Thank you! Of course it is possible to change the cross validation approach from n-splits to one-step ahead forecasts. However that means that your model will undergo train/test operations N-1 times, where N is the total number of observations. Depending on the complexity of your model and the number N, this might be very computationally expensive and really slow to compute. So n-splits, while being a somewhat crude estimate, provides you with a quality assessment much faster.