ETD

• anchoring
• biased learning
• conditional momentum
• cross-sectional
• momentum
• reversal
• S&P 500 stocks
• time-series
• u.s. equities

This thesis offers an empirical investigation of momentum and reversal in U.S. equity markets, focusing on S&P 500 stocks over the period 1999-2019 and adopting the conceptual framework proposed by Antico et al. (2025). Rather than treating “momentum” as a single, self-contained anomaly, the thesis emphasizes that return predictability is definition-dependent and shaped by how information is aggregated. Within a learning-based perspective with misspecification and anchoring, momentum and reversal are approached as equilibrium outcomes that may vary across horizons, empirical definitions, and market environments.
The empirical analysis is conducted at weekly frequency, which helps limit microstructure noise while retaining sufficient resolution to study short- and medium-horizon dynamics relevant for belief updating. Returns are constructed as log differences of adjusted prices. The empirical design mirrors the three definitions highlighted by Antico et al. and implements them in a unified way within the same dataset and sample period.
First, cross-sectional momentum and reversal are examined through winner–loser portfolio strategies based on past cumulative returns over a fixed formation window, with decile sorting and alternative holding horizons. Second, conditional momentum and reversal are studied by conditioning future returns on sequences of past weekly outcomes, using the sign and length of return streaks as a reduced-form proxy for informative histories. Third, time-series momentum and reversal are analyzed through within-asset predictive regressions across multiple horizons, capturing persistence and mean-reversion patterns in individual return dynamics.
A key contribution is a unifying interpretation of why these three notions can coexist without contradiction: cross-sectional evidence is inherently relative (rank-based), time-series evidence is within-asset and absolute, and conditional evidence depends on the structure of recent histories rather than on cumulative performance alone. The thesis also explores heterogeneity in cross-sectional predictability across market regimes and firm characteristics (e.g., size and liquidity proxies), treating these exercises as a way to measure variation in the empirical manifestations of the mechanism rather than as strict invariance checks.
Relative to Antico et al. (2025), the thesis provides a focused empirical implementation of their conceptual framework on a major U.S. equity universe, offering a coherent comparison of the three definitions within a single design and clarifying how definition choice, horizon, and conditioning information jointly shape empirical conclusions.