{"id":4235,"date":"2019-06-22T21:13:00","date_gmt":"2019-06-22T21:13:00","guid":{"rendered":"http:\/\/quantpedia.com\/using-deep-neural-networks-to-enhance-time-series-momentum\/"},"modified":"2025-06-04T16:14:39","modified_gmt":"2025-06-04T14:14:39","slug":"using-deep-neural-networks-to-enhance-time-series-momentum","status":"publish","type":"post","link":"https:\/\/vvv.quantpedia.com\/es\/using-deep-neural-networks-to-enhance-time-series-momentum\/","title":{"rendered":"Using Deep Neural Networks to Enhance Time Series Momentum"},"content":{"rendered":"

A new research paper related to:<\/strong><\/p>\n

#118 – Time Series Momentum<\/strong><\/p>\n

Authors:<\/strong> Lim, Zohren, Roberts<\/p>\n

Title: <\/strong>Enhancing Time Series Momentum Strategies Using Deep Neural Networks<\/p>\n

Link:<\/strong> https:\/\/papers.ssrn.com\/sol3\/papers.cfm?abstract_id=3369195<\/a><\/p>\n

Abstract:<\/strong><\/p>\n

While time series momentum is a well-studied phenomenon in finance, common strategies require the explicit definition of both a trend estimator and a position sizing rule. In this paper, we introduce Deep Momentum Networks — a hybrid approach which injects deep learning based trading rules into the volatility scaling framework of time series momentum. The model also simultaneously learns both trend estimation and position sizing in a data-driven manner, with networks directly trained by optimising the Sharpe ratio of the signal. Backtesting on a portfolio of 88 continuous futures contracts, we demonstrate that the Sharpe-optimised LSTM improved traditional methods by more than two times in the absence of transactions costs, and continue outperforming when considering transaction costs up to 2-3 basis points. To account for more illiquid assets, we also propose a turnover regularisation term which trains the network to factor in costs at run-time.<\/p>\n

Notable quotations from the academic research paper:<\/strong><\/p>\n

“While numerous papers have investigated the use of machine learning for financial time series prediction, they typically focus on casting the underlying prediction problem as a standard regression or classification task \u2013 with regression models forecasting expected returns, and classification models predicting the direction of future price movements. This approach, however, could lead to suboptimal performance in the context time-series momentum for several reasons.<\/p>\n

Firstly, sizing positions based on expected returns alone does not take risk characteristics into account \u2013 such as the volatility or skew of the predictive returns distribution \u2014 which could inadvertently expose signals to large downside moves. This is particularly relevant as raw momentum strategies without adequate risk adjustments, such as volatility scaling, are susceptible to large crashes during periods of market panic. Furthermore, even with volatility scaling \u2013 which leads to positively skewed returns distributions and long-option-like behaviour \u2013 trend following strategies can place more losing trades than winning ones and still be profitable on the whole \u2013 as they size up only into large but infrequent directional moves. The fraction of winning trades is a meaningless metric of performance, given that it cannot be evaluated independently from the trading style of the strategy. Similarly, high classification accuracies may not necessarily translate into positive strategy performance, as profitability also depends on the magnitude of returns in each class. In light of the deficiencies of standard supervised learning techniques, new loss functions and training methods would need to be explored for position sizing \u2013 accounting for tradeoffs between risk and reward.<\/p>\n

In this paper, we introduce a novel class of hybrid models that combines deep learning-based trading signals with the volatility scaling framework used in time series momentum strategies \u2013 which we refer to as the Deep Momentum Networks (DMNs). This improves existing methods from several angles.<\/p>\n

Firstly, by using deep neural networks to directly generate trading signals, we remove the need to manually specify both the trend estimator and position sizing methodology \u2013 allowing them to be learnt directly using modern time series prediction architectures.<\/p>\n

Secondly, by utilising automatic differentiation in existing backpropagation frameworks, we explicitly optimise networks for risk-adjusted performance metrics, i.e. the Sharpe ratio, improving the risk profile of the signal on the whole.<\/p>\n

Lastly, retaining a consistent framework with other momentum strategies also allows us to retain desirable attributes from previous works \u2013 specifically volatility scaling, which plays a critical role in the positive performance of time series momentum strategies. This consistency also helps when making comparisons to existing methods, and facilitates the interpretation of different components of the overall signal by practitioners.<\/p>\n

