How does Commerzbank define internalisation? What is the impact of internalisation in the use of execution algos?
In the specific context of an agency algorithmic execution, we define internalisation as the proportion of volume filled on Commerzbank’s principal liquidity coming from its automated e-FX market making desk. All of Commerzbank’s execution algorithms allow the client to either select Commerzbank liquidity only as the liquidity pool, or a hybrid mode, where both our market making desk’s liquidity and external market sources are used to fill the order. When only our liquidity is selected, the algo’s internalisation rate is 100%.
The natural and important subsequent consideration is what is the internalisation rate of Commerzbank’s eFX principal market making desk? It would be deceiving to say the execution algorithm has a 100% internationalisation rate if the principal market making desk hedged every single child order back-to-back right away on external markets. The precise definition of internalisation for a principal market making desk is more equivocal. Clearly, hedging the market making risk on public venues that disseminate market data and are considered paramount for price discovery constitutes externalisation. But what about dark mid match venues? One can argue it is externalisation as the risk is transferred to another liquidity provider (LP). On the other hand, that other liquidity provider was posting interest because it had the opposite position in its book. Therefore, when thinking about liquidity providers as a group, the client flow ended up being absorbed in one of the LP’s inventories and thus it was internalised.
The main reason liquidity consumers prefer LPs that internalise is to limit market impact. If the flow ends its path with an LP that is happy to sit on that risk for a while, one can argue the flow was internalised and the market did not move because of it. How about skewing, showing an axe to liquidity consumers? Again, one can argue that it is internalisation as the LP exits risk with a liquidity consumer (LC). It is not that simple, skewing to the wrong LC will create market impact. Market participants who take advantage of LP skews to recycle their liquidity will have market impact.
At Commerzbank we think that the internalisation rate should be defined in the simplest way as client only volume divided by total volume traded.
Why does Commerzbank’s approach particularly benefit FX algo users?
At Commerzbank, we think that the internalisation rate should be defined in the simplest way as client only volume divided by total volume traded. LPs with a large and diverse franchise who see bi-directional flows will display the highest internalisation rates. Commerzbank has increased is risk appetite over the last few months and is increasing its internalisation rate. We take pride in being able to provide liquidity even during the most volatile market conditions, when interbank liquidity can be patchier in some currency pairs. Trading with a true market maker that absorbs the flow and can take on large inventory limits market impact and benefits the liquidity consumers. An execution algorithm filling a large proportion on internal liquidity is better to avoid signalling and market impact only if the principal source of liquidity behind it has a very high internalisation rate too.
Can algo analytics help to shape and inform client understanding and further improve their execution performance when using algos?
It is paramount to be entirely transparent on the liquidity sources used to fill an execution algorithm. It is not enough to report what proportion of trades are filled on ‘internal’ liquidity as the market impact benefit will eventually be dictated by the internalisation rate of the internal source of liquidity. A good TCA will clearly give the source of liquidity for every child order. A step further in the right direction can be to give a sense of the potential for signalling or market impact depending on which liquidity source is used by the execution algorithm. The issue of different definitions of internalisation used by algo providers impacts clients when trying to compare key algo performance metrics, but this can be overcome by being explicit in terms of what is the exact definition of internalization used when an LP says ‘we internalise X%’.