Look-Ahead Bias

The “Hindsight is 20-20” bias. When we’re trying to gain some insight into the past performance of an investment, we sometimes run a simulation where we might change some initial factors, run the simulation, and then see what happened. We need to be careful to not use information in running the simulation that we didn’t know or that wasn’t available back when the data was actually taken. If we do use this kind of future-sight info in our simulation, we’re introducing look-ahead bias, and we’re probably altering our simulation to gives us the results we expect (or hope for) rather than real results.

For example, let’s say we’re trying to simulate how the iceberg might have actually damaged the Titanic under the waterline to give us more insight into the actual mechanics of the sinking. If we program a simulation to act with the current knowledge that we have that ignoring ice warnings and then hitting huge icebergs has a tendency to cause ocean liners to sink, the simulated Titanic will just turn south and avoid the ice. We don’t get any data about how the iceberg hits the ship because our look-ahead bias altered the results of the simulation.