Within the scope of the Einstein Telescope Technologies programme, we are currently working on a nice new innovation. To apply the concept of Explainable A.I. to the world of Financial Trading. If it works it will also be useful for the world of Gravitational Wave Detection and the Einstein Telescope.
With Explainable A.I. techniques we are trying to pilot technologies that can help explain the decisions an A.I. model is making. In the picture below I have included a visual model of how this could help in Financial Trading. In the most left picture, you can see a 10-bar Japanese Candlestick image which is automatically drawn based on the price and volume data from the last 10 price/volume bars on a specific asset on the financial market. Our A.I. model which has been trained on large volumes of this data, has assigned a “Buy” flag to this pattern. This means the A.I. is advising us to open a BUY position for this particular asset.
But why does it make this decision? The 3 candlestick images on the right are the closest resembling “similar” patterns that the A.I. has seen before (based on all the historic data it has been trained on). So our A.I. is indicating whether it has seen a similar pattern in history.
Now we can do a couple of cool innovative things:
– We can look at the predicted labels from these previous patterns (did the A.I. at those historic periods also generate a BUY decision?). Can we move to some sort of “majority voting” system from these previous patterns and thereby generate a higher accuracy or confidence level on the current prediction?
– Since the 3 patterns happened in history, we can also determine how accurate these signals were back in history, what the price impact of these signals was back then (and how it helps us to determine for the current signal how big the price movement might be) and it might also help us to predict already when we should close the position (before we even open the position).
Still heavily in the research phase but it is already starting to look very impressive.