Animal Spirits

Having been trained by humans on human information, do Large Language Models (LLM) behave like human investors? In their January 2026 paper entitled “Artificially Biased Intelligence: Does AI Think Like a Human Investor?”, Javad Keshavarz, Cayman Seagraves and Stace Sirmans investigate whether 48 widely used LLMs exhibit any of 11 known cognitive biases in financial decision-making. They speculate that LLMs acquire biases via human-authored training data, statistical learning and responses that reward perceived helpfulness over logical consistency. Specifically, they test whether:

• LLMs reliably exhibit economically meaningful cognitive biases in financial decision problems.
• Any biases vary across LLMs with different levels of intelligence.
• Users can intervene to suppress any biases in real-time LLM use.

Their prompt-pair methodology ensures that findings are causal rather than just correlational. Using 25 prompt-pairs per each of 11 biases across 48 LLMs, they find that: Keep Reading

Do analysts/investors predictably and exploitably misinterpret tones of earnings calls? In their October 2025 paper entitled “Do Investors Get It Right? Reaction Bias to Earnings Calls”, Zhenzhen Fan and Fred Liu study interactions between textual information in earnings conference calls and analyst revisions to their next-quarter earnings forecasts. Specifically, they:

• For each stock and each following analyst: (1) link call transcripts to the closest pre-call and post-call analyst forecasts, and (2) measure analyst reaction as the fraction of the pre-announcement forecast error corrected in the post-announcement revision.
• Use Shapley values to assess the contribution of each model feature to predictive power.
• Across stocks and earnings calls, use a random forest model to test whether earnings call transcripts predict direction and magnitude of reaction bias. Use seven rolling years of the sample for model training/validation and test predictive accuracy each next year.
• Each month sort stocks into fifths (quintiles) based on predicted reaction bias and form a value-weighted, monthly rebalanced portfolio that is long stocks with the highest signal and short the stocks with the lowest. Keep a stock within its quintile until the end of the month of its next earnings announcement.

Using earnings conference call transcripts, analyst earnings forecasts, actual earnings and monthly stock return data during 2006 through 2023, they find that:

Keep Reading

How do exchange-traded-funds (ETF) focused on supplying renewable energy perform? To investigate, we consider nine of the largest renewable energy ETFs, all currently available, as follows:

• • iShares Global Clean Energy (ICLN)
  • Invesco Solar (TAN)
  • First Trust NASDAQ Clean Edge Green Energy Index (QCLN)
  • Invesco WilderHill Clean Energy (PBW)
  • ALPS Clean Energy (ACES)
  • Invesco Global Clean Energy (PBD)
  • SPDR S&P Kensho Clean Power (CNRG)
  • First Trust Global Wind Energy (FAN)
  • Global X Renewable Energy Producers (RNRG)

We use SPDR S&P 500 (SPY) as a benchmark, assuming investors look at renewable energy stocks to beat the market and not to beat the energy sector. We focus on monthly return statistics, along with compound annual growth rates (CAGR) and maximum drawdowns (MaxDD). Using monthly returns for the nine renewable energy ETFs and SPY as available through September 2025, we find that: Keep Reading

Is it reasonable to assume that strong earnings growth and price-to-earnings ratio (P/E) expansion will sustain the unusually strong U.S. stock market returns of the past decade? In his brief September 2025 paper entitled “Expected Stock Returns in Bullish Times”, Javier Estrada decomposes stock returns into: (1) dividend yield, (2) change in earnings and (3) change in P/E. He then employs this decomposition to compare the bullish environments at the end of the 1990s with that of the summer of 2025, including an outlook for the next decade. Using Robert Shiller’s prices, earnings and dividends for the S&P Composite Index during 1872 through June 2025, he finds that: Keep Reading

What variables drive differences in price-to-earnings ratios (P/E) across U.S. stocks? In the September 2025 revision of their paper entitled “The Cross-section of Subjective Expectations: Understanding Prices and Anomalies”, Ricardo Delao, Xiao Han and Sean Myers analyze cross-sectional P/E differences of U.S. common stocks by comparing:

• Professional forecasts of earnings growth, return and P/E over the next four years.
• Actual (realized) earnings growth, returns and P/E over the next four years.

