With k equal to 5 and 1,000 observations, each fold holds 200 test points. You fit five models and average the five test scores, which uses the data more efficiently than a single split and gives a spread as well as a mean.
For time series, use the forward-chaining variant: fold 1 trains on the first 200 and tests on the next 200, fold 2 trains on the first 400 and tests on the next 200, and so on. Otherwise you are training on the future.
Higher k means more training data per fold and more compute, with test sets that overlap more between folds. k equal to 5 or 10 is conventional; the choice rarely matters as much as fixing the leakage.
Related: cross-validation, purged-cross-validation, train-validation-test-split, expanding-window