Franziska Bender

We study how subjective expectations about personal income are shaped by economic news provided by the media. We address two questions: first, does publicly available information in the form of news topics affect households' income expectations? Second, do households use this information efficiently? To answer this, we combine household survey data from the Michigan Survey of Consumers with a corpus of roughly 3 million articles from The Wall Street Journal and The New York Times. We use a Latent Dirichlet Allocation (LDA) model to construct monthly time series for the prominence and tone of 70 economic news topics. Our analysis yields two key findings. First, we establish that news content has significant predictive power for households' future income expectations. Second, and our main contribution, we find compelling evidence that households use this information inefficiently. While household forecasts suffer from large general biases, a model that optimally processes news topics can reduce the remaining forecast error for realized income substantially. This belief distortion stems from households over-relying on their private information or judgment and mis-weighting information from news topics. Our results help explain the origins of systematic errors in personal income expectations, identifying the inefficient processing of public information from the media as a significant driver. This suggests that models of expectation formation may need to incorporate deviations from rational processing to fully capture household behaviour.

Does the inherent uncertainty of innovation make economies vulnerable to persistent stagnation? I study this question in an expanding-variety endogenous growth model with two features absent from standard frameworks: aggregate uncertainty about the returns to R&D, and learning from market outcomes. Firms cannot observe the conditions determining their future profits and instead infer them from the noisy outcomes of other innovators. Because firms do not account for the value of the information their R&D reveals to others, this learning creates an informational externality that can produce growth traps - periods where innovation and growth remain persistently low even after fundamentals have recovered. The mechanism is belief persistence: pessimistic firms invest little, which chokes off the signals that would reveal improved conditions, so beliefs recover only sluggishly. Strikingly, these traps arise even when uncertainty, in itself, encourages R&D, highlighting the crucial role of belief dynamics in shaping long-run growth.

The theoretical and empirical literature provides mixed evidence on the effect of uncertainty on R&D investment. We hypothesize that this ambiguity arises because the effect depends on the type of uncertainty a firm faces. This paper, therefore, investigates how different types of risk shocks affect R&D and capital investment. To do so, we introduce a novel methodology that exploits a specific feature of SEC disclosure regulation: the requirement that firms report “material changes” to their risk factors in quarterly 10-Q filings. Using textual analysis, we leverage this rule to create a new dataset of firm-specific, time-varying shocks to distinct risk categories. Our analysis yields two main findings. First, the effect of risk shocks on R&D is highly heterogeneous: shocks related to regulatory risk decrease R&D, while shocks related to financial and technological risk increase it. Second, the set of risks that matter for R&D is almost entirely different from the set that matters for capital investment. Our results suggest that to better understand the effect of uncertainty on firm decisions, it is important to move beyond a monolithic view and consider the heterogeneous impacts of its different sources.