Here’s a funny thing about algorithmic trading. Most of it depends on statistical arbitrage, which in turn depends on volatility to detect price variations to benefit from.
In which case the following abstract from a paper by Andrew Pole, a Managing Director at TIG Advisors, should be noted with interest:
Statistical arbitrage suffered a performance famine over 2003-2005 in part driven by the rise of the trading algorithms developed by brokerage houses. As other contributing causes exited the field and trading algorithms gained ever wider patronage, statistical arbitrage defied the lament, “Stat.Arb. is dead, killed by low volatility,” and returned to splendid performance in 2006. The resurgence will be sustained by emergent market properties from algorithmic trading.
Emergent market properties, eh? Yes, we’re very curious.
The key point though is clearly that volatility is good for statistical arbitrage, as is not many other people doing the same thing. It’s therefore very much suited to investment classes where most market participants will not have the means or interest to develop comparative quant-based algorithmic strategies.
Successful algorithmic trading meanwhile also depends on the development of market-neutral strategies that achieve success via superior data mining skills, as discussed by Sal L. Arnuk and Joseph Saluzzi in their Themis Trading paper.
One area which therefore particularly lends itself to the practice is the fast growing world of exchange-traded-funds. Not only are most ETFs traded on suitably computerised exchanges, they offer two huge opportunities. Firstly, they’re populated by “uninformed money”. Secondly, their model depends on reverting to mean — a hugely attractive behavioural quality to any statistical arbitrageur.
The concept is nicely explained in the following abstract from a paper entitled “Dynamic modeling of mean-reverting spreads for statistical arbitrage”:
Statistical arbitrage strategies, such as pairs trading and its generalizations, rely on the construction of mean-reverting spreads enjoying a certain degree of predictability. Gaussian linear state-space processes have recently been proposed as a model for such spreads under the assumption that the observed process is a noisy realization of some hidden states.
Real-time estimation of the unobserved spread process can reveal temporary market inefficiencies which can then be exploited to generate excess returns. Building on previous work, we embrace the state-space framework for modeling spread processes and extend this methodology along three different directions. First, we introduce time-dependency in the model parameters, which allows for quick adaptation to changes in the data generating process.
Second, we provide an on-line estimation algorithm that can be constantly run in real-time. Being computationally fast, the algorithm is particularly suitable for building aggressive trading strategies based on high-frequency data and may be used as a monitoring device for mean-reversion.
This paper on dynamic modelling of mean-reverting spreads for statistical arbitrage, meanwhile, adds even more insight:
The underlying assumption of pairs trading is that two financial instruments with similar characteristics must be priced more or less the same. Accordingly, the first step consists in finding two financial instruments whose prices, in the long term, are expected to be tied together by some common stochastic trend. What this implies is that, although the two time series of prices may not necessarily move in the same direction at all times, their spread (for instance, the simple price difference) will fluctuate around an equilibrium level. Since the spread quantifies the degree of mispricing of one asset relative to the other one, these strategies are also refereed to as relative-value.
If a common stochastic trend indeed exists between the two chosen assets, any temporary deviation from the assumed mean or equilibrium level is likely to correct itself over time. The predictability of this portfolio can then be exploited to generate excess returns: a trader, or an algorithmic trading system, would open a position every time a substantially large deviation from the equilibrium level is detected and would close the position when the spread has reverted back to the its mean. This simple concept can be extended in several ways, for instance by replacing one of the two assets with an artificial one (e.g. a linear combination of asset prices), with the purpose of exploiting the same notions of relative-value pricing and mean-reversion, although in different ways; some relevant work along these lines has been documented, among others, by Montana et al. [2009] and Montana and Parrella [2009], who describe statistical arbitrage strategies involving futures contracts and exchange-traded funds (ETFs), respectively.
Did you get that? The key is finding two instruments whose prices in the long-term are expected to be tied together. A little bit like ETFs and their underlying securities.
We went on a little data-mining field trip of our own and discovered some of the following job adverts on the webosphere:
Statistical Arbitrage Hedge Fund which trades ETF products is hiring for a ‘superstar’ quant to join their team. You will have up to 2 yrs experience in the Statistical Arbitrage space although Options Market Making experience will also …
….You will work directly on the trading desk which offers a collegial atmosphere. A strong technical background is essential & you will be tested equally on your programming skills and your math and finance knowledge. You will ideally have a PhD in a hard science from a top University and will have applied your research skills in data cleansing to generate trading signals. Candidates must have quant research experience for ETF trading strategies. If you are looking to join a firm which will allow you to work on researching your own trading strategy and eventually running your own book, send your resume…
And then we found this on Seeking Alpha:
One favorite investment vehicle for the stat arbs are the exchange traded funds (ETF), according to the head of one proprietary trading desk. The stat arbs execute more than 50 percent of the volume in the ETF that is based on the S&P 500, the SPDR. “They’re taking these orders and just shredding them..
We’re not PhD quant-level statisticians here on FT Alphaville, so we won’t be able to explain exactly how they do it; but all those peculiarities in the USO and UNG ETFs do suddenly seem to make a lot of sense. This is especially so if you consider statistical arbitrageurs look for pairs of securities that consistently revert to mean, but offer large amounts of “relative-value” in the trading day.
Relative-value would have been incrementally increased in commodity ETFs by the contango and the volatility stemming from a descending price. So yes, commodity ETFs were not responsible for upswings in the price in 2008! The increased flows they attracted counter-intuitively this year, meanwhile — in a descending market– may have had nothing to do with retail investors being keen to take a view on the underlying market. The flows may have stemmed from huge arbitrage opportunities presented to statistical arbitrage firms via the contango and price volatility.
In which case, if the structurers of these ETFs are aware of the problem, marketing these securities forcefully to retail investors does indeed raise some important questions.
Related links:
Electronic trading and commodity prices – FT Alphaville
The ETF blow-up begins - FT Alphaville
UNG goes OTC - FT Alphaville
The problem with commodity ETFs – FT Alphaville
A self propelled pyramid? -FT Alphaville