Difference between revisions of "Machine Learning/Finance"

From Thalesians Wiki
 
(19 intermediate revisions by the same user not shown)
Line 1: Line 1:
=Machine Learning (ML) and Artificial Intelligence (AI)=
The importance of machine learning in finance continues to grow. This page attempts to collect the most important research papers and developments in this area as mentioned in the Press.


The importance of machine learning in finance continues to grow.
=Research papers=


==Developers now make up a quarter of Goldman Sachs' workforce==
==2021.07.12 A rigorous and robust quantum speed-up in supervised machine learning==


Jia Jen Low. T_HQ. 14 February 2020. <span class="plainlinks">https://techhq.com/2020/02/developers-now-make-up-quarter-of-goldman-sachs-workforce/</span>
'''Yunchao Liu, Srinivasan Arunachalam, and Kristan Temme. ''Nature Physics'' (2021). https://www.nature.com/articles/s41567-021-01287-z'''


Abstract: "The leading finance firm says it's now competing with Silicon Valley tech giants for talent."
<blockquote>
Recently, several quantum machine learning algorithms have been proposed that may offer quantum speed-ups over their classical counterparts. Most of these algorithms are either heuristic or assume that the data can be accessed quantum-mechanically, making it unclear whether a quantum advantage can be proven without resorting to strong assumptions. Here we construct a classification problem with which we can rigorously show that heuristic quantum kernel methods can provide an end-to-end quantum speed-up with only classical access to data. To prove the quantum speed-up, we construct a family of datasets and show that no classical learner can classify the data inverse-polynomially better than random guessing, assuming the widely believed harness of the discrete logarithm problem. Furthermore, we construct a family of parameterized unitary circuits, which can be efficiently implemented on a fault-tolerant quantum computer, and use them to map the data samples to a quantum feature space and estimate the kernel entries. The resulting quantum classifier achieves high accuracy and is robust against additive errors in the kernel entries that arise from finite sampling statistics.
</blockquote>
 
==2021.05.21 Multi-Horizon Forecasting for Limit Order Books: Novel Deep Learning Approaches and Hardware Acceleration using Intelligent Processing Units==
 
'''Zihao Zhang, Stefan Zohren. Submitted on 21 May 2021. https://arxiv.org/abs/2105.10430'''
 
<blockquote>
We design multi-horizon forecasting models for limit order book (LOB) data by using deep learning techniques. Unlike standard structures where a single prediction is made, we adopt encoder-decoder models with sequence-to-sequence and Attention mechanisms, to generate a forecasting path. Our methods achieve comparable performance to state-of-art algorithms at short prediction horizons. Importantly, they outperform when generating predictions over long horizons by leveraging the multi-horizon setup. Given that encoder-decoder models rely on recurrent neural layers, they generally suffer from a slow training process. To remedy this, we experiment with utilising novel hardware, so-called Intelligent Processing Units (IPUs) produced by Graphcore. IPUs are specifically designed for machine intelligence workload with the aim to speed up the computation process. We show that in our setup this leads to significantly faster training times when compared to training models with GPUs.
</blockquote>
 
=The Press=
 
==2021.07.19 Quantum machine learning achieves advantage in IBM research==
 
'''Sam Cox. Silicon Republic. https://www.siliconrepublic.com/machines/quantum-advantage-ibm-machine-learning'''
 
'''Abstract:''' ''In a new paper by IBM, quantum machine learning was able to discern patterns where classical computers missed the signal in the noise.''
 
<blockquote>
Quantum computing is a field full of promise but has yet to prove many of its supposed advantages. IBM is confident that quantum advantage will come to fruition but is still working away to establish the proof in the pudding.
 
Its [https://www.siliconrepublic.com/machines/ibm-quantum-computer-system-one-fraunhofer-germany worldwide machines] have shown quantum superiority [https://www.siliconrepublic.com/machines/ibm-quantum-advantage-experiment in other domains], but machine learning is still in the works. However, [https://www.nature.com/articles/s41567-021-01287-z a new research paper] from IBM Quantum has tackled a central question related to quantum machine learning: which quantum algorithms are currently capable of delivering a provable quantum advantage over classical machine learning algorithms?
 
Proposals in quantum machine learning are often driven by the challenge to find algorithms that can be tested in the near-term with only conventional access to data.
 
One such class of algorithms is the proposal for quantum enhanced feature spaces, also known as quantum kernel methods. In these set-ups, the quantum computer steps in for just part of the process in the algorithm.
 
These were the focus of IBM's latest research.
 
An area of potential use for these kernel methods is classification problems. These are one of the most fundamental problems in machine learning.
 
They begin by training an algorithm on data, called a training set, where data points include one of two labels. Following the training phase is a testing phase where the algorithm needs to classify a new data point that has not been seen before.
 
A standard example is giving a computer pictures of dogs and cats, and from this dataset it classifies all future images as a dog or a cat.
 
Ultimately, the goal of an efficient machine learning algorithm for classification should be to generate an accurate label in an amount of time that scales polynomially with the size of the input.
 
In this case, the researchers started with a conventional machine learning model to learn the kernel function, which finds the relevant features in the data to use for classification.
 
The quantum advantage comes from the fact that the researchers were able to construct a family of datasets for which only quantum computers can recognise the intrinsic labelling patterns.
 
The researchers used a problem that separates classical and quantum computation: computing logarithms in a cyclic group, where you can generate all the members of the group using a single mathematical operation.
 
For classical computers, the dataset looked like meaningless noise, whereas the quantum computers were able to work through the data.
 
The team demonstrated this by constructing a family of classification problems and showed that no efficient classical algorithm could do better than random guessing when attempting to solve these problems.
 
They also constructed a quantum feature map. This is a way to view complicated data in a higher-dimensional space to pull out patterns. When used alongside the corresponding kernel function, the researchers were able to predict the labels with high accuracy.
 
What's more, they could show that the high accuracy persists in the presence of finite sampling noise from taking measurements, a form of noise that needs to be considered even for fault-tolerant quantum computers.
 
IBM [https://www.research.ibm.com/blog/quantum-kernels highlighted] that there will still be many real-life problems for which this quantum algorithm does not perform better than conventional classical machine learning algorithms.
 
To obtain a quantum advantage, the classification problem must stick to this cyclical structure described above. This is an important caveat, as IBM's further research will be aimed at discussing how generalisable this structure is.
 
Nevertheless, the researchers were confident that this proves an end-to-end quantum speed-up for a quantum kernel method implemented fault-tolerantly with realistic assumptions.
 
Another potential sticking point could be the hardware limitations of modern quantum computers.
 
With the field rapidly advancing and ever-bigger offerings being set up across the globe however, solving this limitation might just be a case of when rather than if.
</blockquote>
 
==2021.07.16 Blender Bot 2.0: An open source chatbot that builds long-term memory and searches the internet==
 
'''Jason Weston (Research Scientist), Kurt Shuster (Research Scientist). Facebook AI. July 16, 2021. https://ai.facebook.com/blog/blender-bot-2-an-open-source-chatbot-that-builds-long-term-memory-and-searches-the-internet/'''
 
<blockquote>
* Facebook AI Research has built and open-sourced BlenderBot 2.0, the first chatbot that can simultaneously build long-term memory it can continually access, search the internet for timely information, and have sophisticated conversations on nearly any topic. It's a significant update to the original BlenderBot, which we open-sourced in 2020 and which broke ground as the first to combine several conversational skills&mdash;like personality, empathy, and knowledge&mdash;into a single system.
* When talking to people, BlenderBot 2.0 demonstrated that it's better at conducting longer, more knowledgeable, and factually consistent conversations over multiple sessions than its predecessor, the existing state-of-the-art chatbot.
* The model takes pertinent information gleaned during conversation and stores it in a long-term memory so it can then leverage this knowledge in ongoing conversations that may continue for days, weeks, or even months. The knowledge is stored separately for each person it speaks with, which ensures that no new information learned on one conversation is used in another.
* During conversation, the model can generate contextual internet search queries, read the results, and incorporate that information when responding to people's questions and comments. This means the model stays up-to-date in an ever-changing world.
* Today we're releasing the complete model, code, and evaluation setup, as well as two new conversational data sets&mdash;human conversations bolstered by internet searches, and multisession chats with people that reference previous sessions&mdash;used to trin the model, so other researchers can reproduce this work and advance conversational AI research.
 
[...]
</blockquote>
 
==2021.07.09 China Targets AI Dominance by 2030==
 
'''Bloomberg Quicktake. July 9th, 2021, 9:48 AM GMT+0100. https://www.bloomberg.com/news/videos/2021-07-09/china-targets-ai-dominance-by-2030-video'''
 
<blockquote>
Endless cups of coffee, hi-tech tools to help doctors, a masseuse who never gets tired... Just some of the latest innovations on display at this year's World Artificial Intelligence Conference in Shanghai.
 
"The advantages are first of all its trajectory is a digital record of master-level massage skills. Compared with ordinary masseurs, its techniques are more accurate and precise. Second, with the 3D images it can find different acupuncture points according to different people's body shapes and can massage in a targeted manner."
 
Meet Walker X, he could be your next butler.
 
"The functions that we have been able to achieve in the family setting include bringing you a Coke from the refrigerator, opening a drink, bringing you a newspaper, fetching an umbrella, and watering the flowers, playing chess with you, singing and doing homework with you. It can voice chat and has other emotional interactions. In the future, we will make robots that have a more diversified role in the entire home. Walker will play the role to help the family, accompany you as you grow up and accompany children and the elderly. It can be a teacher of the family as well as a helper who can do a variety of housework in the family.
 
AI is also helping drive medical progress. Morphogo is a bone marrow cytomorphology analysis system. It can make a diagnosis within 1 day compared to the usual 5 to 10 days with traditional hospital methods.
 
Yang Chuhu, ZhiWei Information Technology Co. Ltd.: "This kind of high-efficiency detection is crucial for accurate diagnosis and timely treatment which cannot be achieved only by manual work. Taking China for example, China has vast remote areas and it is impossible for people living in remote villages to go to a big hospital for a rapid morphological examination of bone marrow. Without such a system, it can be said that the life and health of patients with blood diseases in remote areas cannot be effectively guaranteed. This is not a problem for China alone. This is a global problem."
 
China has set its sights on building a domestic AI industry worth nearly $150 billion in the next few years and of becoming the leading AI power by 2030.
</blockquote>
 
==2021.07.07 U.K. Deploys Fastest Supercomputer to Fight Dementia, MS==
 
'''Adeola Eribake. Bloomberg. July 7, 2021, 12:01 AM GMT+1. https://www.bloomberg.com/news/articles/2021-07-06/nvidia-boots-up-u-k-supercomputer-to-boost-drug-research-nhs'''
 
'''Abstract:''' ''Developed in partnership with AstraZeneca, GlaxoSmithKline. Nvidia waiting for approval for its $40 billion takeover of Arm.''
 
<blockquote>
Nvidia Corp. has built a U.K. supercomputer for scientists and drug makers that will allow researchers to model diseases, discover new medicines and advance knowledge of the human genome.
 
The Cambridge-1 will be the country's most powerful and one of the top 50 supercomputers worldwide. It was developed in partnership with AstraZeneca Plc and GlaxoSmithKline Plc, Nvidia said in a statement on Wednesday.
 
Other founding members include Guy's Hospital and St Thomas' Hospital trust, part of the U.K.'s National Health Service, as well as King's College London.
 
"Through this partnership, we will be able to use a scale of computational power that is unprecedented in health-care research," said Sebastien Ourselin, head of the School of Biomedical Engineering &amp; Imaging Sciences at King's College London. "It will be truly transformational for the health and treatment of patients."
 
Oxford Nanopore, a DNA sequencing startup that's preparing to debut on the London Stock Exchange will also get access to Cambridge-1, allowing it to improve its artificial intelligence algorithms in hours instead of days, the startup said in the statement.
 
Rather than a single unit, the Cambridge-1 is made up of several modules working together. This parallel processing lets the computer carry out tasks simultaneously, allowing for greater speeds. It was also faster to build, coming together in about a quarter of the time it takes to construct similar machines.
 
The supercomputer has a computing capability of more than 400 petaflops. That's close to the world's fastest, the Fugaku machine in Japan. Top500, which publishes a twice-yearly list of the fastest supercomputers, included the Cambridge-1 at No. 41 on its most recent ranking, published at the end of June.
 
The speed and capabilities that supercomputers offer scientists are crucial as they try to make sense of unprecedented amounts of data. "We accumulated more data in one quarter in 2020 than in the last 300 years," said Kim Branson, GSK's senior vice president and global head of AI-ML.
 
The machine, which cost $100 million to build, was announced in October, a month after Nvidia launched a $40 billion bid for semiconductor designer Arm Ltd. Arm's technology is ubiquitous in smartphones and other consumer electronics, making the deal controversial among Nvidia's competitors, who rely on the designs. The U.K. has also expressed concerns that selling Arm, which is currently owned by Softbank Group Corp., has national security implications and the country's competition authority is due to submit a report at the end of the month.
 
Nvidia said it's already developing another supercomputer, which is to be made with Arm technology.
 
The Cambridge-1 will be kept at a site run by Kao Data Campus just outside London and about 40 miles south of its namesake city.
 
King's College and London hospital group Guy's and St Thomas' will use the new AI capability to learn from MRI scans and then generate synthetic brain images, creating models to help scientists better understand diseases such as dementia, stroke, brain cancer and multiple sclerosis. That could allow for earlier diagnosis and treatment, the university and teaching hospital said.
 
Nvidia said it would work with its partners to share the insights gained with the wider scientific community. For example, AstraZeneca will open up one of its projects, an AI-based model for chemical structures to help with drug discovery, to other scientists.
 
