I get this question all the time, and it is a very, very hard question to answer. So, I decided to create a list to keep it at hand for the next time somebody has the same question. Then, I decided to publish this list on social media with the goal to have other people adding more value to this list.
So, let’s get into it.
First, some quick introduction, I worked with many hedge funds and proprietary trading shops as HFT startups, helping them to setup their initial infrastructure. From hiring and managing their human-resources to making sure they have all the hardware and tools. So, I have a pretty good idea of the initial costs that such HFT trading shops have.
Now, of course, we are going to have some degree of variance depending on the markets this trading shop: it is not the same as a hedge fund managing $50 Billion, where the market impact will be very different from those funds that manage $10 Million. So, let’s say that we want to be focused on having a minimal HFT operation where will allow us to access the markets they need (based on the strategies they want to run).
As of last, we want to keep the trading capital out of this list, since it doesn’t count as a “cost” (or yes?). Moreover, depending on the markets, brokers we may use some of the margins they offer, so that’s going to be a matter of how much “risk” are we able to hold based on the strategies we want to execute.
Finally, here is the list:
Data: High-frequency strategies are data-intensive, so you need to get the best data providers at the tick level (level 3). That’s expensive. Depending on the market you are in (forex, futures, bonds, etc) the cost could vary. FX is even more complex, because of its highly fragmented nature, so they will need to have a broad view of all different sources. The cost for each provider could start from $5k per month each, up to $50k per month.
If you are running a market-making strategy on FX you will want to make sure you can have “at least” 3 or 4 of the main FX platforms (EBS, CBOE FX, FXAll, Fastmatch) and this could total $70k per month.
Servers: You will need power. A decent dedicated server (please don’t use the clouds), could cost you 20k at least. It needs to have CPU 32-cores at least. You can rent a dedicated server, and its cost could start from $2k per month (this could be a very good starting point).
Some have argued with me that is not necessary for so many CPU cores, and before entering into the same discussion, I would recommend reading my other articles on how to “architect HFT trading/low latency systems” in my blog. If you want to reach the best latency, you will need all those cores. And maybe, extra servers as well… but that could be left once your operation is up and running and you are profitable already.
Collocation: That powerful server you just got, must be placed inside a collocated environment. The idea is to reduce the latency between exchanges or ECNs as much as possible. Being close to the exchanges/venues is the only choice, and it will make the difference between a profitable business and a failure. These data centers will charge you for your server space and for the connectivity you use. The connection between your server and the exchanges or venues are called: cross-connections, and you will need to “cross-connect” to every exchange or venue you are working with. This varies considerably depending on the markets you are in, but the collocation & cross-connections could total $8,000.00 per month
Software: this would be the most expensive piece of your setup. Remember, that the software is the brain of your operation. Not only needs to get ALL the data from the exchanges/venues but normalize it, store it, manipulate it, and prepare it to be consumed by your strategies(s). These strategies will be doing different calculations, will lookup for previous data, and then will generate buy/sell signals. And all that must be done in a fraction of milliseconds (hopefully within 5–30 microseconds)
On top of that, you must be sure, that you will have all the different modules in place: price aggregators (to make sure you have all price sources aggregated in one place), order management systems (OMS), execution management systems (EMS), smart order routing (SOR), liquidity manager (LM), risk management systems (RMS). and any interface you may need (to databases, storage, monitoring systems, reporting, etc)
Cost-wise will depend on what you choose: If you go with an off-the-shelf solution (not recommended, it is cheaper, hard to customize, you don’t own anything, slow), or you start your own development (time to market +1 year, very costly). If you choose the 1st option, the cost varies from provider to provider between $10k to $20K monthly. If you go for the second option, the cost could go way over $1M, but the advantages and the edge you may have over the competition is incomparable.
There is a third option, which is using a hybrid between the two: you can hire experts to develop your entire system, along with your current software development team (if you already have it). Something like this could cost you half of what will cost you if you build it all by yourself.
Full disclosure: our company offers this kind of hybrid service, where we can join into companies’ team, bringing all the expertise and help them to build this kind of system.
People: you will need human resources. This is not a one-guy operation. You will need to have software engineers, quantitative analysts, technical support for your servers, and researchers. Think about 150k /year at the lower end.
Brokers/Prime Brokers: you will need to open up a brokerage account to have access to the trading venues and start to execute your HFT system. They will require you to have a minimum capital to trade (besides the commissions/fees they may charge). So, that adds up to your initial setup cost.
It’s a very lucrative business but is hard to get started. Usually, startups try to start small and grow as they see profits, but that always falls into failure. If you do that, you will fail to have all the above points I’ve listed.
Your initial investment is high, and keeping in mind that after having all these startup costs, all your infrastructure in place, and the software ready to run, your first profitable trades could start to come in after 6 to 12 months of operations.
I hope my question is not as vague as the others…
Please, let me know if I was missing something else, so we can add it to this list?
As last, we launched a personalized service for quantitative trading startups trying to setup their operations. Please see more here.