Igor Kuznetsov

Do Market Prices Reflect Biased Information? Evidence from Voluntary Corporate Disclosures

with Marco Grotteria and Roberto Gómez-Cram

Using a high-frequency identification approach, we establish stock prices don't fully incorporate biased information. We study the announcements of easily quantifiable value-relevant news during voluntary corporate disclosure events where bias has been widely documented. For a large cross-section, we observe that market prices move in the same direction when the manager communicates cash-flow projections and when the same estimates are repeated by analysts the day after. Our findings are consistent with models in which investors have a heterogeneous interpretation of managerial announcements due to unknown reporting bias. We discuss alternative explanations and the implications of our results for market efficiency.

SSRN

Symbolic Regressions: Opening the Black Box of Equity Premium Predictions

Predicting the aggregate equity premium remains a central challenge in asset pricing because standard financial predictors often fail out of sample. I use deep symbolic regression to identify parsimonious nonlinear models linking excess market returns to prices, dividends, earnings and risk measures. Simulations show that the method can recover the true functional form of the data-generating process with sufficient data and appropriate regularization. In monthly U.S. data from 1927 to 2021, a parsimonious model produces lower out-of-sample forecast errors than linear and tuned random-forest benchmarks. Greater complexity improves in-sample fit but generally weakens forecasts. Predictive accuracy also varies across economic episodes: some specifications perform well during the Great Recession but poorly during the pandemic.

SSRN

Agent-Based Trading and Price Discovery

with Svetlana Bryzgalova and Vicente Garcia