"Cambridge-1 will empower world-leading researchers in business and academia with the ability to perform their life's work on the U.K.'s most powerful supercomputer, unlocking clues to disease and treatments at a scale and speed previously impossible in the U.K.," said Jensen Huang, founder and chief executive officer of Nvidia, in a statement.
</blockquote>
 
==2021.07.05 China beats Google to claim the world's most powerful quantum computer==
 
'''Matthew Sparkes. NewScientist. 5 July 2021. https://www.newscientist.com/article/2282961-china-beats-google-to-claim-the-worlds-most-powerful-quantum-computer/'''
 
<blockquote>
A team in China has demonstrated that it has the world's most powerful quantum computer, leapfrogging the previous record holder, Google.
 
Jian-Wei Pan at the University of Science and Technology of China in Hefei and his colleagues say their quantum computer has solved a problem in just over an hour that would take the world's most powerful classical supercomputer eight years to crack, and may yet be capable of exponentially higher performance.
</blockquote>
 
==2021.06.24 Deserve Valuation Tops $500 Million as MasterCard and Ally Ventures Invest==
 
'''Gillian Tan. Bloomberg. June 24, 2021, 12:00 PM GMT+1. https://www.bloomberg.com/news/articles/2021-06-24/deserve-valuation-tops-500-million-as-mastercard-ally-invest'''
 
'''Key points:''' ''Goldman, Mission Holdings and Sallie Mae also backed startup. Palo Alto firm seeks to be profitable in first half of 2023.''
 
<blockquote>
Deserve, a credit-card technology startup, raised $50 million from backers including Mastercard Inc. and Ally Financial Inc.'s strategic investment arm, its chief executive officer said in an interview.
 
Mission Holdings, Goldman Sachs Group Inc., Sallie Mae and other existing investors participated in the funding round, proceeds of which will be used to launch card programs and accelerate growth, CEO Kalpesh Kapadia said. The company is seeking to provide rewards unlike traditional methods of cash-back, points or miles. Already, one of its partners, BlockFi, offers cryptocurrency rewards and another, Seneca Women, rewards spending at women-owned businesses.
 
The founding round values the Palo Alto, California-based company at more than $500 million, according to a person with knowledge of the matter.
 
"We are Instagram to the credit-card industry's Kodak," Kapadia said, referencing Deserve's plastic- and metal-free operations. Its software enables companies to offer digital credit cards that can be applied for and used on iPhones or Android devices. "People may forget their wallets, but they never forget their phones," he added.
 
Deserve and the Apple Card have a huge lead over competitors, said Saurabh Mittal, founder of Mission Holdings, which first invested in the startup's seed round in 2014. "There's an exponential growth story that will play out over the next couple of decades," he said.
 
The company is striving to achieve profitability by the first half of 2023, and will consider options for going public when it's closer to that milestone, Kapadia said. It will explore new lending products that could include payroll advances, "buy now pay later" programs and installment loans, he added.
 
Deserve uses so-called deep machine learning and artificial intelligence, and says its underwriting process, in addition to FICO scores, relies upon income and employment data. The company touts "signle-call resolution" user support which should lead to fewer delinquencies, charge-backs and disputes.
 
"Credit cards are an area of interest for Ally, and Deserve &mdash with its digital-first approach &mdash; is a disruptive company in this space," Ally chief strategy &amp; corporate development officer Dinesh Chopra said. Ally doesn't currently offer a credit card but has expressed inteest in unsecured lending, he added.
 
Ally last year abandoned plans to buy subprime credit card lender CardWorks.
 
''(Updates to add details of a partnership in the second paragraph. An earlier story was corrected to remove a reference to a partner company.)''
</blockquote>
 
==2021.06.21 Former Morgan Stanley Traders Raise $100M to Turn Crypto Startup into Unicorn==
 
''Vishawam Sankaran. Independent. Monday 21 June 2021 06:59. https://www.independent.co.uk/life-style/gadgets-and-tech/cryptocurrency-morgan-stanley-amber-group-b1869555.html''
 
'''Abstract:''' ''Firm with over 300 employees in Hong Kong, Taipei, Seoul and Vancouver now plans to expand''
 
<blockquote>
After a successful fundraiser, Amber Group, a Hong Kong-based cryptocurrency startup founded by former Morgan Stanley traders, has raised $100 million, scoring a pre-money valuation of $1 billion.
 
Michael Wu, co-founder and CEO of the crypto financial services firm with over 300 employees in Hong Kong, Taipei, Seoul and Vancouver, said in a statement that the funding would be used to "extend global operations to meet client demand and develop market solutions for the world's leading crypto investors and companies."
 
"We've had record months over the past quarter across both client flow and on-exchange market-making volumes. Our cumulative trading volumes have doubled from $250 billion since the beginning of the year to over $500 billion," Wu added.
 
In 2017, the founders, including former Morgan Stanley traders Michael Wu, Luke Li, Wayne Huo, Tiantian Kullander, and Tony He, had initially sought to apply machine learning to quantitative trading, but pivoted to crypto in 2019 when the trading volumes for the virtual currency increased.
 
The series B funding, which has brought the several times more than the series A round in 2019, was bankrolled by several big-name financiers including Tiger Brokers, Tiger Global Management, Sky9 Capital, Tru Arrow Partners, Arena Holdings, Gobi Partners, and DCM Ventures.
 
In the series A funding as well the startup was backed by several high-profile investors including Paradigm and Pantera Capital, Polychain Capital, Dragonfly Capital, Blockchain.com, Fenbushi Capital, and Coinbase Ventures, the firm noted.
 
Amber Group serves over 500 institutional clients, trading over "$330 billion across 100+ electronic exchanges," since its inception, expanding also to retail consumers with the launch of its mobile app in 2020.
 
Venture capital firms have shown growing interest in the crypto economy in the last six months.
 
Last month, Babel Finance, another Hong Kong-based crypto asset manager secured funding of $40 million from several institutional investors, including Tiger Global.
 
Matrixport, another cryptocurrency lending service started by Bitmain's influential founder Jihan Wu, is also seeking a new capital injection, Bloomberg reported.
 
Blockchain intelligence platform TRM Labs also secured $14 million in Series A funding from investors including the venture capital arms of PayPal and Salesforce.
 
TRM helps blockchain companies, financial institutions and law enforcement agencies investigate, detect and prevent financial crimes related to cryptocurrencies.
</blockquote>
 
==2021.06.08 Man Group-Oxford Quants Say Their AI Can Predict Stock Moves==
 
'''Amy Thomson (with assistance by Justina Lee, and Julius Domoney). Bloomberg. June 8, 2021, 10:24 AM GMT+1. https://www.bloomberg.com/news/articles/2021-06-08/man-group-oxford-quants-say-their-ai-can-predict-stock-moves'''
 
'''Abstract:''' ''Machine-learning program hits 80% success rate over 30 seconds. Tape bombs and processing power are challenges to deployment.''
 
<blockquote>
Man Group Plc-backed researchers at the University of Oxford say they've created a machine-learning program that can project how share prices move&mdash;notching an 80% success rate for the equivalent of about 30 seconds of live trading.
 
Artificial-intelligence experts at the Oxford-Man Institute of Quantitative Finance exploted principles from natural-language processing to trawl liquidity data across limit order books, a record of buying and selling at preset prices.
 
In a potential step forward for fast-money traders seeking to time markets, the algorithm figured out the direction of a price move over a period of 100 ticks, the equivalent of about 30 seconds to two minutes of trading depending on market conditions.
 
"In the multi-step forecasting, we effectively have a model which is trained to make a forecast at a smaller horizon," said Stefan Zohren, an associate professor at the institute who co-authored the research. "But we can feed this information back into itself and roll forward the prediction to arrive at longer-horizon forecasts."
 
The algorithm, which remains at the testing stage, has a clear appeal for hedge fund managers who typically break up large stock orders into multiple smaller transactions, according to Anthony Ledford, chief scientist at Man AHL.
 
"If we think we're going to enter a position, we may hold that position for several weeks, but actually making the trade that gives you that exposure happens over a much shorter-time period," via a number of smaller trades, he said.
 
Where this model "has much more of an impact for us, it's understanding how to release and trade those smaller pieces into the market&mdash;and each one of those may be done in a timescale of a few minutes," said Ledford, a former winner of the Royal Statistical Society's Research Prize.
 
With endless tape bombs hitting financial markets, accuracy rates for these kind of models are taken with a pinch of salt when it comes to the real world. But the Man-Oxford findings illustrate the allure of using AI to divine complex relationships across data points that in theory can run into the billions.
 
As increasing industry competition whittles down returns in core strategies, quants are vying ever-more to deploy programs that learn statistical patterns in equities in order to cut trading costs and find new investing signals.
 
Predicting share movements one or two milliseconds before everyone else does has been the goal of strategies such as statistical arbitrage and exchange colocation for more than a decade. Yet leveling computational firepower at stock prices is a crowded field, with an entrenched arms race among the biggest shops ensuring that no technical advantage lasts long.
 
The hedge fund, with $127 billion in assets under management, provided initial funding for the institute in 2007, and has committed more than 30 million pounds ($42.5 million). The research center was responsible for introducing Man to graphics processing units, or GPUs, that are able to handle the intensive processing demands of artificial intelligence about a decade ago.
 
Multi-horizon forecast models using statistical analysis have been around for years now, channeling market variables into predictions about how a stock will move over different time periods.
 
The new techniques used in the Oxford-Man Institute's research, which increase the potential accuracy of the predictions over a longer period of time, channeled principles from natural-language processing. Zohren, who worked with research associate Zihao Zhang on the [https://arxiv.org/abs/2105.10430 paper], compared the model to a program that can translate a sentence to French from English by building inferences incrementally.
 
But to make the Oxford-Man model work, the AI has to be able to process a huge amount of data quickly.
 
The researchers turned to Bristol, England-based Graphcore's Intelligence Processing Unit, part of a pizza box-sized system designed specifically to handle the demands of an AI program. In the trials, Graphcore's chip performed about 10-times faster than GPUs.
 
While the research and the Graphcore chips that make the model possible are the "logical next step" in the high-speed computations that Man Group is interested in, the fund hasn't committed to rolling it out, Ledford said.
 
Meanwhile, not every firm would be able to deploy this kind of strategy.
 
"You would not try this model if you did not have access to fast computation," said Zohren.
</blockquote>
 
==2021.04.30 Goldman Sachs is betting on Artificial Intelligence to drive growth==
 
'''Disha Sinha. Analytics Insight. April 30, 2021. https://www.analyticsinsight.net/goldman-sachs-is-betting-on-artificial-intelligence-to-dive-growth/'''
 
<blockquote>
Artificial Intelligence (AI) has taken a major role in acting as the main driver of upcoming hi-tech future in the world. It is shifting the information age to a completely new digital domain where upgraded machines help to solve critical decisions and assist in diverse sectors of a country. The banking and financial sector is very keen in investing in AI to protect their customers against the competitors in this market. Thus depending on the current scenario, Goldman Sachs, a leading American multinational investment bank, is betting on AI to drive growth in the economy. It has a fund of $72.5 million exclusively for the investment in Artificial Intelligence algorithms and data analytics.
 
There is a sudden surge in the cases against cyber-security, cyber-threats, phishing, spamming, hacking and many more unethical behaviours from dark web. It is due to the digital transformation in online banking through websites or mobile apps. Goldman Sachs deals with investment management, securities, asset management, prime brokerage and securities underwriting. This means banks need the best security possible against the online threats. Here comes AI to assist Goldman Sachs in the best possible way.
 
'''Why Goldman Sachs is betting on AI to drive growth?'''
 
# The constant fear of cyber-attacks has been reduced with the help of filtering application of AI. Yes, there are possibilities of receiving fraud applications with unethical motives. AI cybersecurity detects and blocks these applications while collecting real-time data from the users on a large scale.
# AI-enabled Investment Trust has already become a successful test for Goldman Sachs in enabling the best investment options for the customers. This has improved customer engagement while running through various kinds of analyst reports and news reports. The investment trust runs on NLP (Natural Language Processing) to assist asset managers in identifying undervalued shares and probable opportunities for profits.
# Partnership with an AI-based startup, H2O.ai, helps to focus on deep machine learning transparency and model interpretability to predict a better future. It assists in decision-making process for finance department and equity trading floor such as market making, providing liquidity to the bank and many more.
# This is the best opportunity to increase growth in hyper-personalised banking through conversational AI. AI will help in 24*7 two-way communication with innumerable personalised responses and feedbacks to the users.
 
Goldman Sachs prefers to focus on the net profit over the investment cost of the AI. The bank is eager to maintain its brand image through efficient customer experience, upgraded security and service enhancement.
</blockquote>
 
==2021.04.08 Machine learning futures algo trading surges at JP Morgan==
 
'''Hayley McDowell. The TRADE. April 8, 2021 10:02 AM GMT. https://www.thetradenews.com/machine-learning-futures-algo-trading-surges-at-jp-morgan/'''
 
'''Abstract:''' ''Peter Ward, global head of futures and options electronic execution at JP Morgan, tells Hayley McDowell that buy-side adoption of its reinfocement learning FICC futures algorithms has surged in recent years, accelerated by the market volatility in 2020.''
 
<blockquote>
Growth in fixed income futures algorithmic trading at JP Morgan has accelerated rapidly in 2020 as buy-side traders globally turned to the investment bank's machine learning-equipped algos to grapple with intense market volatility.
 
Speaking to The TRADE, Peter Ward, global head of futures and options electronic execution at JP Morgan, explains that while the volatility contributed to recent growth, adoption of futures algo trading has picked up pace with clients significantly in the last few years.
 
Since 2016, futures volumes traded via algos at JP Morgan has increased 40% year-on-year. In fact, algos now comprise of almost 20% of the bank's total futures trading flow, up significantly from roughly 4-5% in 2016 and 2017, figures seen by The TRADE have revealed.
 
