Spontaneous Feedback Effects
When individuals make decisions sequentially, and each individual has imperfect or fuzzy information, it is possible that decisions made by individuals earlier in the sequence will impact those making decisions later, since the latter group will ignore imperfect private information, or revise their private information, and instead follow the early decision makers. This leads to information cascades. The important assumptions here are (1) sequential decision-making where subsequent individuals observe the action but not the private information and (2) private information is imperfect or fuzzy and subject to some misinterpretation.5 The rapid proliferation of novel financial instruments, such as collateralized debt obligations (CDOs) and credit default swaps (CDSs), during the financial crisis of 2008 is an example of this phenomenon.
Deliberate Feedback Due to Ranking
Firms can deliberately generate feedback effects by providing critical information. For example, The Washington Post provides an updated list of most emailed articles, inducing other readers to emulate and thereby increasing the likelihood of a future reader going to these articles. Further, if you were told how many times each article was emailed, this copying effect is likely to be exacerbated since the popularity differential between articles becomes public information. Knowing that article #1 on the list was emailed twice as many times as #2 will send even more traffic to article #1. If, on the other hand, the numerical difference between #1 and #2 is less than 1 %, traffic to both might be equal. Examples of this kind of deliberate feedback prevail in many media markets such as Netflix, with stars to denote rank. In financial markets, this could lead to panic and systemic risk if the Post article suggested some financial instability.
Are rankings themselves subject to some sort of information cascade? Users may start out with their own private rankings based on fuzzy, private information, but then upon observation of others’ behavior, will update their own rankings such that ranking cascades are likely to form. For example, if all prior ranks of a local bank are 1 (on a scale of 1-5, with 5 denoting safety) then I am likely to ignore my own private information of 5 (the bank is safe) thinking that perhaps I was mistaken and withdraw my funds.
There are two significant assumptions being made in this analysis. First, the underlying assumption in a copying model is that all users are identical and trust-worthy. If reliability becomes a factor, then the copying model breaks down since each user’s decision has to be weighed by their trust index. Second, we assume that an insignificant amount of time has lapsed between previous decisions and current ones. If all past users made their decision a long time ago, the product itself might have changed quality so current users may ignore past decisions and act as if they were the initial decision makers. Of course, this may start a fresh cascade in current time.