In density forecasts, respondents are asked to assign probabilities to a response scale with pre-specified ranges of inflation. In two large-scale experiments, one conducted in the US and one in Germany, we show how the specifics of the response scale determine the results: Shifting, compressing, or expanding the scale leads to shifted, compressed, and expanded forecasts. Quantifying this effect using the scale most widely used among central banks, we find that mean forecast, uncertainty, and disagreement can vary by several percentage points. Density forecasts may severely misestimate the "true" inflation expectations, and may generate spurious changes in reported inflation uncertainty. The analysis suggests several ways to reduce this bias.
We study the binary choice problem in a data-rich environment with asymmetric loss functions. The econometrics literature covers nonparametric binary choice problems but does not offer computationally attractive solutions in data-rich environments. The machine learning literature has many algorithms but is focused mostly on loss functions that are independent of covariates. We show that theoretically valid decisions on binary outcomes with general loss functions can be achieved via a very simple loss-based reweighting of logistic regression or state-of-the-art machine learning techniques. We apply our analysis to algorithmic fairness in pretrial detentions.
Seasonal migration is a common strategy to mitigate rural seasonal deprivation, but migrants need to remit money during the lean season to family members facing food shortages. We observe counterintuitively low remittances in rural Nepal during periods of seasonal hunger, and migrants return with remittances later during harvest when food is relatively abundant. To indirectly overcome this apparent constraint in remittance timing, we provide a $90 consumption loan to randomly selected rural households during the pre-harvest lean season. Loan-recipient households increase pre-harvest investments in fertilizer and time spent working on their own farm, smooth consumption, and save more of their migration income to bring it back home. Food security, subjective well-being, rice harvest, and revenues improve. Ninety-eight percent of beneficiaries repay the loan with the increased harvest-period remittance. In a two-period model of household decision making, we show that remittance frictions-a market failure-are necessary to qualitatively match our experimental results.
This paper proposes a new mixed vector autoregression (MVAR) model to examine the relationship between aggregate time series and functional variables in a multivariate setting. The model facilitates a reexamination of the oil-stock price nexus by estimating the effects of demand and supply shocks from the global market for crude oil on the entire distribution of U.S. stock returns since the late 1980s. We show that the MVAR effectively extracts information from the returns distribution that is more relevant for understanding the oil-stock price nexus beyond simply looking at the first few moments. Using novel functional impulse response functions (FIRFs), we find that oil market demand shocks tend to increase returns, while both demand and supply shocks reduce volatility, and have an asymmetric effect on the returns distribution as a whole. In a value-at-risk (VaR) analysis, we also find that the oil market contains important information that reduces expected loss, and that the response of VaR to the oil market demand and supply shocks has changed over time.
Traditional dynamic discrete choice model-Conditional Choice Probability estimator (DDCM-CCP) literature considered identification using the likelihood of longitudinal choice only. However, such an approach has a limitation in identifying flexible dynamics for latent state due to the discrete and typically small-dimensional nature of choice. This paper extends this literature to utilize imperfect measurements, called proxies, for a latent discrete state. I first show that proxies improve identification and discuss how survey design on proxies affects identification conditions. I then extend the estimator from Arcidiacono and Miller (2011) to pool information from an unbalanced panel of noisy proxies, enabling estimation of more flexible latent state dynamics than Markov chains. As an application, I estimate a dynamic model of labor supply and mental health, where dynamic mental health only imperfectly observed-plays a central role. The results reveal more complex dynamics than a standard Markov chain.
