HOW BIG IS “BIG?”
Milan Borkovec and Ian Domowitz, both Managing Directors, and Christopher Escobar, a Director at ITG examine execution performance using evidence from aggregate trading costs in the FX market.(Photos L to R: M.Borkovec, I.Domowitz, C.Escobar)
On July 3 of this year, the courts pronounced caveat emptor with respect to execution performance in the FX market. US District Judge Denise Cote threw out a lawsuit, which accused JPMorgan Chase & Co. of breaching a fiduciary duty to custodial clients by charging “hidden and excessive mark-ups” on currency trades. Judge Lewis Kaplan dismissed a lawsuit directed at officials of Bank of New York Mellon, for ignoring “red flags” or knowing that trades were being processed at the worst or near- worst prices of the day.
In the case of JPMorgan, allegations were rejected that the custodial agreement obligated the bank to process trades at “the best available market rate” or by any other measure. Judge Kaplan said it was improper to hold Bank of New York Mellon executives and directors responsible for alleged currency trading practices leading to the lawsuit. The lawsuit did not create “reasonable doubt that the board’s inaction was a valid exercise of business judgment.”
We have no opinion on such rulings or the legal actions which engendered them. But, what do buyers do when caveat emptor is the way of a market? They look, or at least, they should. An examination of currency trading activity is more involved than Judge Cote believes, when she said, there was “nothing secret about the mark-ups” charged, because they are disclosed in public databases and on trade confirmations. We are still searching for those public data, and simply note that trade confirmations are hardly evidence of execution quality.
Are we looking yet?
Credible examination of FX transaction costs is difficult, but not impossible. Early studies by Russell Investment Group and Record Currency Management used data from sub advisors to measure the contribution of FX to overall portfolio transaction costs.( 1) In 2004, Russell found that the distribution of transactions in major currencies was highly skewed towards the worst rates of the day, with an average cost of nine basis points (bps). Record found costs to be between 10 and 12 bps; they noted that “approximately one half of the audits conducted to date by Record revealed that clients received uncompetitive FX pricing on a routine basis.” A 2010 paper from Russell shows that most FX trades of investors, who rely on the executions of managers or custodians, are executed at prices inferior to the average rate of the day.(2)
In 2007, a survey of 17 transaction cost analysis (TCA) providers revealed that none offered foreign exchange TCA.(3) Today, there are at least 9.
Some ascribe this growth to regulatory pressures. The concept of regulation is tenuous in the global FX market, but mandates for best execution are on the horizon. Legal cases provide some impetus, but TCA providers are shy about appearing on the witness stand. Customer demand, focused on process improvement in pursuit of alpha preservation, is a driver. A survey of FX traders in 2007 suggests that 73 percent would look at reports if available, and 37 percent of respondents would want such reports daily.(4)
Demand for measurement grows with changes in market structure and their consequences. Earlier this year, ITG commissioned a special report from Greenwich Associates in order to assess this evolution, as seen by the global buyside. FX volume executed electronically increased 55 percent from 2011 to 2012. Electronically-traded volume was 65 percent of the total, an increase from 57 percent the year before. Multi-dealer platforms account for the majority of volume, dwarfing the use of single-dealer systems, messaging systems, and dealing on the telephone. Algorithmic trading already is used by 23 percent of respondents.(5)
The survey also touched on TCA for FX, with 44 percent of respondents reporting usage. Early equity TCA was focused on compliance; similarly, 50 percent of users in FX also report compliance requirements. Surprisingly, 92 percent of respondents cited investment process improvements as a driver of FX TCA; this type of impetus was slower to develop in equities. Another 33 percent report client requests for execution information as being important.
The relative magnitude of FX transaction costs
Before looking at a single order from any individual trader, one goal of TCA is to quantify transaction costs prevalent in the aggregate market, for various deal sizes, times of day, and across market conditions. We give a flavour of such analysis for five major currency pairs and six minors.(6) The time period is January 1, 2013 through March 31, 2013.
The key to the analysis is the construction of a consolidated limit order book for each currency pair, based on data from twelve banks and five electronic communications networks (ECNs).7 Tradable quotes are identified, and all statistics are based upon them; we examine indicative quotes in the next section. We limit ourselves to a discussion of spot rates.(8)
Figure 1 illustrates depth of book for one major and one minor currency pair.
