July 24th, 2019: Euribor Z0H1 spread.

A strip of Euribor futures are derivatives on sequential forward starting inter-bank offered rates. As would be expected, each Euribor future is highly and consistently correlated with its adjacent contracts.  For example, a Z0 future (expiring in December 2020) might be ≈98% correlated with the H1 (March 2021) contract.  

This stable relationship between adjacent contracts make spread trading in the form of calendar spreads (future1 -future2) or butterfly’s (future1- 2x future2-future3) very popular as a low volatility trading strategy albeit often requiring high leverage in order to generate enough return.

It is difficult to imagine how Euribor futures could ever become negatively correlated to each other but as is often with risk, reality can be stranger than theory. This happened once in my trading lifetime in 1999 due to a phenomenon subsequently called the millennium fly. This was when the Z0 (December 2000) futures sold off sharply (implied forward rates increased since stir futures are quoted on a 100-r basis) whilst the adjacent U9 (Sept 1999) and H1 (March 2001) contracts continued a upward trend (implied forward rates decreased) in line with other interest rate related product. This effect was later attributed to a US bank allegedly being refused access to funding markets in the new millennium for being deemed to be non-Y2K compliant.  

This effect of Euribor futures becoming negatively correlated has happened again for the first time in 20 years (to my knowledge). During the 2019, up until 24th July, the Z0H1 (December 2020-March 2021) spread traded in a range of 0.02 to 0.07 with an average daily spread volume of ≈4000.

During normal trading hours on 24th July, the Z0H1 spread had traded between 0.02 and 0.03, but from 17:25 hrs to 17:37 hrs on the same day, blew out to 0.14 and then traded back down to 0.03 by 18:37pm on a volume of over 90,000 contracts. This was driven by aggressive buying of ≈60,000 Z0 contracts from a price of 100.565 to 100.640 whilst the H0 sold off from 100.535 to 100.510 on volume of ≈28,000 during the same time period (17:25 hrs to 17:37 hrs).

The charts below graphically display the moves in the both the spread and the individual futures contracts, highlighting the temporary enforced negative correlation between Z0 and H1. Charts by Reuters Eikon.

Z0H1 Spread: 24/07/19 Tick data, Price and volume
Z0 (top) and H1 (middle) futures:
24/07/19 1 min data, Price and volume (Z0 bottom)

This was probably a very expensive 12 minutes from the perspective of proprietary traders and the instigator of the trade. The adverse mark-to-market on leveraged short Z0H1 positions might have led to proprietary traders buying back short spreads to reduce inventory but thereby exacerbating the move. Whoever or whatever was behind the trade would have ultimately lost money since the spread and the individual component futures contracts subsequently normalised back to levels seen earlier in the day. Given that none of the trades violated the Exchange’s tolerances and circuit breakers, all trades stood, and none were cancelled.

The move defies logic. The next day (25th) was an ECB policy meeting day but rates were unchanged and there was no market moving news regarding QE policy. The move was measured in minutes, not milli-seconds, perhaps ruling out a fat finger or flash crash. Someone might have a very specific idea regarding European rates in Q120, but it was awful trade execution to achieve it. Market manipulation? Isn’t the idea the make money from that. This trade would have lost the instigator money. Focus will probably fall on an algorithmic error, but the reality is that we will probably never know for sure.  Maybe it’s a sign of age but when things like this happen, I’m glad that I’m no longer in the game!  

The demise of stir futures?

Stir futures are, of course, futures on short term interest rates, primarily IBORs (interbank offered rates). The Eurodollar and Short Sterling are based on LIBOR (London Interbank Offered Rate) and the Euribor is named after its underlying reference rate – EURIBOR (Euro Interbank Offered Rate).

Because of the LIBOR/EUROBOR fixing scandals, plus operational aspects like the lack of volume and submission and fixing methodology, the FSB (Financial Stability Board) published a report in 2014 on “Reforming Major Interest Rate Benchmarks” which provided the basis for the FCA’s (Financial Conduct Authority) decision to sustain LIBOR until the end of 2021. After that, panel banks will no longer be required to submit LIBOR rates, which likely lead to the demise of LIBOR as a reference rate and remove the demand for derivatives based upon it.

