Automated Trader presented the findings of their annual survey last week at the London Stock Exchange. We attended an extremely interested presentation, panel discussion and Q&A session and wanted to summarise the findings before the official results are published later this year.
The overall consensus was that the continued automation of the trading process is going to continue, the pace of change will increase along with competition and technical complexities. Full end to end automation is still continuing but a mass uptake is going to take longer than expected and people operating a fully automated system will remain in the minority for the foreseeable future. Respondents were predominately from the buy (36.58%) and sell side (20.62%) with the remaining 41.44% being made up of Infrastructure, Vendors etc.
This is a brief summary to give an indication of major trends, as soon as the full results are available online we will post the link.
The key challenges people are currently facing include - finding alpha in increasing competitive markets, managing real-time risk and coping with an ever increasing volume of data. Responses to these challenges include creating smarter not faster strategies, increased use of machine learning algorithms and a quicker than expected take up of FPGA’s (more popular for sell side) or GPU’s (more likely to be used by buy side as they are slightly slower but more flexible).
Latency is still important but people are moving away from latency arbitrage and concentrating on creating alpha that is profitable as long as you ‘one of the fastest’, 35% of respondents fell into this category. Saying that, people are obviously still taking latency seriously and colocation is expected to continue to increase – both to markets and data suppliers.
Decreasing latency equals increased volumes and increasing tick data so everyone is having to deal with more data. This means that people will have to change their systems to deal with these challenges but will also have more data to learn from - swing and roundabouts.
The complexity of systematic trading algorithms is also growing with the mean number of parameters for an algo being 70 with some having up to 1000. With a growing number of parameters it is increasingly hard to recalibrate algorithms ‘manually’ so genetic algorithms and adaptive learning are increasingly common place. Execution algo’s are less complicated with an average of 20 parameters, though there can be up to 25 algo’s that are used in buy side firms.
One of the key areas of discussion was the potential effect of proposed regulation. There are plenty regulations and taxes that have been propositions over the last few years and the general consensus was that increased regulation was not workable and up to 62% of respondents said they would relocate if it got too much. 79% were against Dodd Frank, 80% were against the Financial Transaction tax an d60% said that Trading speed limits would be unworkable.