The period of intense volatility in 2020 due to the global pandemic played a key role in the cumulative buy-side adoption of futures algos as traders became more accustomed to on-screen execution and liquidity.
 
"When liquidity is harder to source and there is more volatility, execution performance becomes challenged," Ward explains. "Clients are driven to look at the problem areas in executions and that's when we consult with them to figure out ways to bring in that performance. Maybe they should consider trading at higher volume at the open or close, or perhaps sitting out the first five minutes on the cash open because of the noise. All of that we can customise for them.
 
"I think the more challenges clients see in execution, the more opportunity there is for us to come in and help them, and the solution is increasingly the customised algorithm."
 
Customised algorithms have become particularly popular with traders in 2020 and in recent years. Volumes on customised algos at JP Morgan have roughly tripled in each of the past three years, alongside a 21% increase in the number of custom algos in 2020 to almost 50 customisations, up from close to zero in 2017.
 
The bank's flagship liquidity-seeking algorithm, known as Aqua, is the most common foundation for modified client parameters and customisations. A classic example of customisation is where a client wants to follow a particular trading pattern but then switch urgency or strategy based upon predefined triggers.
 
JP Morgan rebuilt its algo platform around five years ago to provide the buy-side with more choice about the parameters they can set on their side for algorithms, and there are further customisations that the bank's electronic traders can configure on behalf of clients. Ward adds this has allowed his team to have a "richer" dialogue with clients and demand is clearly there.
 
"There has always been demand for customised algos, even 10 years ago there was a lot of demand," he says. "We just didn't have a scalable way back then to adapt an algo to what a client really wanted. The reason for that is when a client wants something different, we needed developers to code that and then release it for implementation in the client's platform, which takes a lot of time."
 
'''Reinforcement learning'''
 
The Aqua algorithm has been a particular area of focus for JP Morgan recently. It uses a technology referred to as reinforcement learning to create advanced signals on order routing and placement.
 
With reinforcement learning, which is a form of machine learning, the algorithm essentially learns from itself over time by looking back at previous signals that it has generated and evaluates performance. The signals will dictate whether the algo crosses the market or stays passive.
 
Reinforcement learning technology was first applied to a recently launched model of Aqua that is focused on navigating quarterly roll dates when futures contracts expire. It can be a high-volume period and volatile time for traders as everybody is typically rolling in the same week to the next expiration date. In recent years, this activity has evolved from manual, voice-based trading to more electronic, low-touch trading.
 
"Previously, a lot of this business was executed through voice desks and one reason for that was because trading systems out there couldn't handle multi-leg products," he says. "As those systems have been developed in the last few years, we found more of that activity moving to electronic channels.
 
"A lot of volume goes through on calendar rolls and the challenge is around optimising that experience for the clients rather than imposing a model of trading without looking at the particular client objective."
 
In response to the trend and client demand, JP Morgan developed a model of its Aqua strategy, known as the Roll Algo, which went live not long ago for the most recent US treasury roll in February. It has been especially popular with buy-side traders, according to Ward.
 
"The Roll Algo model focuses on maximising liquidity and pricing opportunities by using signals that help it understand when to cross the spread. It's the most important area we are working on and has peaked the greatest interest from clients.
 
"It performed really well in February and there was a lot of client use in that period. With that, the algo learned a lot along the way so we can expect the performance in the next quarter's roll to be improved."
 
The Roll Algo is not the only new addition to JP Morgan's new strategy line-up. Advanced strategies like Target to trade around the cash or futures close, Multi Leg Strategy for trading multiple instruments at the same time across futures and US treasuries, and options algos have also been developed by the bank.
 
Volumes in options on futures surged in 2020 as trading floors at major derivatives exchanges like CME that facilitate options trading were forced to shut down. As a result, liquidity shifted to low-touch and electronic channels and JP Morgan's clients began to ask more questions about trading options through algorithms.
 
"Options on futures volumes have seen significant growth in the industry over the past few years and 2020 was a breakout year for liquidity on-screen," Ward adds.
 
"With that said there are still challenges and nuances to trading them and that's where we see opportunities to innovate and help our clients with their execution. This can be through simpler Peg and Cross type strategies and ultimately more targeted strategies using a delta or volatility reference."
 
JP Morgan expects buy-side adoption of futures algo trading to continue increasing in the near future, having been driven by ongoing market developments and trends over the past few years.
 
Explicit regulatory requirements on best execution and growing appetite among the buy-side to address challenges in futures and options market structure have been instrumental in the growth of this trend. Best execution essentially forces traders to establish benchmarks to measure performance and trading through algorithms can provide an effective way to do this.
 
New products have also entered the market where liquidity is shared on multiple markets, which presents challenges in trading those products. Nifty derivatives, for example, are now tradeable in both Singapore and India after the Singapore Exchange (SGX) and India's National Stock Exchange ended a two-year dispute which put SGX's futures into question.
 
Other developments such as extended hours in futures markets also means there are now more hours to trade what is often the same amount of volume. Add periods of decreased liquidity and increased volatility to the mix, traders have progressively sought algorithmic strategies and automated solutions for consistent execution in volatile products, and when targeting cash settlement periods, for example.
 
It's not just JP Morgan that is doubling down on efforts in futures algo trading. In January, rival investment bank Citi rolled out a suite of execution algorithms, including its flagship Arrival strategy, for futures markets across all major exchanges in the US, Europe, and Asia Pacific.
 
In contrast to JP Morgan, the electronic traders at Citi handle all of the algo customisations on behalf of clients. Head of EMEA futures electronic execution at Citi, Gordon Ball, said at the time clients don't want to enter numerous parameters to execute an order. He added: "the complexity of operating an intelligent algorithm and fine-tuning customisations sits with us, so our clients can focus on their overall investment and trading objectives".
 
Elsewhere, a start-up founded by former global head of trading at AQR Capital Management, Hitesh Mittal, launched its own suite of execution algorithms in early 2020 that aims to reduce costs for the buy-side with customised and high-performance strategies. In December, BestEx Research secured $5 million in funding as it prepares to roll out its algos in futures markets.
 
Amid the arms race in this space, JP Morgan's Ward predicts the pace of fixed income futures algo trading adoption, particularly customised algos, will continue apace in 2021. It remains a significant focus at JP Morgan as different buy-side clients are also now using algorithms to trade futures.
 
In the past few years, the type of buy-side client seeking algorithmic execution has shifted from being a relatively small number of large hedge fund clients to the more traditional managers, including pension funds, asset managers and insurance companies.
 
"Five years ago, there were pockets of interest in executing this way, depending on the specific trader or firm's appetite. It's now become far more mainstream, driven by broader electronification in fixed income markets as well more investment firms adopting more explicit execution benchmarks," Ward concludes.
</blockquote>
 
==2021.02.09 Electronic trading surges with traders eyeing the impact of machine learning==
 
'''Angharad Carrick. CITY A.M. Tuesday 9 February 2021 6:15 am. https://www.cityam.com/electronic-trading-surges-with-traders-eyeing-the-impact-of-machine-learning/'''
 
'''Abstract:''' ''Professional traders are anticipating artificial intelligence and machine learning to be the most influential technology over the next three years.''
 
<blockquote>
JP Morgan's flagship survey reveals more than half of professional and institutional traders anticipate machine learning to lead technology.
 
Currently a third of client traders predict mobile trading applications to be the most influential this year. Certainly the Reddit Gamestop rally powered by low cost trading platform is already testament to just how quickly the environment has changed.
 
Electronic trading picked up last year and all surveyed expect to increase electronic volumes this year. FX electronic trading to increase six per cent over the next two years to 84 per cent while credit should climb 12 per cent to 40 per cent.
 
"Year-on-year, the first half of 2020 saw a 45 per cent increase in volume of transactions and March, perhaps unsurprisingly, saw a new high water mark for notional value traded on the bank’s Execute on Mobile channel," Richard James, JP Morgan's head of macro markets execution said.
 
"The surge in activity was driven by what was also a new high in external client logins, about 30 per cent of the bank's user base were actively transacting over the channel with the balance accessing market information and analytics."
 
Banks and other financial institutions are already starting to use AI to execute trades quicker and more efficiently. The vast majority of surveyed traders &mdash; 71 per cent &mdash; agree that machine learning provides deeper analytics while just over half agree it optimises trade execution.
 
Looking forward to this year, just under half of those surveyed believe the pandemic will continue to have the biggest impact on markets this year. In a sign of just how much the pandemic has taken over market discussions, international trade tensions come just fourth in traders' concerns, with only nine per cent concerned over the prospect of trade wars.
 
When it comes to traders' day-to-day life 55 per cent will continue to work from home for average of four days a week.
</blockquote>
 
==2020.12.29 Nvidia rival Graphcore raises $222 million for AI chips with potential IPO on the horizon==
 
'''Sam Shead. CNBC. Published Tue, Dec 29 2020, 6:25 AM EST, updated Tue, Dec 29 2020, 6:49 AM EST. https://www.cnbc.com/2020/12/29/graphcore-raises-222-million-to-take-on-nvidia-with-ai-chips.html'''
 
'''Key points:'''
 
* Graphcore has raised $222 million as it looks to take on U.S. rivals Nvidia and Intel.
* The Series E funding round, which comes less than a year after Graphcore raised a $150 million extension to its last round, values the company at $2.77 billion.
* Total investment in Graphcore now stands at $710 million.
 
<blockquote>
London&mdash;U.K.-based chipmaker Graphcore announced Tuesday that it had raised $222 million of investment as it looks to take on U.S. rivals Nvidia and Intel.
 
Graphcore said it will use the funding to support its global expansion and to accelerate the development of its intelligence processing units&nbsp;(IPUs), which are specifically designed to power artificial intelligence software. The company has already shipped tens of thousands of its chips to customers including Microsoft and Dell.
 
The Series E funding round, which comes less than a year after Graphcore raised a $150 million extension to its last round, values the company at $2.77 billion, up from $1.5 billion in 2018.
 
Graphcore CEO and co-founder Nigel Toon told CNBC in July: "We're now at the point where we're not really looking for venture investors in the business. We're more interested in companies that would be long term investors and holders of the stock, perhaps, in the public markets, if we ever reach that point."
 
At the time, Toon said going public is "ideally what we would like to do" but he stressed "lots of water still has to flow under the bridge before we get to that point."
 
Total investment in Graphcore now stands at $710 million and the four-year-old company has $440 million of cash on hand.
 
The latest funding round was led by the Ontario Teachers' Pension Plan Board while other new investors included private equity investor Baillie Gifford, venture capital investor Draper Esprit, as well as funds managed by Fidelity International and Schroders.
 
On Tuesday, Toon said in a statement: "Having the backing of such respected institutional investors says something very powerful about how the markets now view Graphcore. The confidence that they have in us comes from the competence we have demonstrated building our products and our business."
 
He added: "We have created a technology that dramatically outperforms legacy processors such as GPUs, a powerful set of software tools that are tailored to the needs of AI developers, and a global sales operation that is bringing our products to market."
 
'''Serial chip entrepreneurs'''
 
Graphcore was founded in June 2016 in Bristol, England, by Toon and Simon Knowles, who sold their previous chip company, Icera, to Nvidia for $435 million in 2011. The pair formed the initial idea for Graphcore in a small pub called the Marlborough Tavern in Bath in January 2012.
 
Today, the company employs around 450 people in Bristol, Cambridge, London, Beijing, Oslo, Palo Alto, Seattle, and Hsinchu in Taiwan. It expects the number to grow to 600 by the end of 2021.
 
But the rapid growth hasn't come cheap. It made a pre-tax loss of $95.9 million on revenues of $10.1 million in 2019, according to an annual report filed on U.K. business registry Companies House.
 
Santa Clara heavyweights Intel and Nvidia are two of the obvious front runners in the AI chip market given their expertise in chip making. The companies haven't disclosed how many of their AI-optimized chips have been sold. However, over a trillion computer chips are expected to be shipped in 2020, according to market data website Statistica. In 2019, Intel's slice of the overall chip market came in at 15.7% and it has been the market leader every year since 2008, with the exception of 2017 when Samsung took the number one spot.
 
Graphcore's Toon criticized Nvidia's plan to buy U.K. chip designer Arm from SoftBank for $40 billion, saying it is bad for competition.
 
"We believe that Nvidia's proposed acquisition of Arm is anti-competitive," he said. "It risks closing-down or limiting other companies' access to leading edge CPU processor designs which are so important across the technology world, from datacenters, to mobile, to cars and in embedded devices of every kind."
 
Google, Amazon and Apple are also working on their own AI chips.
 
'''Sequoia backs Nvidia and Graphcore'''
 
Previous investors in Graphcore include the likes of Microsoft and BMW iVentures, as well as venture firms like London's Atomico and Silicon Valley's Sequoia, which has also backed Nvidia.
 
Last month Sequoia partner Matt Miller told CNBC: Graphcore "are in this position where they always have people coming at them trying to give them more money. So, they do not need funding. They are well funded for the next several years, but they definitely have people trying to invest in the company."
 
He added: "I don't think that you have to take on Nvidia because the market is so huge. Taking on Nvidia is like this huge task. It's a huge company with billions of revenue and incredible teams doing all sorts of wonderful things. I think that what Graphcore has the opportunity to do is be a very strong player in the AI microprocessor market. It continues to have great progress with many of the cloud providers, and many people want to be diversified. They don't want to be all in with one chip."
 
