Although FX markets are always changing, many market characteristics such as spread, volatility or liquidity often fall into predictable ranges. An algo that can recognise and adapt as these market characteristics change is more likely to give clients better outcomes than one that doesn’t. We will consider some talking points around algos in changing markets; and will also consider questions a buyside trader might ask about algo behaviour to assess algo providers.
A lesson learnt from 2008
In late 2008 and early 2009, deleveraging and risk-off sentiment lead to significant intraday movements in currencies. Algo execution was still a nascent product in FX, and although there were sophisticated heuristic algos, it is reasonable to say that algos today generally should have better models to recognise and adapt to variable market conditions.
I remember a US macro fund using an algo for the first time to trade an Aussie cross. Liquidity was heavily imbalanced with multiple sellers of Aussie and few buyers. To reduce impact, this algo trade was restricted to 10% of the main market volume. This slowed down the execution and as the market moved a big figure, what could have been a 20-point slippage, turned into 50. Using an algo to reduce market impact was the wrong approach in these market conditions. A more prudent approach when liquidity on the opposing side of the market is scarce, is to use the execution certainty given by a risk transfer price.
The lesson from this trade, and a number of events since, is that although markets predominantly behave in a reasonably stable manner with predictable liquidity and volatility, when market conditions change, algo users need to understand how the algo will adapt, what slippage estimates change, and alter execution style accordingly. Today, algos and pre-trade analytics are more sophisticated, and just as importantly the way we talk about and use algos has advanced. Picking the correct algo and using it in the right way will significantly improve outcomes.
Market regimes and adapting to changing conditions
A market regime can be characterised by quantitative variables such as direction, spreads, liquidity and volatility. If any of these characteristics substantially change there will be a shift from one regime to another. Two methods for identifying regimes are via market structure or historical events. Identifying structural regimes is usually straight-forward as they are relatively common and due to factors such as the intraday distribution of liquidity. However, identifying regimes through historical events is also important so an algo can be programmed to adapt to market conditions that may only be seen rarely. Examples of this are different regimes that followed sterling devaluing after the Brexit referendum result.
An algo will be most effective if it can adapt to as many market regimes as possible and make informed execution decisions that are appropriate for each situation. Some algos may not explicitly trade according to defined regimes but will still use quantitative models to analyse real-time and historical market data. It is important to note that if these models are only looking at recent history, then they may not capture more historical event behaviours. Because such events could reoccur, it is worthwhile understanding how far back quantitative analysis goes.
Let’s consider changes in algo logic for a TWAP execution. In a low volatility regime, the decision to trade passively is stronger as there is lower risk that the market will move substantially away and capturing spread is beneficial. It follows that for a TWAP algo to trade more passively it will need to have greater flexibility around the TWAP schedule. However, if volatility increases, the risks from trading passively are increased and the algo should keep closer to the TWAP schedule.
Now consider a TWAP execution when the market has a significant liquidity imbalance and the market is trending. Clearly if the trend is towards us (i.e. market is going lower and we are buying) we would want to trade slower and fall behind the TWAP schedule. This will take advantage of trading more of the order in a lower market and thus outperform the TWAP benchmark. On the other hand, if the market is moving away, we should trade quicker, cross the spread more often, and try and be ahead of the schedule to take advantage of better prices now than what we expect later.
Other algo types have characteristics and behaviours for different market regimes to outperform versus benchmarks. An algo with an arrival price benchmark, for example, might determine that the current market regime is likely to result in a higher cost execution than usually expected. In response to this the algo behaviour could have more market impact and cross wider spreads. Of course, every algo provider is going to have different methodology to measure regimes and adapt their algos. An effective way that algo providers test algo behaviour across regimes is to carry out event specific trading simulations. Buyside traders interested in algo behaviour across different regimes, should review these simulation results.
Traders can gain execution savings during normal market conditions, only to give them up when markets start to behave differently and algos underperform. To partly mitigate this, having an execution plan for different market conditions or market regimes is a worthwhile exercise. Traders should therefore seek to understand how algo providers analyse changing markets, how they adapt their algos and what they recommend in different scenarios.
Traders may also consider asking the following questions of their algo providers:
What quantitative metrics do you use in determining market regimes?
The metrics used to characterise market conditions will vary substantially across algo providers.
How do you identify infrequent historical regimes or events?
Identifying past events is important for ensuring an algo is programmed to react effectively.
How does the algo adapt to different regimes or changes in market characteristics?
Changing market conditions should lead to different algo behaviour.
How does the algo behave when the market gaps?
Significant market gaps can often stop algos from executing due to hard limits.
How does the algo trade differently in trending or volatile markets?
More advanced algos will have adaptable behaviour for different market conditions.
How do you ensure algos behave as expected in different regimes?
Simulation testing is an example of preparing for regime changes.
What notifications or alerts do you have in place when market conditions unexpectedly change?
Ideally, there would be an automated or manual notification of a significant change to market conditions when trading an order.
Do you have specific recommendations for algos in different market conditions?
Algo providers might have specific market conditions where they believe their algos have an advantage.