Quantopian's community platform is shutting down. Please read this post for more information and download your code.
Back to Community
Neural Network that tests for mean-reversion or momentum trending
Seong Lee
edited Oct 2, 2013
Machine Learning Show Q

This is a simple neural network, much of which is taken from here

It uses the Hurst Exponent and the Sharpe Ratio as inputs and trains for a small amount of days before actually using stock data. If the output of the neural network is < 0.2 there's hope that it's mean-reverting and if it's greater than 0.8, there's hope that it's momentum trending.

Things to note:

  • From my testing, it doesn't seem to be detecting momentum trending data when used with actual S&P
  • Bias with the S&P as it's been on-rise since the last three years
  • After a while, the output of the neural network seems to converge to -1, almost indefinitely, so limiting the amount of training days would be a good idea.

The algorithm definitely needs a lot of work so please feel free to play around with it and post back your suggestions.

Edit** The output seems to converge to -1 regardless of training depth, if anyone has some input on this, please feel free to contribute!

Much credit goes to Tom Starke
-Seong

Disclaimer

The material on this website is provided for informational purposes only and does not constitute an offer to sell, a solicitation to buy, or a recommendation or endorsement for any security or strategy, nor does it constitute an offer to provide investment advisory services by Quantopian. In addition, the material offers no opinion with respect to the suitability of any security or specific investment. No information contained herein should be regarded as a suggestion to engage in or refrain from any investment-related course of action as none of Quantopian nor any of its affiliates is undertaking to provide investment advice, act as an adviser to any plan or entity subject to the Employee Retirement Income Security Act of 1974, as amended, individual retirement account or individual retirement annuity, or give advice in a fiduciary capacity with respect to the materials presented herein. If you are an individual retirement or other investor, contact your financial advisor or other fiduciary unrelated to Quantopian about whether any given investment idea, strategy, product or service described herein may be appropriate for your circumstances. All investments involve risk, including loss of principal. Quantopian makes no guarantees as to the accuracy or completeness of the views expressed in the website. The views are subject to change, and may have become unreliable for various reasons, including changes in market conditions or economic circumstances.

12 responses
Tom Starke
Oct 4, 2013

Hi Seong,
I haven't had time to look over your algo yet but the returns seem to be way too high. This post here points out what the problem might be.

Tom

Seong Lee
Oct 21, 2013

Limited leverage a bit and the returns have cooled down a lot. Definitely looking for a better way to check the validity of the output, open to suggestions!

-Seong

Peter Cawthron
Dec 23, 2013

Hello Seong,

Does this algo give a different result each time it is run? That is what I find although some returns (23.8%, 70.4%) do re-occur.

Regards,

Peter

Brent Groya
Dec 26, 2013

It is normal to get different results with neural networks. The network can be slightly different every time you train it. Even with the same data.

I guess that itself (consistency of results) might be an indicator of the strength of a model.

JOHN CHAN
Jul 20, 2015

Seong...how can you make this constant... so that it ill be reliable....https://www.quantopian.com/posts/neural-network-that-tests-for-mean-reversion-or-momentum-trending

Thomas Wiecki
Jul 20, 2015

John, you could try to set the random seed with numpy.random.seed() or do a static initialization of the weights.

Disclaimer

The material on this website is provided for informational purposes only and does not constitute an offer to sell, a solicitation to buy, or a recommendation or endorsement for any security or strategy, nor does it constitute an offer to provide investment advisory services by Quantopian. In addition, the material offers no opinion with respect to the suitability of any security or specific investment. No information contained herein should be regarded as a suggestion to engage in or refrain from any investment-related course of action as none of Quantopian nor any of its affiliates is undertaking to provide investment advice, act as an adviser to any plan or entity subject to the Employee Retirement Income Security Act of 1974, as amended, individual retirement account or individual retirement annuity, or give advice in a fiduciary capacity with respect to the materials presented herein. If you are an individual retirement or other investor, contact your financial advisor or other fiduciary unrelated to Quantopian about whether any given investment idea, strategy, product or service described herein may be appropriate for your circumstances. All investments involve risk, including loss of principal. Quantopian makes no guarantees as to the accuracy or completeness of the views expressed in the website. The views are subject to change, and may have become unreliable for various reasons, including changes in market conditions or economic circumstances.

JOHN CHAN
Jul 20, 2015

wat line... do I substitute that numpy.random.seed()??

Thomas Wiecki
Jul 20, 2015

In initialize you would put numpy.random.seed(123). http://docs.scipy.org/doc/numpy/reference/generated/numpy.random.seed.html

JOHN CHAN
Jul 20, 2015

already put numpy.random.seed(123). @ initialize its still random...

Grassland Shaw
Dec 23, 2016

I kinda did that before. The biggest problem here is that it could be quite confusing when the ranges of the output result are small, say, 0.2-0.21, or , 0.6-0.62. How are you gonna distinguish btw a momentum and a trend with these paras?

Jason He
Feb 13, 2018

You use tanh as your activation function but then use the derivative of sigmoid function for back propagation. How is that valid?

Andrew Kamal
Feb 22, 2018

Why have the S&P confirmation bias been on the rise, also is there a way to integrate Bayes' theorem for false positives or trading constraints?

About Quantopian
Careers
Community
Events
Documentation
Academia
Lectures
Status
Facebook
Twitter
LinkedIn
YouTube
Terms of Use
Privacy Policy

The material on this website is provided for informational purposes only and does not constitute an offer to sell, a solicitation to buy, or a recommendation or endorsement for any security or strategy, nor does it constitute an offer to provide investment advisory services by Quantopian.

In addition, the material offers no opinion with respect to the suitability of any security or specific investment. No information contained herein should be regarded as a suggestion to engage in or refrain from any investment-related course of action as none of Quantopian nor any of its affiliates is undertaking to provide investment advice, act as an adviser to any plan or entity subject to the Employee Retirement Income Security Act of 1974, as amended, individual retirement account or individual retirement annuity, or give advice in a fiduciary capacity with respect to the materials presented herein. If you are an individual retirement or other investor, contact your financial advisor or other fiduciary unrelated to Quantopian about whether any given investment idea, strategy, product or service described herein may be appropriate for your circumstances. All investments involve risk, including loss of principal. Quantopian makes no guarantees as to the accuracy or completeness of the views expressed in the website. The views are subject to change, and may have become unreliable for various reasons, including changes in market conditions or economic circumstances.