If you think algos gone wild in stocks is bad, just wait until you see what happens when the same feedback-loop generating robots start frontrunning and churning all cotton, sugar, and other commodity contracts. According to this trader, this has already happened. Next up: plunging liquidity, and surging volatility, just in time for commodity prices to find that extra computerized "oomph" as they explode in expectation of Bernanke's reflation experiment gone wild to blow all fair value concepts to smithereens.
An Angry Sugar Trader Shares His Frustration With The Incursion Of HFT Algos In The ICE
Submitted by reader Menji
In the days before electronic trading, when commodities were traded open-outcry in trading pits, floor brokers kept spread and flat-prices in line as part of their day’s work – offering and bidding one month against another depending on the spread orders they were working. Some of these guys had almost unbelievable skills of mental arithmetic, bidding and offering across the whole board as the flat price and the spread structure fluctuated.
Since the advent of electronic trading, it has been the job of a computer algorithm to generate flat price bids and offers using the spread structure and generate spread bids and offers using the flat price structure. The algorithm is generally termed an “implied engine”, and it does the job of the old floor brokers, although it does it faster and more efficiently.
ICE (Intercontinental Exchange, self described “leading global exchange and OTC market operator”) has a web-page advertising its “multicast price feed” implied engine upon which the implied engine’s benefits are listed:
- improved price discovery from implieds directly from the market
- availability of implied pricing much further out the curve
- more trading opportunities
- greater market transparency
- improved market depth and liquidity
- more efficient hedging of risk
- increased probability orders will be executed
However, although ICE claims that its implied engine gives “improved market depth and liquidity”, in March 2009 ICE Futures US decided to turn off its implied engine on the No11 sugar contract “to improve liquidity [...] and to attract new traders”.
Hey, hang on a sec...
If the “multicast” implied engine is that good for liquidity, how come someone decided to just switch it off on the sugar market, to “improve liquidity”?
Just think about this for a second. This breathtaking display of shameless hypocrisy, writ large in html, is *still* up there on the ICE website. It’s been up there for months, which speaks volumes or, if not, at least a few words: ICE simply doesn’t give a fuck.
It is very difficult to imagine how switching an implied engine off can improve liquidity. Take a closer look at the the screen shot at the end of this article, and you’ll see that the bid/offer spreads on most of the forward contracts are enormous compared to those on the nearby contracts. This is a direct result of their still being no implied engine on the No11 contract. What is less difficult to imagine is who the “new traders” referred to in ICE’s release are: the algo traders were being invited to come and play in the No11 sugar market.
The consequences to ICE’s decision to switch off the implied engine were as expected: The sugar market began to immediately suffer from the lack of what ICE’s multicast price feed” implied engine provides to other markets.
It now had to put up with:
- obscured price discovery
- no availability of implied pricing anywhere on the curve
- fewer trading opportunities
- reduced market transparency
- reduced market depth and liquidity
- less efficient hedging of risk
- decreased probability orders will be executed
Other consequences were greatly increased volatility as algo trading systems unleashed their orders into a relatively small market, and increased exchange revenue for ICE and its shareholders as the algo traders fed their orders directly into the exchange servers. Liquidity plummeted, as many of the market participants who traditionally provided it (market makers and day-traders) packed up in disgust and went elsewhere, sickened by the random walk behaviour generated by computers which had started to push the market the range of an old, pre-algo, day in the space of seconds.
Liquidity is NOT the same thing as volatility, no matter what HFT apologists tell you. Insane volatility of the type generated by HFT “traders” is good for only exchange fee revenue and, usually, the fuckheads running the computers (although occasionally they get their just deserts).
In October 2009, NYSE Liffe decided to turn off the implied engine on their No5 white sugar contract, in an attempt to lure algo traders into this much smaller contract. What was immediately apparent in this experiment was that the newly-arrived algo traders began running their own implied engines, which would do exactly the same job as that done by traditional, exchange-based implied engines, except with a “haircut” cost of $0.20/tonne to whoever traded spreads against it – more revenue for the exchange, more revenue for the algo traders, more costly execution and less transparency for traditional users, and no improvement to liquidity whatsoever.
Then, on 5th January 2010, there was a sugar price spike, which was reported by the Financial Times. The market was already highly volatile, trading at 30-year highs, and (presumably) several buy orders hitting the market at more or less the same time caused sugar to spike from around 28.00 to 29.50 in less than 90 seconds.
ICE invoked its recently-issued “short-term price spike” rule, and simply cancelled all the trades above 28.90. If you look on a chart, that’s the high, but it certainly isn’t the highest it traded that day.
What happened on sugar on the morning of 5th January 2010 would NEVER have happened had the implied engine been switched on. How can you have March No11 trading to at least 325 over the next month on the board when the Mar/May spread was trading around 160 points? There was plenty of (unfilled) selling above the market down the board which a) would have been filled and b) would have added sufficient liquidity to prevent such as spike had the implied engine been functional.
Thousands upon thousands of tons of producer selling was left unfilled; day traders sold near the top and bought back on the collapse only to find that their sales no longer existed, and that their buy-backs were now naked longs way above the market. ICE's claim that the lack of an implied engine somehow promotes liquidity was finally shown for what it really is - corporate doublespeak whose sole aim is a shallow attempt to cover the fact that ICE’s behaviour serves purely to line the pockets of the exchange, its shareholders, and a horde of algorithmic traders at the expense of market transparency, price discovery, and the needs all other market participants.
