Can I utilise time series properties of the data, WITHOUT creating lags?

External question — Data Science Stack Exchange Author: insipidintegrator Original post: https://datascience.stackexchange.com/questions/129669 License: CC BY-SA 4.0 — https://creativecommons.org/licenses/by-sa/4.0/ Adaptation: HTML converted to plain text; contact email addresses removed. I am working on a project where the train and test sets are given to me. The data (stock returns) is time series by nature, but the point is that I cannot create lags because that would mean discarding the top k observations because they would not be able to have the lags. Why is that a problem for me? Because I will get judged by the MSE on the testset, which is calculated by a predefined script which has the true values (which I do not have access to). So that means I cannot drop any values from the test dataset. I would have gladly implemented a LSTM or AR(I)MA on the data, but since that would mean that for prediction I would have to create lags in the testset too, I am stuck with Random Forest and XGBoost. Is there any way I could utilise time series properties of the data, WITHOUT CREATING LAGS, or does there exist any Deep Learning methodologies that will allow me to do that (I do not know of any)? Thanks, I appreciate your kind input.