The advantage to using an empirical order book is the ability to construct size-adjusted spreads for any time of day. Intuitively, the spread should depend on the notional amount available at any given price. The order book quantifies this notion, based on the cost of climbing the book for any given deal size.
The top panels illustrate depth at the best quote and cumulative depth of book for GBP.USD, and the bottom panels for USD.TRY. The median number of price levels from which total depth is calculated is remarkably constant across the day, ranging between 17 and 20 for GBP.USD and 10 to 12 for USD.TRY.
The basic patterns are the same across all currency pairs. Median depth at the best quotes is 1mm, rising at most to roughly 1.5mm for the EUR.USD. The 90th percentile exhibits a bit more fluctuation, but is still relatively constant for GBP. USD at 3mm; for the EUR.USD pair, the upper percentile range hovers around 5mm.
In comparison, cumulative liquidity across available prices rises to 300mm, on average, for EUR.USD, and between 200mm and 250mm for the Pound, depending on time of day. Median cumulative depth for USD.TRY is between 25 and 30 times that available at the best quotes. This minor pair is not completely representative, however. Book liquidity is sparse for all CZK pairs and for EUR.PLN. Median cumulative depth is below 5mm for these pairs. In contrast, USD.PLN exhibits liquidity on the order of 25mm to 30mm.
Based upon these book data, we construct a measure of cost, by currency pair, time and order size.(9) Figure 2 contains the results of the exercise for two major and two minor pairs, which are representative of the eleven currency pairs studied.(10)
Time-of-day effects are small during London trading hours; the 1000GMT and 1600GMT curves virtually lie on top of each other. Off-hours trading is substantially more costly, although such differences are minimised for Asia-Pacific pairs such as AUD. USD and USD.JPY.
Aggregate costs are far lower than previously reported estimates, such as those cited for Russell and Record Currency Management. In those cases, time is a factor, since results date back to 2003. Morgan Stanley reports more current numbers based on their model-based methodology, which are multiples of those shown here.(11) Our liquidity-based estimates for 50mm GBP, for example, are a tenth of the Morgan Stanley estimate of 4.35 bps. For USD.CAD, they report 4.40 bps on average, with minimum cost at 1.82 bps; in contrast, for the same 50mm deal size, the graph suggests about 0.5 bps. In the case of USD.ZAR, Morgan Stanley’s estimate is seven times what we see in the aggregate data.
As is often the case, the truth is probably somewhere in between. Our method takes advantage of real liquidity provision from seventeen sources. The assumption in deriving size-adjusted spreads is a bit heroic, however. We are assuming a trader’s ability to sweep the aggregate book, taking advantage of all liquidity for all deal sizes. While this is closer to reality in equity markets, fragmentation of data sources and the mixed nature of the dealer-ECN markets in FX suggest that our estimates constitute a lower bound, at least in view of current trading practice. Finally, we model size-adjusted spreads only, ignoring latent liquidity and slippage costs. In contrast, the Morgan Stanley numbers are single-sourced, and depend on models, for which waiting times for new orders and volatility are sufficient to determine cost.(12) Depth of book is not taken into account. A true reality check awaits serious examination of individual buyside data with good time-stamps.
Indicative quotes and their tradable counterparts
Indicative quotes are widely disseminated, and underlie most studies of the FX market.(13) A comparison of tradable quotes (TQ) to indications (IQ) is therefore of interest. Although there are some quantitative differences in quote levels across currencies, patterns are similar enough that the relevant points can be illustrated using one major and one minor pair. The first comparison appears in Figure 3.
Spreads calculated from indicative quotes are significantly greater than those computed from tradable quotes. The ratio of the two spreads ranges between five and ten, on average for all currencies. Although the indicative quotes are not updated as quickly as tradable quotes, they track each other fairly closely, albeit at different levels.(14) In other words, the intraday patterns are essentially the same. Although it is not obvious from the USD. JPY example, the difference between tradable and indicative quotes narrows during London trading hours for most currency pairs.
Indicative quotes do not vary by size. A natural question concerns the size of deal for which the indicative spread “correctly” prices a trade relative to what is actually available in the market. Figure 4 contains representative plots, taken from the GBP. USD and USD.ZAR pairs.
