Colocation is the practice of hosting privately-owned networking equipment and servers in a third-party data centre.
We discovered the principle of colocation by accident. We just knew that sometimes, we would see prices on one exchange lagging behind another. The general assumption at the time was that pricing of all assets in the financial markets was the same from one broker to another. While this is generally true for non-cryptocurrency assets, there are some slight variations in price which have more to do with proximity of the trading servers from another. The accidental find came from a certain trade on the GBP/USD in response to particularly awful US data. While the pair responded on the demo trading account of one broker, there was a lag of about 20 seconds on a live account. This lag was enough to see the extent of the response, necessitating an order on the live account that ended up being profitable when the price data caught up.
It took a lot of research to discover what had happened and what financial firms were doing to gain the advantage of getting ahead of any price lags. The price lag, otherwise known as latency, usually occurs because of the geographical distance between the execution servers on the broker’s platform and the server sending the price feed. The distance and speed of the internet network of the trader’s trading station also plays a part.
When a trader clicks the button to execute a buy or sell order, what is the process of data flow that gets this done?
From this short explanation, you can see that interventions to prevent latency will only be effective if they are done on a holistic level and not simply at steps 2 or 3, which is where a lot of the interventions by institutions are done. Institutional trading firms already have the best front end setups, so this may not be an issue. But for the retail trader, any interventions must examine all parts of the chain. It starts from the trader’s computer, internet connection and devices, and finally terminates at the point of the data exchanges that service the broker’s servers.
The institutional firms have been known to site their trading desks next door to where the exchanges housing trade execution servers are located. These data centres are available for member traders to set up their trading stations and algorithms next to the exchange's order matching engines, to get the fastest executions possible. However, the cost of gaining such the advantage of beating rivals to the best possible pricing and fastest executions comes at a huge cost. With many retail traders only able to afford a few hundred dollars to fund their trading accounts, access to such colocated centres is something they can only dream of. With such a disadvantage, retail traders have very little chance of beating the institutional players in the game.
If prices are changing fast, only the earliest get the best executions. A typical news trade can see a currency pair spike by as much as 50 pips in as little as a few milliseconds. These trades are the fast-paced trades that end up being the moneymakers for those who can get in and get out very quickly. A change of $1 in the price of gold on the XAU/USD pair, for an institutional pair trading with 1000 standard lots is $100,000. Imagine getting in and out using an algorithm that is colocated at just the right moment, and $100,000 is in the bank. That is a lot of money to leave under the table. That is why institutions spend a lot to gain colocation advantage that reduced the latency of the trade ordering process to near negligible amounts.
95% of retail traders fail at forex and other forms of financial trading. The lack of a colocation advantage plays a major part. Even without a strategy and just following price action on a colocated server, a retail trader's winning ratio can undergo a dramatic change.
A January 2013 paper titled "The Cost of Latency in High-Frequency Trading" by Ciamac Moallemi of Columbia University and Mehmet Sa˘glam of Princeton University indicates that the cost of one millisecond of latency is $100 million in a year. If firm A has a latency of 1ms and firm B has a latency of 4ms, firm A stands to make $300 million more than firm B in a year.
Retail traders are not in this category, and definitely cannot use the “spend money to make money” approach. Some firms have been known to sponsor the construction of fibreoptic cables to improve the speed of executions from their data exchange centres. A retail trader has no such resources. So any attempts to reduce latency via colocation has to follow some steps to get as close to what would be found on the institutional setups as possible.
Latency is not a one-way street. Latency is actually a round-trip phenomenon. It does not only work on the back-end alone. There is the back end component and there is also the front end component, where the trader has to make adjustments to the setup of the trading station.
There are firms that now offer proximity colocation hosting services, allowing users to enjoy remote network computer linking and cloud connectivity to colocated data exchange centres. These centres would already have fiber connectivity with various liquidity providers in New York, London, Asia and Europe. As much as possible, use facilities that already have connectivity with the pricing servers of the broker you use. Some of the brokers listed on FX-List are among them.
You must choose the right data centre for your broker on the colocation hosting service. For instance, FxPro is one of the brokers listed on FX-List, and has a latency of 0.39ms in its Los Angeles server. On the same server, another broker has a latency of 63.531ms. Which broker would you rather use?
Something as simple as virtualization can help reduce latency. Virtualization means running workloads on dedicated channels. For a trader, this means you cannot be running your trading software and data-heavy programs such as gaming arcades on your computer. If you are going to be trading, use a dedicated trading computer with nothing else running on it.
The trading computer should be one that has access to a high-speed broadband internet connection, and should also have very fast processors. Whatever EAs and algorithms will be used for trading will only be as good as the systems they operate on. It may be necessary to use multiple screens, where a screen can be used to project charts and another can be used to place orders. Unburdening the trade ordering device is the endgame at this level.
