An often overlooked problem in transaction cost analysis is endogeneity. The term arises from econometric analysis. It describes how the results of a regression analysis can be biased when both the dependent and independent variables are affected by the same predictor factors. This problem is most acute when we try to measure market impact. 

According to Collins and Fabozzi (1991)[1] market impact is defined as the difference between a fair price benchmark prior to execution and the actual execution rate. This difference incorporates both a spread and any market movement between these two moments in time. It’s easy to define, but actually identifying market impact is like finding an answer to the zen koan which asks whether a tree falling in the forest makes a noise, if no one is there to hear it.  The difficult question is what would have happened if you had not traded.  

Eureka!  What Archimedes tells us about Market Impact.

Consider Archimedes in his bath, adding bricks to the water and measuring the change in level of the water. The level of the water rises proportionally to the volume of bricks being immersed in the bath. Unfortunately, unlike Archimedes, who was able to mark a line in the bath with each brick being added, traders cannot observe a “true” level of the water (the market price) until after they have placed the brick in the bath. 

The level (the market price) is only revealed via the prices requested and supplied to us by market counterparties. These prices are in the form of bid and ask. Normally, receiving an ex ante two-way price should be neutral; there should be a 50% probability assigned to the value of the bid and the ask, but this assumption depends on anonymity of traders in the market. Unfortunately for the trader, bids and asks are not anonymised and equally, the probability of a trader dealing on the bid or offer is not symmetrical. The price-makers are therefore trying to work out whether the trader wishes to add a brick to the bath or take one away. Because the prices are not anonymised, over time the dealers get better at guessing the likelihood of a given action for each individual client, and these skews are reflected in the prices that are shown.

To further complicate matters,  individual clients are not the only people adding or removing bricks from the bath. And the volume of water in the bath is not constant through the day. But for now, let us admit that the prices observed are reflexive of the client’s user profile.

Heisenberg and the limits of accuracy caused by circularity.

Referring to execution costs, Grinold and Kahn[2] opined that “Market impact is Finance’s analogue of the Heisenberg Uncertainty Principle. Every trade alters the market”. The Heisenberg Principle states that at the quantum level accuracy is bounded. The reason for their pessimism is due to the circularity of most price measurement. To mix metaphors, the client is usually told what the addition or removal of a brick did to the water level by the price maker – a calculation based on the skewed price received by the client. Until quite recently, an anonymised fair price benchmark for FX transactions to measure price levels beforetaking action did not exist.

The difficulty of finding a suitable fair price benchmark stems from the legacy of Foreign Exchange market structure. When we at New Change (NCFX) were awarded our first transaction cost analysis mandate the only tradeable prices we could obtain were from the trading platform on which our client had conducted their trades. This violated a fundamental principle of objective measurement, that the fair price benchmark be independent of the decision to trade. This got us thinking.

A second problem we faced was that prices obtained from one venue are not necessarily representative of the best available price in the market at the time. This sampling error is usually overlooked.  In fact we are often told that a trading platform can create an independent aggregated mid-rate.  But this isn’t true – and this is reflected in the standards required for FX cost measurement in Mifid2.

The error lies in the fact that FX Markets were historically two-tiered with an interbank market and a customer market. Although the distinction between interbank and customer segments no longer exists (Reuters and EBS Volumes combined account for no more than 10% of average daily turnover[3]) some FX participants continue to reference them, even though they may not offer the best rates.   This is no doubt a legacy of the days when EBS and Reuters rates were only available to interbank dealers.   

Due to the unavailability of a fair price benchmark, the earliest forms of FX TCA focused on very broad measures of execution quality (was the dealt rate within bounds of the day’s trading range?!) and used a trading platform’s estimate of mid to provide the ‘objective’ level.  But this doesn’t tell you where the bathwater was before the brick was added.

Breaking the circle.

The New Change FX solution was to interact with the market as a client would, asking for real, streaming, tradeable prices, but we are strange kind of client, a client that never trades. Because there is no trade history to the NCFX user profile, the bids and offers we receive have a 50% probability value, exactly as would be expected from an anonymised stream.

Finding a suitable fair price benchmark is fundamental to being able to employ more sophisticated techniques, such as the Implementation Shortfall method (I/S). First proposed by André Perold (1988) the method decomposes transaction cost into its component parts: explicit cost, delay, market impact and the opportunity cost of rejected or unfilled trades.

The advantage of this method is that it presents transaction costs from the portfolio manager’s or investors perspective. Furthermore, it does allow some form of market impact to be identified.

The disadvantage of the I/S method is that it offers a very crude measure of market impact. It conflates two distinct elements: The price impact of participating in the market and market drift, the direction of the market irrespective of our participation. Buying a rising market will certainly push prices higher, but would the market have risen in any case, without our purchase? And by how much?

A recent report by the ESRB[4]demonstrated that if customers have access to more “exact” measures of what the going price is, dealer profits will be lower. FX transaction costs can be quite low, if investors are discriminatory about how they approach FX.  On the other hand, undiscriminating customers’ FX costs can be very high.[5]What better way to be discerning than to compare actual quotes to normalised ex- ante aggregated data?

We are often led to believe that TCA presents a way of thinking about market impact, but actually it is the reverse. Thinking about market impact can help participants use TCA to be discriminating in their approach to their currency trading, which in turn will lead to lower costs. All trading is about either adding or removing bricks from the bath. Everything is market impact.

[1]Bruce Collins and Frank Fabozzi, A Methodology for Measuring Transaction Costs, Financial Analysts Journal 1991

[2]Grinold and Kahn, Active Portfolio Management, 1999

[3]BIS Market Structure Report September 2018

[4]ESRB working paper Discriminatory Pricing of Over-the-Counter Derivatives, December 2017

[5]HSBC vs Cairn Energy- a case where the customer, Cairn Energy contracted with a  bank to trade at the 3 pm fix rate, which HSBC traders were able to front run, moving the market in favour of the bank, to their detriment of their customer.

 

If you would like further information on how New Change FX can help you, please contact: Andrew Woolmer [email protected]