Graphcore launched its second generation IPU earlier this year despite disruption from the coronavirus pandemic.
</blockquote>
 
==2020.08.09 ADIA Hires Marcos Lopez de Prado as Global Head of Quant Research==
 
'''Sovereign Wealth Fund Institute (SWFI). Posted on 09/08/2020. https://www.swfinstitute.org/news/81369/adia-hires-marcos-lopez-de-prado-as-global-head-of-quant-research'''
 
<blockquote>
The Abu Dhabi Investment Authority (ADIA) hired Marcos López de Prado as global head of quantitative research & development. Prado is a Cornell University professor. Prado is joining a newly-formed investment group at ADIA within the strategy and planning department. This group seeks to apply a systematic, science-based approach to developing and implementing investment strategies.
 
Most recently, Prado was professor of practice at Cornell University’s School of Engineering, teaching machine learning, according to the statement.
 
Prado is also the CIO of True Positive Technologies (TPT). PT is currently engaged by clients with a combined AUM in excess of US$ 1 trillion. Prado launched TPT after he sold some of his patents to AQR Capital Management, where he was a principal and AQR’s first head of machine learning. Marcos also founded and led Guggenheim Partners’ Quantitative Investment Strategies business, where he managed up to US$ 13 billion in assets.
</blockquote>
 
==2020.02.14 Developers now make up a quarter of Goldman Sachs' workforce==
 
'''Jia Jen Low. T_HQ. 14 February 2020. https://techhq.com/2020/02/developers-now-make-up-quarter-of-goldman-sachs-workforce/'''
 
'''Abstract:''' ''The leading finance firm says it's now competing with Silicon Valley tech giants for talent.''


<blockquote>
<blockquote>
Line 33: Line 470:
</blockquote>
</blockquote>


==Finance is headed for AI mass adoption&mdash;and soon==
==2020.02.06 Finance is headed for AI mass adoption&mdash;and soon==


Jia Jen Low. T_HQ. 6 February 2020. <span class="plainlinks">https://techhq.com/2020/02/finance-is-headed-for-ai-mass-adoption-and-soon/</span>
'''Jia Jen Low. T_HQ. 6 February 2020. https://techhq.com/2020/02/finance-is-headed-for-ai-mass-adoption-and-soon/'''


Abstract: "While other industries struggle with AI, finance members are locked in an arms race.
'''Abstract:''' ''While other industries struggle with AI, finance members are locked in an arms race.''


<blockquote>
<blockquote>
Line 63: Line 500:
</blockquote>
</blockquote>


==Rival banks are hiring technologists from Goldman Sachs==
==2020.01.27 Rival banks are hiring technologists from Goldman Sachs==


Sarah Butcher. eFinancialCareers. 27 January 2020. <span class="plainlinks">https://www.efinancialcareers.com/news/2020/01/rival-banks-hiring-goldman-sachs-technologists</span>
'''Sarah Butcher. eFinancialCareers. 27 January 2020. https://www.efinancialcareers.com/news/2020/01/rival-banks-hiring-goldman-sachs-technologists'''


<blockquote>
<blockquote>
Line 81: Line 518:
</blockquote>
</blockquote>


==A Global AI in Financial Services Survey==
==2020.01m A Global AI in Financial Services Survey==


Lukas Ryll, Mary Emma Barton, Bryan Zheng Zhang, Jesse McWaters, Emmanuel Schizas, Rui Hao, Keith Bear, Massimo Preziuso, Elizabeth Sege, Robert Wardrop, Raghavendra Rau, Pradeep Debata, Philip Rowan, Nicola Adams, Mia Gray, Nikos Yerolemou. University of Cambridge Judge Business School. January 2020. <span class="plainlinks">https://www.jbs.cam.ac.uk/faculty-research/centres/alternative-finance/publications/transforming-paradigms/</span>
'''Lukas Ryll, Mary Emma Barton, Bryan Zheng Zhang, Jesse McWaters, Emmanuel Schizas, Rui Hao, Keith Bear, Massimo Preziuso, Elizabeth Sege, Robert Wardrop, Raghavendra Rau, Pradeep Debata, Philip Rowan, Nicola Adams, Mia Gray, Nikos Yerolemou. University of Cambridge Judge Business School. January 2020. https://www.jbs.cam.ac.uk/faculty-research/centres/alternative-finance/publications/transforming-paradigms/'''


Abstract: "This report presents the findings of a global survey on AI in Financial Services jointly conducted by the Cambridge Centre for Alternative Finance (CCAF) at the University of Cambridge Judge Business School and the World Economic Forum in Q2-Q3 2019. Representing one of the largest global empirical studies on AI in Financial Services, a total of 151 respondents from 33 countries participated in the survey, including both FinTechs (54 per cent of the sample) and incumbent financial institutions (46 per cent of the sample). The study was supported by EY and Invesco."
'''Abstract:''' ''This report presents the findings of a global survey on AI in Financial Services jointly conducted by the Cambridge Centre for Alternative Finance (CCAF) at the University of Cambridge Judge Business School and the World Economic Forum in Q2-Q3 2019. Representing one of the largest global empirical studies on AI in Financial Services, a total of 151 respondents from 33 countries participated in the survey, including both FinTechs (54 per cent of the sample) and incumbent financial institutions (46 per cent of the sample). The study was supported by EY and Invesco.''


<blockquote>
<blockquote>
Line 123: Line 560:
</blockquote>
</blockquote>


==Goldman Sachs hunts AI experts for all-important quant team==
==2018.11.23 Goldman Sachs hunts AI experts for all-important quant team==


Paul Clarke. Financial News. Friday November 23, 2018 4:29 am. <span class="plainlinks">https://www.fnlondon.com/articles/goldman-sachs-hunts-ai-experts-for-all-important-quant-team-20180130</span>
'''Paul Clarke. Financial News. Friday November 23, 2018 4:29 am. https://www.fnlondon.com/articles/goldman-sachs-hunts-ai-experts-for-all-important-quant-team-20180130'''


Abstract: "US bank is building its vast strats department by hiring a new generation of machine learning and artificial intelligence specialists."
'''Abstract:''' ''US bank is building its vast strats department by hiring a new generation of machine learning and artificial intelligence specialists.''


<blockquote>
<blockquote>
Line 151: Line 588:
</blockquote>
</blockquote>


==Which?==
==2017.06.19 London currency trader bets on machine learning for high speed trading foray in US stocks==
 
'''Bloomberg. June 19, 2017 1:17 PM'''
 
'''Abstract:''' ''A tiny London firm with no human traders made its name last year beating banks to climb up the currency trading ranks. Now it wants a bite of something new: the $27 trillion US stock market.''


<blockquote>
<blockquote>
Growth in fixed income futures algorithmic trading at JP Morgan has accelerated rapidly in 2020 as buy-side traders globally turned to the investment bank's machine learning-equipped algos to grapple with intense market volatility.
A tiny London firm with no human traders made its name last year beating banks to climb up the currency trading ranks. Now, it wants a bite of something new: the $27 trillion US stock market. XTX Markets Ltd. is only two years old, but its executives say it has what it takes to compete with more established American trading Goliaths in the world's largest, most complex and most saturated equity market. The firm is prepping a new Manhattan office, lining up the necessary regulatory nods and scooping up a big-name hire: Eric Swanson, who helped Bats Global Markets Inc. become the nation's second-biggest stock exchange operator. With Swanson, who joined this month, XTX can "go from having a toehold, to being a more significant player in the US," says Zar Amrolia, co-chief executive officer of XTX who formerly ran digital technology at Deutsche Bank AG. "We are just rolling out what we think is a successful quantitative research-driven approach to market making."
 
It won’t be easy. A Dutch speed trader, Flow Traders NV, that last year kicked off a similar US expansion, is having trouble. The choppy, scattered nature of the market has few parallels. Firms that want to compete will have to connect to 12 national securities exchanges, scour constantly for trading risk, suck in proprietary data feeds and ward off any behavior that could run afoul of regulators, all at once. For the fastest firms, trading strategies can be made or broken by millionths of a second.


Speaking to The TRADE, Peter Ward, global head of futures and options electronic execution at JP Morgan, explains that while the volatility contributed to recent growth, adoption of futures algo trading has picked up pace with clients significantly in the last few years.
'''Smart, Not Fast'''


Since 2016, futures volumes traded via algos at JP Morgan has increased 40% year-on-year. In fact, algos now comprise almost 20% of the bank's total futures trading flow, up significantly from roughly 4-5% in 2016 and 2017, figures seen by The TRADE have revealed.
Amrolia's not fazed by speedier rivals &mdash; he says his goal is to be "smart, not fast." Taking its name from a mathematical expression, XTX uses technology to forecast where prices for securities will be in a matter of minutes or hours. Amrolia contrasts this strategy to some North American firms that rely on the speediest networks for getting information, and carrying out trading decisions based on it. Virtu Financial Inc. and Citadel Securities LLC will soon be among XTX's biggest rivals. XTX's focus on machine learning puts the firm "at the forefront" of trading technology, Swanson said. Their strategies will be put to the test in the U.S., which hosts more than one-third of global equity trading value, and holds a complex web of big-name exchanges and dozens of smaller private dark-pool venues. "That presents a challenge for everyone," says Swanson, Americas CEO at XTX and the former general counsel at Bats, which is now owned by CBOE Holdings Inc. "We're up to managing that challenge."


The period of intense volatility in 2020 due to the global pandemic played a key role  in the cumulative buy-side adoption of futures algos as traders became more accustomed to on-screen execution and liquidity.
XTX made its name last year after managing to leapfrog big banks to place fourth in spot currency trading. The firm repeated the spot-trading feat in the 2017 Euromoney Institutional Investor Plc survey, despite slipping to 12th in this year’s rankings for overall trading.
</blockquote>
 
'''Avoiding Bad Trades'''
 
XTX says its technology and the way it sends orders into the market allows its systems to be alerted quickly when things go wrong, giving them the ability to make lightning-fast pivots to avoid bad trades. The firm employs just 78 people, according to spokesman Tim Moxon. "They have a good combination of ability to manage their own risk and create really good prices," said Steve Grob, global director of group strategy at Fidessa Group Plc. Its Amsterdam rival Flow, the largest trader of European exchange-traded funds, pushed into the US late last year, hoping to profit from buying and selling ETFs that no one else will touch. Addled by calmer markets, Flow's first-quarter profit slumped. Trading income in the Americas over the three-month period dropped 23 percent.


=Reinforcement Learning=
One part of Flow Traders's strategy was clinching regulatory approval to trade directly with large U.S. investors. That may also be a winning option for XTX if the firm can nab similar permissions, which in some cases would allow the firm to bypass public exchanges. "That's where you can be disruptive, if you've got the right technology behind you," Grob said.


=Quantum Computing=
The spread-out nature of the US stock market can pose obstacles for any newcomer to the region, said Michael Beller, chief executive officer of Thesys Technologies LLC, a company that sells market-structure technology to help firms manage that vast trading network. "It's complicated," Beller said in an interview. "You can't walk in, start trading and get results."
</blockquote>

Latest revision as of 22:31, 21 December 2021

The importance of machine learning in finance continues to grow. This page attempts to collect the most important research papers and developments in this area as mentioned in the Press.

Research papers

2021.07.12 A rigorous and robust quantum speed-up in supervised machine learning

Yunchao Liu, Srinivasan Arunachalam, and Kristan Temme. Nature Physics (2021). https://www.nature.com/articles/s41567-021-01287-z

Recently, several quantum machine learning algorithms have been proposed that may offer quantum speed-ups over their classical counterparts. Most of these algorithms are either heuristic or assume that the data can be accessed quantum-mechanically, making it unclear whether a quantum advantage can be proven without resorting to strong assumptions. Here we construct a classification problem with which we can rigorously show that heuristic quantum kernel methods can provide an end-to-end quantum speed-up with only classical access to data. To prove the quantum speed-up, we construct a family of datasets and show that no classical learner can classify the data inverse-polynomially better than random guessing, assuming the widely believed harness of the discrete logarithm problem. Furthermore, we construct a family of parameterized unitary circuits, which can be efficiently implemented on a fault-tolerant quantum computer, and use them to map the data samples to a quantum feature space and estimate the kernel entries. The resulting quantum classifier achieves high accuracy and is robust against additive errors in the kernel entries that arise from finite sampling statistics.

2021.05.21 Multi-Horizon Forecasting for Limit Order Books: Novel Deep Learning Approaches and Hardware Acceleration using Intelligent Processing Units

Zihao Zhang, Stefan Zohren. Submitted on 21 May 2021. https://arxiv.org/abs/2105.10430

We design multi-horizon forecasting models for limit order book (LOB) data by using deep learning techniques. Unlike standard structures where a single prediction is made, we adopt encoder-decoder models with sequence-to-sequence and Attention mechanisms, to generate a forecasting path. Our methods achieve comparable performance to state-of-art algorithms at short prediction horizons. Importantly, they outperform when generating predictions over long horizons by leveraging the multi-horizon setup. Given that encoder-decoder models rely on recurrent neural layers, they generally suffer from a slow training process. To remedy this, we experiment with utilising novel hardware, so-called Intelligent Processing Units (IPUs) produced by Graphcore. IPUs are specifically designed for machine intelligence workload with the aim to speed up the computation process. We show that in our setup this leads to significantly faster training times when compared to training models with GPUs.

The Press

2021.07.19 Quantum machine learning achieves advantage in IBM research

Sam Cox. Silicon Republic. https://www.siliconrepublic.com/machines/quantum-advantage-ibm-machine-learning

Abstract: In a new paper by IBM, quantum machine learning was able to discern patterns where classical computers missed the signal in the noise.

Quantum computing is a field full of promise but has yet to prove many of its supposed advantages. IBM is confident that quantum advantage will come to fruition but is still working away to establish the proof in the pudding.

Its worldwide machines have shown quantum superiority in other domains, but machine learning is still in the works. However, a new research paper from IBM Quantum has tackled a central question related to quantum machine learning: which quantum algorithms are currently capable of delivering a provable quantum advantage over classical machine learning algorithms?

Proposals in quantum machine learning are often driven by the challenge to find algorithms that can be tested in the near-term with only conventional access to data.

