Hi,
I am wondering, is it possible for another users to use machine learning model to learn my algo model (through historical trade record as input) then duplicate the model?
what is the potential ML model that he will use? and to what extend the model can be duplicated? can he copy 100%?
Thanks in advance.
Hello Kheng Lim,
Reverse engineering an ML model is a very challenging task especially if it's a financial ML trading model.
Let's say you want to reconstruct the trading model X using the historical trade sheet.
To accurately reconstruct the model we need to know the exact input features of model X. Since most of the financial features are custom features, it's highly impossible to accurately reverse engineer the features and the model X.
The next challenge would be to predict the hyperparameters. Since model X must have been using custom features with optimised hyperparameters (on undisclosed training data), it's not possible for someone to reconstruct model X without knowing the exact custom features and hyperparameters.
Even with a clue of the input features, it takes high computation power, a huge amount of historical trade data and time to reconstruct the model X even with less accuracy.
However, there are instances in history where some of the licensed image classification neural networks and spam filtering algorithms were reverse-engineered by spammers. You can read this article to know more about it.
You should note that this reverse engineering was possible since the spammers has access to these algorithms and were able to input a huge amount of data and extract the output.
This is different from the case you have presented since nobody can access your algorithm but you.
Hope this helps!
Thank you