Machine learning and short positions in stock trading strategies

David E. Allen, Robert J. Powell, Abhay K. Singh

Research output: Chapter in Book/Report/Conference proceedingChapterResearchpeer-review

Abstract

Any investment strategy requires some form of asset evaluation, that is, determination of the price or fundamental value, and the prediction of likely future price movements using financial, technical, or fundamental indicators. The decision to adopt a long or short position in an asset requires a view on its immediate future price movements. Financial forecasting involves a huge data processing exercise, which may be noisy, nonstationary, and unstructured in nature, for which customary statistical methods, such as linear logistic regression and discriminant analysis are frequently applied. The development of more flexible methods, such as support vector machine classification, offers practitioners potentially better and more powerful solutions. This chapter applies support vector machines (SVM) to predict the direction of price changes for a small set of Dow Jones Industrial Average stocks and tests them against the predictions obtained from logistic regression analysis. SVM is a machine learning algorithm, which is characterized by its particular decision functions and ability to apply linear and non-linear transformations using different kernel functions. The results show that SVM improves on simple logistic regression and provides more accuracy in predicting price changes. As SVM is established on structural risk minimization, it is more resistive to overfitting than other learning methods used for empirical risk minimization and may perform better. It also performs well in comparison to other commonly used forecasting methods such as ARIMA and Artificial Neural Networks.

LanguageEnglish
Title of host publicationHandbook of short selling
EditorsGreg N. Gregoriou
Place of PublicationWaltham
PublisherElsevier
Chapter32
Pages467-478
Number of pages12
Edition1st
ISBN (Electronic)9780123877253
ISBN (Print)9780123877246
DOIs
Publication statusPublished - 2012
Externally publishedYes

Fingerprint

Machine learning
Trading strategies
Support vector machine
Logistic regression analysis
Price changes
Risk minimization
Futures prices
Assets
Prediction
Investment strategy
Discriminant analysis
Kernel
Nonlinear transformation
Forecasting method
Overfitting
Learning methods
Logistic regression
Learning algorithm
Statistical methods
Evaluation

Keywords

  • book-to-market ratio
  • dividend yield
  • hyperplane
  • investment returns
  • logistic regression
  • price-to-earnings ratio
  • radial basis function
  • Sharpe ratios
  • support vector machines
  • traded volume factor

Cite this

Allen, D. E., Powell, R. J., & Singh, A. K. (2012). Machine learning and short positions in stock trading strategies. In G. N. Gregoriou (Ed.), Handbook of short selling (1st ed., pp. 467-478). Waltham: Elsevier. https://doi.org/10.1016/B978-0-12-387724-6.00032-5
Allen, David E. ; Powell, Robert J. ; Singh, Abhay K. / Machine learning and short positions in stock trading strategies. Handbook of short selling. editor / Greg N. Gregoriou. 1st. ed. Waltham : Elsevier, 2012. pp. 467-478
@inbook{163dcd3f92224a4b9c4fafb3b53160ae,
title = "Machine learning and short positions in stock trading strategies",
abstract = "Any investment strategy requires some form of asset evaluation, that is, determination of the price or fundamental value, and the prediction of likely future price movements using financial, technical, or fundamental indicators. The decision to adopt a long or short position in an asset requires a view on its immediate future price movements. Financial forecasting involves a huge data processing exercise, which may be noisy, nonstationary, and unstructured in nature, for which customary statistical methods, such as linear logistic regression and discriminant analysis are frequently applied. The development of more flexible methods, such as support vector machine classification, offers practitioners potentially better and more powerful solutions. This chapter applies support vector machines (SVM) to predict the direction of price changes for a small set of Dow Jones Industrial Average stocks and tests them against the predictions obtained from logistic regression analysis. SVM is a machine learning algorithm, which is characterized by its particular decision functions and ability to apply linear and non-linear transformations using different kernel functions. The results show that SVM improves on simple logistic regression and provides more accuracy in predicting price changes. As SVM is established on structural risk minimization, it is more resistive to overfitting than other learning methods used for empirical risk minimization and may perform better. It also performs well in comparison to other commonly used forecasting methods such as ARIMA and Artificial Neural Networks.",
keywords = "book-to-market ratio, dividend yield, hyperplane, investment returns, logistic regression, price-to-earnings ratio, radial basis function, Sharpe ratios, support vector machines, traded volume factor",
author = "Allen, {David E.} and Powell, {Robert J.} and Singh, {Abhay K.}",
year = "2012",
doi = "10.1016/B978-0-12-387724-6.00032-5",
language = "English",
isbn = "9780123877246",
pages = "467--478",
editor = "Gregoriou, {Greg N. }",
booktitle = "Handbook of short selling",
publisher = "Elsevier",
address = "Netherlands",
edition = "1st",

