$20k/year for EOD stock prices??? $50k/year for a friggin 15 minute delayed feed???!!! Guys, if you have too much money you don't know what to do with, I can provide you with an account number where you can send it. Seriously, these prices are almost 50x too high. Quandl EOD is 50 bucks a month. Polygon.io gives you the realtime SIP for $199/mo. Quandl is not bad quality. If they are missing 1 in 10000 of their splits or dividends, this is not going to make a bit of difference in your quant trading. If your methods aren't robust to data errors you are looking for trouble no matter how much you pay for data. I would not be surprised if your $20K solution has more data errors than the cheap one.
There's a lot more to just price and dividend data in a basic market data $20+k/year feed more traditionally used by trading companies. It is always surprising that super standard daily frequency equities data costs that much, but when I take into account total costs I can see why even a small company would pick that. To your low cost solutions, you could add IEX for recent history since it is free.
There's a lot more, such as..? I am running a small trading firm, please enlighten me what am I missing out on with my cheap data
Equity metada, non-price daily data points, data for non-US markets, pooled pricing with other datasets (daily price data is rarely the only thing a trading company needs). Algoseek is not meant to be a complete solution for daily equity market data. I would add generally support and quality from a larger company - I know for example the error rate for quandl for dividend data was quite high as of a few years ago and development time needed to investigate or fix issues is a cost to add when comparing costs.
Hi Laplacian, This is exactly what Norgate Data offers in its US Stocks Platinum subscription (back to 1990) - it is survivorship-bias free (so it includes delisted stocks). We also offer historical index constituents so you might be able to use as a universe for your "top stocks". For example, Russell 1000 back to 1990 has over 3300 stocks that are currently or were in the index at some stage. Using supported environments you can determine a stock's membership on a time-series basis. We also have a Diamond-level package back to 1950. Here's a stock in that universe selected at random: FMER-201608 FirstMerit Corp Common. It was in the Russell 1000 from mid-1992 to mid-1993 then again from mid-1997 to mid-2006. It delisted in 2016.
algoseek has eod historical data adjusted for corporate actions such as splits, dividends, M&As, etc.