Internalisation by electronic FX spot dealers

Abstract

Dealers in over-the-counter financial markets provide liquidity to customers on a principal basis and manage the risk position that arises out of this activity in one of two ways. They may internalise a customer's trade by warehousing the risk in anticipation of future offsetting flow, or they can externalise the trade by hedging it out in the open market. It is often argued that internalisation underlies much of the liquidity provision in the currency markets, particularly in the electronic spot segment, and that it can deliver significant benefits in terms of depth and consistency of liquidity, reduced spreads, and a diminished market footprint. However, for many market participants, the internalisation process can be somewhat opaque, data on it are scarcely available, and even the largest and most sophisticated customers in the market often do not appreciate or measure the impact that internalisation has on their execution costs and liquidity access. This paper formulates a simple model of internalisation and uses queuing theory to provide important insights into its mechanics and properties. We derive closed form expressions for the internalisation horizon and demonstrate—using data from the Bank of International Settlement's triennial FX survey—that a representative tier 1 dealer takes on average several minutes to complete the internalisation of a customer's trade in the most liquid currencies, increasing to tens of minutes for emerging markets. Next, we analyse the costs of internalisation and show that they are lower for dealers that are willing to hold more risk and for those that face more price-sensitive traders. The key message of the paper is that a customer's transaction costs and liquidity access are determined both by their own trading decisions as well as the dealer's risk management approach. A customer should not only identify the externalisers but also distinguish between passive and aggressive internalisers, and select those that provide liquidity compatible with their execution objectives.

Keywords:

• Internalisation
• Queuing theory
• FX liquidity
• Execution costs

Acknowledgements

The authors would like to thank two anonymous referees, colleagues at DB, Rama Cont, Søren Johansen, Vasco Leemans, Richard Louth, Michael Melvin, and seminar participants at the London School of Economics, The London Quantitative Finance Seminar at the CFM-Imperial Institute of Quantitative Finance, INET ‘Microstructure workshop’ at Trinity College Cambridge, the MathFinance Conference 2017 in Frankfurt, TradeTech FX 2017 in Miami, the BIS Working Party on Markets in Buenos Aires, and Jim Gatheral's 60th Birthday Conference at the Courant Institute New York University for helpful comments.

Disclaimer

The views and opinions rendered in this paper reflect the authors' personal views about the subject and do not necessarily represent the views of Deutsche Bank AG or any part thereof. This article is necessarily general and is not intended to be comprehensive, nor does it constitute legal or financial advice in relation to any particular situation.

Disclosure statement

The authors are employed within the electronic FX spot trading division of Deutsche Bank A.G. Deutsche Bank (DB) is an industry recognised world leader in the foreign exchange business, and offers a full spectrum of foreign exchange products and services, including the trading of foreign exchange products through its Autobahn electronic trading platform. This paper was prepared within the Sales and Trading function of DB, and was not produced, reviewed or edited by the DB Research Department. The views and opinions rendered in this paper reflect the authors' personal views about the subject. No part of the authors' compensation was, is, or will be directly related to the views expressed in this paper.

Notes

1 An example of externaliser behaviour is found in Virtu Financial, Inc (2014, p. 2) ‘Our strategies are also designed to lock in returns through precise and nearly instantaneous hedging, as we seek to eliminate the price risk in any positions held’. The practice of internalisation is referenced in Bank of England, H.M. Treasury, and Financial Conduct Authority (2014, p. 59) ‘Market participants have indicated that some dealers with large enough market share can now internalise up to 90% of their client orders in major currency pairs’. The notion that dealers are either perfect internalisers or perfect externalisers is of course too constraining and in practice they will use to varying extent a mix of both to manage their risk. This is likely to be true even at a more granular level, e.g. by individual customer, currency pair, time-zone, etc. The distinction between internalisers and externalisers is nevertheless an important one.

2 The highest rates of internalisation are attained in electronic spot, particularly by larger dealers in more active trading centres.

