High-Frequency Equity Index Futures Trading Using Recurrent Reinforcement Learning with Candlesticks

Gabrielsson, Patrick; Johansson, Ulf

doi:10.1109/ssci.2015.111

Cited by 15 publications

(17 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The baseline model used were fixed asset allocation strategies compared with the dynamic asset allocation strategy tested on the Korea Composite Stock Price Index [28]. Gabrielsson et al in 2015 attempted to combine recurrent reinforcement learning with lagged time series information from Japanese candlesticks in short, one minute intervals to create a trading algorithm using data from the Standard and Poor 500 index futures market [29]. The daily trading returns over the 31-trading-day period and Sharpe ratio were used as the performance benchmarks [29].…”

Section: Advanced Learning Strategies In Reinforcement Learningmentioning

confidence: 99%

“…Gabrielsson et al in 2015 attempted to combine recurrent reinforcement learning with lagged time series information from Japanese candlesticks in short, one minute intervals to create a trading algorithm using data from the Standard and Poor 500 index futures market [29]. The daily trading returns over the 31-trading-day period and Sharpe ratio were used as the performance benchmarks [29]. In Zhang et al in 2014 the first idea, also called the average elitist method, was aimed to improve the returns on out-of-sample testing data [30].…”

Section: Advanced Learning Strategies In Reinforcement Learningmentioning

confidence: 99%

“…The results showed that over the prescribed test period, the meta policy trading strategy produced a return of around 258% that was more than twice the profits generated by the baseline fixed asset allocation strategies [28]. The algorithm from Gabrielsson et al in 2015 was compared with three different benchmark models including a buy-and-hold model, a zero intelligence model and a basic recurrent reinforcement learning model under two separate settings: one with no trading costs and one with trading costs included [29]. The candlestick-based reinforcement learning algorithm significantly outperformed the benchmark models when transaction costs were non-existent but was not able to repeat the feat when transaction costs were included with an average overall return of 0 [29].…”

Section: Advanced Learning Strategies In Reinforcement Learningmentioning

confidence: 99%

“…The algorithm from Gabrielsson et al in 2015 was compared with three different benchmark models including a buy-and-hold model, a zero intelligence model and a basic recurrent reinforcement learning model under two separate settings: one with no trading costs and one with trading costs included [29]. The candlestick-based reinforcement learning algorithm significantly outperformed the benchmark models when transaction costs were non-existent but was not able to repeat the feat when transaction costs were included with an average overall return of 0 [29]. Within Zhang et al in 2014 the data used to test the trading system were stocks on the Standard and Poor 500 index over the four-year period of 2009 to 2012 [30].…”

Section: Advanced Learning Strategies In Reinforcement Learningmentioning

confidence: 99%

“…The article by Jangmin O et al in 2006 had the additional advantages of using a relatively complex fixed asset allocation strategy as the baseline strategy with the additional disadvantage being that it used a very old Stock Price Index from 1998 to 2001 that was most likely outdated [28]. The article by Gabrielsson et al in 2015 had the additional benefits of using three different benchmark models for comparison as well as analysing the profitability with and without transaction costs [29]. An additional strength was the zero overall rate of return once transaction costs were included [29].…”

Section: Advanced Learning Strategies In Reinforcement Learningmentioning

confidence: 99%

See 4 more Smart Citations

Reinforcement Learning in Financial Markets

Meng

Khushi

2019

Data

105

View full text Add to dashboard Cite

Recently there has been an exponential increase in the use of artificial intelligence for trading in financial markets such as stock and forex. Reinforcement learning has become of particular interest to financial traders ever since the program AlphaGo defeated the strongest human contemporary Go board game player Lee Sedol in 2016. We systematically reviewed all recent stock/forex prediction or trading articles that used reinforcement learning as their primary machine learning method. All reviewed articles had some unrealistic assumptions such as no transaction costs, no liquidity issues and no bid or ask spread issues. Transaction costs had significant impacts on the profitability of the reinforcement learning algorithms compared with the baseline algorithms tested. Despite showing statistically significant profitability when reinforcement learning was used in comparison with baseline models in many studies, some showed no meaningful level of profitability, in particular with large changes in the price pattern between the system training and testing data. Furthermore, few performance comparisons between reinforcement learning and other sophisticated machine/deep learning models were provided. The impact of transaction costs, including the bid/ask spread on profitability has also been assessed. In conclusion, reinforcement learning in stock/forex trading is still in its early development and further research is needed to make it a reliable method in this domain.Google Scholar was used to search for the reinforcement learning articles for this systematic review. By typing into Google Scholar the key phrases "reinforcement learning forex" and "reinforcement learning stock trading" and then all the results were filtered according to the selection process given below in Figure 1. The search results returned from Google Scholar was automatically sorted by relevance, therefore only the first few pages are selected for manual inspection for eligibility. Afterwards, the results of the searched phrases were manually inspected without opening the article links to determine the most relevant articles for this systematic review. The selected articles from this step were then opened and their content read through to determine the final list of articles to be reviewed. Of 27 articles reviewed, 20 articles implemented or simulated trades to maximise profit and 7 articles were only interested in forecasting future financial asset prices. Of 20 trading articles 11 articles provided comparison with other models and 10 did not.

show abstract

Section: Advanced Learning Strategies In Reinforcement Learningmentioning

confidence: 99%

Section: Advanced Learning Strategies In Reinforcement Learningmentioning

confidence: 99%

Section: Advanced Learning Strategies In Reinforcement Learningmentioning

confidence: 99%

Section: Advanced Learning Strategies In Reinforcement Learningmentioning

confidence: 99%

Section: Advanced Learning Strategies In Reinforcement Learningmentioning

confidence: 99%

See 3 more Smart Citations

Reinforcement Learning in Financial Markets

Meng

Khushi

2019

Data

105

View full text Add to dashboard Cite

show abstract

The Technological Shift: AI in Big Data and IoT

Sharma¹,

Singh

Singhal

2021

The Smart Cyber Ecosystem for Sustainable Development

View full text Add to dashboard Cite

Adapting deep learning models between regional markets

Tonkin

Gepp

Harris

et al. 2022

Neural Comput & Applic

View full text Add to dashboard Cite

This paper extends a series of deep learning models developed on US equity data to the Australian market. The model architectures are retrained, without structural modification, and tested on Australian data comparable with the original US data. Relative to the original US-based results, the retrained models are statistically less accurate at predicting next day returns. The models were also modified in the standard train/validate manner on the Australian data, and these models yielded significantly better predictive results on the holdout data. It was determined that the best-performing models were a CNN and LSTM, attaining highly significant Z-scores of 6.154 and 8.789, respectively. Due to the relative structural similarity across all models, the improvement is ascribed to regional influences within the respective training data sets. Such unique regional differences are consistent with views in the literature stating that deep learning models in computational finance that are developed and trained on a single market will always contain market-specific bias. Given this finding, future research into the development of deep learning models trained on global markets is recommended.

show abstract

High-Frequency Equity Index Futures Trading Using Recurrent Reinforcement Learning with Candlesticks

Cited by 15 publications

References 30 publications

Reinforcement Learning in Financial Markets

Reinforcement Learning in Financial Markets

The Technological Shift: AI in Big Data and IoT

Adapting deep learning models between regional markets

Contact Info

Product

Resources

About