Three essays on empirical asset pricing

Abstract

This thesis is a collection of three essays regarding three research projects that I undertook in the area of empirical asset pricing, with a special focus on economic network and textual analysis.

In the first essay titled ``Link-Firm Characteristics Are Covariances'', I study how information embedded in economically linked firm (link-firm) shapes risk compensation. Specifically, I adopt the Instrumented Principal Component Analysis (IPCA) approach and use link-firm characteristics to proxy for dynamic risk loadings. Empirical results are three-fold. First, IPCA based on link-firm characteristics (Link-firm-IPCA) delivers high in- and out-of-sample time-series R^2. Second, Link-firm-IPCA yields a high annualized out-of-sample Sharpe Ratio of 1.9, and combining link-firm and self-firm information can further improve mean-variance efficiency. Third, Link-firm-IPCA factors explain seven lead-lag momentums documented in the economic linkage literature by over 50% in economic magnitude, suggesting a novel risk-based interpretation of lead-lag momentums.

In the second essay titled ``Cross-sectional Inflation and Stock Returns'', we develop a “Weighted Production Price Index” (WPPI) using micro-level data behind headline inflation. WPPI captures the composite effect of multiple industry-level inflationary changes on each focal industry. We show: (a) WPPI explains cross-sectional variations in same-quarter earnings and same-month stock returns, (b) inflationary shocks captured by WPPI propagate across industries along supply chains and production complementary networks, and (c) WPPI has strong predictive power for industry and firm returns, yielding industry-level alphas of 1.4% per month. Further analyses of earnings forecast errors and information frictions point to sluggish price adjustment as the likely driver of these return predictability patterns.

In the third essay titled ``Conditional Asset Pricing with Text-Managed Portfolios'', we construct managed portfolios that exploit information extracted from firms' earnings call transcripts and examine their asset pricing implications. Returns on these text-managed portfolios correlate with investor sentiment and predict macroeconomic outcomes. The exposure of individual stocks to the text-managed portfolios explain as much return variation as those to the characteristics-sorted portfolios. Adding earnings call information to firm characteristics increases mean-variance efficiency, although it does not improve stock-level return predictability. Consistent with the insights from Kozak and Nagel (2024) on mean-variance spanning, our results suggest that earnings calls provide information about return covariances beyond what is captured by firm characteristics alone.