We develop a model of optimal monetary policy in an economy where firms' price-setting decisions are distorted by a signal-extraction problem: they cannot perfectly distinguish aggregate from idiosyncratic shocks based on noisy local information. This systematic misattribution of aggregate nominal disturbances to firm-specific factors generates an additional indirect price response, implying that strict price stability is no longer optimal. Instead, the optimal policy fully stabilizes the output gap, which necessarily requires accommodating fluctuations in the price level. The cyclicality of the optimal price level depends critically on the source of firms' incomplete information: learning from local demand signals implies a procyclical price level, whereas learning from local productivity signals yields a countercyclical one. We show that these results are robust to extensions featuring elastic attention and sentiment shocks.

We construct a firm-level text-based measure of macroeconomic perceptions from the annual reports of Chinese listed firms. The measure exhibits substantial cross-firm dispersion, with this heterogeneity systematically related to firm- and industry-level local information. Macroeconomic perceptions are more optimistic among firms with stronger own-business conditions and those in faster-growing industries, whereas regional conditions show little systematic association with these perceptions. We also find that firms whose macroeconomic perceptions are less aligned with the aggregate state exhibit greater investment inefficiency in the following year.

We revisit the canonical tax-smoothing model when fiscal authorities have limited capacity to process information about permanent government spending. Using annual central-government data for 20 OECD economies, we document two empirical patterns: tax rates are highly persistent and predominantly nonstationary, while lagged changes in government expenditure help predict subsequent tax-rate changes. We introduce rational inattention by allowing the government to process information about permanent government spending subject to an information-processing constraint. The tax rate remains a martingale with respect to the government's endogenous information set, but limited attention generates delayed tax adjustment from the econometrician's perspective. Limited attention attenuates the initial tax response to government spending innovations, shifts more financing toward debt, and spreads tax adjustment over time. Lower information-processing capacity, through more persistent filtering errors and endogenous signal noise, also raises the relative volatility of tax-rate changes. Cross-country evidence provides partial support for these implications. Rational inattention improves the relative-volatility prediction in 13 of 20 countries and, in a VAR for government spending and the budget deficit, brings the model-implied coefficients closer to their empirical counterparts in up to 15 countries.