Using the specified forecasted and actual data during 1999 (1982 for some inputs) through 2020, they find that: Keep Reading

Do individual investors trade carefully and cautiously, or do they make snap judgments? In their March 2025 paper entitled “The Research Behavior of Individual Investors”, Toomas Laarits and Jeffrey Wurgler employ browser histories for an approximately representative sample of U.S. individual investors to address:

• How much time do they spend on stock research?
• Which sites do they use?
• On which stocks do they focus?
• How proximate is their research to their trades or to corporate events?
• What types of information do and do not attract their attention?

Using browser and trading histories, with confidential/personal data removed, from 484 volunteer households that make 2,911 individual U.S. stock trades via online brokerage accounts during January, February, March and June of 2007, they find that:

Keep Reading

Is proximity to doom good or bad for the U.S. stock market? To measure proximity to doom, we use the Doomsday Clock “Minutes-to-Midnight” metric, revised intermittently in late January via the Bulletin of the Atomic Scientists, which “warns the public about how close we are to destroying our world with dangerous technologies of our own making. It is a metaphor, a reminder of the perils we must address if we are to survive on the planet.” Using the timeline for the Doomsday Clock since inception in 1947 and contemporaneous end-of-year levels of the S&P 500 Index through 2024, we find that:

Keep Reading

Are there “trigger” words in risk sections of annual U.S. firm 10-K reports that materially influence buying and selling of associated stocks? In his December 2024 paper entitled “Risky Words and Returns”, Sina Seyfi tests a way to predict stock returns by analyzing the text of risk disclosures in respective firm 10-K reports. Specifically, he searches for words in the risk sections that predict the cross-section of stock returns by regressing future returns on specific words. He measures the import of findings by each month reforming an equal-weighted hedge portfolio that is long (short) firms with in the highest (lowest) tenth, or decile, of emphasis on predictive risk words. Using 10-K reports, firm characteristics and returns for a broad sample of U.S. stocks and stock factor returns during 2005 (when the SEC started requiring risk sections) through 2023, he finds that: Keep Reading

Experimental (researcher-imagined) asset markets provide a controlled environment for testing hypotheses about investor behaviors. Do limits on abilities of researchers to model markets realistically, and researcher incentives/motivations, jeopardize the credibility of associated studies? In their December 2024 paper entitled “Do Experimental Asset Market Results Replicate? High-powered Preregistered Replications of 17 Claims”, Christoph Huber, Felix Holzmeister, Magnus Johannesson, Christian König-Kersting, Anna Dreber, Juergen Huber and Michael Kirchler attempt to replicate findings on behavioral drivers of asset prices from four prominent studies based on experimental asset markets.  Specifically, they seek to replicate 17 published findings on: (1) associations between asset prices and emotions, self-control, experience and gender; and, (2) trader characteristics (cognitive reflection, fluid intelligence and theory of mind) that explain trading success. 14 of the 17 published findings are reported as statistically significant. To mitigate snooping bias, they fully document (preregister) study design, analyses and statistical tests before collecting any data. When evaluating replicability, they consider statistical significance (confirmation of prior finding with p < 0.05) and the ratio of effect size in the replication to that in the original study. Using new data from 166 experimental asset markets with 1,544 participants (more than used in the original studies), they find that:

Keep Reading

Do some stocks entail emotional relationships that alter investor perceptions of risk and return? Is the effect exploitable? In their November 2024 paper entitled “Investor Emotions and Asset Prices”, Shehub Bin Hasan, Alok Kumar and Richard Taffler develop and test a measure of the emotional state of the market and assess its implications for individual stocks. Specifically, they each month:

1. Use a bag-of-words approach encompassing 295 emotion words to construct a market-level emotion index as the ratio of emotion words to total number of words in newspaper articles about the S&P 500 Index.
2. Estimate for each stock an emotion beta by regressing monthly excess returns versus the market emotion index over the last 60 months.
3. Sort stocks into tenths (deciles) based on last-month emotion beta and compute monthly value-weighted returns of the decile portfolios.

Using 65,825 news articles about the S&P 500 Index from 21 national and local newspapers, monthly returns and firm/stock characteristics for a listed U.S. stocks and monthly returns for various stock factors during January 1990 through September 2022, they find that:

Keep Reading