Working Papers

“Bias and Predictability in Analysts’ Beliefs” - Job Market Paper

Abstract: I study the relationship between sell-side analysts’ forecast bias and stock returns by comparing three forecast families—price targets (PTG), earnings per share (EPS), and long-term growth (LTG)—to ex-ante machine-learning (ML) benchmarks. Bias in return expectations co-moves positively with bias in cash-flow expectations, suggesting a common behavioral source of belief distortions. Analysts’ deviations from ML benchmarks across all forecast families predict returns in a systematic and nonlinear manner. Across horizons, forecast bias predicts returns positively in the short term but negatively in the long term, with LTG-based return-predictability regressions exhibiting similar patterns. Across the sign of bias, optimistic (pessimistic) forecast bias positively (negatively) predict returns. Finally, I document a common tendency to anchor expectations on consensus forecasts. An asset-pricing model with asymmetric information and positively skewed fundamental shocks generates these dynamics: analysts anchor on consensus and misperceive the informativeness of signals, producing the observed bias–return relation.

“Time-Series and Cross-Section of Risk Premia Expectations: Evidence from Financial Analysts”

2nd Round R&R, Journal of Financial Economics

Abstract: I show that—despite potential bias—sell-side analysts’ return expectations are the closest to rational, most informative, and relevant among survey forecasts. They efficiently incorporate available information, exhibit less predictable forecast errors, and align with a multi-factor model in the cross-section. Out-of-sample tests highlight their predictive ability, and a machine learning benchmark trained to predict single-stock outcomes assigns significant importance to their beliefs. Analysts’ expectations are strongly positively correlated with mutual fund institutional flows and allocations, and intraday returns, underscoring their relevance for institutional investment decisions. A multi-asset, heterogeneous-agent, two-period model identifies transmission mechanisms that account for these empirical findings.

Award: John A. Doukas Doctoral Best Paper Award, 2023 EFMA.

“The Term Structure of Return Expectations” with Cameron Peng

Abstract: Using a long-running survey of U.S. investors, we examine subjective return expectations for the aggregate stock market across horizons from one month to ten years. Short-term expectations are extrapolative, whereas forward return expectations are contrarian. The term structure of return expectations is upward-sloping in bad times and relatively flat in good times. We introduce perceived mispricing as a complementary belief measure and show that it helps explain the dynamics of the term structure: it is negatively correlated with short-term return expectations but positively correlated with forward return expectations. Examining the information content of these belief measures, and how they are reflected in free-text investment rationales and related to both stated and actual trading decisions, we show that both long-term expectations and perceived mispricing are central to investment behavior.

“From Numbers to Words: Breaking Down Institutional Beliefs” with Andrea Andolfatto

Abstract: We examine how large asset managers form and justify long-horizon beliefs by analyzing their Capital Market Assumptions (CMAs)—articulated through tables, figures, and narratives. We develop a novel method that transforms CMA text narratives into quantifiable causal networks using large language models, capturing both the complexity of managers’ mental models and their allocation of attention across macro-financial topics. Our analysis reveals substantial heterogeneity in asset managers’ beliefs, both quantitative and narrative. Using granular numerical data on the building blocks of managers' return expectations, we identify multiple drivers of cross-sectional dispersion in expectations. Text-based measures show that the average coefficient of variation in cognitive complexity exceeds 0.7, while that in topic attention exceeds 1, indicating pronounced dispersion in both how managers reason and how they allocate attention to economic relationships. We further document systematic biases in asset managers’ beliefs using both quantitative and textual evidence. Return expectations deviate predictably from objective benchmark forecasts: greater cognitive complexity is associated with larger ex-ante forecast errors, and differences in attention to key building blocks affect the degree of over- or underreaction. Finally, we find evidence of historically anchored expectations in second moments. Overall, while institutional expectations are economically meaningful and linked to objective return predictors, they nonetheless exhibit systematic and predictable deviations from objective benchmarks.

“Global Fund Managers' Beliefs, Perceived Mispricing, and Asset Allocation” with Cameron Peng

Abstract: We investigate global fund managers' subjective beliefs about stock market mispricing using a newly compiled dataset from a 30-year survey. Fund managers perceive the market as underpriced when retail investor sentiment is pessimistic, markets have recently declined, risks to market stability are elevated, and expectations for fundamental growth are low. Perceptions of underpricing correlate weakly but positively with equity risk premium expectations from asset managers' long-term Capital Market Assumptions. Perceived mispricing significantly predicts realized excess returns in the aggregate stock market over a three-year horizon, even after controlling for established predictive factors. Perceived mispricing also strongly correlates with fund managers' future investment plans, as well as with future inflows into equity mutual funds.

“On the Dynamics of Subjective Forecasts”

Abstract: Many macroeconomic and financial survey forecasts are reported in levels rather than in growth rates. I analyze forecasts in levels to better understand belief dynamics. I find widespread cointegration relationships between the macro-financial forecasts in levels and the most recent realized historical observation of the same variable. In the long run this implies a heuristic rule which economists rely on when forming their forecasts. In the short run the corresponding conditional error-correction models (ECM) indicate that forecasters’ beliefs display both extrapolative behavior and reversal towards the long-run equilibrium heuristic. I further show how cointegration relationships can be used in predictability tests and Coibion & Gorodnichenko (2015) tests of information rigidity to avoid potential distortions induced by normalizations.

In Preparation

“Trading in Prediction Markets: Evidence from Micro-Level Data” with Cameron Cohen and Marco Sammon