One such class of algorithms is the proposal for quantum enhanced feature spaces, also known as quantum kernel methods. In these set-ups, the quantum computer steps in for just part of the process in the algorithm.

These were the focus of IBM's latest research.

An area of potential use for these kernel methods is classification problems. These are one of the most fundamental problems in machine learning.

They begin by training an algorithm on data, called a training set, where data points include one of two labels. Following the training phase is a testing phase where the algorithm needs to classify a new data point that has not been seen before.

A standard example is giving a computer pictures of dogs and cats, and from this dataset it classifies all future images as a dog or a cat.

Ultimately, the goal of an efficient machine learning algorithm for classification should be to generate an accurate label in an amount of time that scales polynomially with the size of the input.

In this case, the researchers started with a conventional machine learning model to learn the kernel function, which finds the relevant features in the data to use for classification.

The quantum advantage comes from the fact that the researchers were able to construct a family of datasets for which only quantum computers can recognise the intrinsic labelling patterns.

The researchers used a problem that separates classical and quantum computation: computing logarithms in a cyclic group, where you can generate all the members of the group using a single mathematical operation.

For classical computers, the dataset looked like meaningless noise, whereas the quantum computers were able to work through the data.

The team demonstrated this by constructing a family of classification problems and showed that no efficient classical algorithm could do better than random guessing when attempting to solve these problems.

They also constructed a quantum feature map. This is a way to view complicated data in a higher-dimensional space to pull out patterns. When used alongside the corresponding kernel function, the researchers were able to predict the labels with high accuracy.

What's more, they could show that the high accuracy persists in the presence of finite sampling noise from taking measurements, a form of noise that needs to be considered even for fault-tolerant quantum computers.

IBM highlighted that there will still be many real-life problems for which this quantum algorithm does not perform better than conventional classical machine learning algorithms.

To obtain a quantum advantage, the classification problem must stick to this cyclical structure described above. This is an important caveat, as IBM's further research will be aimed at discussing how generalisable this structure is.

Nevertheless, the researchers were confident that this proves an end-to-end quantum speed-up for a quantum kernel method implemented fault-tolerantly with realistic assumptions.

Another potential sticking point could be the hardware limitations of modern quantum computers.

With the field rapidly advancing and ever-bigger offerings being set up across the globe however, solving this limitation might just be a case of when rather than if.

2021.07.16 Blender Bot 2.0: An open source chatbot that builds long-term memory and searches the internet

Jason Weston (Research Scientist), Kurt Shuster (Research Scientist). Facebook AI. July 16, 2021. https://ai.facebook.com/blog/blender-bot-2-an-open-source-chatbot-that-builds-long-term-memory-and-searches-the-internet/

  • Facebook AI Research has built and open-sourced BlenderBot 2.0, the first chatbot that can simultaneously build long-term memory it can continually access, search the internet for timely information, and have sophisticated conversations on nearly any topic. It's a significant update to the original BlenderBot, which we open-sourced in 2020 and which broke ground as the first to combine several conversational skills—like personality, empathy, and knowledge—into a single system.
  • When talking to people, BlenderBot 2.0 demonstrated that it's better at conducting longer, more knowledgeable, and factually consistent conversations over multiple sessions than its predecessor, the existing state-of-the-art chatbot.
  • The model takes pertinent information gleaned during conversation and stores it in a long-term memory so it can then leverage this knowledge in ongoing conversations that may continue for days, weeks, or even months. The knowledge is stored separately for each person it speaks with, which ensures that no new information learned on one conversation is used in another.
  • During conversation, the model can generate contextual internet search queries, read the results, and incorporate that information when responding to people's questions and comments. This means the model stays up-to-date in an ever-changing world.
  • Today we're releasing the complete model, code, and evaluation setup, as well as two new conversational data sets—human conversations bolstered by internet searches, and multisession chats with people that reference previous sessions—used to trin the model, so other researchers can reproduce this work and advance conversational AI research.

[...]

2021.07.09 China Targets AI Dominance by 2030

Bloomberg Quicktake. July 9th, 2021, 9:48 AM GMT+0100. https://www.bloomberg.com/news/videos/2021-07-09/china-targets-ai-dominance-by-2030-video

Endless cups of coffee, hi-tech tools to help doctors, a masseuse who never gets tired... Just some of the latest innovations on display at this year's World Artificial Intelligence Conference in Shanghai.

"The advantages are first of all its trajectory is a digital record of master-level massage skills. Compared with ordinary masseurs, its techniques are more accurate and precise. Second, with the 3D images it can find different acupuncture points according to different people's body shapes and can massage in a targeted manner."

Meet Walker X, he could be your next butler.

"The functions that we have been able to achieve in the family setting include bringing you a Coke from the refrigerator, opening a drink, bringing you a newspaper, fetching an umbrella, and watering the flowers, playing chess with you, singing and doing homework with you. It can voice chat and has other emotional interactions. In the future, we will make robots that have a more diversified role in the entire home. Walker will play the role to help the family, accompany you as you grow up and accompany children and the elderly. It can be a teacher of the family as well as a helper who can do a variety of housework in the family.

AI is also helping drive medical progress. Morphogo is a bone marrow cytomorphology analysis system. It can make a diagnosis within 1 day compared to the usual 5 to 10 days with traditional hospital methods.

Yang Chuhu, ZhiWei Information Technology Co. Ltd.: "This kind of high-efficiency detection is crucial for accurate diagnosis and timely treatment which cannot be achieved only by manual work. Taking China for example, China has vast remote areas and it is impossible for people living in remote villages to go to a big hospital for a rapid morphological examination of bone marrow. Without such a system, it can be said that the life and health of patients with blood diseases in remote areas cannot be effectively guaranteed. This is not a problem for China alone. This is a global problem."

China has set its sights on building a domestic AI industry worth nearly $150 billion in the next few years and of becoming the leading AI power by 2030.

2021.07.07 U.K. Deploys Fastest Supercomputer to Fight Dementia, MS

Adeola Eribake. Bloomberg. July 7, 2021, 12:01 AM GMT+1. https://www.bloomberg.com/news/articles/2021-07-06/nvidia-boots-up-u-k-supercomputer-to-boost-drug-research-nhs

Abstract: Developed in partnership with AstraZeneca, GlaxoSmithKline. Nvidia waiting for approval for its $40 billion takeover of Arm.

Nvidia Corp. has built a U.K. supercomputer for scientists and drug makers that will allow researchers to model diseases, discover new medicines and advance knowledge of the human genome.

The Cambridge-1 will be the country's most powerful and one of the top 50 supercomputers worldwide. It was developed in partnership with AstraZeneca Plc and GlaxoSmithKline Plc, Nvidia said in a statement on Wednesday.

Other founding members include Guy's Hospital and St Thomas' Hospital trust, part of the U.K.'s National Health Service, as well as King's College London.

"Through this partnership, we will be able to use a scale of computational power that is unprecedented in health-care research," said Sebastien Ourselin, head of the School of Biomedical Engineering & Imaging Sciences at King's College London. "It will be truly transformational for the health and treatment of patients."

Oxford Nanopore, a DNA sequencing startup that's preparing to debut on the London Stock Exchange will also get access to Cambridge-1, allowing it to improve its artificial intelligence algorithms in hours instead of days, the startup said in the statement.

Rather than a single unit, the Cambridge-1 is made up of several modules working together. This parallel processing lets the computer carry out tasks simultaneously, allowing for greater speeds. It was also faster to build, coming together in about a quarter of the time it takes to construct similar machines.

The supercomputer has a computing capability of more than 400 petaflops. That's close to the world's fastest, the Fugaku machine in Japan. Top500, which publishes a twice-yearly list of the fastest supercomputers, included the Cambridge-1 at No. 41 on its most recent ranking, published at the end of June.

The speed and capabilities that supercomputers offer scientists are crucial as they try to make sense of unprecedented amounts of data. "We accumulated more data in one quarter in 2020 than in the last 300 years," said Kim Branson, GSK's senior vice president and global head of AI-ML.

The machine, which cost $100 million to build, was announced in October, a month after Nvidia launched a $40 billion bid for semiconductor designer Arm Ltd. Arm's technology is ubiquitous in smartphones and other consumer electronics, making the deal controversial among Nvidia's competitors, who rely on the designs. The U.K. has also expressed concerns that selling Arm, which is currently owned by Softbank Group Corp., has national security implications and the country's competition authority is due to submit a report at the end of the month.

Nvidia said it's already developing another supercomputer, which is to be made with Arm technology.

The Cambridge-1 will be kept at a site run by Kao Data Campus just outside London and about 40 miles south of its namesake city.

King's College and London hospital group Guy's and St Thomas' will use the new AI capability to learn from MRI scans and then generate synthetic brain images, creating models to help scientists better understand diseases such as dementia, stroke, brain cancer and multiple sclerosis. That could allow for earlier diagnosis and treatment, the university and teaching hospital said.

Nvidia said it would work with its partners to share the insights gained with the wider scientific community. For example, AstraZeneca will open up one of its projects, an AI-based model for chemical structures to help with drug discovery, to other scientists.

"Cambridge-1 will empower world-leading researchers in business and academia with the ability to perform their life's work on the U.K.'s most powerful supercomputer, unlocking clues to disease and treatments at a scale and speed previously impossible in the U.K.," said Jensen Huang, founder and chief executive officer of Nvidia, in a statement.

2021.07.05 China beats Google to claim the world's most powerful quantum computer

Matthew Sparkes. NewScientist. 5 July 2021. https://www.newscientist.com/article/2282961-china-beats-google-to-claim-the-worlds-most-powerful-quantum-computer/

A team in China has demonstrated that it has the world's most powerful quantum computer, leapfrogging the previous record holder, Google.

Jian-Wei Pan at the University of Science and Technology of China in Hefei and his colleagues say their quantum computer has solved a problem in just over an hour that would take the world's most powerful classical supercomputer eight years to crack, and may yet be capable of exponentially higher performance.

2021.06.24 Deserve Valuation Tops $500 Million as MasterCard and Ally Ventures Invest

Gillian Tan. Bloomberg. June 24, 2021, 12:00 PM GMT+1. https://www.bloomberg.com/news/articles/2021-06-24/deserve-valuation-tops-500-million-as-mastercard-ally-invest

Key points: Goldman, Mission Holdings and Sallie Mae also backed startup. Palo Alto firm seeks to be profitable in first half of 2023.

Deserve, a credit-card technology startup, raised $50 million from backers including Mastercard Inc. and Ally Financial Inc.'s strategic investment arm, its chief executive officer said in an interview.

Mission Holdings, Goldman Sachs Group Inc., Sallie Mae and other existing investors participated in the funding round, proceeds of which will be used to launch card programs and accelerate growth, CEO Kalpesh Kapadia said. The company is seeking to provide rewards unlike traditional methods of cash-back, points or miles. Already, one of its partners, BlockFi, offers cryptocurrency rewards and another, Seneca Women, rewards spending at women-owned businesses.

The founding round values the Palo Alto, California-based company at more than $500 million, according to a person with knowledge of the matter.

"We are Instagram to the credit-card industry's Kodak," Kapadia said, referencing Deserve's plastic- and metal-free operations. Its software enables companies to offer digital credit cards that can be applied for and used on iPhones or Android devices. "People may forget their wallets, but they never forget their phones," he added.

Deserve and the Apple Card have a huge lead over competitors, said Saurabh Mittal, founder of Mission Holdings, which first invested in the startup's seed round in 2014. "There's an exponential growth story that will play out over the next couple of decades," he said.

The company is striving to achieve profitability by the first half of 2023, and will consider options for going public when it's closer to that milestone, Kapadia said. It will explore new lending products that could include payroll advances, "buy now pay later" programs and installment loans, he added.

Deserve uses so-called deep machine learning and artificial intelligence, and says its underwriting process, in addition to FICO scores, relies upon income and employment data. The company touts "signle-call resolution" user support which should lead to fewer delinquencies, charge-backs and disputes.

"Credit cards are an area of interest for Ally, and Deserve &mdash with its digital-first approach — is a disruptive company in this space," Ally chief strategy & corporate development officer Dinesh Chopra said. Ally doesn't currently offer a credit card but has expressed inteest in unsecured lending, he added.

Ally last year abandoned plans to buy subprime credit card lender CardWorks.

(Updates to add details of a partnership in the second paragraph. An earlier story was corrected to remove a reference to a partner company.)

2021.06.21 Former Morgan Stanley Traders Raise $100M to Turn Crypto Startup into Unicorn

Vishawam Sankaran. Independent. Monday 21 June 2021 06:59. https://www.independent.co.uk/life-style/gadgets-and-tech/cryptocurrency-morgan-stanley-amber-group-b1869555.html

Abstract: Firm with over 300 employees in Hong Kong, Taipei, Seoul and Vancouver now plans to expand

After a successful fundraiser, Amber Group, a Hong Kong-based cryptocurrency startup founded by former Morgan Stanley traders, has raised $100 million, scoring a pre-money valuation of $1 billion.

Michael Wu, co-founder and CEO of the crypto financial services firm with over 300 employees in Hong Kong, Taipei, Seoul and Vancouver, said in a statement that the funding would be used to "extend global operations to meet client demand and develop market solutions for the world's leading crypto investors and companies."

"We've had record months over the past quarter across both client flow and on-exchange market-making volumes. Our cumulative trading volumes have doubled from $250 billion since the beginning of the year to over $500 billion," Wu added.

In 2017, the founders, including former Morgan Stanley traders Michael Wu, Luke Li, Wayne Huo, Tiantian Kullander, and Tony He, had initially sought to apply machine learning to quantitative trading, but pivoted to crypto in 2019 when the trading volumes for the virtual currency increased.

