Trading Bitcoin and Online Time Series Prediction

Muhammad Amjad, Devavrat Shah
Proceedings of the Time Series Workshop at NIPS 2016, PMLR 55:1-15, 2017.

Abstract

Given live streaming Bitcoin activity, we aim to forecast future Bitcoin prices so as to execute profitable trades. We show that Bitcoin price data exhibit desirable properties such as stationarity and mixing. Even so, some classical time series prediction methods that exploit this behavior, such as ARIMA models, produce poor predictions and also lack a probabilistic interpretation. In light of these limitations, we make two contributions: first, we introduce a theoretical framework for predicting and trading ternary-state Bitcoin price changes, i.e. increase, decrease or no-change; and second, using the framework, we present simple, scalable and real-time algorithms that achieve a high return on average Bitcoin investment (e.g. 6-7x, 4-6x and 3-6x return on investments for tests in 2014, 2015 and 2016), while consistently maintaining a high prediction accuracy (> 60-70%) and respectable Sharpe Ratio (> 2.0). Furthermore, when trained on a period eight months earlier than the test period, our algorithms performed nearly as well as they did when trained on recent data! As an important contribution, we provide a justification for why it makes sense to use classification algorithms in settings where the underlying time series is stationary and mixing.

Cite this Paper

BibTeX
@InProceedings{pmlr-v55-amjad16, title = {Trading Bitcoin and Online Time Series Prediction}, author = {Amjad, Muhammad and Shah, Devavrat}, booktitle = {Proceedings of the Time Series Workshop at NIPS 2016}, pages = {1--15}, year = {2017}, editor = {Anava, Oren and Khaleghi, Azadeh and Cuturi, Marco and Kuznetsov, Vitaly and Rakhlin, Alexander}, volume = {55}, series = {Proceedings of Machine Learning Research}, address = {Barcelona, Spain}, month = {09 Dec}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v55/amjad16.pdf}, url = {https://proceedings.mlr.press/v55/amjad16.html}, abstract = {Given live streaming Bitcoin activity, we aim to forecast future Bitcoin prices so as to execute profitable trades. We show that Bitcoin price data exhibit desirable properties such as stationarity and mixing. Even so, some classical time series prediction methods that exploit this behavior, such as ARIMA models, produce poor predictions and also lack a probabilistic interpretation. In light of these limitations, we make two contributions: first, we introduce a theoretical framework for predicting and trading ternary-state Bitcoin price changes, i.e. increase, decrease or no-change; and second, using the framework, we present simple, scalable and real-time algorithms that achieve a high return on average Bitcoin investment (e.g. 6-7x, 4-6x and 3-6x return on investments for tests in 2014, 2015 and 2016), while consistently maintaining a high prediction accuracy (> 60-70%) and respectable Sharpe Ratio (> 2.0). Furthermore, when trained on a period eight months earlier than the test period, our algorithms performed nearly as well as they did when trained on recent data! As an important contribution, we provide a justification for why it makes sense to use classification algorithms in settings where the underlying time series is stationary and mixing.} }
Endnote
%0 Conference Paper %T Trading Bitcoin and Online Time Series Prediction %A Muhammad Amjad %A Devavrat Shah %B Proceedings of the Time Series Workshop at NIPS 2016 %C Proceedings of Machine Learning Research %D 2017 %E Oren Anava %E Azadeh Khaleghi %E Marco Cuturi %E Vitaly Kuznetsov %E Alexander Rakhlin %F pmlr-v55-amjad16 %I PMLR %P 1--15 %U https://proceedings.mlr.press/v55/amjad16.html %V 55 %X Given live streaming Bitcoin activity, we aim to forecast future Bitcoin prices so as to execute profitable trades. We show that Bitcoin price data exhibit desirable properties such as stationarity and mixing. Even so, some classical time series prediction methods that exploit this behavior, such as ARIMA models, produce poor predictions and also lack a probabilistic interpretation. In light of these limitations, we make two contributions: first, we introduce a theoretical framework for predicting and trading ternary-state Bitcoin price changes, i.e. increase, decrease or no-change; and second, using the framework, we present simple, scalable and real-time algorithms that achieve a high return on average Bitcoin investment (e.g. 6-7x, 4-6x and 3-6x return on investments for tests in 2014, 2015 and 2016), while consistently maintaining a high prediction accuracy (> 60-70%) and respectable Sharpe Ratio (> 2.0). Furthermore, when trained on a period eight months earlier than the test period, our algorithms performed nearly as well as they did when trained on recent data! As an important contribution, we provide a justification for why it makes sense to use classification algorithms in settings where the underlying time series is stationary and mixing.
RIS
TY - CPAPER TI - Trading Bitcoin and Online Time Series Prediction AU - Muhammad Amjad AU - Devavrat Shah BT - Proceedings of the Time Series Workshop at NIPS 2016 DA - 2017/02/16 ED - Oren Anava ED - Azadeh Khaleghi ED - Marco Cuturi ED - Vitaly Kuznetsov ED - Alexander Rakhlin ID - pmlr-v55-amjad16 PB - PMLR DP - Proceedings of Machine Learning Research VL - 55 SP - 1 EP - 15 L1 - http://proceedings.mlr.press/v55/amjad16.pdf UR - https://proceedings.mlr.press/v55/amjad16.html AB - Given live streaming Bitcoin activity, we aim to forecast future Bitcoin prices so as to execute profitable trades. We show that Bitcoin price data exhibit desirable properties such as stationarity and mixing. Even so, some classical time series prediction methods that exploit this behavior, such as ARIMA models, produce poor predictions and also lack a probabilistic interpretation. In light of these limitations, we make two contributions: first, we introduce a theoretical framework for predicting and trading ternary-state Bitcoin price changes, i.e. increase, decrease or no-change; and second, using the framework, we present simple, scalable and real-time algorithms that achieve a high return on average Bitcoin investment (e.g. 6-7x, 4-6x and 3-6x return on investments for tests in 2014, 2015 and 2016), while consistently maintaining a high prediction accuracy (> 60-70%) and respectable Sharpe Ratio (> 2.0). Furthermore, when trained on a period eight months earlier than the test period, our algorithms performed nearly as well as they did when trained on recent data! As an important contribution, we provide a justification for why it makes sense to use classification algorithms in settings where the underlying time series is stationary and mixing. ER -
APA
Amjad, M. & Shah, D.. (2017). Trading Bitcoin and Online Time Series Prediction. Proceedings of the Time Series Workshop at NIPS 2016, in Proceedings of Machine Learning Research 55:1-15 Available from https://proceedings.mlr.press/v55/amjad16.html.

Related Material