For the major currency pairs, size-adjusted spreads cross the indicative quote at deal sizes between 80mm and 120mm. Indicative spreads overstate cost for all sizes below that range, illustrated here by the GBP.USD pair. For the minors, where deal sizes tend to be smaller, the crossing point is much lower, at about 50mm in the example above and for pairs such as EUR.PLN (roughly 30mm).
Linking FX costs to institutional equity demand
One of the grander aspirations of TCA is to provide the “all-in” cost of a transaction. Appropriate linking of orders is an ongoing issue even in straight equity TCA, depending on the workflow of any individual buyside institution. Making the connection between global equity trades and their corresponding FX costs eventually will require information from the buyside, which enables the connection to be made.
We can provide an idea as to what might be expected. For the first quarter of 2013, we select the ten most active equity trading firms from our TCA Peer database. All equity orders requiring a foreign exchange transaction are identified.(15) At the end of each trading day, the size of the FX transaction is calculated based on the aggregated executed sizes of all equity transactions in a country.
We contrast three polar outcomes. The first is immediate execution of the aggregated FX volume of all equity trades at the time the last equity order is completed. The second and third are FX executions at prices which deliver the best and worst outcome of the day, excluding the period 21:30-22:30 GMT, during which prices are not representative.(16) The results of this exercise are contained in Figure 5.
FX order sizes tend to be relatively small, on average, but have some sizable outliers for certain days. As expected, the magnitude of the ITG Peer client order sizes varies greatly across currency pairs. Euro and Pound lead the pack with the largest order sizes of 550mm and 600mm, respectively. The average order sizes at 100mm and 170mm are also substantial for both pairs.
Further relating the results to costs in Figure 2, orders in Canadian currency have order sizes up to 200mm (with an average around 60mm) and for the Polish Zloty, only about 50mm (9.5mm). The last is not representative of order sizes in all the minors, however. Equity-linked FX order size for the South African Rand is in the range of 30mm on average with a maximum around 90mm.
How much does it cost the average firm to implement the FX leg of an equity transaction? The answer from this sample is an annualised $13.8 million.
How much could it have cost, if executions were consistently at poor prices? The annualised figure per firm would be $40.8 million. Enough said.
Buyside data and the way forward
The purpose of this article is to outline available evidence with respect to aggregate FX transaction costs. There are both caveats and opportunities associated with the exercise.
Our estimates are derived from liquidity information based on seventeen data sources, all providing tradable quotes, and permitting the construction of an order book. It is no surprise that indicative quotes are generally useless in judging levels of cost, for which we provide evidence. In effect, however, we present a lower bound on transaction costs, with results being a fraction of those presented in other published sources. The reason for this is an assumption that a trader can sweep the book in a market fragmented not only by time and space, but also by the proliferation of dealers and ECNs. The difference between our aggregate estimates and realised cost represents the opportunity to save money. Hence the rationale for FX TCA, and some motivation for changes in market structure and sellside applications, which would permit such liquidity aggregation.
We note that a reality check awaits a serious look at a cross-section of buyside firms’ FX dealings, using data with good time-stamps. A preliminary examination of buyside trading in our own files suggests that process improvement can lower costs. Trading in EUR.USD, for example, appears to cost roughly three times what we would have predicted based on the order book. For the AUD.USD pair, the factor is four. For the ten firms for which we match equity transactions with their FX counterparts, the cost is $3.5 million, per firm per quarter, based on our lower bound estimates of size-adjusted spreads. Poor execution multiplies this figure three-fold. These are serious numbers, which call for a serious attempt at measurement and analysis.
There is much more to do, and many more questions than answers at this stage. Forwards and swaps constitute part of our individual buyside analyses, and similar aggregate information would be useful, especially for minor currencies. The impact of the common practice of netting currencies is certainly a topic, especially since any residuals from that process are executed by a single dealer, as opposed to being exposed to the type of liquidity described here. The effects of volatility are not yet well understood. In preliminary work, we find that volatility, per se, may not be as strong an effect as commonly believed. Volatility surprises, deviations from expectations, constitute a powerful driver and can be quantified, not only for forensic analysis, but also as a pre-trade tool. Explicit links between pre-trade and post-trade analysis have been shown to reduce costs in equity markets. We believe the same to be true in FX.
