Cryptocurrency portfolio management with deep reinforcement learning

Cryptocurrencies are a hot topic in the investing world, but is it possible to create an investment methodology combining modern Reinforcement Learning with classical indicators? Along this blog we have covered topics such as how to automate cryptocurrencies investment or whether reinforcement learning is suitable for trading. In this post, we try to combine Reinforcement Learning with a cryptocurrencies investment methodology more information about the methodology here. Reinforcement Learning RL is a subfield of Machine Learning that aims to train an agent to determine under which conditions is better to perform a given action.

We are searching data for your request:

Cryptocurrency portfolio management with deep reinforcement learning

Databases of online projects:
Data from exhibitions and seminars:
Data from registers:
Wait the end of the search in all databases.
Upon completion, a link will appear to access the found materials.

WATCH RELATED VIDEO: Cryptocurrency Portfolio Management - Advance Techniques - #Crypto management - Buy Sell Techniques


The thesis itself can be downloaded from this link. My master's thesis explored deep reinforcement learning in algorithmic trading.

I implemented a trading computer program that balances a portfolio of cryptocurrencies. The program tries to outperform an equally weighted strategy. More specifically, the program uses a convolutional neural network CNN built with Tensorflow. This repo requires Python 3. For installing the dependencies, please run pip install -r requirements. You do not need to worry about downloading datasets since they are automatically fetched from Poloniex API.

I used this CLI to run the actual backtests of my thesis across different machines. Each backtest of my thesis has its' own keyword argument. At the core of the program is the algorithmic trading agent — a computer program powered by deep reinforcement learning.

The agent follows some pre-determined instructions and executes market orders. Traditionally a human trader determines these instructions by using some technical indicators. I instead gave the trading agent raw price data as input and let it figure out its instructions. The algorithmic trading agent has two goals. First, it chooses initial weights, and then it rebalances these weights periodically.

Choosing proper initial weights is crucial since transaction costs make trade action costly. I evaluated the trading agent's performance in these two tasks by using two distinct agents: a static and a dynamic agent. The static agent only does the weight initialization and does not rebalance. The dynamic agent also rebalances.

I found that the agent does a poor job in choosing initial weights. In reinforcement learning terminology, the goal of the agent is to maximize the cumulative reward based on market actions. The cumulative reward is the final value of the portfolio at the end of the test period. The actions are portfolio weights. The trading agent chooses the next set of weights with a convolutional neural network CNN policy. The neural network is implemented with Tensorflow.

I chose cryptocurrencies as my underlying asset class. They are interesting to analyze due to high volatility and lack of previous research. The availability of data is also exceptional. Nevertheless, these same techniques could be utilized in other asset classes as well. I used the Poloniex API as my data source. I evaluated the performance of the agent in seven different backtest stories. Each backtest story reflects some unique and remarkable period in cryptocurrency history.

One backtest period was from December when Bitcoin reached its all-time high price. Another one is from April when Bitcoin almost lost its place as the most valued cryptocurrency. The stories show the market conditions where the agent excels and reveals its risks. I found that the algorithmic trading agent closely follows an equally weighted strategy.

This finding suggests that the agent is unavailable to decipher meaningful signals from the noisy price data. The machine learning approach does not provide an advantage over an equally weighted strategy. Nevertheless, the trading agent excels in volatile and mean-reverting market conditions. In these periods, the dynamic agent has lower volatility and a higher Sharpe ratio. However, it has a dangerous tendency to overinvest in a plummeting asset. I also wanted to find out the optimal time-period for rebalancing for the dynamic agent.

Therefore, I compared rebalancing periods from 15 minutes to 1 day. To make our results robust, I ran over a thousand simulations. I found that 15 — 30 minute rebalancing periods tend to work the best. This is visualized in the figure below where we can see that the 30 minute period had the highest Sharpe ratio.

