This article was written by David Shabotinsky, a Financial Analyst at I Know First, and enrolled at the undergraduate Finance program at the Interdisciplinary Center, Herzliya.

Deep Learning Finance

Summary

• How Deep Learning developed from AI
• The evolution of Deep Learning in the market
• How the finance sector has begun to further take advantage of Deep learning
• I Know First implementation of Deep Learning to better forecast financial markets

Background on Deep Learning

Deep learning, a specific offset branch of machine learning that is becoming more popular by the day as more scientists and people begin understanding the mass array of capabilities that lie within it. Essentially, it truly emulates the idea that machines, or algorithmic technologies can become self-learning after being given a specific set of inputs. For example, if an algorithm is given a picture of a cat it will be able to identify future pictures, and determine whether it fits the qualifications to as well be a cat. The main idea is that scientists would like to be able to expand the realm of market the capabilities of Artificial Intelligence, or AI, how it revolutionizes the world. Deep learning is meant to mimic the human neural network, which itself is massively complex.

The first recorded use of Deep Learning dates back to the early 1940s when Warren McCulloh along with Walter Pitts created threshold logic, which is a computational model neural networks based on algorithm and advanced mathematics. Fast forward to today, Deep Learning has made huge strides in the markets as its being more developed by large corporations such as Alphabet and Facebook. For example, Facebook has implemented into its operations an AI based algorithm using deep learning, called DeepFace. The program allows profiles to automatically tag and identify users in recently added Facebook pictures. Additionally, Google developed AlphaGo, which uses deep learning neural networks to beat what is considered the world’s most ancient and complex game of Go, under the DeepMind division.

In the past, mathematicians and experts have unable to explain the rationale behind how deep learning is able to exist in its development of neural networks; today, students from Harvard University and MIT have cracked the code. It relates to the idea that math alone cannot explain the existence of deep learning rather physics is needed to explain it as it relates to understanding the nature of the universe. The answer lies in the fact that the neural networks that algorithms developed to identify objects and voice recognition are using a tiny subset of the vast possible functions derived from mathematics. This is found even more clear in describing how polynomial function in math. Being that the possibilities for them are technically infinite, in regards to the size of the exponent over ‘x’.

As explained, “For reasons that are still not fully understood, our universe can be accurately described by polynomial Hamiltonians of low order,” say Lin and Tegmark, the researchers who had conducted the study. Additionally, the laws of physics help understand the recognition. Since there is the idea of symmetry that rotating a picture of an animal 360 degrees would be the same as doing so 10 or a 100 meters

An additional element used in deep learning that is explained by physics is the idea of a hierarchy of structure, whereas the complex neural networks are really just sequences of a much smaller set of steps. An example given by the researchers is the Big Bang, which began from a single particle and then advanced through a more orderly sequence of orders. “Elementary particles form atoms which in turn form molecules, cells, organisms, planets, solar systems, galaxies, etc.,” say Lin and Tegmark.

Applications In The Financial Sector

Machine learning trading has become an even more popular phenomenon, especially as traders continue to experience historical precedents with negative interest rates and an evermore globalized world. Many investors believe markets are becoming more efficient, and opportunities for arbitrage are becoming harder to find by the minute. Therefore, large asset managers ranging from Goldman Sachs to Blackstone have begun trying to implement the latest AI-based algorithms in response. “Over the last five years, we’ve seen enormous advances in automated trading technology,” said Alfred Eskandar, chief executive of trading systems provider Portware. “Advanced front-end solutions have introduced massive efficiencies, reduced operational risk and given traders unprecedented access to global liquidity.” In the same way that artificial intelligence, more specifically machine learning, can be used to recommend movies by Netflix or web searches by Google; so too can it be used to predict the trends in the financial markets for underlying assets.

Large institutional investors, such as hedge funds, are able to deploy high-end technology due to their large capabilities. They have begun shifting focus on more algorithmic based trading technology, as a result of more and more scrutiny on the industry that their large fees result in them underperforming the S&P 500 Index, as a benchmark tool. Passive index funds, such as Vanguard, charge low fees and have become much more popular. Therefore, these large asset managers have begun heavily recruiting individuals with PhDs and masters related to the field of AI and machine learning.

