Abstract: Using text from 200 million pages of 13,000 US local newspapers and machine learning methods, we construct a 170-year-long measure of economic sentiment at the country and state levels, that expands existing measures in both the time series (by more than a century) and the cross-section. Our measure predicts GDP (both nationally and locally), consumption, and employment growth, even after controlling for commonly-used predictors, as well as monetary policy decisions. Our measure is distinct from the information in expert forecasts and leads its consensus value. Interestingly, news coverage has become increasingly negative across all states in the past half-century.
Abstract: This paper examines the idea of risk aversion in artificial intelligence (AI), using large language models (LLMs) such as ChatGPT. We explain how AI can exhibit risk-averse or risk-loving behavior and what this means for decision-making processes. We use a new method to measure if an AI has a risk preference based on its general knowledge through imitation. We give AIs psychological tests to collect data on their traits and risk tolerance. We discover that advanced AIs are generally risk-loving, and they differ a lot from the general population in how they handle risk. Additionally, we find that AIs tend to be optimistic, impatient, trust financial institutions, and have financial literacy. We demonstrate that AIs' behavioral traits such as optimism and trust may influence their risk tolerance.
Abstract: Economic models develop conceptual frameworks for fundamental decisions but rarely prescribe a specific estimation approach. Using novel data on the inputs and assumptions in professional stock valuations, we study how financial analysts address estimation ambiguity when calculating a firm’s cost of capital. Analysts use the same return-generating model (CAPM) but diverge in their estimation choices for key inputs, such as equity betas. Such estimation choices are driven by idiosyncratic analyst-specific criteria, persist throughout their career and across brokerages, and generate large cross-analyst variation in discount rates for the same stock. The dispersion in discount rates is associated with higher market measures of investor disagreement, such as trading volume. Overall, we provide micro evidence on how financial experts resolve estimation uncertainty.
Discussant: Steffen Meyer, University of Southern Denmark, Danish Finance Institute
Abstract: We explore strategies employed by traders in the Iowa Electronic Markets’ 2020 Presidential Election Winner-Takes-All Market. We replicate previous research on trader mistakes while documenting behavior consistent with two new biases: a disposition effect and an endowment effect. We explore how markets populated by mistake-prone and biased traders can result in efficient pricing. Efficiency arises from interactions between many biased and mistake prone traders, the market structure, and a smaller number of significantly more rational price-determining traders. The dynamics are not explained fully by current theories on efficient markets, market microstructure, or behavioral finance.
Discussant: Will Cassidy, Washington University in St. Louis
