The first set of papers in this update deal with the ‘curse of dimensionality’. One of the biggest issues with modeling stock returns, ironically, is the great amount of data we have available to throw at the problem. A lot of this is relevant information and needs to be added to a good model. However, many of these individual pieces of data will either have statistically significant relationships to each other or turn out to have no relationship to stock returns at all. Both cases will give extra degrees of freedom to an algorithm that will seriously affect its ability to find the meaningful information in a data set and its relationship to stock returns. Worse, even if we are only including good information in a model, many algorithms scale poorly as the size of their problem grows. With a limit on our computing power and time, this is a big issue.

There are two major solutions to this problem that involve pre-processing data before it is sent to a classification algorithm. The first is called feature selection and involves identifying the set of features in a data set that are important for prediction and discarding the rest. This is the approach taken by Eugene Tuv et al in their November 2009 paper “Feature Selection with Ensembles, Artificial Variables, and Redundancy Elimination”. Improvising on this theme, Krupka, Navot, and Tishby choose a subset of features with the novel idea of learning the meta-properties of a good feature in “Learning to Select Features using their Properties”. The second major way to deal with a high-dimensional data set is to project the data onto a lower dimension manifold – trying to maintain as many properties of the full-sized data set as possible. This is typically achieved using methods like principal component analysis and random projections. Yann LeCun, of convolutional neural network fame, has a paper in this update written with Raia Hadsell and Sumit Chopra that introduces a new method called Dimensionality Reduction by Learning an Invariant Mapping (DrLIM) that seems promising.

The second set of papers in this update deal with Ensembling techniques. It is increasingly apparent that the best practical approach to large, messy domains like the stock market is to combine the predictions of many heterogeneous individual prediction models in one ‘meta-model’. To prove the point, ensembling techniques have won the Netflix Prize and nearly every other machine learning competition that we are aware of in recent memory. Though this is an active field of development, there are a few heuristics emerging to help guide us. For one example, that unanimity seems to trump majority voting in our domain, see “Combining Heterogeneous Classifiers for Stock Selection” by Albanis and Batchelor.

Read past the break for citations for a few of the most interesting papers, or continue to references for the entire set.

1. Arnak S. Dalalyan, Anatoly Juditsky, and Vladimir Spokoiny, “A New Algorithm for Estimating the Effective Dimension-Reduction Subspace,” Journal of Machine Learning Research 9 (August 2008): 1647-1678.
2. Raia Hadsell, Sumit Chopra, and Yann LeCun, “Dimensionality Reduction by Learning an Invariant Mapping,” February 2006.
3. Eugene Tuv et al., “Feature Selection with Ensembles, Artificial Variables, and Redundancy Elimination,” Journal of Machine Learning Research 10 (July 2009): 1341-1366.
4. Eyal Krupka, Amir Navot, and Naftali Tishby, “Learning to Select Features using their Properties,” Journal of Machine Learning Research 9 (October 2008): 2349-2376.
5. Sivan Sabato and Shai Shalev-Shwartz, “Ranking Categorical Features Using Generalization Properties,” Journal of Machine Learning Research 9 (June 2008): 1083-1114.
6. Jianqing Fan, Richard Samworth, and Yichao Wu, “Ultrahigh Dimensional Feature Selection: Beyond The Linear Model,” Journal of Machine Learning Research 10 (September 2009): 2013-2038
7. Luciana Ferrer, Kemal Sonmez, and Elizabeth Shriberg, “An Anticorrelation Kernel for Subsystem Training in Multiple Classifier Systems,” Journal of Machine Learning Research 10 (August 2009): 2079-2114.
8. George T. Albanis and Roy A. Batchelor, “Combining Heterogeneous Classifiers for Stock Selection,” September 1999.
9. Hugo Jair Escalante, Manuel Montes, and Luis Enrique Sucar, “Particle Swarm Model Selection,” Journal of Machine Learning Research 10 (February 2009): 405-440.

