I have a theory I plan to test on some earnings announcements and analyst reports, but in advance of that, I wanted to get it out on my instablog. The theory is that all estimates should be discounted by 30%. This number is my first guess of how to incorporate all the cognitive biases that are known to exist (overconfidence, confirmation bias, hindsight bias, and many others) into a rough estimate.
Interestingly, in my own work in biotechnology research which involve clinical trials, the discount should be much, much more than 30%. If a collaborator says a trial should take one year to complete, you can bet that it will take two. In fact, I budget it that way. If they say they can accrue 200 subjects (patients), you better start looking for a publication that will accept as few as 50. The benefit to the organization is that we control our enthusiasm and also avoid the smaller, riskier projects that probably would never have even obtained publishable results.
The theory is a way of using the research that other people perform, while applying a discounting mechanism to account for the high likelihood that even if they are directionally correct, they have a high likelihood of being numerically incorrect.
This theory also tangentially gets into value investing theory, in that a margin of safety is sought, for similar reasons: even the best analysis is still flawed in some way.
That's all for now, but I hope to write more on this once I find some good examples based on stock analyst reports and the ongoing earnings season. Who knows - maybe the discount should be 20 or 50%. Thirty percent was just my first guess at what it should be.
Disclosure: I am long INTC.