Why high frequency trading works well in today’s markets
By Irene AldridgeWhen all else fails, blaming high-frequency traders has become a sport of sorts among a certain set of investors and their brokers. High-frequency trading (HFT) has been fingered as the source of market crashes, price volatility, and general decline of the U.S. economy. Despite the negative press, high-frequency trading strategies continue to deliver strong positive returns well in excess of broader market indices, further fueling conspiracy theories and, now, subpoenas from the SEC.
In reality, the persistent success of most high-frequency trading is based not on some secretive large-scale manipulation of the markets, but on thorough understanding of what exactly makes the markets “tick,” literally. As I detail in my latest research, forthcoming in Frank Fabozzi’s and Harry Markowitz’ (the Nobel Prize Recipient) book, “Equity Valuation and Portfolio Management,” one of the key features of the markets used in high-frequency trading is the lack of correlations observed between any securities at very high frequencies. For example, daily correlations between the S&P 500 and stocks can reach 90%, implying that on the day-to-day basis, when the S&P 500 rises, so do other stocks; the same pattern persists when the S&P 500 falls – most stocks fall along. At the frequency of the so-called ticks carrying quote and trade information in the intraday data, however, the observed inter-security correlations are considerably lower, approaching 0 at second and sub-second time intervals. Correlation of 0 means that securities move independently of one another, allowing for return-generating portfolio diversification methods, no longer possible at daily frequencies due to high correlations observed then.
The lack of dependency between securities at high frequencies is not a recent phenomenon. In fact, this feature of the markets was first documented by T.W. Epps in (shocker!) 1979. This little fact is still obscure to many traders, yet has proven a boon to others who have used it to harness the powers of the classical portfolio optimization theory to generate abnormal returns. Mathematics, not conspiracy, are at the heart of profitability of high-frequency trading strategies.
Irene Aldridge is a quantitative portfolio manager at ABLE Alpha Trading, LTD. She is also a co-author of the Quant Investor’s Almanac 2011: A Roadmap to Investing (Wiley, 2010), an up-to-the-minute compendium of economic announcements upcoming in 2011, and a summary of responses of various equities, foreign exchange and other securities to different economic announcements. She will teach a brand-new course on high-frequency trading in New York and Chicago in September 2011. More information about the class can be found here: ev923.eventive.incisivecms.co.uk/. She can be reached at ialdridge@ablealpha.com
August 9th, 2011 at 5:04 am
My concern is not with with high-frequency trades, but with bid/asks/cancels that are several orders of magnitude greater than actual mercantile activity. Don’t these bid/asks/cancels actually lead the market without really making a market ?!?
The claim by the HFLL community has been they provide liquidity, but surely hundreds of algorithms suddenly all on the offers with tepid, if not phantom bids, can exaggerate the price discovery process.
August 9th, 2011 at 1:48 pm
Interesting, so if you have a portfolio on for fractions of a second your portfolio risk is reduced. Now what do you do when you want to change the composition of the portfolio and cannot guarantee access to liquidity in fractions of a second? You are stuck with a portfolio for more than a fraction of a second, whereupon the correlations do matter. Moreover, how do you measure correlation at the subsecond interval, using bid/ask/last price? I could generate any number of different ways of looking at price which may generate a variety of different correlations. Simply name dropping and referencing the word “math” doesn’t guarantee something has been thought through. Good luck with the book launch.
August 9th, 2011 at 2:13 pm
Dear Scott,
There has been much negative commentary alleging that high-frequency traders create noise by “stuffing” the exchange order books with limit orders they have no intention to fulfill. Well, academic studies from the times when HFT was hardly prominent, show that the number of limit bid and ask orders typically exceeds the number of market orders by a factor of ten (10), in perfectly normal market conditions. In other words, for every one transaction, it is normal on average to observe ten bid and ask orders that have been cancelled (see, for example, Bouchaud, Mezard, Potter (2002), downloadable here: http://www.ablemarkets.com/Research/BouchaudMezardPotters2002.pdf).
Alternatively, are you referring to the so-called stub quotes? Stub quotes are limit buy and limit sell quotes far away from the market that some practitioners say generate lots of noise and should be, therefore, banned. Several academic studies have come out on the subject over the years that all point to the same fact: stub quotes are a normal occurrence in the markets that can be predictably modeled. Furthermore, stub quotes are not malicious and play a normal useful role of allowing patient traders to trade at the prices of their choosing.