Deep Learning Algorithms To Forecast Financial Markets 

In this article, we have a history of algorithm trading and how it has advanced today. Artificial Intelligence (AI), was first created in 1956 by Arthur Samuel, who wanted his computer to be able to beat him at checkers. Samuel then programmed his computer to play against itself thousands of times to the extent that the program accumulated sufficient knowledge of the game. Today, of course AI has advanced to much higher degree, and is used in most business, as well as financial institutions.

In general, there are two different types of algorithms and algorithmic trading. There is a basic formatting type algorithm that is programed to do specific functions. Then there is a self-learning algorithm, that takes historical data (empirical) and is able to adjust accordingly for a trade.

The new type of self-learning AI based algorithms are able to build trends that are often unknown even to the most intelligent analyst. The reason is because they are able to eliminate “noise” in the market. This refers to short-term (daily or intra-day) fears, worries, and negative fueled perception regarding the price of a security or general market atmosphere. By ignoring it once is able to identify trends in the market. The AI based algorithms, are able to adapt as a result of neural networks built to allow for deep learning, which then allows the algorithm to adapt accordingly. More on how specifically it goes about adjusting can be found here, in a past I Know First article.

I Know First’s Deep Learning Based Algorithm

I Know First, Ltd. is a FinTech company that brings science and math to the financial world by providing daily investment forecasts based on an advanced self-learning algorithm. Developed by CTO Lipa Roitman (PhD from Weizmann Institute – 35 years of experience), the algorithm utilizes artificial intelligence and machine learning techniques through which I Know First is able to analyze, model and predict the stock market.

Predictions are generated daily for a growing universe of over 10,000 securities, including stocks, world indices, ETFs, interest rates and more for the short, medium and long term horizons. The algorithm is applied to discover the best investment opportunities and used as a decision support system for existing investment processes or to develop systematic trading strategies. It is adaptable and scalable allowing comprehensive customized algorithmic solutions including integration of additional markets depending on clients’ needs – family offices, wealth management firms and hedge funds – as well as fund management partnerships. Furthermore, by offering top-notch technology to retail clients, I Know First also empowers private investors to identify opportunities in markets and manage their portfolios with more confidence.

The algorithm, generates two main indicators that represent the competitive advantage offered in the market place by the algorithm.

1. Signal- The sign of the signal tells in which direction the asset price is expected to go (positive = to go up = Long, negative = to drop = Short position), the signal strength is related to the magnitude of the expected return and is used for ranking purposes of the investment opportunities.
2. Predictability- the actual fitness function being optimized every day, and can be simplified explained as the correlation based quality measure of the signal. This is a unique indicator of the I Know First algorithm, allowing the user to separate and focus on the most predictable assets according to the algorithm. Ranging between -1 and 1, one should focus on predictability levels significantly above 0 in order to fill confident about/trust the signal.

The algorithm utilizes neural network, based off deep learning to detect trends in the market, for underlying financial assets around the world. It does so by inputting empirical data on daily basis to generate heat maps. Like most trend detecting technologies, its able to detect price movements more accurately over a longer time frame, ie three-months or a year. The reason is because it involves a new type of self-learning AI based algorithms that are able to build trends that are often unknown even to the most intelligent analyst. The reason is because they are able to eliminate “noise” in the market. This refers to short-term (daily or intra-day) fears, worries, and negative fueled perception regarding the price of a security or general market atmosphere. By ignoring it once is able to identify trends in the market. The AI based algorithms, are able to adapt as a result of neural networks built to allow for deep learning, which then allows the algorithm to adapt accordingly. More on how specifically it goes about adjusting can be found here, in a past I Know First article.

Conclusion

As the world searches to further improve operational functions with technology, scientists and big businesses are devoting massive amounts of resources to develop Deep Learning technology, a branch of AI. Companies such as Google and Facebook have made massive strides in improving the function of their firms using Deep Learning to better analyze how to improve operations. The financial sectors has well been moving towards a greater implementation of these types of technologies as well, in order to better capture opportunity sets regarding underlying assets in the financial markets.