Trading Setup - Do it all yourself or have it all?
DISCLAIMER: The views expressed in this article are my own and not of the individual companies I work with. Nor are they necessarily reflective of any one trading firm I have worked with in the past.
Working in the low latency trading technology space for the last 10 years has brought with it insight in to how trading firms operate from both a technology and business context. There seems to be two camps of thought in the space when it comes to building everything yourself or utilising vendor solutions to facilitate low latency strategies.
The business side is all about about the profitability, time to market and liquidity exposure they can achieve. They want to achieve the objectives of their traders and stakeholders in a consistent and quality driven way. The technology side is looking to deliver on this by ensuring the solutions are optimal, effortless for the users and well maintained.
Before assessing which is best for your setup, it is important to understand the motivation for each of the approaches.
When it comes to doing everything yourself, the main (sensible) aims are to realise the 'perceived' performance benefits of having your own dedicated and custom setup, to keep the proprietary nature of the strategy so secret that you are not willing to have the remotest chance of anyone discovering anything and lastly you have been burned by poor service and performance by vendors in the past and want to be the master of your own destiny.
Vendors and service providers (off the shelf solutions) are usually utilised when a trading firm wants to reduce their cost, development work involved to get trading, maintaining the setup with minimal resource and want to be able to scale quickly.
So which is better for you? Well, like most situations, it depends what you need. A couple questions need to be asked:
If you have answered 'yes' to most of these, then having a custom or dedicated setup might well be of benefit. When you utilise a vendor solution or service, they are catered for the masses; not the few. Their solutions are designed to cover as many use cases as possible in a cost effective and efficient way. When your strategy demands the most optimal and custom setup, there is no better way to realise this than doing it on your own terms. The good vendors and providers will tell you this themselves. But remember, doing this has an impact on your time to market, cost base and resource pool.
In contrast, if you have answered 'no' to two or more of the above questions then using a vendor or service will be of significant benefit. Let's look at each of the questions.
Have you stripped out all inefficiencies in your algo/system and hardware optimisation is all that is left?
There are a handful of trading firms in the world that have done this effectively. They have ripped the algo apart and optimised it to the level where not much more is possible. They are then looking at custom FPGAs or ASICs or other super specialised hardware to get down to double digit nanoseconds tick to trade. They want to serialise the data as fast as possible and they don't want to have their traffic queued behind any traffic. There can be around 100 nanosecond impact if you are using a service provider for colo and exchange access and 100s of nanoseconds to single digit microsecond for an off the shelf trading stack. If you are not optimising to this level, using an off the shelf will not make an material impact to your strategy.
领英推荐
Does a 100 nanosecond saving (or less) make a difference to your Alpha?
This refers to the queuing of your trade execution traffic in a provider's network or a normalised interface to a vendor's software or FPGA stack. Remember, it has been designed for the masses and ease of use. For market data, this is usually multicast. Meaning, the network replicates the market data in hardware switches so there is no queuing and if you need this delivered with minimal delay, firms such as Transaction Network Services (TNS) offer Layer 1 market data in as little as 5 nanoseconds for all of the major markets. Even for order execution they offer latencies of 45 nanoseconds, however there will be some queuing but providers manage this carefully.
When we look at the feed handler and execution stack which consumes the market data or generates the orders, there are solutions out there which are ultra fast, but are designed in a way to reduce the burden of exchange driven changes and developing to them. Firms such as Celoxica enable this with their QDS feed handler and GXA execution stack which delivers a tick to trade for your strategy in single digit microseconds.
The benefit of using such solutions, is to reduce time to market, simpler maintenance and development cost reduction.
Are costs not as important, as the Profit and Loss of the strategy more than cover the operating costs of a dedicated and custom setup?
In building trading setups, the sky's the limit (literally!). As well as the more commercially available traditional Co-location builds, there is a whole world of technology that addresses the connectivity between liquidity centres such as Microwave, Short-Wave and Low earth orbiting satellites. But that is for another day :-).
Cost, the variable in a setup everyone wants to reduce but with no performance hit! Firms such as TNS and Celoxica have spent considerable amounts of money and time in order to offer their customers a wide breadth of products and capabilities. The model is simple, make it more cost effective and efficient for trading firms to trade than them doing it themselves. The ethos of TNS and similar firms is mutualisation of services, exchange access, data and hosting space. Some exchanges charge up to $30,000 per month for just the low latency connectivity to their market, you can slash that to a fraction of the cost by using a providers connection with a minimal latency impact.
A fantastic benefit of using a provider for all trading firms and counterparts is where you need to consume a large set market data or execute from a centralised location such as LD4 in Slough, NY4 in New Jersey or HKEX in Hong Kong. Low latency circuits between the various co-located markets are very costly and are always evolving. Service providers pride themselves on being competitive in this space and can provide the market data and execution path over their existing low latency network. And believe me, it will be a fraction of the cost than doing it yourself.
Do you need complete secrecy of your setup?
This is somewhat contested in the trading world. Afterall, trading firms are trying to make money from edge in their trading strategy and they do not want anyone to have a look at their crown jewels. The reality is, vendors and providers have no interest in stealing ideas, intellectual property or strategies from their customers; they want to build long term relationships with the trading firms so they take more services from them. Think about it, if they divulged details of your setup, they are endangering their revenue stream and opening themselves up to legal action. If it is a real concern, you can mitigate by selecting vendors which are hands off, and hand you products for you to place your algo on after the initial setup with no access to the systems or visibility of traffic.
In summary, whether you build yourself or use off the shelf products and services really does depend on what is your appetite for nanosecond performance, cost and privacy. Unless you are in that handful of elite trading firms where money is no object, you can always benefit from using off the shelf. Look at what you need to achieve and blend dedicated with off the shelf where it makes sense, but make it work for you.
If you would like help or advice on any of the above or even discuss a few ideas, please reach out to me at [email protected]
Principal Specialist - Real Time Data Feeds and Tick History at FactSet
1 年Nice one!
Chief Executive Officer | GTN Middle East | DFSA Regulated officer (SEO) | Executive Director Board of GTN ME | Chairman of Management Committee
1 年Good read Himesh