects of recently introduced regulation that mandates soft commitment.

Ben-David, Li, Rossi, and Song show that correlated demand that is driven by performance chasing creates positive feedback in stock returns, and explains a substantial fraction of the premia of asset pricing factors. Between 1991 and 2018, mutual fund investors chased Morningstar's fund ratings regardless of methodology. Until mid-2002, funds in the best-performing styles received high ratings, and as a result, investors' flows introduced style-level price pressures and positive feedback loops in the underlying equity market. A 2002 revision to Morningstar's methodology equalized investors' demand across styles. The researchers show that the decline in correlated demand explains half of factor profitability, especially for factors that were most exposed to the change in Morningstar's methodology and for momentum.

Bordalo, Gennaioli, La Porta, and Shleifer revisit several leading puzzles about the aggregate stock market by incorporating into a standard dividend discount model survey expectations of earnings of S&P 500 firms. Using survey expectations, while keeping discount rates constant, explains a significant part of "excess" stock price volatility, price-earnings ratio variation, and return predictability. The evidence is consistent with a mechanism in which good news about fundamentals leads to excessively optimistic forecasts of earnings, especially at long horizons, which inflate stock prices and lead to subsequent low returns. Relaxing rational expectations of fundamentals in a standard asset pricing model accounts for stock market anomalies in a parsimonious way.

Long-term treasury yields are known to be: i) excessively volatile (Giglio and Kelly (2018)) ii) highly sensitive to short rate movements (Hanson et al. (2018)) as well as iii) highly predictable from non priced factors (Duffee (2013)). D'Arienzo assesses the possibility that these puzzles may be due to non-rational investor beliefs. Using survey data as well as data on market beliefs recovered from observed yields, he documents that expectations about long rates over-react to news relative to expectations about short rates. The researcher show that introducing diagnostic expectations into an affine term structure model yields such maturity increasing over-reaction and reconciles all three puzzles. When benchmarked to external data on diagnostic distortions, this model accounts for: i) roughly 80% of the excess volatility puzzle ii) for 40% of the excess sensitivity of long rates, and iii) for additional excess bond returns predictability coming from past forecast revisions.

The behavioral finance literature has provided over a dozen explanations for the so-called excessive trading puzzle - retail investors trade a lot even though more trading hurts their performance. It is difficult to use transaction data to differentiate these explanations as they share similar predictions by design. To confront this challenge, Liu, Peng, Xiong, and Xiong design and administer a nation-wide survey among retail investors to elicit their responses to an exhaustive list of trading motives. By merging survey responses with account-level transaction data, they validate survey responses with actual trading behaviors and compare the power of survey-based and transaction-based measures of trading motives. A horse race among survey-based trading motives suggests that overconfidence in having information advantage and gambling preference quantitatively dominate other explanations. Moreover, other popular arguments such as neglect of trading cost do not contribute to excessive trading.