La complicidad voluntaria y activa; la objeción de conciencia del profesorado

When designing statistical experiments on ultra-high-frequency data, understand- ing of the institutional arrangements within the transactional market matters. Eurodollar futures are traded in two ways, first via open-outcry and second via an electronic centrally cleared market. The most actively traded Eurodollar futures have quarterly deliveries and extend out to ten years from contract rollover to maturity. These contracts mature during the months of March, June, September, or December, extending outward 10 years into the future. Since the Eurodollar time deposits cannot be transferred or used as collateral for loans, the Eurodollar future cannot be made by physical delivery at maturity – settlement is always to cash. CME is the electronic and open-outcry clearing house for these contracts. The exchange operates continuous electronic trading from 5 pm (Chicago time) Sunday to 4 pm (Chicago time) Fridays. Open-outcry runs from 7:20 am (Chicago time) to 2 pm (Chicago time) Monday to Friday. The electronic trading market is the CME GLOBEX (centrally cleared limit order book) located centrally in Chicago. Unlike energy futures with physical delivery, no Eurodollars are cleared

1_See

http://www.cftc.gov/idc/groups/public/@lrenforcementactions/documents/

legalpleading/enfbarclaysorder062712.pdfpp. 6–11 for the Barclays’ Bank report. Page 6

specifically refers to “Three-Month Eurodollar Futures” and this forms the basis of the CFTC’s remit for this section of the charge.

by over-the-counter (ClearPort) transactions, so my dataset is the complete his- tory of the inside-quotes. One interesting issue is that of “iceberg” orders – these are permitted for this contract and are labelled by CME as a maxShow option. Cleared iceberg trades appear on the trades tapes as a cleared trade (often within the inside-spread) with volume attached (200 contracts is a very common iceberg trade size) much larger than the surrounding quotes. Whilst some anecdotal evi- dence on the price impact implication of iceberg trades exists within out dataset, I leave a comprehensive analysis of these trades for future work.

Most Eurodollar Futures are settled quarterly; the top code is ED and this covers both open-outcry and electronic settlement1_{. The codes for Eurodollars are de-}

noted H (March), M (June), U (September) and Z (December). The last trading day is the second business day prior to the third Wednesday of the settlement month. Positions in excess of 10,000 contracts (recall that each contract is one million dollars) have limited trading (i.e. have position accountability) to the exchange and positions in excess of 850 contracts are reportable to the weekly CFTC commitment of traders report. As shown by the descriptives of my dataset, individual trades of in excess of 10,000 contracts are actually fairly common so the analysis of the Commitments of Traders (COT) report is relatively useful for this contract.

For each settlement quarter (H, M, U and Z), there are 10 maturities ranging from years ending in 0 to 9. Therefore 40 types of contract are simultaneously traded from EDH0 to EDZ9. The nearest maturity contract is the one delivering to coded quarter closest to the current date with a year ending in the same number as the present year. For instance, if today is March 11, 2014, then the nearest

1_{Note that Bloomberg tickers restrict the code ED to only electronically traded futures;}

my access to the historical CME tapes is by Thomson Reuters where GE reflects the electronic trades only.

delivery is EDH4 delivering on Wednesday, March 19, 2014. On Thursday, March 20, 2014, the EDH4 contract will then become the longest maturity contract, with settlement on Wednesday, March 20, 2024. The actual settlement is termed as being T + 2, i.e. the rate is the rate on the last business day two days prior to the third Wednesday (this must be a business day in both New York and London, so very occasionally this moves). Rates are rounded to the nearest 1/100 of a basis point and are settled in cash. CME timestamps the Eurodollar futures prices using millisecond stamps and the tapes retain these tick-times. Table 2.1

illustrates the EDH0 quotes data over 4 min around noon GMT (6 am EST) on January 3, 2006 with the GLOBEX millisecond timestamps included. More recently a monthly contract for the “near years” has been issued and traded at a substantially lower volume than the nearest maturity quarterly contracts. My analysis indicates that the pattern for these contracts is consistent with the quarterly contracts in spite of the lower volume of trades – results can be found in my online appendix.1