\"performance<\/p>\n

Referring to the cumulative returns plots for the rescaled portfolios in Exhibit 4, the benefits of direct outputs with Sharpe ratio optimisation can also be observed \u2013 with larger cumulative returns observed for linear, MLP and LSTM models compared to the reference benchmarks. Furthermore, we note the general underperformance of models which use standard regression and classification methods for trend estimation \u2013 hinting at the difficulties faced in selecting an appropriate position sizing function, and in optimising models to generate positions without accounting for risk.<\/p>\n\n\n

\n\n\n\n

Are you looking for more strategies to read about? Sign up for our newsletter<\/a> or visit our Blog<\/a> or Screener<\/a><\/strong>.<\/p>\n\n\n\n

Do you want to learn more about Quantpedia Premium service? Check how Quantpedia works<\/a>, our mission<\/a> and Premium pricing offer<\/a>.<\/strong><\/p>\n\n\n\n

Do you want to learn more about Quantpedia Pro service? Check its description<\/a>, watch videos<\/a>, review reporting capabilities<\/a> and visit our pricing offer<\/a>.<\/strong><\/p>\n\n\n\n

Are you looking for historical data or backtesting platforms? Check our list of Algo Trading Discounts<\/a><\/strong>.<\/p>\n\n\n\n

Would you like free access to our services<\/a>? Then, open an account with Lightspeed<\/a> and enjoy one year of Quantpedia Premium at no cost.<\/strong><\/p>\n\n\n\n

\n\n\n\n

Or follow us on:<\/strong><\/p>\n\n\n\n

Facebook Group<\/a>, Facebook Page<\/a>, Twitter<\/a>, Linkedin<\/a>, Medium<\/a> or Youtube<\/a><\/strong><\/p>","protected":false},"excerpt":{"rendered":"

\n\tA new research paper related to:<\/strong><\/p>\n

\n\t#118 – Time Series Momentum<\/strong><\/p>\n

\n\tAuthors:<\/strong> Lim, Zohren, Roberts<\/p>\n

\n\tTitle: <\/strong>Enhancing Time Series Momentum Strategies Using Deep Neural Networks<\/p>\n

\n\tLink:<\/strong> https:\/\/papers.ssrn.com\/sol3\/papers.cfm?abstract_id=3369195<\/a><\/p>\n

\n\tAbstract:<\/strong><\/p>\n

\n\tWhile time series momentum is a well-studied phenomenon in finance, common strategies require the explicit definition of both a trend estimator and a position sizing rule. In this paper, we introduce Deep Momentum Networks — a hybrid approach which injects deep learning based trading rules into the volatility scaling framework of time series momentum. The model also simultaneously learns both trend estimation and position sizing in a data-driven manner, with networks directly trained by optimising the Sharpe ratio of the signal. Backtesting on a portfolio of 88 continuous futures contracts, we demonstrate that the Sharpe-optimised LSTM improved traditional methods by more than two times in the absence of transactions costs, and continue outperforming when considering transaction costs up to 2-3 basis points. To account for more illiquid assets, we also propose a turnover regularisation term which trains the network to factor in costs at run-time.<\/p>\n

\n\tNotable quotations from the academic research paper:<\/strong><\/p>\n

\n\t"While numerous papers have investigated the use of machine learning for financial time series prediction, they typically focus on casting the underlying prediction problem as a standard regression or classification task – with regression models forecasting expected returns, and classification models predicting the direction of future price movements. This approach, however, could lead to suboptimal performance in the context time-series momentum for several reasons.<\/p>\n

\n\tFirstly, sizing positions based on expected returns alone does not take risk characteristics into account – such as the volatility or skew of the predictive returns distribution — which could inadvertently expose signals to large downside moves. This is particularly relevant as raw momentum strategies without adequate risk adjustments, such as volatility scaling, are susceptible to large crashes during periods of market panic. Furthermore, even with volatility scaling – which leads to positively skewed returns distributions and long-option-like behaviour – trend following strategies can place more losing trades than winning ones and still be profitable on the whole – as they size up only into large but infrequent directional moves. The fraction of winning trades is a meaningless metric of performance, given that it cannot be evaluated independently from the trading style of the strategy. Similarly, high classification accuracies may not necessarily translate into positive strategy performance, as profitability also depends on the magnitude of returns in each class. In light of the deficiencies of standard supervised learning techniques, new loss functions and training methods would need to be explored for position sizing – accounting for tradeoffs between risk and reward.<\/p>\n