What happened on sugar that morning - indeed, the need for ICE to invent "price spike" rules in order to deal with situations entirely due to the lack of liquidity that they themselves have helped create - should be a warning to exchanges which sacrifice market efficiency for the sake of exchange fee revenue, and a heads-up to the bodies which oversee their activities.
Interestingly, a mere week later, on 13th January 2010, NYSE Liffe decided to switch its implied engine back on for the No5 white sugar contract, saying that “having implied prices will help to lower some of the risk of short term price spikes”. NYSE Liffe hadn’t even *had* a price spike, but what they saw happen to the ICE No11 market was enough for them.
But, as I said earlier, ICE simply doesn’t give a fuck. Its shareholders are happy, the algo thugs are happy, and those unfortunates who need to use the market for genuine hedging purposes don’t count. It doesn’t look as though the CFTC gives a fuck, either.
This has to change.
APPENDIX – how spreads and futures work
You can trade sugar futures for four delivery months a year, going to around three years into the future. So the market, as represented on electronic trading screens, market reports and financial newspapers looks something like this:
BID ASK HIGH LOW LAST
MAY10 19.11 19.12 20.07 19.06 19.11
JUL10 18.50 18.44 19.28 18.31 18.32
OCT10 17.45 18.11 18.65 17.79 17.83
MAR11 17.38 17.57 18.02 17.27 17.38
MAY11 16.60 17.00 17.28 16.67 16.70
JUL11 16.12 16.45 16.75 16.20 16.25
OCT11 15.92 17.00 16.45 15.88 15.98
MAR12 15.45 15.75 15.88 15.30 15.45
MAY12 15.35 15.65 15.55 15.14 15.46
JUL12 15.30 15.70 15.50 15.40 15.44
OCT12 15.45 15.55 15.50 15.25 15.45
(THESE ARE ACTUAL PRICES FROM THE ICE SUGAR No11 CONTRACT AT CLOSE OF BUSINESS ON 11 MARCH 2010)
Each delivery period is a market in its own right. On the screen shown above, May 2010 is worth roughly three quarters of a cent per pound more than July 2010, but the differential between the two delivery periods isn't carved in stone - both contracts have their buyers and sellers, and each moves according to its own order flow. However, the relationship between the various delivery periods is a closely-followed and much-traded aspect of the market.
Producers will roll short hedges from one delivery month to the next depending on the timing of the crop, and their valuations of their merchandise. A steep enough carry could pay a producer to leave his sugar in a warehouse until later in the year. A steep enough backwardation (nearby month at a premium to forward) could pay a producer to bring forward sugar he intended to deliver later. Depending on the relative structure between the various delivery months, consumers can either decide to bring forward purchases intended for later, or squeeze their pipelines and roll prompt purchases further down the board. Speculators can bet on the fact that, no matter what happens to the market price, such and such a delivery month will be worth more (or less) compared to another delivery month further down the futures board.
All this is known as spread trading: a spread is the differential between one delivery month on the board and another, and producers, consumers, trade houses and speculators trade an awful lot of them every day, buying one delivery month whilst simultaneously selling a second. As I write this, ICE No11 sugar has traded a total of 44580 contracts in 6 1/2 hours business of which over 7000 were spread trades. Almost 20% of the total volume traded on the first and most-traded month on the board, May10, traded against something else on the board.
It is easy to imagine a spread matrix (every trader has one displayed on his screen) where the relationship between each delivery month on the board is shown. Part of the one I have on my screen at present looks something like this:
JUL10 OCT10 MAR11 MAY11
MAY10 0.85/0.86 1.38/1.41 1.87/1.92 2.56/2.67
JUL10 0.53/0.54 1.01/1.06 1.70/1.80
OCT10 0.49/0.51 1.19/1.25
Looking at the top left-hand corner of the matrix above, we can see that someone is willing to simultaneously buy May 2010 and sell July 2010 at a differential of 0.85 cents/lb – whatever they pay for the May contract, they’ll sell a July contract 0.85 cents/lb cheaper. Similarly, someone is willing to sell May 2010 and buy July 2010 at 0.86 cents/lb.
It is important to understand the connection between these spread quotes and the flat-price values of each individual delivery month. Again, taking our example of the May10/Jul10 spread, if the May 2010 contract is quoted
19.66 bid/19.68 offered, then the buyer of the spread (the buyer May10 seller Jul10 at 0.85) should be willing to offer July 2010 outright at 18.83 – he can buy May10 at the offered priced of 19.68 and sell July at 0.85 cents/lb discount to this to fill his order. Similarly, the seller of the spread should be willing to buy July 2010 at the bid price on May minus the 0.86 cents/lb he wants to sell his spread at. So, even if no one is trading July 2010 outright, as long as May is quoted 19.66/19.68 and the May/Jul10 is 0.85/0.86, July should be quoted 18.80/18.83.
Now, let’s look at this the other way round. Imagine that no one is offering the May/Jul10 spread. It’s still 0.85 bid, but no one is willing to sell it as a spread. Imagine that May 2010 is still quoted 19.66 bid/19.68 offered but that, this time, there *is* an outright quote on July 2010. Let’s say it’s 18.80 bid/18.83 offered.
MAY10 19.66 19.68
JUL10 18.80 18.83
You could simultaneously buy a May contract at 19.68 and sell a July at 18.80, which means that you could buy a May/Jul10 spread at 0.88. So, although no one may be willing to sell May/Jul10 as a spread, the quote should still be 0.85/0.88.