Looking too far: macro attention and analyst forecast bias

Purpose This study examines whether and how macro-level attention in sell-side analyst reports affects earnings forecast bias in China's equity market. We focus on a setting where policy-sensitive firms and a large retail investor base give analysts strong incentives to embed macroeconomic narratives in firm-specific research. Design/methodology/approach We assemble 17,440 in-depth equity reports on Chinese listed firms from 2007 to 2023. We partition each report into 200-word segments and use GPT-4o-mini to classify macroeconomic content into six topics, which are aggregated into a report-level Macro Attention Index and a no-double-counting measure (Macro Attentionrobust). We estimate baseline regressions linking macro attention to both price-scaled absolute earnings forecast errors (Bias) at horizons t, t+1, and t+2 and the signed forecast error (Optimism Bias) in period t, controlling for rich firm and analyst characteristics and fixed effects. To examine heterogeneity, we decompose macro attention by topic and interact it with measures of the information environment, analyst coverage, analyst skill and textual detail. We further address endogeneity by exploiting the 2015 exchange rate reform in a DID design using two treatment definitions (manufacturing firms and firms with pre-2015 overseas revenue disclosure). Finally, we examine market consequences using post-report institutional and retail order imbalance. Findings Greater macro attention is associated with larger current-year forecast bias, while the relation becomes statistically insignificant at longer horizons, and it is also positively related to optimism bias. The effect is driven by monetary, regulatory, national and global macroeconomic and stock market narratives, is weaker where public data are more open, analyst coverage is higher and for star analysts, and is stronger when reports contain more named-entity detail. The results are qualitatively unchanged when using Macro Attentionrobust, and the DID estimates show a larger post-reform decline in forecast bias for treated firms under both treatment definitions. Macro-intensive reports increase retail, but not institutional, order imbalance. Originality/value The study develops a Macro Attention Index and shows that macro narratives coincide with larger forecast errors and retail-biased trading responses. The findings highlight that the content of analyst discussion, rather than report length, systematically shapes forecast quality.