}

Allen, DE, Powell, RJ & Singh, AK 2012, Machine learning and short positions in stock trading strategies. in GN Gregoriou (ed.), Handbook of short selling. 1st edn, Elsevier, Waltham, pp. 467-478. https://doi.org/10.1016/B978-0-12-387724-6.00032-5

Machine learning and short positions in stock trading strategies. / Allen, David E.; Powell, Robert J.; Singh, Abhay K.

Handbook of short selling. ed. / Greg N. Gregoriou. 1st. ed. Waltham : Elsevier, 2012. p. 467-478.

Research output: Chapter in Book/Report/Conference proceedingChapterResearchpeer-review

TY - CHAP

T1 - Machine learning and short positions in stock trading strategies

AU - Allen, David E.

AU - Powell, Robert J.

AU - Singh, Abhay K.

PY - 2012

Y1 - 2012

N2 - Any investment strategy requires some form of asset evaluation, that is, determination of the price or fundamental value, and the prediction of likely future price movements using financial, technical, or fundamental indicators. The decision to adopt a long or short position in an asset requires a view on its immediate future price movements. Financial forecasting involves a huge data processing exercise, which may be noisy, nonstationary, and unstructured in nature, for which customary statistical methods, such as linear logistic regression and discriminant analysis are frequently applied. The development of more flexible methods, such as support vector machine classification, offers practitioners potentially better and more powerful solutions. This chapter applies support vector machines (SVM) to predict the direction of price changes for a small set of Dow Jones Industrial Average stocks and tests them against the predictions obtained from logistic regression analysis. SVM is a machine learning algorithm, which is characterized by its particular decision functions and ability to apply linear and non-linear transformations using different kernel functions. The results show that SVM improves on simple logistic regression and provides more accuracy in predicting price changes. As SVM is established on structural risk minimization, it is more resistive to overfitting than other learning methods used for empirical risk minimization and may perform better. It also performs well in comparison to other commonly used forecasting methods such as ARIMA and Artificial Neural Networks.

AB - Any investment strategy requires some form of asset evaluation, that is, determination of the price or fundamental value, and the prediction of likely future price movements using financial, technical, or fundamental indicators. The decision to adopt a long or short position in an asset requires a view on its immediate future price movements. Financial forecasting involves a huge data processing exercise, which may be noisy, nonstationary, and unstructured in nature, for which customary statistical methods, such as linear logistic regression and discriminant analysis are frequently applied. The development of more flexible methods, such as support vector machine classification, offers practitioners potentially better and more powerful solutions. This chapter applies support vector machines (SVM) to predict the direction of price changes for a small set of Dow Jones Industrial Average stocks and tests them against the predictions obtained from logistic regression analysis. SVM is a machine learning algorithm, which is characterized by its particular decision functions and ability to apply linear and non-linear transformations using different kernel functions. The results show that SVM improves on simple logistic regression and provides more accuracy in predicting price changes. As SVM is established on structural risk minimization, it is more resistive to overfitting than other learning methods used for empirical risk minimization and may perform better. It also performs well in comparison to other commonly used forecasting methods such as ARIMA and Artificial Neural Networks.

KW - book-to-market ratio

KW - dividend yield

KW - hyperplane

KW - investment returns

KW - logistic regression

KW - price-to-earnings ratio

KW - radial basis function

KW - Sharpe ratios

KW - support vector machines

KW - traded volume factor

UR - http://www.scopus.com/inward/record.url?scp=84884432681&partnerID=8YFLogxK

U2 - 10.1016/B978-0-12-387724-6.00032-5

DO - 10.1016/B978-0-12-387724-6.00032-5

M3 - Chapter

SN - 9780123877246

SP - 467

EP - 478

BT - Handbook of short selling

A2 - Gregoriou, Greg N.

PB - Elsevier

CY - Waltham

ER -

Allen DE, Powell RJ, Singh AK. Machine learning and short positions in stock trading strategies. In Gregoriou GN, editor, Handbook of short selling. 1st ed. Waltham: Elsevier. 2012. p. 467-478 https://doi.org/10.1016/B978-0-12-387724-6.00032-5