11 evidence that artificial intelligence — the future of trading digital currencies
A team of journalists specialized in topics such as blockchain, artificial intelligence, Internet of things, financial technology and news kriptonyte, prepared for Hacker Noon list of trends, confirming that AI is the future of cryptocurrency trading.
Artificial intelligence and machine learning is relatively young, but they have stormed the world of Finance and cryptocurrency. Below are 11 trends that perfectly illustrate it.
1. Computer algorithms are already being applied in most part of the trading operations
To achieve success in trading using the discretionary (intuitive) methods, it becomes harder and harder. After all, in order to remain competitive, today we have to take decisions with speed.
In accordance with the assessments of the regulatory authorities and the findings of academic research, with the help of computers today is 50— 70% of all trades on the stock markets, 60% on the futures market and more than 50% of the jewelery market.
At the conference on financial technology at the law School of Michigan it was noted that machine learning and artificial intelligence are used increasingly in the analysis of data in the securities trading and investment consulting.
2. The artificial intelligence necessary to process huge streams of digital data
Currently, the amount of digital data doubles every two years.
Artificial intelligence is not just important, but essential tool for analyzing the vast amount of digital data produced in the world today. According to estimates by International Data Corporation, by 2020 the global volume of digital data will reach 44 zetabytes (one zettabyte, or STS, is a trillion gigabytes). If you load the data into the memory of the iPad Air and put them in a row, they will make a chain six times longer than the distance from the earth to the moon (in 2013, the volume of data was 4.4 zettabyte two-thirds of the distance from the earth to the moon).
3. The effectiveness of hedge funds using artificial intelligence, higher than that of traditional funds
The application of artificial intelligence in the hedge Fund industry is still at an early stage: today, some hedge Fund managers are turning to AI as an additional means of continuing to use intuitive methods in investment and risk management. At the same time, many funds have already used the engine management of the technical aspects of how proper trading and risk management with minimal involvement of the Fund managers.
A study conducted Eurokahedge shows that funds that use AI, superior to the results of hedge funds with the traditional approach:
From the graph we see that for two—, three – , and five-year segments hedge funds with AI demonstrated greater efficacy than traditional, and a better result than global hedge funds on average return for these periods, the profit of 8.35, 9 and 57 and of 10.56%, respectively.
4. Data neural networks allow to build a strategy on the next trading day
The analysis of the data for 1995— 2000 and forecast for 2001, is made using AI, have shown that neural networks can give up to 150% more information for building future trading strategies compared to the traditional approach of buy— and— hold.
Below is a graph of efficiency of the use of artificial neural networks to build a trading strategy on the market of high-tech companies of Taiwan:
5. Artificial intelligence helps to identify cases of market manipulation
In may 2017, the Economist published an article reviewing apps that use machine learning. In addition to mention of the fact that, beginning in 2019 to pass a professional exam for financial analysts will be required to pass the examination with the use of AI, the article has some interesting insights regarding the use of AI in trading.
Here is one of the examples given in this article. Castle Ridge Asset Management, one of the companies on asset management, since 2013 have been able to gross an average income of 32%, using complex machine learning systems. Such a high income is partly due to the fact that AI has obtained data on 24 transactions before they had been announced. The AI algorithms have identified these transactions on control signals that indicate a low volume of insider trading.
In this year’s draft RoninAI-oriented algorithms of the AI for the cryptocurrency, revealed numerous manipulations in the market because of the unusual behavior of the indicators of social mood.
6. Artificial neural networks show greater efficacy compared to the passive strategy buy-and-hold
There have been numerous studies to test the effectiveness of using neural networks the backpropagation (Backpropagation Neural Network, BPN) in forecasting stock prices. The purpose of these studies would be to assess how effectively operate a trading strategy on the basis of AI in comparison with the strategies of buy— and— hold.
Here a neural network model Backpropagation to predict at the exchange outlets:
7. Artificial intelligence works better during the financial crisis
Studies have shown that the algorithms based on artificial intelligence that is able to help you make more profitable investment decisions. For example, in the case of their application to the components of the S&P 500 index from 1992 to 2015, selected by the neural network stock portfolio showed an annual income in double figures, the highest profit was achieved in periods of financial turmoil.
Initially, the algorithms of AI have shown the greatest annual yield (334%) in 1999, a year before the maximum value of the dot-com bubble. This figure was exceeded in 2000 (annual return of 545%) when the dot-com bubble burst, and the shares of technology companies lost billions in market capitalization.