How big is “Big?” – regardless of disparities in alternative estimates, FX trading costs, if not measured and managed correctly, can be a meaningful drag on investment performance. Solutions now exist, which permit leverage to achieve better performance. Investors and traders are beginning to expect counterparty accountability in terms of execution. Focus and measurement are the first necessary steps.Footnotes:
1. Robert Collie, “It’s Time for More Choice in FX,” Russell Investment Group Viewpoint, December 2004; Record Currency Management, “Paying Heed Pays Off,” Record Research Summary #5, July 2003; Record Currency Management, “Report to Frank Russell on Currency Transaction Costs,” February 2005.
2. See https://investment.russell.com/public/pdfs/Consulting/ Asset Class Strategy/0110 RR FX Fees.pdf.
3. Michael DuCharme, “First Steps in Foreign Exchange Transaction Cost Analysis,” Journal of Performance Measurement, Spring 2007.
4. Tabb Group, “Just What is Best Execution in FX?” Tabb Group Perspective, July 2008.
5. There also is nascent dark pool activity; see Foreign Exchange Trading Creeps into Dark Pools, Wall Street Journal, October 11 2012.
6. The majors are EUR.USD, GBP.USD, AUD.USD, USD.CAD, and USD.JPY. The minors are USD.PLN, EUR.PLN, USD. CZK, EUR.CZK, USD.TRY, and USD.ZAR.
7. This approach was discussed at length in work by Morgan Stanley, but not acted upon, due to lack of available data. Instead, aggregate cost models were built based on options pricing formulae and an assumption of Poisson arrivals of orders. See, “A Guide To FX Transaction Cost Analysis, Parts I and II,” Morgan Stanley White Paper Series, October, 2009 and February, 2010.
8. Space constraints preclude discussion of forwards and swaps.
9. The measures to follow are based on five-minute intervals and adjusted for daylight saving time regimes. Costs are computed for six deal sizes: 0.1mm, 2.5mm, 7.5mm, 15mm, 35mm, 75mm, and 200mm; remaining data points are simply interpolated. Median values of the size-adjusted spreads are illustrated in the graphs.
10. Although extrapolation produces reasonable results for large deal sizes for the CZK pairs and EUR.PLN, the lack of substantial liquidity on those books precludes reliable estimates past the 5mm deal mark.
11. See, for example, “How Much Does It Cost to Trade 50M?” Morgan Stanley Fixed Income and Trading white paper, June 2013.
12. Earlier work by Morgan Stanley cites EBS as the single data source. The 2013 document from which we make the comparisons simply gives the source as their own Quantitative Solutions and Innovations group. Their characteristic waiting time can be approximated by the notional amount of the order, divided by the average size of arriving orders, times the average rate of order flow in the market during the transaction period. This follows from a statistical distribution assumption for order arrivals.
13. An early example is by Ian Domowitz and Tim Bollerslev, “Trading Patterns and Prices in the Interbank Foreign Exchange Market,” Journal of Finance 48, 1993.
14. IQ quotes tend to lag the tradable quotes consistently by a few seconds for major currency pairs.
15. For reasons idiosyncratic to our own database, the analysis is restricted to USD pairs, e.g., AUD.USD and USD.TRY.
16. Even excluding this period, the worst prices of the day still are temporally close to this interval.
© 2013 Investment Technology Group, Inc. All rights reserved. Not to be reproduced or retransmitted without permission. 91613-17020
These materials are for informational purposes only, and are not intended to be used for trading or investment purposes or as an offer to sell or the solicitation of an offer to buy any security or financial product. The information contained herein has been taken from trade and statistical services and other sources we deem reliable but we do not represent that such information is accurate or complete and it should not be relied upon as such. No guarantee or warranty is made as to the reasonableness of the assumptions or the accuracy of the models or market data used by ITG or the actual results that may be achieved. These materials do not provide any form of advice (investment, tax or legal). ITG Inc is not a registered investment adviser and does not provide investment advice or recommendations to buy or sell securities, to hire any investment adviser or to pursue any investment or trading strategy. The positions taken in this document reflect the judgment of the individual author(s) and are not necessarily those of ITG.
©Best Execution 2013