Why is quantitative research so important in algorithmic trading and becoming more significant, especially in FX?
People use algos to reduce costs and control risk. Algos can do the job effectively because they process realtime information and make automated trading decisions in an efficient way, in some cases better than human beings. This process determines that quantitative research, as opposed to qualitative or discretionary decisions, is the most dominant factor. Algos work best in efficient and transparent markets like equity and FX because most of the information required in the decision making process is available in a quantitative format in real time.
What range of skillsets are usually required to be an FX Quant?
To me, the most important skill is having a quantitative mind set to solve problems, using the right approach. A good quant needs to be able to consume data and find the story hidden behind the numbers. In other words, I feel the research ability is more important than the applications of certain techniques, e.g. a Machine Learning library.
Another very important skill is communication. One of quants’ main value-adds is to provide state-of-art research. The key difference between doing research in the business and academic environments lies in the fact that the former requires turning a piece of work into commercial value. It, in turn, requires the acceptance and collaborations from multiple teams, senior management and clients. Communication is clearly a key part of it.
Why are many buyside firms who utilise bank algos looking for “quantitative excellence” from their providers?
Investors are looking for quantitative excellence because the performance of algos depends on how good the models are. There is a real need for banks to do the legwork to help normalise the market and remove some of the uncertainty around executing across multiple venues. Understanding implicit and explicit costs and risks based on a client decision in their execution medium is an important factor. Having empirical facts that explain the mechanics of the market are key in helping to develop that conversation and further strengthen trust with clients.
In what ways does the contribution of Quants in the development of FX algorithms help to reduce execution costs?
It can improve the performance of algos, i.e. making less market impact with the same amount of risk given certain constraints, e.g. a TWAP algo. It can also help human traders to make optimised decisions based on real market conditions, e.g. choosing the optimal duration of a TWAP. Sometimes people undervalue this second aspect.
Actually a big part of trader’s job is risk management, which heavily depends on the realtime market conditions. Choosing the right duration of a TWAP is probably more important than choosing the best TWAP on the street because duration contributes most to the risk. If you look at the sports world, while you get individuals who are extremely talented, it’s those players that work as a team, and pull expertise and resources from a number of areas, who perform the best. The macro decisions from traders when executing algos, e.g. choice of timing and durations, are very important for cost and risk savings.
What are the greatest challenges facing Quants when it comes to trying to understand the microstructure of the FX market and the associated risks of market impact?
One of the greatest challenges is modelling the market, which is a very complicated and dynamic machine. There is a lot of detective work in being a quant, you have a few pieces of information and the question is how to stitch this information together to figure out the whole picture - essentially what’s going on in the market. You need to lose the noise, distil the information and then connect it using logic. This is similar to a voice trader’s expertise. They are trained to quickly pick up and connect market information to form a story. The quants’ job is to automate this process so that the machine can process the information in a more efficient way.
One of the comments we heard from a client is that “there are a lot of liquidity metrics on the market, but most of them give counter-intuitive indications in reality.” It shows there is still a gap in a lot of quantitative tools to catch intuitions. I think the actual difficulty does not lie in how we model or predict, for example, market volume, but how to model and tie in all the different bits and pieces in a consistent manner into a machine. Eventually your machine should be able to tell a market story which traders feel intuitively.
Trading analytics is an important part of algorithmic FX trading. Why are Quants key to delivering these toolsets?
Trading analytics provide two value-adds to users: convenience and insights. By convenience, it’s typically an IT solution which helps traders to improve their workflow.
For example, a few realtime TCA solutions on the market fall into this category, bringing realtime updates of an algo and providing information and convenience to traders.
However, personally I feel the more challenging part is the “insight”, where the analytics can provide data and information hidden behind the scenes. It requires in-depth analysis of data, i.e. quantitative work. Creative thinking is the key here because real value is delivered through new angles and insights
What techniques are Quants using to extract new insights from FX data and where are they looking for more comprehensive data sets?
To win in the market, you either need to have more information or a better way to utilise information. In reality, most real world applications use low dimensional data. Typically an FX algo uses FX market data and makes decisions based on the important features selected by the model.
The availability of more datasets and the ability to process larger amount of data in real time make it possible for algos to understand larger datasets across different markets. This is again a voice trader’s expertise. They follow multiple markets at the same time and their brain is a high dimensional processor.
Essentially this is the information advantage of human vs machine. We want machines to learn from human expertise and process information in a more efficient way. The challenge is how to digest multiple correlated data sources in a coherent framework. This leads to the wider applications of non-parametric approaches.
Essentially it replaces handcrafted models, which will be very complicated and potentially biased, by the functions that the data chooses. We found it very valuable in practice. There has also been a lot of focus on Machine learning techniques across the industry.
We have not seen too much ground breaking work in this field yet. Given the very low signal to noise ratio in the financial data, it won’t be an easy task. However I do see the value of the framework and systematic approach to increase the efficiency and ability to tackle a larger dataset.
What role are Quants playing in helping to engineer a new generation of FX algorithms that can respond to microstructure signals more dynamically?
Personally I think a quant should be the architect, who designs the mechanism of the algorithm. They also need to be the engineer, who implements the components of the machine.
Lastly they need to be the promoter, explaining the intuitions behind the scene. Algos are often associated with “black box”, which is not a helpful description because people are less likely to use something they are not fully comfortable with.
I think it is part of quant’s job to explain the intuitions embedded in the algo and showcase its excellence backed with statistics. Ultimately, algos can only produce value when people are comfortable to use them.