As the academics in us struggle to ignore how close some of this research comes to the kind of technical analysis we are taught to distrust from the first day of Finance 101, we would be remiss to not mention exactly how we use this kind of research.  First, a quick review of Andrew Lo’s “Foundations of Technical Analysis” in the August 2000 issue of The Journal of Finance will go a long way to assuage the reader’s fear of technical analysis.  Yes, the subject is given some ‘validation by association’ here by the distinction of the author and publication, but we also agree with the argument that price dynamics encode some of the same sort of human behavioral bias that is central to the mainstream academics who study Behavioral Economics.  Regardless, we would stress that Helix does not use this kind of research as prescriptive for our positions, but rather as a way to cluster market environments.  We are interested in partitioning the historical record into different regimes so that we may attempt to have our models build themselves against the environments they are most likely to encounter instead of just simply what has been the recent past.

As an example, consider Hynek Mlnarik, Subramanian Ramamoorthy, and Rahul Savani’s February 2009 paper, “Multi-strategy trading utilizing market regimes”.  This paper illustrates this idea that different models and parameterizations can be appropriate for different market regimes and the modeller who exploits this information can improve their investment process.

Read past the break for citations for a few of the most interesting papers, or continue to references for the entire set.

1. Valeriy V. Gavrishchaka and Valery Bykov, “Market-Neutral Portfolio of Trading Strategies as Universal Indicator of Market Micro-Regimes: From Rare-Event Forecasting to Single-Example Learning of Emerging Patterns,” May 2007.
2. Hynek Mlnarik, Subramanian Ramamoorthy, and Rahul Savani, “Multi-strategy trading utilizing market regimes,” February 2009.
3. Gary Anderson, “The Janus Factor,” 2003.
4. Kevin Lapham, “Using IPOs to Identify Sector Opportunities,” 2009.
5. Samuel L. Tibbs, Stanley G. Eakins, and William DeShurko, “Using Style Index Momentum to Generate Alpha,” 2008.

Visit the Magazine section to check it out!

Now, thanks to the ongoing innovation of our favorite feed aggregator, Feedly, we can push some of that information to our website.  The development team at Feedly just launched a concept that allows users to embed feed content into curated ‘magazines’ on their own web pages.  Thanks to this work, we’ve introduced a Magazine section to our website that pulls information from our feeds marked as related to Finance and Machine Learning.

We’re excited to share more information with our website users and will be expanding this section of the website soon.  New ‘magazines’ will bring more information from our feeds on other topics like Quantitative Finance, General Finance, Economics, and Math.

Factor research papers present data items that have some sort of explanatory power of future stock returns in one of two ways: either by describing a novel piece of information or by combining or decomposing factors already familiar to the industry. That both approaches can be equally valuable highlights some of the difficulty inherent in this type of research. Not only is the search space over possible factors basically infinite, but we can not just be content to try to find novel data items because the non-linear and conditionally dependent relationships between data and returns makes combining and decomposing known factors an equally compelling research avenue. With this in mind, it is essential that any factor research thoughtfully discuss the robustness of its approach and the possibility that any abnormal relationship is simply a spurious result.

In this update, we would like to highlight two novel factor approaches. In the first, Umut Gocken uses price and volume data to construct a proxy for information revelation to show negative abnormal returns to low information environments. The second paper, by Ioannis V. Floros and Travis R. A. Sapp, discusses the abnormal returns to shell companies involved in reverse mergers.

We also highlight three papers that revisit factors well-known to the investment community. Xuemin (Sterling) Yan and Zhe Zang decompose institutional ownership in a stock to examine only those holdings of short-term institutions, which are thought to be better informed. They find confirming evidence that the positive relationship between institutional ownership and stock returns is actually driven by the holdings of these short-term institutions. In other work, Tim Loughran and Jay W. Wellman improve on the classic measure of value, the book-to-market ratio, by presenting an enterprise multiple that shows better cross-sectional explanatory power, especially in large market cap stocks. Finally, Zhipeng Yan and Yan Zhao compose post earnings announcement drift with the value anomaly to great success.

Read past the break for citations for a few of the most interesting papers, or continue to references for the entire set.

