I have around 200 unique time series, each with its own set of exogenous variables. However, I'm worried that this dataset of 200 time series might be too small for the models to learn effectively. Is it a reasonable idea to combine these exogenous variables with an external dataset, like M4, to increase the training data and enhance model performance? If this technique exists, could you recommend any literature that I can read to better understand its advantages and disadvantages?