What does it take to start a data feed company?

Discussion in 'Data Sets and Feeds' started by m4a1, Aug 28, 2006.

  1. m4a1


    Lets say someone wants to do things right and provide a reliable data feed. What does it take to start a company like this? What does it cost?
  2. bjg


    A lot of cash for the following:

    * Access to exchange data

    * Hire a few DB and software devs to develop a platform/mechanism for data storage and retrieval

    * Ensure data can always be accessed in case of a disaster, i.e. a fire or something. So you're obviously going to need many servers dispersed across the country if not the world

    * Lots of redundant bandwidth. You'd likely use data centres to hold servers, so they'd manage all this

    * People to provide support

    * People to do ongoing server maintenance and backups etc.

    Probably more...

    Doing it 'right' isn't going to be cheap. ;)
  3. It depends on a lot of things:

    - How many exchanges?
    - How reliable do you want the feed to be?
    - What does your typical customer use your feed for (satellite delivery of all symbols, or watch a few hundred?)
    - Will you build your own ticker plant or buy someone elses?
    - If the former, how many engineers will you have to deal with options switching to penny-multiples, for example?
    - How much history will you maintain?
    - How much tick history?
    - How much will you pay to backfill data?
    - Will you keep external data? (Commitment of Traders, Implied Volatility of options)
    - How many people will you have to enter new futures and options contracts? (Expirations for options are entirely different per futures contract. Things like, "the last London banking business day on a Thursday followed by at least 3 week days.")

    Feeds alone may cost $50k/month (I'm extrapolating from the known prices of CME, CBOT, and CBOEs feeds. Throw in NYBOT, NYMEX, LIFFE, Forex, etc., and it's getting pretty steep) Add tier 1 or 2 hosting, redundantly distributed across the country (or world, depending on your customer base), and add another $10k/mo+.

    There are then exchange fees which you are required to pass onto your customers. DTN pointed out recently that a feed company has a different set of rules than a broker for distributing data. For example, IB can hand out CBOT electronic contract data in realtime for free. DTN is required to charge the "full" CBOT fee of $55/mo.

    Add to all of that the fact most customers want a complete (and free) technical analysis package as part of their data feed.

    If you were putting together a business plan, you'd have to do something extremely different from everyone else, and people would have to be willing to pay a premium for it. This couldn't be a "low cost" competitive service.

    I'm no expert, but I did look into the costs for getting a full CBOE feed at one point. So, take my comments as a semi-informed party. :)
  4. m4a1


    suppose you go small first and the only data you provide is ES. please give some numbers instead of just saying it's expensive.

    and lets say you limit the number of customers at first until you can handle more. so no need to spend so much money on overhead at first.

    and lets say there's no charting package. reliable data feed only for as many customers as it can handle. there are plenty of charting packages (e.g., ensign) that people can use to connect to the datafeed themselves.
  5. I gave you numbers. I also gave you at least a dozen things you have to consider. If you're going to keep back tick data, you'll have to buy it from somewhere, and you'll have to store it in a pretty massive database. Figure a tick has 20-40 bytes, multiply that times 2-3 million per day. Look at Dell's web site to get a price.

    A quick google search turns up CME's feed prices:

    For hosting, run through rackspace.com's bandwidth utilization tool to figure out your monthly costs.

    Lastly, how are you going to get Ensign to be compatible with your feed? Are you going to provide them an API and hope they port? Are you going to develop the glue logic yourself? Are you going to knock off someone elses' API?

    Figure a base price of $50k/mo just for hosting, connectivity, and bandwidth charges. Then add engineers, accounting, computer equipment, etc.
  6. m4a1


    thanks. i was not criticizing your post. okay, to get the feed from the exchange costs $50K. Edit: the cme website says $9000 per year.

    no tick history will be provided. let people get their backtesting data from historical data providers. maybe provide 1 week's worth of data at most just so people can connect their historical data with the current data.

    hosting should be real cheap. i am limiting the number of customers until i can handle more. say i buy a $3000 server and $250/month hosting plan. how many customers can this handle if only ES feed is provided?

    forget engineers, accountants, etc. it will not need all these extra people at first. support is online forum, but there shouldn't be many problems because customers are limited and everything will be done right before it's released.

  7. yahoo finance quotes are free

    scottrade has RTQ for free as well but some charge $10/month
  8. In the last hour and 15 minutes, I've downloaded 4MB from DTN for primarily the ES feed. And their feed is highly compressed. In an uncompressed form (that my application actually reads), I've read 40MB so far today. Assume you have a nicely compressed API as well. 7 hours of active trades per day, 20 days per month comes to 560MB/user/month. Don't forget latency and realtime bandwidth used. At FOMC time, will you fall behind? Traders hate that. :)

    As for "forgetting engineers"... How are you going to get Ensign to be compatible? Develop your API? Publish it so users can access it? Support it when there are problems?

    Take a look at opentick.com and do some ET searches on them. Despite being free (except for exchange fees), people are extremely reluctant to use them. They've been down frequently, are often slow, etc. To "do it right" requires reliability. That requires people.

    In the end, you're proposing a feed with no history (i.e. Ensign can't draw a chart beyond when the user logs in), an as-yet-defined API with no engineers to support it, non-redundant servers, supporting only one contract, and you'll bill everyone yourself.

    If I were you, I'd happily go with someone like DTN. They have the bandwidth, redundancy, support, and are awfully cheap for what you're proposing. :)

    Still, if you are determined to do what you propose, take a look at DTN's NxCore product. They'll feed you as few or as many contracts as you like over the Internet, no direct fees to connect to CME, etc, and you can redistribute.