How firms’ information choices amplify micro shocks
In this paper, Zhen Huo, Minghao Li, Jieran Wu, and Huan Xiong combine a novel dataset, a theoretical model, and a quantitative exercise.
Data. The authors use SEC server logs to track which listed firms actively download the regulatory filings of other listed firms. This produces a directed attention network for over 7,000 US public firms, capturing, for each firm, how much attention it receives from others and how much it sends out.
Empirics. They then run a horse-race: does the output growth of the most watched firms predict GDP better than the output growth of the largest firms? They test this against professional GDP forecasts and a broad set of realized macro outcomes.
Theory and quantification. To explain the mechanism, they build a general equilibrium model where firms make production decisions based only on the subset of others they observe. A heavily-watched firm’s shock gets simultaneously picked up by many other decision-makers, coordinating their responses and amplifying into aggregate fluctuations. The model is calibrated to the observed attention network to isolate how much of aggregate volatility the attention channel generates, independently of firm size.
Findings
- Attention received by firms is highly concentrated as a small number attract a disproportionate share of total monitoring, and this concentration is largely independent of firm size.
- Attention predicts macro outcomes, size does not. Attention-weighted sales growth of the top 100 most-watched firms explains 55% of the variation in one-quarter-ahead Survey of Professional Forecasters GDP forecasts and 45% of realized GDP variation, while the conventional sales-weighted measure of the 100 largest firms explains only 10% of forecasts and 26% of realized GDP. This gap holds across GDP, industrial production, consumption, investment, and unemployment.
- Attention heterogeneity alone generates granularity. In a model without firm-size heterogeneity, the information network still generates sizable aggregate fluctuations: the top 100 most-watched firms account for 16% of aggregate volatility, versus just 2% under symmetric information. When size heterogeneity is added, the attention channel still accounts for about one-third of total granular effects.
- Sorting matters. When size and attention are positively correlated (large firms are also the most watched), the two channels reinforce each other maximally.
Micro Shocks and Macro Fluctuations in the Information Network
Authors: Zhen Huo, Minghao Li, Jieran Wu, Huan Xiong
From: Yale University, Peking University, Zhejiang University, Harbin Institute of Technology
How firms’ sentiment shapes macro fluctuations
Similarly to the Huo et al. above, Rustam Jamilov, Alexandre Kohlhas, Oleksandr Talavera, and Mao Zhang, in this paper, combine a novel dataset, a theoretical model, and a quantitative exercise.
Data and sentiment measurement. The authors build firm-level sentiment indices for all S&P 500 firms from 2006 to 2021 using three complementary sources: earnings call transcripts analyzed with a standard finance dictionary of positive and negative keywords; a more sophisticated natural language processing approach using FinBERT, a language model pre-trained on financial text; and financial analysts’ one-quarter-ahead earnings forecasts from IBES. These three measures serve as cross-checks on each other.
Empirical strategy. For each firm, they estimate how closely its sentiment tracks aggregate US output, producing thus a firm-level “informativeness” ranking. They then test what firm characteristics predict belonging to the most informative group, running regressions against a wide array of candidates including firm size, leverage, market beta, and sectoral position (upstreamness/downstreamness in the input-output network). Finally, using local projections, they trace the dynamic macroeconomic consequences of shocks to the sentiment of the most informative firms.
Theory and quantification. The authors build a multi-sector general equilibrium model with endogenous attention choice, calibrated to 27 US industries from BEA production tables. Firms choose how much attention to allocate to different sectors, and the model explains why attention gravitates toward downstream firms.
Findings
- Fifty firms drive the aggregate. The average sentiment of roughly 50 firms, around 8% of the sample, accounts for over 60% of variation in US output and economy-wide sentiment. Beyond the 50th firm, adding more firms introduces noise that outweighs the informational gain.
- Downstreamness, not size, is the defining characteristic. Among all firm characteristics tested, sectoral downstreamness i.e. proximity to the final consumer, is the only variable that systematically and robustly predicts which firms are macro-informative. Firm size, leverage, cyclicality, and network centrality are all largely irrelevant.
- Downstream firms are natural information agglomerators. The model shows two reasons why attention gravitates toward downstream firms: they are more central in the production network, so their productivity matters more for everyone’s optimal decisions; and because they combine inputs from multiple upstream providers, their output acts as a natural summary statistic for dispersed information in the economy.
- Quantitative impact. Sentiment shocks to the 20% most downstream sectors account for around 70% of sentiment-driven output fluctuations, and approximately 20% of all aggregate output fluctuations.
Conclusion
Both papers find that a small number of firms drive aggregate dynamics through informational rather than size-based channels, and both find that firm size is a weak predictor of macro importance. The key difference is in what makes a firm informationally central: Huo et al. point to received attention in a bilateral information graph (who watches whom), while Jamilov et al. point to downstream position in the production network (who aggregates information naturally).
Granular Sentiments
Authors: Rustam Jamilov, Alexandre Kohlhas, Oleksandr Talavera, Mao Zhang
From: University of Oxford, University of Birmingham, University of St Andrews
