The Index Investor Blog

Most people are already overwhelmed with information and besieged by constant demands for their scarce attention. That is why we believe in publishing more in-depth, long-form content only once per month. In our experience, publishing too frequently with too little analysis and synthesis only adds to noice, and provides little or no signal.

There are, however, exceptions to this general rule. Broadly speaking, forecasters encounter four different types of information. "Indicators" are pieces of high value evidence that have been anticipated in one or more previous estimates (forecasts), and have thus been either actively monitored or sought. They are deemed to be "high value" because they are much more likely to be observed (or not observed, as in Sherlock Holmes dog that didn't bark) when hypothesis is true than when competing ones are.

In contrast, "surprises" are pieces of information that initially trigger an emotional response because they are outside the range of outcomes automatically anticipated by your existing (if usually subconscious) mental models and beliefs about the world. In Shannon terms, surprises are information with high entropy that significantly increase your uncertainty. The problem with surprises, however, is that our mind rebels against them, and will subconsciously work hard and fast to fit them into our existing views. For that reason, the initial feeling of surprise tends to rapidly dissipate, and its source can easily be forgotten if not quickly written down.

In the forecasting context, indicators and surprises provide us with information about either our mental model(s)/hypotheses about how a system works, or the values of those models' parameters.

This yields a rough 2x2 matrix the can be used to grade information value. The most important information is a surprise that causes us to question the accuracy or completeness of our existing mental model or models of the world. The next most important information is a surprise that causes us to increase the range of values that one or more model parameters can take.

The third most important type of information is an indicator that changes the probability that one of our mental models is the correct one. In relative terms, the least important information is an indicator that narrows the range of possible values for one or more model parameters.

Unless they are very urgent, we seek to incorporate indicators in our monthly publication. In the case of surprises, however, we may (depending on how important we judge them to be), publish an immediate update on this blog, and send a Twitter alert that we have done so, under our handle @indexllc. We hope this doesn't happen very often — but if it does, it will be because we think the information in question is surprising and important, and worthy of your immediate attention.