As an alternative, the following have been adopted as reference rates

USD: Secured Overnight Financing Rate (SOFR)

EUR: Euro Short-term Rate (ESTER)

JPY: Tokyo Overnight Average Rate (TONAR)

GBP: Sterling Overnight Index Average (SONIA)

CHF: Swiss Average Rate Overnight (SARON)

SOFR is a fully transaction-based reference rate, derived from a composition of repo rates – General Tri-Party (Bank of New York Mellon), DTCC cleared General Collateral and Bilateral FICC cleared. SOFR daily trading volume is around $800 B which is about 1500 times the daily LIBOR transaction volume.

In the US, SOFR futures were introduced by CME in May 2018 and because they are based on a secured financing rate (in contrast to LIBOR being based on unsecured financing), the SOFR futures trade at a higher price (lower rate) that Eurodollar futures.

Source: Bloomberg

Volumes are increasing

Source: Bloomberg

And a developing market in inter contract spreads

Source: Bloomberg

Based on a presentation by Numerix

Causation between 3M Euribor fixings and Euribor futures rates

Are changes in Euribor fixings correlated with movements in the Euribor implied forward rates?

Intuitively, we might think so but empirical evidence suggests otherwise. Given a data set from January 1999 to May 2017, there were 2740 days when a change in the Euribor fixing was replicated with a similar movement in the implied forward rates of the white pack and 1964 days when they were not. This results in a low correlation of just 17.6%. Breaking the data into smaller 5-6 year buckets does not demonstrate causality between the direction of Euribor fixings and movements in the Euribor futures implied forward curve.

causal

Policy rate tightening

With all this talk about Fed tightening in the autumn with UK rises to follow in early 2016, it might be useful to remind ourselves of the strong inverse relationship between stir futures and policy rates. Of course, stir futures are LIBOR linked derivatives but LIBOR is very closely correlated with policy rates making stir futures ideal for the systematic trading of policy rate changes.

The first chart shows the front month continuous short sterling contact (orange RHS) and the BOE Base Rate (purple LHS) over the last 20 years.

UK Source: Reuters Eikon

The inverse relationship is clear and most importantly shows that when policy rates start changing, stir futures start trending.

Same with the US. (Eurodollar front month continuous in purple (LHS) and Fed Funds Effective Rate in orange (RHS)

US

Source: Reuters Eikon

It can also be observed that the US often acts preemptively to the UK. When America sneezes, the UK catches a cold.

 

The year -end turn effect

The year-end effect in STIR futures is a legacy from the 1980’s. There are often high borrowing requirements at year end as Banks look to bolster their cash reserves at the end of a fiscal year or quarterly period. This requirement and the fact that the days straddling the end of one year and the beginning of the next fall in the holiday season and create an overnight borrowing period that can be two or four days in duration due to bank holidays.

In December 2013, the year end turn effect would have been muted since money could have been borrowed on Tuesday 31st December 2013 and returned on Thursday 2nd January 2014. This would have been just two days but if those days had straddled a weekend, it could have been four. It can have an effect on STIR futures because year end is included in the forward period covered by the December contract.

2

 

If the forward deposit period rate was 1%, then the futures implied rate should be

3

 

However, if rates jumped to 1.5% over the turn, the futures implied rate would be:

4

 

Clearly this is insignificant when the turn premium is 0.50% and the turn period two days so it is off most STIR trader’s radar apart from the most exceptional periods like run up to December 1999 when the markets were spooked by the millennium bug (Y2K) fears. This was most clearly observed in the Sterling Dec butterfly.

