I have conducted several new tests where I used K-fold cross-validation (blocked) instead of time series cross-validation. Additionally, I switched from Z-score to ranking. This consistently yields better results for me. Is there a natural explanation for this, or could it just be coincidence?