Irrational Economist_ Making Decisions in a Dangerous World - Erwann Michel-Kerjan [52]
The individual may also need to define how much new information the event provides. It is often convenient to think of such situations in terms of binomial trials. Although there are some cases in which an event occurs and knowledge of the number of trials in the data can be taken as given, such as an adverse reaction to a drug in a defined group of patients, often it is unclear how many “trials” have occurred. It is difficult to determine how many people have been exposed to a particular chemical, or how many potential terrorist attacks have been prevented, especially if one is not privy to proprietary or classified information. For all these reasons, then, Bayesian updating is unlikely to be used. How, then, do individuals incorporate new information from the occurrence of an event?
UPDATING AS A FUNCTION OF PREVIOUS EXPERIENCE AND PREVIOUS CONTEMPLATION
We conjecture that individuals will excessively update their assessment of virgin risks after one occurs, and fail to (or barely) update their assessment of experienced risks, even when significant updating is warranted. Some experienced risks are well-understood systems with a large amount of frequency data, justifying little updating. For others, however, we have far fewer observations than we think and/or the risk is changing over time. Then, we suspect, individuals update their risk assessments less than they should.
For example, Carolyn Kousky recently examined the changes in property sales prices in St. Louis County, Missouri, before and after the devastating 1993 flood involving the Missouri and Mississippi rivers, both in and out of the floodplain, to determine how homeowners’ assessment of the risk changed after the occurrence of an extreme event.4 During this flood, 100-year floodplains were inundated, as were many 500-year floodplains, particularly those behind failed levees. In our classification, 100-year floodplains represent experienced risks: Floods have previously occurred in these areas and the flood hazard is widely recognized, with national regulations requiring disclosure of the flood risk to homeowners and those with a mortgage required to purchase a flood insurance policy. By contrast, 500-year floodplains represent virgin risks: They had not been recently flooded, and the risk in these areas was also out-of-mind, with no information-disclosure or insurance requirements. After the 1993 flood, property values in the 500-year floodplains declined 2 to 5 percent on average (whether or not they had been flooded in 1993). All property values in municipalities located on one of the flooded rivers declined 6 to 10 percent compared with property in the interior of the county. (These declines were statistically significant.) This finding suggests dramatic updating of the risk. However, there was almost no change in property values in the 100-year floodplains, many of which were flooded, suggesting little to no updating of this experienced risk.
And here is the paradox. Homeowners in the 100-year floodplains should have updated their risk assessments, for three reasons. First, there is insufficient experience to know the probability of a flood with any precision. While there are data on flood events in St. Louis going back to the Lewis and Clark expedition in the early 1800s, this time series is not nearly long enough to provide a tight estimate on the likelihood of an event thought to occur perhaps once a century. Second, the risk is evolving over time. Flood risk has been shown to be increasing over time in the St. Louis area due to structural changes to the Missouri and Mississippi rivers, increased development in the watershed, and possibly climate change. Third,