5

 LIFFE Short Sterling U9Z9H0 butterfly Apr 98 to Sep 99 (price differential)

Worries that that non Y2K compliant Banks would be refused year end funding caused the Z99 future to fall sharply relative to the surrounding U99 and H0 contracts, causing the usually staid butterfly spread to blow out by around 80 ticks (bps). This also happened to Eurodollar and Euribor and was probably a once in a generation occurrence but watch out for potential repeats driven by, for instance, rumours of Banks failing regulatory requirements.

Looking at the latest set of Euribor Dec butterfly’s in the chart below, there are no discernible Dec premium reflecting year-end turn premiums, mainly because Central Banks are very adept at financing these days and post the financial crisis, Banks have relatively easy access to central bank funding

6

 LIFFE Euribor Butterfly’s (Dec fly’s are coloured and surrounding fly’s are greyed for contrast)

However, despite the fact that the year-end turn effect seems to be diluted by the relatively small turn premiums in Dec fly’s these days, there is another aspect to consider relating to year end liquidity:

Stricter liquidity regulation caused a significant reduction in the liquidity available in the overnight market during December 2013, which fuelled sharp upward pressure on the overnight cost of cash. Having started December at 11.2bp, EONIA moved up to 20.6bp on 17 December, then declined a little to 17.1bp on 24 December before spiking in the last week of the year to 44bp in an illiquid market.

This effect can be observed on changes in the Euribor strip.

7

 LIFFE Euribor Strips November to December 2013

The strip yield increases, almost in parallel as condition tighten in the EONIA market during December and the causal link is observable between 3M OIS and Euribor Z3.

8

Indeed, there seems to be a general pattern of tighter overnight rates in the euro zone and a quick study reveals that in seven out of the last ten years (2004 to 2013), Euribor futures closed lower (rates higher ) at the end of December compared to where they started December.

Interest rate change probabilities

STIR futures are not the ideal instrument for determining policy rate change probabilities since they are LIBOR or EURIBOR linked derivatives. These 3 month interbank fixings are not the same as policy rates and there is often substantial disparity between them. However, STIR futures are often used to speculate on or hedge against changes in official interest rates due to their liquidity and transparency since all short rate products will generally reflect changes in policy rates.

There has recently been much in the UK press about the Bank of England’s (BOE) forward guidance policy.  Market expectations of base rate increases have been pulled forward in response to improving economic conditions, particularly the rapid decline in unemployment.

It is possible to determine the probabilities of base rate changes within a particular time frame by the use of a specific instrument called a Meeting Dated Overnight Index Swap. 

IR prob 3

 The table below shows the BOE MPC meeting dates for the next year and corresponding meeting dated OIS rates.

table2

 

Example

The rate for the MPC meeting on 8th May 2014 is 0.499%. This rate is the market rate of a one month forward starting OIS, starting on 8th May and maturing on 5th June. 0.499% is the fixed OIS rate quoted against receiving compounded daily SONIA for one month between these dates. If base rates were increased on or before 8th May from the present level of 0.5% to 0.75%, then the OIS fixed payer would be in the money, paying just 0.499% for a month against receiving SONIA which would have increased to around 0.75%.

The probability of a 0.25% (25bps) increase in the base rate on or before 8th May would be calculated as:

IR prob

This is basically saying that the market puts 0% probability of a 25bps base rate increase occurring by May 2014. This probability has increased to 80% by February 2015 reflecting the current market consensus for the first 25bps increase in base rates to occur early in Q2 2015.

The graph below illustrates the term structures of MPC dated OIS, market consensus base rate forecast and the LIFFE Short Sterling STIR futures strip (expressed as implied forward rates).

graph2

Term structures of UK BOE MPC dated OIS, market consensus UK BOE base rate forecasts and the LIFFE Short Sterling STIR futures strip (expressed as implied forward rates): Late January 2014

The Short Sterling futures should not be interpreted as being more aggressive in terms of interest rate expectations, the difference between the rates just reflecting their  types: forward SONIA OIS as compared to  forward 3M LIBOR. However, the gradient of the Short Sterling term structure suggests uniformity with the OIS expectations, suggesting equilibrium between products in terms of future interest rate expectations.

