DO HFTs REALLY ‘GAME’ BUYSIDE ORDERS?

Darren Toulson, head of research, LiquidMetrix explains to Best Execution how to dissect an order to see what is really going on.

In recent years, a significant proportion of liquidity provision and trading on lit markets, usually estimated to be around 40% of matched volumes in Europe, is conducted by ‘HFT’ firms.

A common buyside complaint is that the trading styles and strategies of HFT firms exacerbate the market impact of ‘real’ orders being sent to the market, making it difficult for them to access lit liquidity without being ‘gamed’ by faster HFT participants. This fear has lead to an ‘arms race’ with increasingly elaborate algorithms, faster connections and faster machines being used to slice orders and hide intent when accessing lit markets.

To what extent are these fears justified?

Example: SOR order

We’ll look in detail at the lit market reaction to a specific buyside order. The example chosen here is typical of the behaviour seen when analysing thousands of similar orders.

The order in question is a ‘SOR slice’ executed in a Swedish stock by a London-based algorithm. The order’s target volume is too large to execute with a single aggressive order at EBBO on one venue and so was split into chunks and sent aggressively, and simultaneously, to 3 London-based MTFs (CHIX, BATE, TRQX) and the primary Swedish market.

The entire algorithm executed over a period of 58ms and resulted in over 70 individual fills at different price levels on different execution venues.

So what happened?

Diagram. Orderbook across all venues showing orders available at the beginning of the SOR trade

The diagram (above/right/left/below?) shows all on-book offer liquidity (this was a buy order) available on CHIX/TRQX/BATS/BURG/XSTO at the first millisecond of the SOR order. Each rectangle represents a piece of resting liquidity on a lit venue. The background colour depicts the venue.

The volume text colour shows what happened to the available liquidity over the next 58ms as the SOR slice executed:

• Black: the SOR algorithm successfully ‘hit’ the liquidity; for instance, the SOR algorithm successfully hit all of the London MTF liquidity at 174.1.

• Orange: the liquidity was pulled (cancelled) before the aggressive orders sent by the SOR slice reached the venue. So, at the first price level of 174.1, 7 out of 11 of the resting orders on Stockholm were cancelled just before the SOR orders could execute.

• Red: another market participant sent an aggressive order that executed ahead of the SOR slice. In other words, it was ‘stolen’ by a faster participant.

What’s striking about this picture is the amount of red / orange activity. There is nothing special about the timing of this order; market activity on this stock in the seconds preceding the order was negligible.

So to tell the story of what happened with this SOR slice:

• The first orders sent by the SOR to London MTFs successfully hit their targets.

• HFT liquidity providers, reacting to these trades, immediately cancelled most of the orders resting on XSTO. This implies that the HFTs were able to react to the executions on London and send messages to cancel orders in XSTO faster than time taken for the SOR orders to reach Stockholm.

• Other HFT market participants (not necessarily the same firms) aggressively traded ‘in front’ of the SOR slice.

• A similar story unfolds as the SOR order executes further down the book.

The analysis

What does this example tell us? Firstly, it illustrates the degree to which liquidity and trading really do react at millisecond timescales. Buyside orders attempting to access such liquidity must be precise in their timing and sequencing otherwise they may be ‘gamed’.

However, it also demonstrates that much of the resting liquidity on lit venues is currently being provided by HFT players. The improvements in average spreads and liquidity seen when MTFs appeared in Europe may well be in part due to such HFT liquidity providers feeling able to offer tight prices as long as they can “dodge the falling knife” when they see large incoming buyside orders.

Finally, what of current proposals for minimum 500ms resting times? This rule would primarily target the ‘orange’ HFT liquidity providers, i.e. those trying to avoid adverse selection but who do add liquidity. It would do nothing to prevent, and in fact would encourage, aggressive (red) HFT activity using lower latency to trade ahead of slower participants.

Perhaps not the regulators’ intended target?