The largest rebound occurred in 2008, when the annual profit of 681% fell during the peak of the financial crisis.
In particular, the largest decline (over 100%) was in October 2008, a month after the Lehman Brothers collapse, and this is the strongest decline over the period from December 1992 to October 2015. Finally, in October 2011, positive income amounted to 35%, which coincided with the peak of the crisis, the European debt market.
Thus, we can rightly claim that machine learning algorithms are particularly effective during periods of strong market turmoil.
8. Artificial intelligence maximizes the percentage of winning trades
To understand whether the neural network to surpass the results of traditional technical analysis, a study was conducted. To obtain the empirical results were used daily closing prices of five stocks traded on the Singapore stock exchange.
A series of prices from January 1991 to December 2000 (ten years) was used for network training, and a series of prices from January 2001 to December 2004 (four years) for testing.
And this is what I found:
Empirical results demonstrated that neural networks can outperform traditional benchmarking— trading strategies due to their ability to weed out false or incorrect signals and to capitalize on fluctuations in the counters of the shares.
The proposed trading system has also increased the percentage of profitable trades to more than 90% loss of transactions was very small. Moreover, in this case a losing trade preventive in nature, as they arise, with the inevitable losses, and all losing trades are associated with transaction costs.
9. The use of artificial intelligence is perfectly located for financial instruments pricing
In the space of cryptocurrencies there are problems with attributing the correct fundamental value most popular currencies, such as bitcoin, litecoin, and Ethereum. Although there are a number of theories potentially correct pricing methodologies, widely adopted standard still has not come up.
The solution can be found using neural networks.
The first and most well-known model of option pricing was proposed by the economists Fischer black and Myron Soulzon in 1973 to determine the prices of European options. For this formula Scholes and Robert Merton received the 1997 Nobel prize in Economics (black died in 1995).
Given certain shortcomings of the black — Scholes formula in the valuation of real options, it would be interesting to see if the neural network to improve its effectiveness.
Study Bennell and Sutcliffe for improving the application of the black — Scholes formula using neural networks for option pricing of the FTSE 100 UK stock exchange (Black — Scholes Versus Artificial Neural Networks in Pricing FTSE 100 Options) was published in 2005.
In this work we compared the efficiency of using the black — Scholes formula when pricing options call for the FTSE 100 and the use of artificial neural networks. For options out of the money (out— of— the— money) the results of AI were clearly higher than when using the traditional formula.
The researchers noted that the superiority of the neural networks was quite unexpected, considering that the field of European options, the stock market is the traditional field for the application of the black — Scholes formula.
This study suggests that AI may play an important role in pricing other options for which there is no final model or the final model less effective than the same formula.
10. Artificial intelligence successfully predicts rates for all types of traditional and new asset classes
Many scientific studies show that AI can significantly exceed the effectiveness of existing trading strategies, for example strategies buy— and— hold in a wide range of asset classes.
The stock market
Researchers believe that machine learning algorithms generate a much higher absolute returns in combination with a higher Sharpe ratio (a measure of efficiency of investment portfolio).
The study of Lucas Schulze— Rebecca (Lukas Schulze— Roebbecke) showed that artificial neural network can show significantly better results with low root mean square deviation for futures copper market.
The currency market
Another study, conducted Zsidisin Han Gould (Jinxing Han Gould) from the University of Oklahoma, showed that the indexes of the Forex market can be predicted by the neural network utilizing the backpropagation that will allow you to get the maximum profit.
In an interesting article published in the Emerald Journal, are the reasons advanced approaches such as artificial neural networks and fuzzy logic (fuzzy logic), more efficient than traditional.
The article is a table summarizing the strengths of some machine learning algorithms that are used as advanced evaluation methods for the properties.
11. Profitability in the use of artificial intelligence far beyond the average level of profitability of the market
Magnus Eric Hvass Pedersen (Eric Magnus Hvass Pedersen), fellow of the University of Southampton conducted a study on “the Use of artificial intelligence for long-term investments” in January 2016. The aim of the study was to determine the optimal portfolio composition when using AI long-term investment.
In the period from 1995 to 2015 model AI exceeded the S&P 500 index by an average of approximately 18% per year. It worked particularly well in the period when the shares were highly overvalued or, as during the peak of the dot-com bubble in 2000, or undervalued, as during financial crises.