4 dimension for forecasting portfolio with LSTM

I am leveraging the LSTM for forecasting portfolio. The input is of shape (batch_size, sequences_length, number_assets, features), in which features refers to Open price, Closing price, Volume and etc. The expected output should be of shape (batch_size, 1, number_assets, closing_price), in which closing price is the forecasting target. However, LSTM only supports inputs of 3-dimension. How to resolve this issue that troubled me for so long time? Thanks!