In Figure2.2(a) and Figure2.2 (b), I provide two daily snapshots of the best-bid best-offer, the mid-price and the executed trade prices for the March 2010 matu- rity Eurodollar future. Figure2.2(a) illustrates the trading in a 2.5-year maturity futures contract for September 10, 2007. We can see that the spread during the early part of the day is at nearly two basis points – very few transactions occur at this stage. As the day progresses, quote updates increase and so do the number of trades, some of which are outside the best-bid best-offer (the magenta trade markers). Quote updating at the inside-spread during this day is measured in seconds and even minutes. By contrast, Figure 2.2(b) shows that when the same coded contract is observed within one year to maturity, the update speed declines

1_{The online appendix contains supplemental information about this thesis. As it is too}

large to upload to the submission system, you can find it at https://www.dropbox.com/s/

Table 2.1: Inside Quotes Data Structure

Notes: This table displays the raw inside quotes data structure of EDH0 quotes on January 3, 2006. I gather Eurodollar futures contract Quotes and Trades bid price and volume, ask price and volume, and trades price and volume in milliseconds. The data are sourced from the Thomson Reuters Tick History database (February 2014).

to sub-seconds and the number of transactions becomes higher. To my eyes, the spread does not vary much over the day, however we can see that quoting activity and density of transactions appear concomitant through the density of the trade markers. However, the actual variance and kurtosis of the spread over this day is far higher than is visible in the simple plot. There are a number of trades outside the best-bid best-offer, in Figure 2.2 (b) but the proportion is much lower than in the first trading epoch.

The extent of algorithmic trading is relatively unknown in this market. However, two facts support the conjecture that algorithmic trading of these contracts is both possible and present. The first rationale is the Globex market platform itself. The fastest access is via the GLink system which provides co-location options for traders and as such there is no technical reason to prevent high- frequency algorithmic trading. The second observation comes from the data. For contracts near maturity there is evidence of level of quoting and re-quoting at speeds beyond the reaction capacity of a human being (less than 200 milliseconds).

00:00 01:00 02:00 03:00 04:00 05:00 06:00 07:00 08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 00:00 95.38 95.4 95.42 95.44 95.46 95.48 95.5 95.52 95.54

Trade Inside BB−BO Trade Outside BB−BO Asks

Bids Mid Price

(a) Trades and inside quotes in EDH0 (March, 2010 maturity) on September 10, 2007

00:00 01:00 02:00 03:00 04:00 05:00 06:00 07:00 08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 98.2 98.3 98.4 98.5 98.6 98.7 98.8 98.9 99

Trade Inside BB−BO Trade Outside BB−BO Asks

Bids Mid Price

(b) Trades and inside quotes in EDH0 (March, 2010 Maturity) on June 5, 2009

Figure 2.2: Trades and Inside Quotes in EDH0 (March, 2010 maturity)

Notes: Figure (a) is trades and inside quotes in EDH0 on September 10, 2007: the best bid, best ask, the middle price and the executed price. The total volume on this day is $8.113 billion with 63,520 observations of asks and 63,534 observations of bids. Total trades observations are 1,016 with 14 trades outside the best bid and best ask. NOTE that the images are enlarged for review purposes only and are for illustrative purposes only. Figure (b) is trades and quotes in EDH0 on June 5, 2009, which has the largest daily trade observations for EDH0 with $449,988 billion volume of trades. Total trades observations are 30,030 with 34 trades outside the best bid and best ask. 435,611 asks and 435,544 bids are observed on this day. For data purposes I treat a day as being 00:00:00.001 CST to 11:59.59.999 CST, however, my plots follow the scientific literature and I plot the day from 00:00:00.001 UCT to 11:59.59.999 UCT.

I find a large number of reacting bids/asks that occur at “identical” millisecond stamps, indicating that the trade initiation is in fact faster than one millisecond. I will now review some of the current literature relating to algorithmic trading at high frequencies.

As a subset of algorithmic trading, high-frequency trading is usually defined as computer-based transactions that are designed and initiated at speeds faster than any human being could manage (sub-200 ms is a standard benchmark). The rise of HFT follows two developments: first, the relatively steep drop in the cost of high performance computing environments; and second, competition between exchanges resulting in increased speed of execution and far lower latency times.