The results of the thesis contribute to the field of algorithmic finance. I showed that frequent rebalancing is a useful tool in the risk management of highly volatile asset classes.

Further investigation is required to extend these findings beyond cryptocurrencies. For more details, please refer to the complete work. They have released their source code on GitHub. Also, I found this IPython Notebook by selimamrouni very helpful. Skip to content.

Star This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository. Branches Tags. Could not load branches. Could not load tags. Latest commit. Git stats commits. Failed to load latest commit information. View code. About No description, website, or topics provided. Releases No releases published. Packages 0 No packages published. You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window.

Article Info.

To give you the best possible experience, this site uses cookies. Review our Privacy Policy and Terms of Service to learn more. Got it! Search a title or topic. Waymo Podcasts. Download the App! Conversations about science, technology, history, philosophy and the nature of intelligence, consciousness, love, and power.

This paper presents a model-less convolutional neural network with historic prices of a set of financial assets as its input, outputting portfolio weights of.

A Deep Reinforcement Learning Framework for the Financial

Andrew W. William F. Sharpe, David P. Warmuth, Thomas M. Cover,


cryptocurrency portfolio management with deep reinforcement learning

There are millions of trades made in the global financial market every day. Data grows very quickly and people are hard to understand. This list contains the research, tools and code that people use to beat the market. Awesome AI in Finance There are millions of trades made in the global financial market every day. Bachelier, - The influences which determine the movements of the Stock Exchange are.

SlideShare uses cookies to improve functionality and performance, and to provide you with relevant advertising.

"Cryptocurrency Portfolio Management with Deep Reinforcement Learning."

In this project, we would like to manage portfolio by distributing our investment into a number of stocks based on the market. We define our environment similar to this paper Jiang et al. Without losing generality, at initial timestamp, we assume our total investment volume is 1 dollar. As the time goes by, the impact of history data decreases. Note that this is a continuous state and action space problem. We try to directly solve it in continuous space instead of using discretization in previous work Du et al and Jin et al.

A deep Q-learning portfolio management framework for the cryptocurrency market.

The third cryptocurrency in my portfolio is Polkadot, with DOT currently representing exactly 9. Market cap. The total dollar value of all Ripple transactions over the past 24 hours. Digital asset trading robots, like Bitcoin Era, have been making it easy for people with little to no trading experience to buy, sell, and earn benefits from cryptocurrency. Hallo ihr Lieben! Wie sieht der gesamte Kryptomarkt aus? Es gibt letzte Tage unseren ersten Podcast aufgenommen. Das Volumen gibt hierbei einen entscheidenden Einblick, wenn andere Wahrnehmen und verkaufen.

A novel deep Q-learning portfolio management framework composed of a set of local agents that learn assets behaviours and a global agent.

To protect your privacy, all features that rely on external API calls from your browser are turned off by default. You need to opt-in for them to become active. All settings here will be stored as cookies with your web browser. For more information see our F.

Skip to search form Skip to main content Skip to account menu You are currently offline. Some features of the site may not work correctly. DOI: Cryptocurrencies are electronic and decentralized alternatives to government-issued money, with Bitcoin as the best-known example of a cryptocurrency.

The thesis itself can be downloaded from this link.

DOI: Portfolio management is the decision-making process of allocating an amount of fund into different financial investment products. Cryptocurrencies are electronic and decentralized alternatives to government-issued money, with Bitcoin as the best-known example of a cryptocurrency. This paper presents a model-less convolutional neural network with historic prices of a set of financial assets as its input, outputting portfolio weights of the set. The network is trained with 0.

According to rspadim , functions in entropy. Financial portfolio optimisation in python, including classical efficient frontier, Black-Litterman, Hierarchical Risk Parity. A list of online resources for quantitative modeling, trading, portfolio management. A program for financial portfolio management, analysis and optimisation.

Comments: 0
Thanks! Your comment will appear after verification.
Add a comment

  1. There are no comments yet.