If you're a software developer like me, then the phenomenon of a so-called bug is a well know thing. The other day, I ran some tests and discovered one in my neural network which has significantly degraded the performance in the last 6 months.
The Greed and Fear indicator is a neural network which I've build myself and have improved step by step over the years. To see how valuable a neural network is, you need to run tests. In this case, it meant replaying history and see how well it performed in calling the right direction of the S&P 500. This routine should then lead to consecutive upgrades to the neural network, as for example here and here, and more to come.
In any model development, there's always the risk of not having a correct simulation. For instance, the simulation may look perfectly ok, but when the model is applied in real-live situations, it does not meet the expectations. And that's what happened with the neural network, known as the Greed and Fear indicator.
When a simulation run was done starting Jan 1st 2018, it clearly outperformed the actual performance this year (2018) multiple times! And it did so on each and every simulation, which was very suspicious. Either the simulation had a bug or the real time usage had one. 'Luckily' it was the latter one. The simulation still looks fantastic, but in real time usage in day-to-day analysis, a bug interfered causing the serious degradation in performance. The bug wasn't so much of an issue in 2017, but now with 2018, patterns have clearly changed. These new market circumstances allowed the bug to cause more interference.
Now that the bug is fixed, let's hope trading result will pick up soon.