The Role of Risk in a Modern FX Business

In the latest Voice of Experience column, Steve Flanagan highlights a risk to the FX markets that is external to LPs, and maybe needs more attention

Risk can be analysed from a variety of angles, market risk, technology risk and regulatory risk to name just a few. For the purpose of this discussion, I will focus upon liquidity and model risk from the perspective of a liquidity provider.

Most of the obvious risk elements today have been double and triple covered by LPs and controls implemented after the many internal audits, checks and reviews. I think it is safe to say that internal risk models are continuously being updated and covered. The evidence to this today is that market liquidity has remained constant – even during recent geopolitical events, there has been a price to trade upon. Spreads may widen during market uncertainty to reflect the risk event, but pricing has always remained firm. Liquidity has been tested in the past and has held up quite well in the marketplace. There has always been a price to execute upon, and that means FX liquidity gets the grade of an “A”.  Kudo’s to the LP side of the business.

Nevertheless, liquidity availability is always a concern in traders’ minds. Could we encounter a liquidity risk event today and if so, how would it possibly play out? It would have to be a shock to the system, because as I stated, recent events have proved that an LPs liquidity and pricing have been very resilient.

As I see today’s FX market, there is one area of risk the stands out as its Achilles Heel and is external to an LP, that is market data risk. Market data affects both liquidity and models, it is consumed by every pricing and execution algorithm, and its accuracy represents a significant risk to today’s FX marketplace. To understand my point of view, I will digress for a moment and take you under the hood of an LP’s engine.

Just as any machine today consumes gasoline or electricity as an input for its power output, an LP consumes market data as its input for its price and hedging. Whether an LP hedges via internalisation or direct market pass-through, it relies on accurate market data to triangulate its risk position and how and where to price its risk. Market data comes in the form of price history – last paid or given in the marketplace.

An important influence on that market data price is news, whether it is a scheduled economic release or a sudden news release. We have all seen recent data releases, such as US CPI and the sudden price change it can have in the FX market. Within milliseconds market data is consumed and passed through to price which is critical to every algorithm in today’s market.

Market participants, many of whom are armed with new high-tech models using the latest influences of AI and ML, can be fooled by errant market data

Most of today’s LPs’ and market participants’ risk management tools were built and back tested against the 2015 SNB, 2016 Brexit and 2020 Covid events, and will respond immediately to new incoming market data. These risk models are unproven in the 2024 market where millisecond speed limits have been eclipsed and broken as microsecond speeds exist.

A sudden shock, referred to as a new data event (NDE), sees liquidity impacted immediately in two ways. First, the top of book (TOB) pricing immediately widens, and second, depth of book is reduced, dispersed or thinned out. These are two prudent reactions by an LP and allow for a quick risk assessment and new model weighting to this NDE. Whether you are a LP trying to manage your incoming one-way risk or a market participant reacting to the NDE, executing trades, slippage and rejections characterise the initial impact of the event. Who is faster, the price maker or the price taker? LPs, responding to new market data, may automatically reduce potential incoming risk via reducing outgoing liquidity.

Market participants, many of whom are armed with new high-tech models using the latest influences of AI and ML, can be fooled by errant market data.

In a risk event, everything trades top of book as depth of market liquidity is reduced by 50-75%. In a perfect environment, the ebbs and flows at TOB, allow an LP to deploy an internalisation model for risk management. In an NDE, internalisation ceases to exist as the one-way nature of flow overwhelms and swamps the top of book space where today more than 75% of all transaction occur. LPs and market participants all turn to a depleted marketplace to execute and unload risk.

Each and every day new execution algorithms, new models, are deployed to the market, all seeking to be faster than the next, and all seeking a price to execute upon. LPs must constantly be improving upon the accuracy and speed at which they can price, however, somewhere there must be a speed limit, a vanishing point, that present day machinery hits. After all, how fast can pricing machinery react and re-price an NDE before being flooded by incoming trades? View a move like this in slow motion, think about how many sell or buy orders are being executed on a handful or more of LP prices within milliseconds of the NDE (visualise a watermelon coming down a garden hose).

Present-day markets have not been tested against an AI cyber-attack that corrupts multiple new data events across several asset classes and includes errant headlines

The point is there is a limit to how much risk a liquidity provider can, or is willing to, take at a given price. What is the response time to a NDE by an LP and market participant? Herein lies the source point for market dislocation. Everything relies upon market data. It must be reliable, timely and accurate.

Colin Lambert articulately argued in his interview with Galen Stops at 360T earlier this year that a cyber-attack is a risk to markets in 2024 and I agree. I believe a cyber-attack targeting market data, which is an external risk factor to an LP, is the number one risk to a modern FX business today.

Consider the cross-asset trading nature of our markets today, when one asset class moves other asset classes react in accordance to the models’ new risk assessment. But, when they all move, correlations completely break down. Our present-day markets have not been tested against an AI cyber-attack that corrupts multiple new data events across several asset classes and includes errant headlines involving geopolitical events. The key is it only takes one errant price to trade, to create a source of new market data which could legitimise the price movement. Uncertainty and the worst trading emotion of all now permeates all marketplaces – FEAR.

Even if we can assume central authorities act rapidly and the errant headlines are officially denied, the damage is done. The entire event may only take 10 to 15 minutes, but I think you see the risk that market data plays in today’s highly electronically-connected marketplace. Data transmits instantaneously and speed limits from previous market tested stress points have been broken and are now measured in microseconds. A move like this could initially have characteristics of a flash crash but can’t be because there would be thousands of traded price points throughout the event. There can be no re-papering of execution rates as they occur against all asset classes. You get the picture.

Accurate data is the lifeblood of a healthy marketplace, but it only takes one or two headlines followed by trade executions to create an errant market data print for all to consume and be forced to follow. Machinery does not have the luxury of questioning whether the market data is real, it must react. Every machine has its limit. Think about Formula 1 or NASCAR racing. You can only go so fast around a turn or on a straightaway, the engine can only give you so much power and speed. There are mechanical restrictions. Machines have limitations, and then break. It’s under what conditions they break.

I believe the risk in today’s FX market lies in the potential corruption of market data which all algorithmic models consume. Technology speed limits have advanced so much in the last few years, but I feel remain untested as to the market data they consume. In a NDE all machinery must respond and be the first to execute due to the uncertainty of the situation and the next price.

This means market participants must know their technology risk limits and continue to support relationships, in the form of automated or voice trading, for that is where firm pricing will always be found in a potential event. In other words, make sure you still have a phone number to connect to your LP’s sales desk – it’s always the best back up plan!