A couple of studies on stub quotes that readily come to mind: the same Bouchaud, Mezard, Potter (2002) (http://www.ablemarkets.com/Research/BouchaudMezardPotters2002.pdf), and Zovko and Farmer (2002).
August 9th, 2011 at 2:15 pm
A typo in the previous comment:
“In other words, for every one transaction, it is normal on average to observe ten bid and ask orders that have been cancelled”
should read as ” In other words, for every one transaction, it is normal on average to observe NINE bid and ask orders that have been cancelled.” (One bid or ask order out of ten was matched with the incoming market order.)
August 9th, 2011 at 2:19 pm
Davey,
“Good luck with the book launch.” — thank you.
“Simply name dropping and referencing the word “math” doesn’t guarantee something has been thought through.” — not sure how this is relevant.
You are correct in that correlations can be measured in a variety of ways, as my chapter in the book illustrates. Why don’t you read the chapter and then let me know if you have any follow up questions, as I am pretty sure the chapter may address many of your concerns.
August 9th, 2011 at 6:49 pm
This is a response I posted earlier today; it seem to have gone into moderation queue because of the embedded link, which I deleted in the copy below.
Dear Scott,
There has been much negative commentary alleging that high-frequency traders create noise by “stuffing” the exchange order books with limit orders they have no intention to fulfill. Well, academic studies from the times when HFT was hardly prominent, show that the number of limit bid and ask orders typically exceeds the number of market orders by a factor of ten (10), in perfectly normal market conditions. In other words, for every one transaction, it is normal on average to observe ten bid and ask orders that have been cancelled (see, for example, Bouchaud, Mezard, Potter (2002) — cannot post the link here).
Alternatively, are you referring to the so-called stub quotes? Stub quotes are limit buy and limit sell quotes far away from the market that some practitioners say generate lots of noise and should be, therefore, banned. Several academic studies have come out on the subject over the years that all point to the same fact: stub quotes are a normal occurrence in the markets that can be predictably modeled. Furthermore, stub quotes are not malicious and play a normal useful role of allowing patient traders to trade at the prices of their choosing.
A couple of studies on stub quotes that readily come to mind: the same Bouchaud, Mezard, Potter (2002) (cannot post the link), and Zovko and Farmer (2002).
August 9th, 2011 at 7:04 pm
[...] Hedge Funds | HedgeWorld | The Definitive Hedge Fund Community Good article from Irene Aldridge. (tags: highfrequencytrading) [...]
August 9th, 2011 at 10:46 pm
Dear Irene,
All above comments have related to equities or futures markets with centralised exchanges. The spot forex cash market has multiple marketmaker liquidity sources competing in a OTC best bid best ask enviroment and is not a centralised market. Have you researched or know of research of pricing models that extrapolate possible executed trade volumes of Prime brokerage Banks and ECN Bank aggregation platforms? With regards to the majority of daily global volumes the top 20 banks and top 7 ECN mulitbank aggregators by there stated monthly volumes attest for 90+% of the entire daily spot market volumes.
Ill be very interested in reading your reply
Thanks
Brad
August 11th, 2011 at 9:30 am
Dear Brad,
My first quant trading desk experience was in foreign exchange, so I am happy to contribute on the subject.
Forex data actually works better with equity HFT models than does equity data. This is likely due to the market inefficiencies that are inherent in forex: 1) lack of volume transparency you mention and 2) the presence of corporate traders who need to transact for business reasons (for example, to exchange cash flows resulting from foreign sales).
One way forex is different from equities at tick level is that forex data typically does not contain last trade information. Many HFT equity models use last trade data to make inferences about the imminent direction of the market price. One way to overcome this challenge is to assume that the last trade variable is the size-weighted average between the last best bid and the last best offer. Many practitioners find that this size-weighted quote average is a fair substitute for the last trade quote. A simple average of best bid and best offer sizes, following a change in either, can serve as a proxy for the last trade size value.
My chapter in the new book “Equity Valuation and Portfolio Management” (ISBN: 978-0470929919) discusses statistical properties of such averages and how they differ from regular last trade values.
Will be happy to discuss more offline.
Best Regards.