At present, a prominent item on many trading groups' wish lists is 'faster time to market': in a highly competitive market environment, profitable inefficiencies are transitory, so rapidly deploying automated trading strategies to capture these is a primary point of competitive advantage. Therefore anything that expedites the transition from initial model concept to its deployment in live trading is of value in enhancing productivity and making the best possible use of the organisation's intellectual capital.
A common obstacle to this rapid deployment is the tendency of some quantitative research groups to use a development environment or framework that is only suitable for development. For some, this may be because they find a particular development environment convenient for prototyping trading models. For others, it may be no more than force of habit; they have always used a particular environment, so continue to do so.
Whatever the motivation, this approach has the significant downside of having to port model code written in the development environment to the trading environment before any live trading can begin. Apart from the inevitable delays this causes (and thereby the risk that others will exploit and exhaust the intended market inefficiency first) it also carries the very significant risk that errors will be introduced during the code translation processes.
However, even when quantitative traders are working in an environment where development and production code are one and the same, this is still only part of the picture. The final piece of the jigsaw is the actual connectivity to the markets in which the automated model will trade. This has become a far more important issue over the past few years for a number of reasons.
The most fundamental point here is whether the application includes integrated market connectivity, which is often not the case. In these situations, the development team has to become involved with the plumbing, by writing to assorted market APIs. However, this means they are no longer entirely focused upon delivering alpha, so their intellectual capital is effectively being misused and competitive productivity reduced.
If the application does include integrated market connectivity, additional considerations are its quality and reliability. Can trading models be immediately deployed via a connection of industrial grade resilience? If not, the situation is not much better than if no connectivity is included, as the quantitative trading group has to expend time and resources fire fighting rather than focusing on their core expertise.
Figure 1: Optimising parameters via genetic algorithm
The corollary to the quality of the connectivity that the development/trading environment can access is its scale. As the amount of analytical firepower that traders have at their disposal continues to increase, it has become increasingly necessary to cast the net wider in search of profitable inefficiencies. One of the most obvious manifestations of this is the multi-market application of trading models. This makes it imperative that any model can be immediately deployed across all necessary markets via high quality connectivity. It is apparent that the diversification and return benefits of deploying multiple instances of a strategy across a range of markets would be eroded if connectivity is only available to a subset of these markets.
These considerations were very much to the fore in the development of RTS Realtime Systems' new RTD Tango Quant, which provides a single environment that can be used to develop, optimise and analyse automated trading strategies. But it also includes low latency connectivity to more than 135 trading destinations via 65+ hosted destinations in data centres and seven data centres with global reach. As a result, model development and deployment really can become a straight through process.
It is one of the ironies of strategy development that in many cases the process of optimising a trading strategy's parameters is in itself severely sub-optimal. As the competition for alpha capture has intensified so that variants of the same model are increasingly traded across multiple markets/assets, the scale of the task of optimising their free parameters has also grown. As a result, solution search spaces often become vast. Unfortunately, many development platforms have not adapted to this change and their optimisation tools can only offer exhaustive ('brute force') optimisation methods. Furthermore, these development platforms often also constrain their optimisation routines to running on a single workstation. While some strategy developers struggle along regardless by diligently subdividing the search space and trying to schedule optimisation runs for overnight when they are not using their workstation, this is still an appallingly inefficient working process. Collectively, the end result is still a major bottleneck in the model development and deployment process, which almost guarantees that a model is obsolete by the time it is deployed and that extremely inefficient working processes are enforced on the development team.
RTD Tango Quant addresses these issues in two ways: through the use of genetic algorithms (see Figure 1) and the offloading of optimisation tasks to separate machines that can even be on the far side of the planet. Genetic algorithms allow the much larger search spaces of today to be traversed far more efficiently and quickly than brute force optimisation. They use evolutionary techniques to quickly identify promising versus less promising areas of the total search space. 'Good genes' hitting 'good areas' reproduce and pass their characteristics on to further generations, while 'bad genes' are selectively culled. While this in itself hugely assists productivity, it is further enhanced by a high degree of user control, whereby users can define 'good' or 'bad' in the context of the optimisation. Furthermore, they can do this by deploying the unique ability of combining multiple optimisation targets using so-called transfer functions, which translate units of specific statistical outputs into a uniform scale. The resulting fitness values are then fed automatically back to the optimisation engine.
Even with the advantages of genetic optimisation, a productive development group can still find the number/scale of optimisation tasks too large for efficient deployment on their own workstations. For this reason RTD Tango Quant provides a Distributed Optimisation architecture, which allows multiple users to queue multiple optimisation tasks to a cluster of separate machines, thereby accelerating the optimisation process. Furthermore, optimisation clusters within the Distributed Optimisation architecture do not have to be local, so workstations and desktops in other time zones can also be used overnight when they are not otherwise required.
The end result is that each user can continue working on other tasks such as strategy modelling, definition of optimisation tasks and results analysis while optimisations are being simultaneously performed elsewhere. As a result, users are able to work productively throughout the day, rather than having optimisation routines effectively dictating their schedule.
Figure 2: Context-based visualisation of equity curves versus the underlying instrument
Once optimisations are complete, RTD Tango Quant also offers comprehensive visualisation and analytics. Browsing features such as context-based visualisation (see Figure 2), distribution graphs and 3D visualisation enable users to comprehend and manipulate optimisation results. The robustness of models across a wide range of parameters when applied to both seen and unseen data is quickly apparent, as is the performance of predetermined parameter sets on other instruments.
Code reuse and collaborative productivity
A complete automated trading strategy need not necessarily be the brainchild of a single developer. In fact many model development teams work on a collaborative basis with certain members specialising in the development of specific components, such as entry/exit signals or filters, position sizing, money management or risk management rules. While code reuse (such as through object orientation) is obviously common practice, RTD Tango Quant takes this to a further level of componentisation.
Users of RTD Tango Quant can combine multiple interacting strategy elements such as Signals, Filters and PositionSizers (see Figure 3). These individual elements can be dragged and dropped together to create strategies and can also be circulated throughout the organisation for others to use in building their own strategies. Elements can be distributed with or without preset and locked parameter values or with a defined possible range of parameter values.
For additional security of intellectual property, developers can opt to deploy their strategies (which can be written in any programming language compatible with the .NET framework) in the form of a dynamic link library (DLL) file, so the source code is invisible to unauthorised individuals. The drag and drop capabilities of these strategy components obviously deliver considerable flexibility around model development, as well as boosting code reuse and productivity.
RTD Tango Quant can be described as an 'end to end' solution, but in fact it is perhaps more accurately characterised as 'end to end and back again' technology. This is because in common with any dedicated trading platform it has the ability to receive details of completed order fills and incorporates risk, order and execution management systems. However, because it also incorporates quantitative analytics, this order fill data can be used to adjust a model's risk and money management parameters (such as the size traded).
Figure 3: Componentised architecture
Conclusion: alpha productivity
From the practitioner's perspective, RTD Tango Quant delivers 'full circuit' automated strategy trading: the design, optimisation, analysis, execution and online real time tuning of fully automated trading strategies. From the business manager's perspective - whether operating within a hedge fund, corporation, proprietary trading group or bank prop desk - it delivers exceptional productivity gains.
By combining a sophisticated quantitative trading system platform with outstanding execution infrastructure, the organisation's intellectual capital will always be deployed to the best possible effect in delivering alpha. RTD Tango Quant - one platform, endless possibilities.