In most academic and practical settings, Betas are calculated using at the very least monthly returns over an interval thats pretty flexible. This is because there are a lot more noise within weekly and daily returns than actual information content. The standard is five years or 60-month returns. It is good practice to ensure that the number of data points is atleast over 30 (heuristic) to ensure that the parameter converges well. Furthermore, many adjustment methods are available in form of bayesian shrinkage that improve the out of sample estimates (See Bloom/Vasicek). Depending on your strategy, id say the interval varies, if you are trading everyday and if your algo depends on beta, then the beta won't change over atleast quite a few trades, then you might need to consider different intervals at the cost of potential estimation risk.