1. Umut Gokcen, “Information Revelation and Expected Stock Returns,” September 14, 2009.
2. Ioannis V. Floros and Travis R. A. Sapp, “Shell Games: On the Stock Price Performance of Shell Companies,” September 30, 2009.
3. Xuemin (Sterling) Yan and Zhe Zhang, “Institutional Investors and Equity Returns: Are Short-term Institutions Better Informed?,” January 3, 2007.
4. Tim Loughran and Jay W. Wellman, “The Enterprise Multiple Factor and the Value Premium,” October 1, 2009.
5. Zhipeng Yan and Yan Zhao, “When Two Anomalies meet: Post-Earnings-Announcement Drift and Value-Glamour Anomaly,” September 2009.
6. Eric Gettleman and Joseph M. Marks, “Acceleration Strategies,” April 27, 2006.
7. Matteo P. Arena, K. Stephen Haggard, and Xuemin (Sterling) Yan, “Price Momentum and Idiosyncratic Volatility,” The Financial Review 43 (2008): 159-190.
8. Andy Puckett and Xuemin (Sterling) Yan, “Short-term Institutional Herding and Its Impact on Stock Prices,” March 2008.
9. Brian Walkup, “The Cost of Illiquidity: Evidence from After-Hours Trading,” September 9, 2009.
10. Josh Cherry, “The Limits of Arbitrage: Evidence from Exchange Traded Funds,” December 1, 2004.

One of the themes in this selection of papers is multilayered classifier architecture, where an algorithm is really the results of cascading modules.  A particularly compelling example is described in Pradip Ghanty’s contribution to the 10th volume of the Journal of Machine Learning.  The paper proposes the NEUROSVM architecture, which glues together a MLP neural net for feature extraction and a SVM for classification.  In the process, the algorithm significantly reduces the impact of the choice of kernel on the SVM performance.  While Helix has been using neural nets, SVMs, and even hybrid structures of the two for a while, this installment of citations includes our first introduction to the closely related work of Yann LeCun on convolutional networks.  His research represents a potential future avenue of investigation for our team.

Other highlights in this set of updates include a paper that introduces a Mahalanobis distance metric for k-nearest neighbor models (Weinberger 2009) and the presentation of a novel framework called Prototype Ranking in Yan and Ling, 2007.

Read past the break for citations for a few of the most interesting papers, or continue to references for the entire set.

1. Pradip Ghanty, Samrat Paul, and Nikhil R. Pal, “NEUROSVM: An Architecture to Reduce the Effect of the Choice of Kernel on the Performance of SVM,” Journal of Machine Learning Research 10 (March 2009): 591-622.
2. Fu Jie Huang and Yann LeCun, “Large-scale Learning with SVM and Convolutional Nets for Generic Object Categorization,” March 2006.
3. Patrice Simard et al., “Boxlets: a Fast Convolution Algorithm for Signal Processing and Neural Networks.”
4. Kilian Q. Weinberger and Lawrence K. Saul, “Distance Metric Learning for Large Margin Nearest Neighbor Classification,” Journal of Machine Learning Research 10 (February 2009): 207-244.
5. Robert J. Yan and Charles X. Ling, “Machine Learning for Stock Selection,” August 2007.

Helix was also covered by the European Banking News Network.  Read the Article

Read past the break for a few citations of the most interesting papers.

1. Clifford S. Asness, R. Burt Porter, and Ross L. Stevens, “Predicting Stock Returns Using Industry-Relative Firm Characteristics,” February 2000.
2. Steven E. Christophe, Michael G. Ferri, and Jim Hsieh, “Informed Trading Before Analyst Downgrades: Evidence from Short Sellers,” October 2008, http://ssrn.com/abstract=1108162.
3. Hemang Desai, Srinivasan Krishnamurthy, and Kumar Venkatamaran, “The Role of Fundamental Analysis in Information Arbitrage: Evidence from Short Seller Recommendations,” May 2007.
4. Andy C. W. Chui, Sheridan Titman, and K. C. John Wei, “The Cross-Section of Expected REIT Returns,” March 3, 2003.
5. Avanidhar Subrahmanyam, “The Cross-Section of Expected Stock Returns: What Have We Learnt from the Past Twenty-Five Years of Research?,” August 31, 2009.
6. Bradford Cornell, “Investment Research: How Much is Enough?,” June 2009, http://ssrn.com/abstract=1439951.
7. John Dai and Suresh Sundaresan, “Risk Management Framework for Hedge Funds Role of Funding and Redemption Options on Leverage,” July 2009, http://ssrn.com/abstract=1439706.
8. Leland Wilkinson, Anushka Anand, and Dang Nhon Tuan, “Linf: An L-infnity Classifier.”
9. Zhihua Zhang, Guang Dai, and Michael I. Jordan, “A Flexible and Efficient Algorithm for Regularized Fisher Discriminant Analysis.”