https://www.youtube.com/parttimelarry
It covers popular services and libraries like Alpaca, Robinhood Private API, TD Ameritrade option prices, Yahoo Finance, TradingView alerts, backtesting, and a lot more.
I'm asking because I've used semi-pro and pro infra (paid services) and it was hard to get consistent returns. It was easy to get returns, but rarely beyond, say, S&P 500 returns (at which point, I might as well just invest in the S&P 500 index.)
What you really want to know is "how problematic is the noise in this data"?
One way to answer it is to create your own FF sorts [1] and regress your Yahoo-derived factor returns against Ken French's (which come from CRSP/Compustat). If the intercepts are small/statistically insignificant and the R2's reasonable, then you're all set.
The big problem you will hit is survivorship bias in Yahoo!. My own research suggests that the quality is perfectly acceptable, provided you back-fill an unbiased universe (e.g. Russell 3K) from another source.
I'm actually surprised you didn't outperform the market with the survivorship bias. Back test must have covered a very small time period.
[1] https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data...
You need a set of companies that lacks this bias, and the set of Russell 3000 index constituents is a good set to use.
Some of the older Russel 3000 companies will be missing from Yahoo, because they are delisted or bankrupt. You need to find a secondary source for them, or "back-fill" your securities master.