The Gateway to Algorithmic and Automated Trading

High Frequency Trading: One Size Does Not Fit All

Published in Automated Trader Magazine Issue 22 Q3 2011

High frequency trading has already evolved to the point where it actually covers multiple sub-disciplines. For some of these, latency is still the be all and end all, while for others it is just one of many factors that will determine a trading strategy's success. This evolution means that the sellside challenge has also evolved. As Andrew Morgan, European Head of autobahn Equity at Deutsche Bank explains, the sheer diversity of high frequency traders means that a far more flexible approach is now required.

The growth in high frequency trading has gone hand in hand with an expansion in the way in which the general concept is applied. Accommodating this change now requires a level of sellside service, involvement and consultation as high as in any area of equities or electronic trading. However, for this to be effective from the client perspective, the critical sellside discipline is flexibility - both in terms of tailoring solutions to client needs, as well as responding to structural market change. Delivering this flexibility necessitates combining the most appropriate elements from four key areas - access, franchise, pricing and speed - to arrive at a solution that precisely matches the specific needs of each client.


As cross-asset and cross-market high frequency trading continue to expand, only being able to deliver market access to a subset of the venues in a region is no longer adequate. For market access to be meaningful, it now has to be comprehensive.

But there's access and then there's access; the ability to connect into the venue is a starting point, but no more. The delivery of real value and timely product often requires the provider to have a local physical presence. An office staffed by personnel engaging in regular dialogue with local regulators, exchanges and local technology providers is much better positioned to deliver real market intelligence and valuable insight into the probable future geography of the local trading landscape.

The corollary to this is how good the provider is at patching gaps in that landscape. For long/short strategies, access to a market that does not permit shorting is pointless - unless of course the access provider can offer a legitimate mechanism for providing synthetic short exposure.

Andrew Morgan

Andrew Morgan


High frequency trading is also not about providing a single isolated trading component; the client value achieved depends heavily on how tightly the trading offering is integrated with the other parts of the provider's business. Prime brokerage is a classic example; global financing and full service, cost-efficient stock loan are essential requirements for many high frequency traders.

The ability to join the dots across the franchise is also essential. Banks in particular have an unenviable reputation for silo based thinking. In an environment as rapidly evolving as high frequency trading, a fleeting cross asset alpha opportunity cannot afford to wait for a bank's inter-departmental bureaucracy and demarcation lines.

A successful high frequency franchise must therefore be able to deliver flexibility and choice. The sellside provider's own standardised infrastructure might be a highly cost-effective option, but that does not justify it being the only one. Any worthwhile high frequency provider will also have an extensive presence across multiple independent data centres and exchange co-location facilities. The fundamental point is that this is an area where a rigid framework that attempts to constrain the client's business model for the convenience of the provider is valueless.


Pricing is another area where flexibility is vital because the sheer variety of client trading strategies in terms of both characteristics and volume make it essential. A provider that is completely transparent as regards its own costs is able to offer a 'cost plus' pricing model that hands control of explicit trading costs back to the end client.

For example, if a client trades across 20 venues a flat fee pricing model might be sub-optimal; leading to missed opportunities and an inefficient allocation of capital. Over time, as pricing models, liquidity and market structure change, it can also lead to delays in required reconfiguration of trading algorithms.

Helping clients to determine the most appropriate transaction pricing model for their business requires coherent data - and plenty of it. Any credible high frequency sellside provider should be able to feed back the most granular information on the client's cost performance even down to the individual transaction level. Armed with this, the client can quickly evaluate the most efficient way to deploy trading capital and the most cost-effective transaction pricing model for their business.

This data also enables the client to conduct this evaluation on a real time basis, so that the transaction pricing model used is always correctly attuned to their business model. For example, the data will reveal if the characteristics of an order book have changed in relation to the client's strategy and order size, perhaps because other participants are now on average sending different size orders in comparison with a month or a quarter ago.

An important aspect of transaction pricing is flexible smart order routing. High frequency traders who are the most latency sensitive prefer to do their own routing, but others may take a different stance. The traditional assumption is that a router should route to the venue that delivers the optimum fill ratio in terms of price and liquidity. However high frequency traders on a cost plus model might prefer to leverage the mode of a smart order router that routes them to the cheapest venue to maximise their rebates. That requires a smart order router that can deliver a user-configurable 'cost saver' mode.


The pace with which high frequency trading has changed from a competition over milliseconds to one over microseconds is remarkable. Nevertheless, while there are still those who insist upon the absolute minimisation of latency, there are plenty of others who feel that 'fast enough' is sufficient for their requirements - provided it is consistent. Field programmable gate arrays (FPGAs) are a technology that can satisfy both camps; for example, Deutsche's FPGA solution consistently takes 1.25 µs to traverse the card - including the risk check. The inherent scalability and parallelism of FPGAs also means that the average latency performance can also be the minimum and maximum.

However, as mentioned earlier, speed is only one ingredient in the high frequency trading recipe. For some it will be the most critical, but for others pricing or access/franchise range will be a greater priority. Furthermore, these priorities are almost certainly not static; as trading operations change and evolve, the four primary components will also shift in terms of their relative priority.

Conclusion: mix and match

Given the dynamic nature of high frequency trading at both an industry and individual level, it is not unreasonable to expect sellside providers to be able to deliver solutions that can adapt to clients' evolution. That might encompass a migration from an on-site software-based solution, to co-location, to sponsored access.

It might also involve parallel migration paths that expand the number of trading venues accessed, or the number of instrument types. It is therefore entirely reasonable for the buyside to expect the sellside to be able to cover all these bases, both as they stand now - and as they may stand in the future.