Thursday, 2 May 2013

Well known trading secrets become public


CME has been in the news as their order latencies and market data latencies don't align.

Major Stock Exchange Admits High Speed Traders Are Exploiting A Loophole In Their Computer Systems

Shock! Horror!


Not quite. Almost all exchanges operate the same way.  Nothing to see here.  Move along. What a ridiculous beat up. Not sure why the CME should be singled out in such an expose.

The true news in the article is really the public revelation that there are techniques to gain information from this fact.  Just as traders in investment banks can trade off flow, so can a market maker benefit from order executions on an exchange as they learn about prices "early" by their orders being hit.

It's always a bit disappointing, but also a relief, when well known trading secrets become public.  It is disappointing as it is always cool to know something not everyone knows, even if most do.  It is a relief as then you don't have to fear accidentally disclosing such information inappropriately. Still the safest policy is to say nothing unless you are promoting trend following.

As far as secrets go, the revelation from the Wall Street Journal was not a big one. The basic gripe captures a feature that has been part of markets for hundreds of years.

Let's review with the following poor graphic of the essential features of trading:



Traders or brokers send orders to the exchange and they receive fills back when they trade. The other market participants also get notified when a trade takes place by the market data feed.

Call the time the fill gets sent, t_fill. Call the time the market data gets sent, f_market. Now if t_fill < t_market, that is the fill notification is prior to the publishing in the market data feed, then the trader gets the knowledge before the other punters.

This is the way most market places, but not all, work. It is mainly a logical consequence of structuring systems.  The market data feed needs extra processing and thus market data is normally a bit slower than the steps involved in notifying the trader via the execution system.

If you're a market maker this is useful. When you get hit, you get a chance to hedge before it is widely known that you've traded. This is arguably a fair situation. The distinction the WSJ article is referring to is that you may use sacrificial orders a bit away from the bid and ask to get notification in advance of the market data feed that the market has moved. I've often referred to those types of orders as canaries. Such orders are like canaries in the coal mine that get snuffed out by gas giving you advanced warning that something is up. They are more useful when the fill to market data latency gap is “bigger” where bigger is a quite a complex balance of relativities of jitter, latency and load. They are particularly useful as an early warning system when an exchange suffers delays under load as those stress conditions often lead to greater advantage in fills but sometimes, though rarely, the fill systems suffer and the market data is more reliable. Canary trades are losers generally, unless the market has a high mean reversion propensity, and the role of the canary is to given you an informational benefit to reduce risk or take on correlated risk that may benefit.

The vast majority of exchanges work this way and traders have structured their market making, algo and high frequency strategies with this in mind for many years. A few exchanges, very few I might add, actually have faster market data delivery methods than fill methods. Nasdaq is one of those notably different exchanges where market data arrives first. Itch is faster than Ouch. That is, t_market for their ITCH feed is before t_fill on the order OUCH protocol with a suitable transport. The arrival of market data first usually complicates design as systems are typically built for the common case rather than relying on fill notification from public data.  In this regard Nasdaq can trip up a few newbie HFT market makers especially as they'll be slow to hedge if they're unaware.

Perhaps an ideal design for an exchange would be such that data only flowed one way on the order channel and all acknowledgements were from the public feed. This is because even if there was a similar speed to t_market and t_fill, there would likely be variation, as nothing is perfect, and thus some of the time there would be advantage to t_market or t_fill and it just amounts to how significant that may be. If you eliminate the feedback on the order channel then you solve the problem. This would also make the system design a bit easier too as it would be a simpler system overall.

I don't really understand why CME is copping flack over this. Many exchanges around the world have much worse situations in their technology and latency.  CME is a good exchange. In fact, the worse the exchange is, the better the situation for market making as long as there is reasonable execution jitter and infrastructure stability.  The benefits of trading on an exchange are improved if the exchange has more meaningful variations to benefit from.

Since the days of tulip mania and the day of Rothschild's pigeon latency[1] driven bond coup from Waterloo there have been advantages to be sought. The playing field is now more level that it ever has been.  There will always be an arb because the real world exists.  Nothing is perfect and nothing is perfectly omnisciently synchronous in the real world. It will never be a flat deck thanks to the speed of light and resource limitations, but nor does it need to be.  It just needs to be efficient, and it is.  There has never been a better time to trade as a retail participant or asset manager despite the scaremongering in the press.

Wireless broke as a big latency story last year even though it has been used for some years. Now canaries are a breaking story but have been used for many years as well. Hopefully further recipes in the secret trading culinary delights almanac will not be torn out and put into public view but all seems to be revealed eventually. Perhaps traders don't just need confidentiality agreements but a code of ethics similar to magicians where the tricks should never be revealed.

Twenty years ago I thought that in ten years the playing field would be level as all the latency games and strategies would be commonplace. I was wrong. I have similar thoughts again now but will be wrong again.  It is not wise to bet against progress. Fortunately for us trader types, the handbook of trading tricks, treats and strategies remains a thick one.

Happy trading.

--Matt. 


_____________________________
[1] Whilst the Rothschilds did use pigeons for latency advantage, it is disputed if pigeons were indeed used as part of the Waterloo Consol trade or if it was an impressively well organised fast courier system with boats, fleet feet, etal.