Has anyone read the paper 'Pricing of High-Dimensional American Options by Neural Networks' by M. Kohler et al. (2010) and tried to program the proposed method in Python?
I have been trying that for more than 2 weeks. According to plots, the predictions I make with the neural networks in each timestep are good and the neural networks seem to converge. But at the end I still get a price range from like 4 to 4.6 for the put option with strike price 90 (it should be like something between 3.6 and 4.2). I am using Keras.
I don't have any more ideas why I do not get the same results. Maybe someone who has programmed it can give me a hint or maybe someone else has another idea.