Herding. “Individuals’ choices are often influenced by other people's behavior, especially under conditions of uncertainty” (Xxxxxxxxxx, 2021, p. 322). In the context of climate insurance, this implies that agents’ decision to purchase insurance coverage depends on the degree to which other agents in their social network have or have not purchased it. In a situation of low baseline diffusion of climate insurance, such a herding mechanism can reinforce the protection gap. While a couple of papers fail to detect a significant effect of social networks and social norms on risk mitigation decisions and flood insurance demand (Xxxxxxx, 2012; Xxxxxxx et al., 2014), several studies do find support for the presence of such an effect. In one of the first assessments of the relevance of social connections and social norms for insurance purchase decisions, Xxxxxxxxxx et al. (1978) find that homeowners in flood- and earthquake-prone areas prioritised discussions with friends and neighbours over considerations of the likelihood and consequences of a future disaster for their choice to purchase insurance coverage against those hazards. In a series of studies, Xxxx X. Lo investigates the impact of social norms on the demand for household flood insurance. In particular, it is found that individuals are more likely to insure themselves against flooding if they expect that people similar to them (e.g., neighbours) or from intimate social circles (e.g., relatives and friends) will do the same action or will approve their choice (Lo, 2013a). Also, the beliefs that family or friends “want me to buy insurance”, or that “other people would buy insurance”, have a positive effect on both the take-up of flood insurance as well as belief that insuring is important (Lo, 2013b). In addition, Xx & Xxxx (2017), in a survey on the intention to adopt flood prevention measures, including insurance, among British households, argue that community engagement and trusted social networks can enhance the motivation to prepare against flooding and the adverse impacts of climate change. More recently, Xxx et al (2020a), in an experimental analysis of investment in self-insurance and insurance purchase to cope with flood risk, find that social norms have a positive effect on both the WTP for and the level of coverage of flood insurance. The authors also detect a positive impact on the decision to invest in self-insurance, which is mainly driven by cautious individuals perceiving this measure as more effecti...
Herding. Herding in the USDA Baselines - Chandio and Katchova Objective ▶ To examine whether the baseline projections are grouped together for certain crops across different countries (i.e. herding behavior), producing similar projection trends, and whether that contributes to bias. Data ▶ Yield, area harvested, ending stocks, imports, exports, and total consumption for three major commodities: corn, soybeans, and wheat from USDA International Baseline Projections since 2002. Methods ▶ To assess accuracy, we use root mean squared percentage error: RMSPE = , 1 Σ(100(Yˆ t − Y )/Y )2! (13) rcvh T rcvth rcvth rcvth ▶ Yrcvth is the actual value realized for the projection Yˆrcvth ▶ We use the dynamic time warping (DTW) algorithm to compute the distances between all available country pairs for each crop-variable-year-horizon and determine whether herding occurs. Figure 12: RMSPE for corn yield projections of various countries Corn yield projections errors for the US remain substantially low for all horizons Projections for other countries show higher bias, which increases for longer horizons Figure 13: Dynamic Time Warping distance between corn yield projections of various countries from US, China, and Brazil When comparing projections for other countries to the US, all confidence intervals overlap 0. That is, projections for all countries show herding behavior when compared to the US. For the countries where realized values are not herded, this increases bias. When mapping a relationship between projections correlation with US and bias, we find a positive association for multiple crop-variable combinations. Deep Learning Baselines using Deep Learning - Bora and Katchova What is the issue? ▶ Previous studies show that many variables in the baselines are biased. ▶ As prediction error increases with horizon, and the predictive content diminishes, and the projections stop being informative. ▶ Current baseline models do not utilize information efficiently. ▶ Baseline projections process is time-consuming. What did the study find? ▶ This study compares the performance of various deep learning methods against USDA baseline and a na¨ıve benchmark. ▶ Findings suggest that while current baselines perform well in shorter horizons, the deep learning methods perform well in longer horizons. How was the study conducted? ▶ Deep neural networks were trained using past history of commodity indicators. ▶ Performance of the deep neural networks were compared with USDA baseline and na¨ıv...