3 For example, queuing theory is used in the analysis of shopping queues or traffic congestion, design of factory assembly lines, allocation of staffing levels in a hospital's emergency department, and the scheduling and load balancing of large scale calculations or high volumes of search queries across a cluster of servers. See, e.g. Harris (2010), Hopp and Spearman (2000) and Wolff (1989).

4 Throughout the paper, we use the terms ‘dealer’ and ‘liquidity provider’ interchangeably, and the same for ‘trader’ and ‘customer’.

5 This assumption is made to simplify exposition. In practice, the dealer reserves the right to reject incoming trade requests if a set of pre-defined validation checks is not passed. This is commonly referred to as ‘last-look’, see Oomen (2017b) for further discussion. Also, a trader would routinely place more than two dealers in competition for their flow. Oomen (2017a) studies such an execution setup. Generalisations along those lines do not change our basic findings.

6 For example, the dependence of order flow and price distance to mid is assumed to be linear in Ho and Stoll (1981), while Avellaneda and Stoikov (2008) assume an exponential or power law obtained from empirical studies like Maslov and Mills (2001), Potters and Bouchaud (2003), and Gabaix et al. (2006).

7 Hasbrouck and Levich (2017) propose a correction for the double counting of prime brokerage volumes in the raw BIS figures. If applied here, it would reduce volume numbers by about 17%.

8 In the BIS survey customers include financial institutions (e.g. non-reporting commercial and investment banks, security houses, leasing companies, financial subsidiaries of corporate firms), real money (e.g. mutual funds, pension funds, asset and wealth managers, currency funds, money market funds, building societies, insurance and reinsurance companies, endowments), hedge funds and proprietary trading firms (e.g. commodity trade advisers, high frequency trading firms, global macro funds), official sector (e.g. central banks, sovereign wealth funds, development banks and agencies, financial public sector institutions), non-financial (e.g. corporates, non-financial public sector institutions), and other (e.g. retail aggregators). The BIS defines the internalisation ratio as ‘the percentage of reported total foreign exchange turnover which was internally matched against offsetting trades by other customers.’ (BIS 2015, Table 9). Applying this definition to the aggregate BIS numbers for spot we infer an market-wide internalisation ratio of 1,047/1,652 = 63%. This coincides perfectly with the average internalisation ratio across FX spot dealers as reported by Moore et al. (2016).

9 The precise distribution for q is specified as follows: trades of sizes 1 and 10 k occur with a probability of 15% each, trades of sizes 50 and 250 k occur with a probability of 20% each, and then trades of sizes 1, 5, 10, and 25 mn occur with a probability 25%, 3%, 1.5%, and 0.5% respectively.

10 With unit trade sizes, the position volatility is equal to R as per equation (7(7) ${\phi }_n=\frac{1}{\sqrt{2\pi R^2}}{\mathrm{e}}^{-\left(1/2\right)n^2/R^2}.$(7) ) in Example 2.2. With variable trade sizes; however, the distribution is not normal anymore and also there is no well defined relationship between the parameter R and the LP position volatility.

11 It can also be further illustrated with a queuing theory analogy. Consider a group of, say, 15 friends queuing together at airport security where they will be individually and sequentially served by a single custom's officer on a first-come-first-serve or FIFO basis. If the queue ahead of them is short (e.g. 2 people) then the waiting time for the group of friends to all pass security is primarily driven by the size of the group itself. However, if the queue ahead is large in comparison to the size of the group (e.g. 200 people), then the waiting time for them to all pass security is primarily driven by the length of the queue.

12 For instance, with exponential skewing, the dealer's position distribution is Gaussian. The average risk position is therefore $E\left(|X|\right)\propto R$ which after multiplying it with the asset's volatility gives a measure of market risk.

13 This relies on the assumed absence of trade rejects via the last look mechanism, see Oomen (2017b) for further discussion.