Neural Networks In Trading: Goldman Sachs Has Fired 99% of Traders Replacing Them With Robots
Recently, it became known that Nasdaq is going to launch a new tool that is based on machine learning for its analytical hub. This tool will process the users data from the social networks, providing institutional investors with a new market analysis tool. However, Nasdaq specialists didn’t give any direct answer for Bitnewstoday.com but only said noncommittally about the “interesting experience” and then referred to a subscription about nondisclosure. Apparently, this can only mean that Nasdaq AI is being tested now.
Earlier in March this year, Thomson Reuters media company launched the updated MarketPsych Indices service, which makes the market forecasts by tracking for more than 2000 news sources and 800 social networks. This tool which is based on AI in its analysis uses not only quantitative indicators, but also emotional metrics of traders, like behavioral economics. Thus, the service promises its customers to create more accurate forecasts and choose the best strategy.
Traders tools and strategies
Traders usually use several technical indicators for their analysis: from resistance and support level, moving average, total trading volume to relative strength index and stochastic indicator. Among the technical indicators, trend lines are also used, which display the information about the characteristic direction of the crypto asset movement to predict the further trend.
However, due to the high volatility of the cryptocurrency, these trends are extremely difficult to determine, which is why traders rarely take this indicator into account. But in general, nowadays the technical market analysis is no longer sufficient for the forecast, as it does not take into account the political and socio-economic situation in the world.
Mark LYND, a Top-15 ranked influencer, and practitioner for AI, Digital Transformation, DLT, Crypto and influencer for companies like IBM in an exclusive interview for Bitnewstoday.com confirmed that since 2017, the use of AI for database analysis has become the main trend in the race for large companies profit: “More companies are looking to see if these AI technologies can help them realize more value from their huge stores of data. Algorithms based on biology, more specifically Artificial Neural Networks (ANNs) and Genetic Algorithms are considered the primary types used for trading analysis, risk measurement and price predictions”.
Whether all traders use the neural networks
Even today the financial and socio-economic system can not be accurately predicted by humans because of too many extenuating factors. Then AI entered the arena: it is capable of analyzing a variety of life spheres and conduct an independent analysis even better than a financial professional could do. AI-based systems have one unique advantage – a lack of emotion. But then the question arises: then why don’t all the traders use AI in their forecasts, and whether the neural network will become a universal tool for traders in the near future?
According to Mark LYND ”the financial services and tech industries itself is pushing how these technologies can be used to gain a competitive advantage and create new revenue streams, which has the big vendors like IBM using Watson to push out more sophisticated tools to help their clients realize these benefits”.
Nevertheless, Mr. LYND believes that neural networks at this stage of development are still not optimal and human interaction is still required at some level: “Ultimately the results are still largely approximations for predictions and still require some human judgment or monitoring before utilizing for large trades”.
We conducted a small survey among private traders and found that the vast majority are very indirectly familiar in their analysis with AI systems. Nevertheless some noted that they use AI-based services in their forecasts, and Dr. Gordon JONES, the founder of the DLT-project in South Carolina, said: “We focus on the machine learning aspect of all NN and AI where the system is designed to track actions and whether those actions lead to efficiencies or deficiencies and ultimate a positive or negative ROI”.
Traders are afraid to be replaced soon
Top hedge funds managers earned $1 billion in 2015, that creates a great motivation to replace them with neural networks and reduce the cost of employees who earn $500 per hour. And the company Goldman Sachs went this way by reducing the number of advisers from 600 traders to 2. Now the rest of the work is done by robots using machine learning that give out invaluable algorithms for the company’s customers. It is easy to imagine that many large companies are rapidly following this example to reduce their costs.
In an exclusive interview for Bitnewstoday.com Amardeep SINGH, PhD candidate in AI use in Ethereum and at Nasdaq analyst, who has been developing strategies for the introduction of AI in the economy for many years, said that traders are well aware that their place will soon be taken by artificial intelligence and desperately resist to the future changes.“Economic analysis is a complete picture that traders look at in the wider world: events, announcements from the governments, that influence the price of what they are analyzing.
In trading, you have something called technical analysis. They look at things like moving averages, they look at hyperparameter optimization, volatility forecasting, multivariate time series – all this, basically, really really complicated models, that are superimposed upon a moving graph, that they can then kind of make predictions. And technical analysis is also based on human emotion. So, it’s based on how you interpret what the market has done previously. This is why AI is incredibly more powerful than a human”.
How neural networks operate and get trained
Thanks to machine learning neural networks allow much more efficient construction of nonlinear relationships in comparison with linear methods of statistics, such as linear regression, autoregression, and linear discriminant. Any analyst that uses the technical analysis will make a more successful forecast that based on the preliminary work of AI systems. Nonlinear mapping and data visualization by neural networks in the space of fewer nonlinear principal components optimize their processing.
The main advantage of neural networks is their ability to simulate the behavior of the economy on the basis of social overtones comparing to existing algorithms that are based on the specified and determined parameters.
However, in this regard, not all the experts agree that neural networks have to be totally autonomous in the decisions that are based on their forecasts. Thus, the co-founder of the DLT platform in India Bibin BABU believes that the neural network is not very successful in forecasting the digital currency. “The problem is that most NN (neural networks) only has access to few types of data from within the market and not much from outside. Thus, the forces of crowd psychology, consumer confidence, good or bad headlines, political or regulatory decisions, the size of the Cryptocurrency network, the number of users, and merchant adoption are not completely visible to it. It may simply be the case that internal market data alone is not sufficient to make any kind of long term predictions, or foresee short term price fluctuations.
Mark LYND believes that the ability to use neural networks in trading is a kind of art, as the determining factor for qualitative analysis is the ability to determine the context and volume of data and their relevance to the type of results required. “Simply at a high-level, it depends on whether you need a searching or sorting algorithm and then determining what data sets to use. It is actually more complicated than that, but at a high-level, it brings some clarity.
You use mathematical functions called neurons that takes several numbers as inputs and then use a linear combination formula to multiply by the corresponding weights and then sum it up. Essentially, these networks transform data until they can classify it as an output. This is done by a bunch of connected neurons that create outputs that are then used iteratively as inputs for other neurons, hence the network. Then a loss function is used to determine how good the neural network is for solving/working out a certain task or problem. Initially, it will not provide strong results, but over time and use practices, it will provide higher accuracy.
In the next article we will talk about predictions that experts make about the future of trading with the neural networks: whether the AI will manage the market on its own or the final decisions will still remain for traders. Read about it and other exciting issues in the next article.