Brad M. Barber, University of California, Davis
Resolving a Paradox: Retail Trades Positively Predict Returns but are Not Profitable
Retail order imbalance positively predicts returns, but on average retail investor trades lose money. Why? Order imbalance tests equally weight stocks, but retail purchases concentrate in attention-grabbing stocks that subsequently underperform. Long-short strategies based on extreme quintiles of retail order imbalance earn dismal annualized returns of -15.3% among stocks with heavy retail trading but earn 6.8% among other stocks. Our results reconcile the literatures on the performance of retail investors, the predictive content of retail order imbalance, and attention-induced trading and returns. Smaller retail trades concentrate more in attention-grabbing stocks and perform worse.
Sanjiv Das, Santa Clara University
Algorithmic Fairness in Finance
Research on fairness in machine learning is extensive, yet has seen little application in industry. From time to time we see obvious examples of AI/ML bias but there is little systematic regulation that encourages fairness in decision systems. Fairness in ML seems derivative of legal statutes for fairness, whereas it may be useful to design ML fairness systems bottom up. This talk will focus on a framework for ML fairness in finance, develop the simple arithmetic metrics that may be used, and how bias may be mitigated. The goal of the seminar is to develop simple concepts, offer empirical examples, and engage in a discussion of how fairness may be implemented in automated financial algorithms.
Professor John M. Mulvey, Princeton University
Regime-Aware Portfolio Models
Historically, traded markets display changing patterns of performance, such as large increases in volatility and correlation during market crashes. These conditions are caused by a variety of behavioral, political, and economic forces. A natural approach is to identify financial regimes – time periods with relatively homogeneous behavior. This talk describes a recent formal model for labelling regimes through a combination of clustering and a jump penalty across time. We show how the two objectives can be blended for both discrete and continuous states. Behavioral issues, among others, can impact the size of the penalty hyperparameter.
We develop a multi-step process wherein regimes are labelled over a training/validation period, followed by a forecasting model for specifying the probability of regimes ahead. The final step is to construct a regime-aware portfolio model and evaluate over an independent test period. The approach is interpretable, has foundations in financial and behavioral economics, and leads to excellent empirical performance. We compare the process to end-to-end conditional methods such as deep neural networks – which are difficult to interpret.
Terrance Odean, University of California, Berkeley
Disposed to be Overconfident
We show that the disposition effect – the tendency of investors to hold losers and sell winners – can be a source of overconfidence. We find experimental evidence that individuals update beliefs about their own investment ability based on realized gains and losses rather than the overall performance of their portfolio. We also find supporting field evidence. Dutch retail investors who realized more gains than losses believe they have higher portfolio performance relative to other investors, even after controlling for their actual portfolio performance. We develop a formal model demonstrating how the disposition effect leads to overconfidence and examine model implications for investors’ trading behavior and expected profit.
Christine A. Parlour, University of California, Berkeley
Battle of the Bots: Flash Loans, Miner Extractable Value and Efficient Settlement
Settlement on decentralized ledgers is transparent and batched. The settlement also allows settlement agents to expropriate profitable arbitrage trades. Arbitrage may be socially beneficial or wasteful. We model the effect of an alternate, private settlement on arbitrage. We document payments from arbitrageurs to private settlers that exceed 1 million USD per day.
Meir Statman, Santa Clara University
Behavioral Finance: The Third Generation Finding Well-being: From Financial Well-Being to Life Well-Being
Financial well-being comes when we can meet current and future financial obligations, able to absorb financial setbacks, and keep driving toward financial goals. Life well-being comes when we live satisfying lives, full of meaning and purpose. Life well-being is at the center of the third generation of behavioral finance, broadening its lens beyond standard finance and the first and second generations of behavioral finance. It portrays people as whole persons and depicts then in life well-being domains beyond finances, including those of family, friends, health, work, education, religion, and society. The third generation is also explicit in describing life well-being as people’s overall want, clarifying tradeoffs within each domain and among domains.
The domain of finances is only one of the domains of well-being yet it has a special place in life well-being because it is important on its own and because it underlies well-being in all other domains. We need finances to support ourselves and our families, paying for food, shelter, and schooling. We need finances to maintain our own health and that of our families, paying for the services of physicians and hospitals. We need finances to pay for education that would qualify us for well-paying and satisfying jobs and careers. We even need finances to experience and express our religion.