The series B funding, which has brought the several times more than the series A round in 2019, was bankrolled by several big-name financiers including Tiger Brokers, Tiger Global Management, Sky9 Capital, Tru Arrow Partners, Arena Holdings, Gobi Partners, and DCM Ventures.

In the series A funding as well the startup was backed by several high-profile investors including Paradigm and Pantera Capital, Polychain Capital, Dragonfly Capital, Blockchain.com, Fenbushi Capital, and Coinbase Ventures, the firm noted.

Amber Group serves over 500 institutional clients, trading over "$330 billion across 100+ electronic exchanges," since its inception, expanding also to retail consumers with the launch of its mobile app in 2020.

Venture capital firms have shown growing interest in the crypto economy in the last six months.

Last month, Babel Finance, another Hong Kong-based crypto asset manager secured funding of $40 million from several institutional investors, including Tiger Global.

Matrixport, another cryptocurrency lending service started by Bitmain's influential founder Jihan Wu, is also seeking a new capital injection, Bloomberg reported.

Blockchain intelligence platform TRM Labs also secured $14 million in Series A funding from investors including the venture capital arms of PayPal and Salesforce.

TRM helps blockchain companies, financial institutions and law enforcement agencies investigate, detect and prevent financial crimes related to cryptocurrencies.

2021.06.08 Man Group-Oxford Quants Say Their AI Can Predict Stock Moves

Amy Thomson (with assistance by Justina Lee, and Julius Domoney). Bloomberg. June 8, 2021, 10:24 AM GMT+1. https://www.bloomberg.com/news/articles/2021-06-08/man-group-oxford-quants-say-their-ai-can-predict-stock-moves

Abstract: Machine-learning program hits 80% success rate over 30 seconds. Tape bombs and processing power are challenges to deployment.

Man Group Plc-backed researchers at the University of Oxford say they've created a machine-learning program that can project how share prices move—notching an 80% success rate for the equivalent of about 30 seconds of live trading.

Artificial-intelligence experts at the Oxford-Man Institute of Quantitative Finance exploted principles from natural-language processing to trawl liquidity data across limit order books, a record of buying and selling at preset prices.

In a potential step forward for fast-money traders seeking to time markets, the algorithm figured out the direction of a price move over a period of 100 ticks, the equivalent of about 30 seconds to two minutes of trading depending on market conditions.

"In the multi-step forecasting, we effectively have a model which is trained to make a forecast at a smaller horizon," said Stefan Zohren, an associate professor at the institute who co-authored the research. "But we can feed this information back into itself and roll forward the prediction to arrive at longer-horizon forecasts."

The algorithm, which remains at the testing stage, has a clear appeal for hedge fund managers who typically break up large stock orders into multiple smaller transactions, according to Anthony Ledford, chief scientist at Man AHL.

"If we think we're going to enter a position, we may hold that position for several weeks, but actually making the trade that gives you that exposure happens over a much shorter-time period," via a number of smaller trades, he said.

Where this model "has much more of an impact for us, it's understanding how to release and trade those smaller pieces into the market—and each one of those may be done in a timescale of a few minutes," said Ledford, a former winner of the Royal Statistical Society's Research Prize.

With endless tape bombs hitting financial markets, accuracy rates for these kind of models are taken with a pinch of salt when it comes to the real world. But the Man-Oxford findings illustrate the allure of using AI to divine complex relationships across data points that in theory can run into the billions.

As increasing industry competition whittles down returns in core strategies, quants are vying ever-more to deploy programs that learn statistical patterns in equities in order to cut trading costs and find new investing signals.

Predicting share movements one or two milliseconds before everyone else does has been the goal of strategies such as statistical arbitrage and exchange colocation for more than a decade. Yet leveling computational firepower at stock prices is a crowded field, with an entrenched arms race among the biggest shops ensuring that no technical advantage lasts long.

The hedge fund, with $127 billion in assets under management, provided initial funding for the institute in 2007, and has committed more than 30 million pounds ($42.5 million). The research center was responsible for introducing Man to graphics processing units, or GPUs, that are able to handle the intensive processing demands of artificial intelligence about a decade ago.

Multi-horizon forecast models using statistical analysis have been around for years now, channeling market variables into predictions about how a stock will move over different time periods.

The new techniques used in the Oxford-Man Institute's research, which increase the potential accuracy of the predictions over a longer period of time, channeled principles from natural-language processing. Zohren, who worked with research associate Zihao Zhang on the paper, compared the model to a program that can translate a sentence to French from English by building inferences incrementally.

But to make the Oxford-Man model work, the AI has to be able to process a huge amount of data quickly.

The researchers turned to Bristol, England-based Graphcore's Intelligence Processing Unit, part of a pizza box-sized system designed specifically to handle the demands of an AI program. In the trials, Graphcore's chip performed about 10-times faster than GPUs.

While the research and the Graphcore chips that make the model possible are the "logical next step" in the high-speed computations that Man Group is interested in, the fund hasn't committed to rolling it out, Ledford said.

Meanwhile, not every firm would be able to deploy this kind of strategy.

"You would not try this model if you did not have access to fast computation," said Zohren.

2021.04.30 Goldman Sachs is betting on Artificial Intelligence to drive growth

Disha Sinha. Analytics Insight. April 30, 2021. https://www.analyticsinsight.net/goldman-sachs-is-betting-on-artificial-intelligence-to-dive-growth/

Artificial Intelligence (AI) has taken a major role in acting as the main driver of upcoming hi-tech future in the world. It is shifting the information age to a completely new digital domain where upgraded machines help to solve critical decisions and assist in diverse sectors of a country. The banking and financial sector is very keen in investing in AI to protect their customers against the competitors in this market. Thus depending on the current scenario, Goldman Sachs, a leading American multinational investment bank, is betting on AI to drive growth in the economy. It has a fund of $72.5 million exclusively for the investment in Artificial Intelligence algorithms and data analytics.

There is a sudden surge in the cases against cyber-security, cyber-threats, phishing, spamming, hacking and many more unethical behaviours from dark web. It is due to the digital transformation in online banking through websites or mobile apps. Goldman Sachs deals with investment management, securities, asset management, prime brokerage and securities underwriting. This means banks need the best security possible against the online threats. Here comes AI to assist Goldman Sachs in the best possible way.

Why Goldman Sachs is betting on AI to drive growth?

  1. The constant fear of cyber-attacks has been reduced with the help of filtering application of AI. Yes, there are possibilities of receiving fraud applications with unethical motives. AI cybersecurity detects and blocks these applications while collecting real-time data from the users on a large scale.
  2. AI-enabled Investment Trust has already become a successful test for Goldman Sachs in enabling the best investment options for the customers. This has improved customer engagement while running through various kinds of analyst reports and news reports. The investment trust runs on NLP (Natural Language Processing) to assist asset managers in identifying undervalued shares and probable opportunities for profits.
  3. Partnership with an AI-based startup, H2O.ai, helps to focus on deep machine learning transparency and model interpretability to predict a better future. It assists in decision-making process for finance department and equity trading floor such as market making, providing liquidity to the bank and many more.
  4. This is the best opportunity to increase growth in hyper-personalised banking through conversational AI. AI will help in 24*7 two-way communication with innumerable personalised responses and feedbacks to the users.

Goldman Sachs prefers to focus on the net profit over the investment cost of the AI. The bank is eager to maintain its brand image through efficient customer experience, upgraded security and service enhancement.

2021.04.08 Machine learning futures algo trading surges at JP Morgan

Hayley McDowell. The TRADE. April 8, 2021 10:02 AM GMT. https://www.thetradenews.com/machine-learning-futures-algo-trading-surges-at-jp-morgan/

Abstract: Peter Ward, global head of futures and options electronic execution at JP Morgan, tells Hayley McDowell that buy-side adoption of its reinfocement learning FICC futures algorithms has surged in recent years, accelerated by the market volatility in 2020.

Growth in fixed income futures algorithmic trading at JP Morgan has accelerated rapidly in 2020 as buy-side traders globally turned to the investment bank's machine learning-equipped algos to grapple with intense market volatility.

Speaking to The TRADE, Peter Ward, global head of futures and options electronic execution at JP Morgan, explains that while the volatility contributed to recent growth, adoption of futures algo trading has picked up pace with clients significantly in the last few years.

Since 2016, futures volumes traded via algos at JP Morgan has increased 40% year-on-year. In fact, algos now comprise of almost 20% of the bank's total futures trading flow, up significantly from roughly 4-5% in 2016 and 2017, figures seen by The TRADE have revealed.

The period of intense volatility in 2020 due to the global pandemic played a key role in the cumulative buy-side adoption of futures algos as traders became more accustomed to on-screen execution and liquidity.

"When liquidity is harder to source and there is more volatility, execution performance becomes challenged," Ward explains. "Clients are driven to look at the problem areas in executions and that's when we consult with them to figure out ways to bring in that performance. Maybe they should consider trading at higher volume at the open or close, or perhaps sitting out the first five minutes on the cash open because of the noise. All of that we can customise for them.

"I think the more challenges clients see in execution, the more opportunity there is for us to come in and help them, and the solution is increasingly the customised algorithm."

Customised algorithms have become particularly popular with traders in 2020 and in recent years. Volumes on customised algos at JP Morgan have roughly tripled in each of the past three years, alongside a 21% increase in the number of custom algos in 2020 to almost 50 customisations, up from close to zero in 2017.

The bank's flagship liquidity-seeking algorithm, known as Aqua, is the most common foundation for modified client parameters and customisations. A classic example of customisation is where a client wants to follow a particular trading pattern but then switch urgency or strategy based upon predefined triggers.

JP Morgan rebuilt its algo platform around five years ago to provide the buy-side with more choice about the parameters they can set on their side for algorithms, and there are further customisations that the bank's electronic traders can configure on behalf of clients. Ward adds this has allowed his team to have a "richer" dialogue with clients and demand is clearly there.

"There has always been demand for customised algos, even 10 years ago there was a lot of demand," he says. "We just didn't have a scalable way back then to adapt an algo to what a client really wanted. The reason for that is when a client wants something different, we needed developers to code that and then release it for implementation in the client's platform, which takes a lot of time."

Reinforcement learning

The Aqua algorithm has been a particular area of focus for JP Morgan recently. It uses a technology referred to as reinforcement learning to create advanced signals on order routing and placement.

With reinforcement learning, which is a form of machine learning, the algorithm essentially learns from itself over time by looking back at previous signals that it has generated and evaluates performance. The signals will dictate whether the algo crosses the market or stays passive.

Reinforcement learning technology was first applied to a recently launched model of Aqua that is focused on navigating quarterly roll dates when futures contracts expire. It can be a high-volume period and volatile time for traders as everybody is typically rolling in the same week to the next expiration date. In recent years, this activity has evolved from manual, voice-based trading to more electronic, low-touch trading.

"Previously, a lot of this business was executed through voice desks and one reason for that was because trading systems out there couldn't handle multi-leg products," he says. "As those systems have been developed in the last few years, we found more of that activity moving to electronic channels.

"A lot of volume goes through on calendar rolls and the challenge is around optimising that experience for the clients rather than imposing a model of trading without looking at the particular client objective."

In response to the trend and client demand, JP Morgan developed a model of its Aqua strategy, known as the Roll Algo, which went live not long ago for the most recent US treasury roll in February. It has been especially popular with buy-side traders, according to Ward.

"The Roll Algo model focuses on maximising liquidity and pricing opportunities by using signals that help it understand when to cross the spread. It's the most important area we are working on and has peaked the greatest interest from clients.

"It performed really well in February and there was a lot of client use in that period. With that, the algo learned a lot along the way so we can expect the performance in the next quarter's roll to be improved."

The Roll Algo is not the only new addition to JP Morgan's new strategy line-up. Advanced strategies like Target to trade around the cash or futures close, Multi Leg Strategy for trading multiple instruments at the same time across futures and US treasuries, and options algos have also been developed by the bank.

Volumes in options on futures surged in 2020 as trading floors at major derivatives exchanges like CME that facilitate options trading were forced to shut down. As a result, liquidity shifted to low-touch and electronic channels and JP Morgan's clients began to ask more questions about trading options through algorithms.

"Options on futures volumes have seen significant growth in the industry over the past few years and 2020 was a breakout year for liquidity on-screen," Ward adds.

"With that said there are still challenges and nuances to trading them and that's where we see opportunities to innovate and help our clients with their execution. This can be through simpler Peg and Cross type strategies and ultimately more targeted strategies using a delta or volatility reference."

JP Morgan expects buy-side adoption of futures algo trading to continue increasing in the near future, having been driven by ongoing market developments and trends over the past few years.

Explicit regulatory requirements on best execution and growing appetite among the buy-side to address challenges in futures and options market structure have been instrumental in the growth of this trend. Best execution essentially forces traders to establish benchmarks to measure performance and trading through algorithms can provide an effective way to do this.

New products have also entered the market where liquidity is shared on multiple markets, which presents challenges in trading those products. Nifty derivatives, for example, are now tradeable in both Singapore and India after the Singapore Exchange (SGX) and India's National Stock Exchange ended a two-year dispute which put SGX's futures into question.

Other developments such as extended hours in futures markets also means there are now more hours to trade what is often the same amount of volume. Add periods of decreased liquidity and increased volatility to the mix, traders have progressively sought algorithmic strategies and automated solutions for consistent execution in volatile products, and when targeting cash settlement periods, for example.

It's not just JP Morgan that is doubling down on efforts in futures algo trading. In January, rival investment bank Citi rolled out a suite of execution algorithms, including its flagship Arrival strategy, for futures markets across all major exchanges in the US, Europe, and Asia Pacific.

