Failed ideas

I want the failed tests to stay visible.

This is where I keep ideas that looked promising but failed because of a null result, a bad design, a failed holdout, weak economic magnitude, or a data problem. If only the successful experiments survive, the rest of the site becomes much harder to trust.

Failed idea #001
Look-ahead bias

Short the rebound after extreme relative weakness

This initially looked like a strong shorting pattern. I later realized I had selected the rebound peak using future information inside the next 20 sessions. I rejected the result and rebuilt the experiment around fixed landmarks.

RejectedRedesigned

What can put an idea here

A failed idea does not need to be a bad idea.

Look-ahead bias, overfitting, regime instability, a failed locked holdout, economically trivial magnitude, data-quality problems, and genuine null results all qualify. Sometimes the most useful outcome is finding out exactly why an appealing story does not survive.

Read the research log →See how I separate discovery and validation →