From a traders perspective, if rates were thought to increase sooner and faster than expected, then the curve would steepen (buy Short Sterling calendar spreads) and if rates were thought to be on hold and increase at a much slower rate than forecast, then the curve would flatten (sell Short Sterling calendar spreads). Watch this space…

Understanding open interest

Open interest (OI) is the total number of outstanding futures contracts that are not closed or delivered on a particular day. It is a useful measure of liquidity of in a futures contract and many technical practitioners consider it a useful barometer for trend detection.

The following table illustrates how open interest is determined.

table

Open interest comes in two forms, individually per contract (e.g. H14) or for all contracts (e.g. all Euribor futures). The open interest for all contracts, often termed Market Open Interest is a useful barometer of liquidity and activity within a product.  Technical practitioners generally expect open interest to increase when a trend is establishing and reduce when the trend is coming to an end.

The charts below shows market open interest and price (continuous) for the Eurodollar and Euribor.

1

Eurodollar market open interest (LHS) and close (continuous contract) (RHS) 2002 to 2013 

2

Euribor market open interest (LHS) and close (continuous contract) (RHS) 2002 to 2013

There does not seem to be an obvious link between increasing open interest and price action but it might be useful to keep a watch on contract levels to gauge any potential flows.

LIFFE’s STIR matching algorithm

Matching algorithms are the mathematical equations within the computer code of the Exchange matching engine, known as CONNECT on LIFFE.  These algorithms allocate incoming market orders amongst the resting order depth using a methodology somewhere between a pure pro rata and pure time priority, also known as first in-first out (FIFO).

LIFFE operates two matching algorithms for its STIR futures: one for Euribor and the other for Short Sterling and Euroswiss. They combine both pro-rata and time priority but the Short Sterling /Swiss version has a higher bias towards time priority.

Both algorithms have the same mathematical structure. The current algorithm for the Euribor is:

a

The current algorithm for Short Sterling and Euroswiss is similar but to a higher order*.

Capture

*The Time Exponent employed within the Gradual Time Based Pro Rata (GTBPR) matching algorithm will change from 4 to 2 effective on Monday 3 December 2018

An easy way to consider these algorithms is as if the order is set to 1 (i.e. there are no powers). In this case, any incoming order of reasonable size is split amongst resting orders in an equal percentage weighting.

Example

The table shows an incoming order of 200 lots into a market with a resting order depth of 2405 lots made up of 10 individual orders of different size and ranked by time priority.   It could be considered as a market order to sell 200 lots at best when the market is 99.00 bid for 2405 lots. The 2405 lots on the bid are made up of 10 separate orders all bidding the same price but entered at different times. The order with the oldest timestamp will be T1 and the newest T10.

algo1

It can be seen that all orders (T1 through to T10) get 8% of their resting order size. In absolute terms, the larger the bid, the more futures would have been bought.  It might be noted that only 194 futures were transacted and so the algorithm would then make another pass to allocate the remaining 6.

This is an example of pure pro rata and would advantage those over quoting by flashing large orders into the market depth momentarily before the order hits the market and would provide no advantage to consistent resting liquidity providers.  To circumvent this, the actual algorithms operate to a power; two in the case of Euribor and four in the case of Short Sterling and Euroswiss.

The same order and market depth then experiences a very different allocation. The Euribor algorithm allocates on a broadly linear but decreasing profile (in percentage terms) whereas the Sterling/Swiss algorithm would allocate a much higher proportion to the first three time priority orders (T1-T3) with sharply decreasing density thereafter for the remainder. In both cases, orders towards the back of the queue get no allocations.

algo2

In graphical terms the effect is more observable.

algo3

Both algorithms, but particularly the Sterling one, benefit orders with time priority; the idea being to promote and encourage resting liquidity. Traders are encouraged to supply liquidity early into price points.

The spreadsheet containing both algorithms is available to download here. Now you can figure out for yourself why your fills might not be what you hoped for and perhaps find a way to improve your allocations.