In contrast to JP Morgan, the electronic traders at Citi handle all of the algo customisations on behalf of clients. Head of EMEA futures electronic execution at Citi, Gordon Ball, said at the time clients don't want to enter numerous parameters to execute an order. He added: "the complexity of operating an intelligent algorithm and fine-tuning customisations sits with us, so our clients can focus on their overall investment and trading objectives".

Elsewhere, a start-up founded by former global head of trading at AQR Capital Management, Hitesh Mittal, launched its own suite of execution algorithms in early 2020 that aims to reduce costs for the buy-side with customised and high-performance strategies. In December, BestEx Research secured $5 million in funding as it prepares to roll out its algos in futures markets.

Amid the arms race in this space, JP Morgan's Ward predicts the pace of fixed income futures algo trading adoption, particularly customised algos, will continue apace in 2021. It remains a significant focus at JP Morgan as different buy-side clients are also now using algorithms to trade futures.

In the past few years, the type of buy-side client seeking algorithmic execution has shifted from being a relatively small number of large hedge fund clients to the more traditional managers, including pension funds, asset managers and insurance companies.

"Five years ago, there were pockets of interest in executing this way, depending on the specific trader or firm's appetite. It's now become far more mainstream, driven by broader electronification in fixed income markets as well more investment firms adopting more explicit execution benchmarks," Ward concludes.

2021.02.09 Electronic trading surges with traders eyeing the impact of machine learning

Angharad Carrick. CITY A.M. Tuesday 9 February 2021 6:15 am. https://www.cityam.com/electronic-trading-surges-with-traders-eyeing-the-impact-of-machine-learning/

Abstract: Professional traders are anticipating artificial intelligence and machine learning to be the most influential technology over the next three years.

JP Morgan's flagship survey reveals more than half of professional and institutional traders anticipate machine learning to lead technology.

Currently a third of client traders predict mobile trading applications to be the most influential this year. Certainly the Reddit Gamestop rally powered by low cost trading platform is already testament to just how quickly the environment has changed.

Electronic trading picked up last year and all surveyed expect to increase electronic volumes this year. FX electronic trading to increase six per cent over the next two years to 84 per cent while credit should climb 12 per cent to 40 per cent.

"Year-on-year, the first half of 2020 saw a 45 per cent increase in volume of transactions and March, perhaps unsurprisingly, saw a new high water mark for notional value traded on the bank’s Execute on Mobile channel," Richard James, JP Morgan's head of macro markets execution said.

"The surge in activity was driven by what was also a new high in external client logins, about 30 per cent of the bank's user base were actively transacting over the channel with the balance accessing market information and analytics."

Banks and other financial institutions are already starting to use AI to execute trades quicker and more efficiently. The vast majority of surveyed traders — 71 per cent — agree that machine learning provides deeper analytics while just over half agree it optimises trade execution.

Looking forward to this year, just under half of those surveyed believe the pandemic will continue to have the biggest impact on markets this year. In a sign of just how much the pandemic has taken over market discussions, international trade tensions come just fourth in traders' concerns, with only nine per cent concerned over the prospect of trade wars.

When it comes to traders' day-to-day life 55 per cent will continue to work from home for average of four days a week.

2020.12.29 Nvidia rival Graphcore raises $222 million for AI chips with potential IPO on the horizon

Sam Shead. CNBC. Published Tue, Dec 29 2020, 6:25 AM EST, updated Tue, Dec 29 2020, 6:49 AM EST. https://www.cnbc.com/2020/12/29/graphcore-raises-222-million-to-take-on-nvidia-with-ai-chips.html

Key points:

  • Graphcore has raised $222 million as it looks to take on U.S. rivals Nvidia and Intel.
  • The Series E funding round, which comes less than a year after Graphcore raised a $150 million extension to its last round, values the company at $2.77 billion.
  • Total investment in Graphcore now stands at $710 million.

London—U.K.-based chipmaker Graphcore announced Tuesday that it had raised $222 million of investment as it looks to take on U.S. rivals Nvidia and Intel.

Graphcore said it will use the funding to support its global expansion and to accelerate the development of its intelligence processing units (IPUs), which are specifically designed to power artificial intelligence software. The company has already shipped tens of thousands of its chips to customers including Microsoft and Dell.

The Series E funding round, which comes less than a year after Graphcore raised a $150 million extension to its last round, values the company at $2.77 billion, up from $1.5 billion in 2018.

Graphcore CEO and co-founder Nigel Toon told CNBC in July: "We're now at the point where we're not really looking for venture investors in the business. We're more interested in companies that would be long term investors and holders of the stock, perhaps, in the public markets, if we ever reach that point."

At the time, Toon said going public is "ideally what we would like to do" but he stressed "lots of water still has to flow under the bridge before we get to that point."

Total investment in Graphcore now stands at $710 million and the four-year-old company has $440 million of cash on hand.

The latest funding round was led by the Ontario Teachers' Pension Plan Board while other new investors included private equity investor Baillie Gifford, venture capital investor Draper Esprit, as well as funds managed by Fidelity International and Schroders.

On Tuesday, Toon said in a statement: "Having the backing of such respected institutional investors says something very powerful about how the markets now view Graphcore. The confidence that they have in us comes from the competence we have demonstrated building our products and our business."

He added: "We have created a technology that dramatically outperforms legacy processors such as GPUs, a powerful set of software tools that are tailored to the needs of AI developers, and a global sales operation that is bringing our products to market."

Serial chip entrepreneurs

Graphcore was founded in June 2016 in Bristol, England, by Toon and Simon Knowles, who sold their previous chip company, Icera, to Nvidia for $435 million in 2011. The pair formed the initial idea for Graphcore in a small pub called the Marlborough Tavern in Bath in January 2012.

Today, the company employs around 450 people in Bristol, Cambridge, London, Beijing, Oslo, Palo Alto, Seattle, and Hsinchu in Taiwan. It expects the number to grow to 600 by the end of 2021.

But the rapid growth hasn't come cheap. It made a pre-tax loss of $95.9 million on revenues of $10.1 million in 2019, according to an annual report filed on U.K. business registry Companies House.

Santa Clara heavyweights Intel and Nvidia are two of the obvious front runners in the AI chip market given their expertise in chip making. The companies haven't disclosed how many of their AI-optimized chips have been sold. However, over a trillion computer chips are expected to be shipped in 2020, according to market data website Statistica. In 2019, Intel's slice of the overall chip market came in at 15.7% and it has been the market leader every year since 2008, with the exception of 2017 when Samsung took the number one spot.

Graphcore's Toon criticized Nvidia's plan to buy U.K. chip designer Arm from SoftBank for $40 billion, saying it is bad for competition.

"We believe that Nvidia's proposed acquisition of Arm is anti-competitive," he said. "It risks closing-down or limiting other companies' access to leading edge CPU processor designs which are so important across the technology world, from datacenters, to mobile, to cars and in embedded devices of every kind."

Google, Amazon and Apple are also working on their own AI chips.

Sequoia backs Nvidia and Graphcore

Previous investors in Graphcore include the likes of Microsoft and BMW iVentures, as well as venture firms like London's Atomico and Silicon Valley's Sequoia, which has also backed Nvidia.

Last month Sequoia partner Matt Miller told CNBC: Graphcore "are in this position where they always have people coming at them trying to give them more money. So, they do not need funding. They are well funded for the next several years, but they definitely have people trying to invest in the company."

He added: "I don't think that you have to take on Nvidia because the market is so huge. Taking on Nvidia is like this huge task. It's a huge company with billions of revenue and incredible teams doing all sorts of wonderful things. I think that what Graphcore has the opportunity to do is be a very strong player in the AI microprocessor market. It continues to have great progress with many of the cloud providers, and many people want to be diversified. They don't want to be all in with one chip."

Graphcore launched its second generation IPU earlier this year despite disruption from the coronavirus pandemic.

2020.08.09 ADIA Hires Marcos Lopez de Prado as Global Head of Quant Research

Sovereign Wealth Fund Institute (SWFI). Posted on 09/08/2020. https://www.swfinstitute.org/news/81369/adia-hires-marcos-lopez-de-prado-as-global-head-of-quant-research

The Abu Dhabi Investment Authority (ADIA) hired Marcos López de Prado as global head of quantitative research & development. Prado is a Cornell University professor. Prado is joining a newly-formed investment group at ADIA within the strategy and planning department. This group seeks to apply a systematic, science-based approach to developing and implementing investment strategies.

Most recently, Prado was professor of practice at Cornell University’s School of Engineering, teaching machine learning, according to the statement.

Prado is also the CIO of True Positive Technologies (TPT). PT is currently engaged by clients with a combined AUM in excess of US$ 1 trillion. Prado launched TPT after he sold some of his patents to AQR Capital Management, where he was a principal and AQR’s first head of machine learning. Marcos also founded and led Guggenheim Partners’ Quantitative Investment Strategies business, where he managed up to US$ 13 billion in assets.

2020.02.14 Developers now make up a quarter of Goldman Sachs' workforce

Jia Jen Low. T_HQ. 14 February 2020. https://techhq.com/2020/02/developers-now-make-up-quarter-of-goldman-sachs-workforce/

Abstract: The leading finance firm says it's now competing with Silicon Valley tech giants for talent.

With the emergence of technologies such as artificial intelligence (AI), blockchain, machine learning (ML), and big data, finance is one of the most disrupted sectors. The innovations mushrooming from the effect are digital wallets, chatbots with financial knowledge, and smart contracts.

The finance sector is one of the keenest investors in emerging technology. Just shy of two-thirds (64 percent) of financial services leaders expect to be mass AI adopters within the next two years. The industry is also ahead of others with blockchain—enabling faster processing and quicker settlement of trades.

However, these digital innovations require teams of developers, data scientists, and tech specialists.

At Goldman Sachs' Technology and Internet Conference in San Francisco, the Wall Street giant's co-Chief Investment Officer, George Lee, explained how the firm is on a tech hiring spree in a bid to rapidly expand its engineering talent. The investment bank currently staffs 10,000 developers, making up a quarter of its total workforce.

Last year, the leading financial firm celebrated a memorable 150th birthday announcing fourth-quarter profit that exceeded expectations by US$1.59 per share. Now the investment bank is turning up the knob which means it must go head-to-head with tech giants—such as Amazon, Microsoft, and Google—for future talent.

"The trend is very much toward technology companies and we need to compete at that level," said Lee.

Last year, Goldman achieved several milestones by releasing new products like the Apple Card, the credit card it launched with the iPhone maker last year. It has also made headway into automated support, introducing a robo advisor to help with consulting small clients.

But the firm is keenly aware that the finance sector is at continued risk of disruption, with new fintechs reimagining customer experience, delivering enhanced services and new security solutions.

Lee explained the road to hiring top engineers required some shifts in the firm's rigid organizational structure. Wanting to leverage software development's "open-source" approach, a more distributed workforce is being adopted, with developers working from cities across the globe—some of the key locations include India, Poland, and Dallas (the US).

This has been a challenge for the Goldman Sachs' strict working operations—Lee said the firm had to navigate and facilitate these preferences while remaining mindful of company policies.

As part of the aggressive hiring strategy, the bank is tapping into the graduate talent pool by focusing on recruiting efforts at college campuses, but it isn't the only global financial firm that is hungry for tech talents, with JPMorgan, Bank of America and Citi also getting their fair share of digitally skilled workforce.

2020.02.06 Finance is headed for AI mass adoption—and soon

Jia Jen Low. T_HQ. 6 February 2020. https://techhq.com/2020/02/finance-is-headed-for-ai-mass-adoption-and-soon/

Abstract: While other industries struggle with AI, finance members are locked in an arms race.

Despite the hype, many organizations are facing an AI "reality check" this year. Everyone is buying into the technology's promise, but difficulties in implementation are leading many firms to roll back plans.

The finance sector might be the exception to the rule though. According to a new report by the World Economic Forum (WEF) and Cambridge Centre for Alternative Finance (CCAF), organizations here are confident they are already reaping the advantages.

Just shy of two-thirds (64 percent) of financial services leaders expect to be mass AI adopters within the next two years, exploding from just 16 percent today. Aside from cost reduction, applications span revenue generation, process automation, risk management, customer service, and client acquisition.

More than 150 industry leaders from both fintech and incumbent financial institutions took part in the report, Transforming Paradigms: Global AI in Financial Services Survey. Findings painted a picture of an industry already well ahead with the technology. But they also highlighted a distinction between the use of AI by the new wave of fintech disruptors and industry incumbents.

A majority of fintech firms are developing AI-powered products and services, with the aim of automating decition-making and offering more variety in cloud solutions. Legacy firms, meanwhile, are using AI to strengthen financial services and systems, and expect their employment rates to drop by 9 percent within the next 10 years as a result of automation.

According to research by Accenture, banks adopting AI expect the technology to help cut IT operations costs by between 20-25 percent. As more operations are automated, AI will also allow bank employees to spend more time on "exceptional work", or the 20 percent of non-routine tasks that drive 80 percent of value creation.

From a customer experience (CX) standpoint, AI could greatly enhance products. Predictive analytics could track spending patterns and help banks set credit limits; real-time sentiment analytics could provide customer support cues, while AI can also be used to identify fraudulent activities. TechHQ recently spoke to Revolut on how it uses machine learning to tackle FinCrime.

While other industries struggle to get to grips with AI and machine learning technology, particularly in discovering viable applications, the finance industry is already flexing its AI muscle. Indeed, 77 percent of respondents are anticipating AI to have significant importance in their businesses within two years.

With so many use cases for the technology—in improving products, experience, and internal operations—there's plenty of room to innovate. But the growing digitization of the industry opens doors for further disruption: nearly half of respondents saw a new and significant competitive threat emerging in tech firms.

On the study's findings, WEF Head of Financial and Monetary Systems Matthew Blake, said: "The comprehensive and global study confirms that AI is affecting the financial system at an accelerating pace."

"With the rising trend of mass adoption of the technologies throughout financial services, those firms that implement AI quickly look set to sprint ahead."

