Thanks for reading, you’re absolutely right, most well known and industry accepted models take into consideration moneyness in the pricing of options, however the distinction I’ll make here is that the prices are for two different instruments, although one derives it’s value from the other there is still a basis. My comment is more in reference to autoregressive models like ARIMA and sequential neural networks like RNN/LSTM/GRU that use the price of the same asset at lagged time steps. These models effectively capture some form of momentum, which is a factor that has already been well covered in literature. There are reasons this is good and bad, which might require a post in itself to cover. But this comment should serve as a guide(from experience in markets and machine learning) rather than a hard and fast rule.
Hope that helps,