\n\tIn this paper, we introduce a novel class of hybrid models that combines deep learning-based trading signals with the volatility scaling framework used in time series momentum strategies – which we refer to as the Deep Momentum Networks (DMNs). This improves existing methods from several angles.<\/p>\n

\n\tFirstly, by using deep neural networks to directly generate trading signals, we remove the need to manually specify both the trend estimator and position sizing methodology – allowing them to be learnt directly using modern time series prediction architectures.<\/p>\n

\n\tSecondly, by utilising automatic differentiation in existing backpropagation frameworks, we explicitly optimise networks for risk-adjusted performance metrics, i.e. the Sharpe ratio, improving the risk profile of the signal on the whole.<\/p>\n

\n\tLastly, retaining a consistent framework with other momentum strategies also allows us to retain desirable attributes from previous works – specifically volatility scaling, which plays a critical role in the positive performance of time series momentum strategies. This consistency also helps when making comparisons to existing methods, and facilitates the interpretation of different components of the overall signal by practitioners.<\/p>\n

\n\t\"performance<\/p>\n

\n\tReferring to the cumulative returns plots for the rescaled portfolios in Exhibit 4, the benefits of direct outputs with Sharpe ratio optimisation can also be observed – with larger cumulative returns observed for linear, MLP and LSTM models compared to the reference benchmarks. Furthermore, we note the general underperformance of models which use standard regression and classification methods for trend estimation – hinting at the difficulties faced in selecting an appropriate position sizing function, and in optimising models to generate positions without accounting for risk."<\/p>\n

\n

\n\tAre you looking for more strategies to read about? Check http:\/\/\\\/\\\/new-fmhwbzh6ghd9hede.swedencentral-01.azurewebsites.net\/Screener<\/a><\/strong><\/p>\n

\n\tDo you want to see performance of trading systems we described? Check<\/strong> http:\/\/\\\/\\\/new-fmhwbzh6ghd9hede.swedencentral-01.azurewebsites.net\/Chart\/Performance<\/a><\/strong><\/p>\n

\n\tDo you want to know more about us? Check<\/strong> http:\/\/\\\/\\\/new-fmhwbzh6ghd9hede.swedencentral-01.azurewebsites.net\/Home\/About<\/a><\/strong><\/p>\n

\n

\n\tFollow us on:<\/strong><\/p>\n

\n\tFacebook:<\/strong> https:\/\/www.facebook.com\/quantpedia\/<\/a><\/strong><\/p>\n

\n\tTwitter:<\/strong> https:\/\/twitter.com\/quantpedia<\/a><\/strong><\/p>\n

\n\tYoutube: https:\/\/www.youtube.com\/channel\/UC_YubnldxzNjLkIkEoL-FXg<\/a><\/strong><\/p>\n

\n

\n\t <\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[147,148,46,146],"class_list":["post-4235","post","type-post","status-publish","format-standard","hentry","category-uncategorized","tag-factor-investing","tag-machine-learning","tag-momentum","tag-smart-beta"],"_links":{"self":[{"href":"https:\/\/vvv.quantpedia.com\/es\/wp-json\/wp\/v2\/posts\/4235","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/vvv.quantpedia.com\/es\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/vvv.quantpedia.com\/es\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/vvv.quantpedia.com\/es\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/vvv.quantpedia.com\/es\/wp-json\/wp\/v2\/comments?post=4235"}],"version-history":[{"count":0,"href":"https:\/\/vvv.quantpedia.com\/es\/wp-json\/wp\/v2\/posts\/4235\/revisions"}],"wp:attachment":[{"href":"https:\/\/vvv.quantpedia.com\/es\/wp-json\/wp\/v2\/media?parent=4235"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/vvv.quantpedia.com\/es\/wp-json\/wp\/v2\/categories?post=4235"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/vvv.quantpedia.com\/es\/wp-json\/wp\/v2\/tags?post=4235"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}