2020.01.27 Rival banks are hiring technologists from Goldman Sachs

Sarah Butcher. eFinancialCareers. 27 January 2020. https://www.efinancialcareers.com/news/2020/01/rival-banks-hiring-goldman-sachs-technologists

Goldman Sachs might want to keep a tighter grip on its technology talent. Since the start of this year, various of its senior technologists have found new jobs elsewhere.

The latest big name to move is Gavin Leo Rhynie, the former head of platform technology at Goldman in New York City, who has just joined JPMorgan as head of engineering and architecture for the corporate and investment bank (CIB) according to a memo sent by CIB technology head Mike Grimaldi. Leo Rhynie isn't JPM's only ex-Goldman hire though: JP also just poached James Kirby, a London-based vice president who spent seven years at Goldman with a focus on enterprise architecture and technology implementation.

Morgan Stanley has been checking out Goldman's talent too. As we reported earlier this month, the U.S. bank hired Michael Ballard, a VP in digital product at Goldman in New York who joined as an executive director in product strategy.

The exits come as Goldman itself ramps up technology hiring while preparing to cut costs by moving as many as half its technology jobs outside of London. Goldman's technology business is in a state of flux after the departure of leaders like Marty Chavez and Elisha Wiesel last year. However, Goldman technologists told us last week that they're super happy working for the bank, which is less political and pressured than big tech firms, gives them plenty of flexibility and is a better environment than most other places in finance.

Banks are big spenders on technology with JPMorgan, Bank of America and Citi spending the most. Citi is also hiring senior technologists externally: the bank just recruited James Linnett as CIO of global functions technology from Bank of America, where he spent 18 years.

Not all the new technology hires have technology backgrounds. JPMorgan is setting up a new London machine learning centre run by Chak Wong, a former trader and structurer at SocGen, Barclays, Morgan Stanley, Goldman and UBS. Wong, too, is hiring associates. Goldman especially may want to keep a strong grip on its machine learning people in the City.

2020.01m A Global AI in Financial Services Survey

Lukas Ryll, Mary Emma Barton, Bryan Zheng Zhang, Jesse McWaters, Emmanuel Schizas, Rui Hao, Keith Bear, Massimo Preziuso, Elizabeth Sege, Robert Wardrop, Raghavendra Rau, Pradeep Debata, Philip Rowan, Nicola Adams, Mia Gray, Nikos Yerolemou. University of Cambridge Judge Business School. January 2020. https://www.jbs.cam.ac.uk/faculty-research/centres/alternative-finance/publications/transforming-paradigms/

Abstract: This report presents the findings of a global survey on AI in Financial Services jointly conducted by the Cambridge Centre for Alternative Finance (CCAF) at the University of Cambridge Judge Business School and the World Economic Forum in Q2-Q3 2019. Representing one of the largest global empirical studies on AI in Financial Services, a total of 151 respondents from 33 countries participated in the survey, including both FinTechs (54 per cent of the sample) and incumbent financial institutions (46 per cent of the sample). The study was supported by EY and Invesco.

Highlights from the report

The key findings of this empirical study are as follows:

  • AI is expected to turn into an essential business driver across the financial services industry in the short run, with 77 per cent of all respondents anticipating AI to possess high or very high overall importance to their businesses within two years. While AI is currently perceived to have reached a higher strategic relevance to FinTechs, Incumbents are aspiring to catch up within two years.
  • The rising importance of AI is accompanied by the increasingly broad adoption of AI across key business functions. Approximately 64 per cent of surveyed respondents anticipate employing AI in all of the following categories—generating new revenue potential through new products and processes, process automation, risk management, customer service and client acquisition—within the next two years. Only 16 per cent of respondents currently employ AI in all of these areas.
  • Risk management is the usage domain with the highest current AI implementation rates (56 per cent), followed by the generation of new revenue potential through new AI-enabled products and processes, adopted by 52 per cent. However, firms expect the latter to become the most important usage area within two years.
  • AI is expected to become a key lever of success for specific financial services sectors. For example, it is expected to turn into a major driver of investment returns for asset managers. Lenders widely expect to profit from leveraging AI in AI-enabled credit analytics, while payment providers anticipate expanding their AI usage profile towards harnessing AI for customer service and risk management.
  • With the race to AI leadership, the technological gap between high and low spenders is widening as high spenders plan to further increase their R&D investments. These spending ambitions appear to be driven by more-than-linear increases in pay-offs from investing in AI, which are shown to come into effect once AI investment has reached a "critical" mass of approximately 10 per cent R&D expenditure.
  • FinTechs appear to be using AI differently compared to Incumbents. A higher share of FinTechs tends to create AI-based products and services, employ autonomous decision-making systems, and rely on cloud-based offerings. Incumbents predominantly focus on harnessing AI to improve existing products. This might explain why AI appears to have a higher positive impact on FinTechs' profitability, with 30 per cent indicating a significant AI-induced increase in profitability compared to seven per cent of Incumbents.
  • FinTechs are more widely selling AI-enabled products as a service. Successful real-world implementations demonstrate that selling AI as a service may allow large organisations to create "AI flywheels"—self-enforcing virtuous circles—through offering improved AI-driven services based on larger and more diverse datasets and attracting talent.
  • AI Leaders generally build dedicated corporate resources for AI implementation and oversight—mainly a data analytics function—to work with their existing IT department. On average, they also use more sophisticated technology to empower more complex AI use cases.
  • Leveraging alternative datasets to generate novel insights is a key part of harnessing the benefits of AI with 60 per cent of all respondents utilising new or alternative forms of data in AI applications. The most frequently used alternative data sources include social media, data from payment providers, and geo-location data.
  • Incumbents expect AI to replace nearly nine per cent of all jobs in their organisation by 2030 while FinTechs anticipate AI to expand their workforce by 19 per cent. Within the surveyed sample, this implies an estimated net reduction of approximately 336,000 jobs in Incumbents and an increase of 37,700 jobs in FinTechs. Reductions are expected to be highest in investment management, with participants anticipating a net decrease of 10 per cent within five years and 24 per cent within 10 years.
  • Regardless of sectors and entity types, quality of and access to data and access to talent are considered to be major obstacles to implementing AI. Each of these factors is perceived to be a hurdle by more than 80 per cent of all respondents, whereas aspects like the cost of hardware/software, market uncertainty, and technological maturity appear to represent lesser hindrances.
  • Almost 40 per cent of all respondents feel that regulation hinders their implementation of AI, whereas just over 30 per cent perceive that regulation facilitates or enables it. Organisations feel most impeded by data sharing regulations between jurisdictions and entities, but many also deem regulatory complexity and uncertainty to be burdensome. Firms' assessments of the impact of regulation tend to be more positive in China than in the US, the UK, or mainland Europe.
  • Mass AI adoption is expected to exacerbate certain market-wide risks and biases, and at least one in five firms do not believe they are well placed to mitigate those. Firms are particularly wary of the potential for AI to entrench biases in decision-making, or to expose them, through shared resources, to mass data and privacy breaches. Nevertheless, many firms are involving risk and compliance teams in AI implementation, and those who do tend to be more confident in their risk mitigation capability as a result.
  • Long-established, simple machine learning algorithms are more widely used than complex solutions. Nonetheless, a large share of respondents is planning to implement natural language processing (NLP) and computer vision, which commonly involve deep learning, within two years.
  • Nearly half of all participants regard "Big Tech" leveraging AI capabilities to enter financial services as a major competitive threat.

2018.11.23 Goldman Sachs hunts AI experts for all-important quant team

Paul Clarke. Financial News. Friday November 23, 2018 4:29 am. https://www.fnlondon.com/articles/goldman-sachs-hunts-ai-experts-for-all-important-quant-team-20180130

Abstract: US bank is building its vast strats department by hiring a new generation of machine learning and artificial intelligence specialists.

Goldman Sachs is pitting itself against tech giants like Google and Facebook to attract scarce artificial intelligence talent as it places a renewed emphasis on ensuring its sales and trading staff are given cutting-edge IT.

The US investment bank, which has been described by CEO Lloyd Blankfein as a "technology company", is building its strats department, a huge division comprised of quant and technology professionals and which serves its trading businesses among other functions.

Thalia Chryssikou, co-head of global sales strats and structuring across fixed income currencies and commodities and equities, said the bank has shifted its focus towards recruiting a "new generation" of strats.

"The strats we hired 10 to 15 years ago typically specialized in modeling risk and pricing analytics," she said during an interview in an email newsletter from the bank. "Today, we're focused on hiring a new generation of strats who specialize in data management and analytics, including machine learning (ML), artificial intelligence (AI), program management and digital product design, in addition to quantitative sciences."

Goldman's strats division is made up of quantitative finance, engineering and technology professionals spanning various departments across the bank's trading functions, as well as back office and compliance. It has been growing since then-chief information officer Marty Chavez was handed control of both technology and strats in 2014.

Strats now comprise 27% of total headcount within Goldman Sachs's securities business, according to Chryssikou, up from 18% five years ago. Strats made up 27% of experienced hires within Goldman's FICC division last year, according to a September presentation by chief operating officer Harvey Schwartz. The team has also doubled within its investment bank since 2014.

Strats have become increasingly important to a range of functions at Goldman Sachs as it looks to automate more processes and equip its markets business with cutting-edge technology. Last year, for example, a team of 75 programmers at the bank introduced software to reduce some of the grunt work of junior investment bankers.

More broadly, however, Goldman's strats and technology teams have been creating systems to help their sales and trading staff make more informed decisions for their clients. In its fourth quarter results presentation, Chavez said the real growth in headcount last year was within "engineering" as the bank is "digitising our plarform generally".

Goldman recently hired Jeff Wecker as chief data officer and Matthew Rothman as head of data and client service. Rothman is currently hiring quantitative researchers within the securities division at Goldman.

"Almost nothing we do to service our clients—from trade execution, regulatory compliance and the advanced quantitative analysis we touched on earlier—could be possible without investments in technology and engineering," said Chryssikou. "It is essential and our hiring in the business reflects that."

2017.06.19 London currency trader bets on machine learning for high speed trading foray in US stocks

Bloomberg. June 19, 2017 1:17 PM

Abstract: A tiny London firm with no human traders made its name last year beating banks to climb up the currency trading ranks. Now it wants a bite of something new: the $27 trillion US stock market.

A tiny London firm with no human traders made its name last year beating banks to climb up the currency trading ranks. Now, it wants a bite of something new: the $27 trillion US stock market. XTX Markets Ltd. is only two years old, but its executives say it has what it takes to compete with more established American trading Goliaths in the world's largest, most complex and most saturated equity market. The firm is prepping a new Manhattan office, lining up the necessary regulatory nods and scooping up a big-name hire: Eric Swanson, who helped Bats Global Markets Inc. become the nation's second-biggest stock exchange operator. With Swanson, who joined this month, XTX can "go from having a toehold, to being a more significant player in the US," says Zar Amrolia, co-chief executive officer of XTX who formerly ran digital technology at Deutsche Bank AG. "We are just rolling out what we think is a successful quantitative research-driven approach to market making."

It won’t be easy. A Dutch speed trader, Flow Traders NV, that last year kicked off a similar US expansion, is having trouble. The choppy, scattered nature of the market has few parallels. Firms that want to compete will have to connect to 12 national securities exchanges, scour constantly for trading risk, suck in proprietary data feeds and ward off any behavior that could run afoul of regulators, all at once. For the fastest firms, trading strategies can be made or broken by millionths of a second.

Smart, Not Fast

Amrolia's not fazed by speedier rivals — he says his goal is to be "smart, not fast." Taking its name from a mathematical expression, XTX uses technology to forecast where prices for securities will be in a matter of minutes or hours. Amrolia contrasts this strategy to some North American firms that rely on the speediest networks for getting information, and carrying out trading decisions based on it. Virtu Financial Inc. and Citadel Securities LLC will soon be among XTX's biggest rivals. XTX's focus on machine learning puts the firm "at the forefront" of trading technology, Swanson said. Their strategies will be put to the test in the U.S., which hosts more than one-third of global equity trading value, and holds a complex web of big-name exchanges and dozens of smaller private dark-pool venues. "That presents a challenge for everyone," says Swanson, Americas CEO at XTX and the former general counsel at Bats, which is now owned by CBOE Holdings Inc. "We're up to managing that challenge."

XTX made its name last year after managing to leapfrog big banks to place fourth in spot currency trading. The firm repeated the spot-trading feat in the 2017 Euromoney Institutional Investor Plc survey, despite slipping to 12th in this year’s rankings for overall trading.

Avoiding Bad Trades

XTX says its technology and the way it sends orders into the market allows its systems to be alerted quickly when things go wrong, giving them the ability to make lightning-fast pivots to avoid bad trades. The firm employs just 78 people, according to spokesman Tim Moxon. "They have a good combination of ability to manage their own risk and create really good prices," said Steve Grob, global director of group strategy at Fidessa Group Plc. Its Amsterdam rival Flow, the largest trader of European exchange-traded funds, pushed into the US late last year, hoping to profit from buying and selling ETFs that no one else will touch. Addled by calmer markets, Flow's first-quarter profit slumped. Trading income in the Americas over the three-month period dropped 23 percent.

One part of Flow Traders's strategy was clinching regulatory approval to trade directly with large U.S. investors. That may also be a winning option for XTX if the firm can nab similar permissions, which in some cases would allow the firm to bypass public exchanges. "That's where you can be disruptive, if you've got the right technology behind you," Grob said.

The spread-out nature of the US stock market can pose obstacles for any newcomer to the region, said Michael Beller, chief executive officer of Thesys Technologies LLC, a company that sells market-structure technology to help firms manage that vast trading network. "It's complicated," Beller said in an interview. "You can't walk in, start trading and get results."