Investment Analytics in the Dawn of Artificial Intelligence

Bernard Lee

• Publisher: WSPC
• Year: 2019
• eBook ISBN: 9789814725378
• Topic: Business
• Subtopic: Trading
• Index: Business

A class of highly mathematical algorithms works with three-dimensional (3D) data known as graphs. Our research challenge focuses on applying these algorithms to solve more complex problems with financial data, which tend to be in higher dimensions (easily over 100), based on probability distributions, with time subscripts and jumps. The 3D research analogy is to train a navigation algorithm when the way-finding coordinates and obstacles such as buildings change dynamically and are expressed in higher dimensions with jumps.

Our short title 'ia≠ai' symbolizes how investment analytics is not a simplistic reapplication of artificial intelligence (AI) techniques proven in engineering. This book presents best-of-class sophisticated techniques available today to solve high dimensional problems with properties that go deeper than what is required to solve customary problems in engineering today.

Dr Bernard Lee is the Founder and CEO of HedgeSPA, which stands for Sophisticated Predictive Analytics for Hedge Funds and Institutions. Previously, he was a managing director in the Portfolio Management Group of BlackRock in New York City as well as a finance professor who has taught and guest-lectured at a number of top universities globally.

Contents:

• Introduction
• Navigation and Vocabulary
• Construct Portfolios:
  • Understanding Risk
  • Objective Functions in Portfolio Construction
  • Risk and Return Attribution
  • Portfolio-Level Factor Analysis
  • A Hedging Use Case
• Select Assets:
  • Alpha Selection Using Factors
  • Standard Derivative Instruments
• Decide and Execute:
  • Rebalancing
  • Forward Scenarios and Historical Simulations
  • Combining Upside with Black Swan Scenarios
• Deliver Reports:
  • Customary Back Office Reporting
  • Additional Reporting
  • Compliance Analysis
  • Data Integrity Validation
• Deploy:
  • Deployment Best Practices
  • Implications of a Post-IA/AI Society

Readership: Professionals looking for a step-by-step 'cookbook' on algorithms to build and deploy their own investment analytics processes; C-level executives at leading investment firms to technologists doing hands-on deployments. Financial Technology;Investment Analytics;Artificial Intelligence;Asset Managers;Investment Management;Asset Liability Management;Asset Selection;Multi-asset Investing;Risk Management;Market Risk;Extreme Risk;Tail Risk;Scenarios;Historical Simulation;Black Swan;Compliance;Maximizing Returns;Retirement;Wealth Management;Investment Advisor;Investment Company;Investment Objective;Long-Term Investment;Portfolio;Portfolio Construction;Rebalancing;Reinvestment;Relative Risk and Potential Return;Investment Horizons;Valuation;Total Return;Indexing;Alpha;Beta;Asset Allocation;Diversification;Investment Reporting;Performance Reporting;Performance Attribution;Risk Attribution0 Key Features:

• This book is designed as practical teaching materials to go with a master's level or professional education course, with one on-line version is already accessible
• Rigorous use cases that were accepted as stand-alone papers at some of the best recognized academic conferences such as the Joint Statistical Meetings
• Step-by-step algorithms with the clarity required for customary compliance clearance
• Quick and logical navigation to relevant topics for those in different asset management functions without assuming logical chapter-by-chapter progress
• Supplementary teaching aids will be made available to instructors

INVESTMENT ANALYTICS IN THE DAWN OF ARTIFICIAL INTELLIGENCE

1. Introduction

In the coming decade, business school case studies will be written on the sheer amount of media coverage on artificial intelligence (AI) and financial technology (FinTech) recently, especially after Bitcoin reached almost US$20,000 by December 2017 and then crashed back to the US$3,000+ level in 2018. Initially, the popular media was preparing the global audience for the second coming of a messiah in finance, and then quickly reversed to reminding them of how those who could not remember financial bubbles were condemned to repeat the past.

Tesla founder Elon Musk told the National Governors Association of a looming AI “threat”, which could even spark a war. So, he proposed the creation of a regulatory body to guide its development.¹

“I was surprised by your suggestion to bring regulations before we know what we are dealing with,” Arizona Governor Doug Ducey said in response.

Very few of us doubt that the rapid advance, and the public’s growing acceptance of advanced technology in finance, will change the landscape of the global financial industry. AI techniques are not new, and Wall Street has always been searching for better technology and methodologies year after year.

New firms and new players emerge with each generation of new technology. What seems different this time is the public’s recognition (and its fear) of the change and the eagerness in proposing a drastic response by overhauling the existing regulatory framework, which may change the fundamental relationships among the buyers, sellers, and overseers of financial services completely. Compliance departments will find it much more challenging to deny approval “when the customers ask for the solution.” This book’s publication is timed to create impact.

1.1The Fourth Industrial Revolution

The term FinTech describes the application of technology to finance, but what exactly is AI-powered FinTech? The defining characteristics of AI-powered technology are that the machine should be able to make some of its decisions autonomously, as in a self-driving car finding its way from downtown to the airport. The sudden acceptance of AI in finance may have less to do with the hard facts of recent technological advances, since certain underlying techniques have been in use for decades, but more to do with the emotional acceptance of letting a Tesla vehicle drive one’s family from home to airport on autopilot.

A machine that can learn from its own mistakes is one key to public acceptance, but many useful, day-to-day tasks are deterministic and do not require machine learning. Today’s machines still have a long way to go before coming up with insights; at best, machines can emulate known human decision patterns, which is why the best AI-powered investment analytics engine today can only perform basic investment management tasks in an autonomous manner. Such an outcome is not necessarily undesirable because a machine that can perform routine tasks similar to what a human portfolio manager does but on a much large amount of data and with speed and precision can still make a meaningful difference. This observation also means that model parameters should still be set by humans, or that certain analytics that resembles data-mining should be avoided by human operators. The best AI-powered investment analytics should allow an investment professional to use essentially the same front-end user interface today, perhaps without being aware of that AI-powered machine learning algorithms have been added to the back-end – just like today a Tesla driver can still drive with a steering wheel without activating its self-driving features.

Why suddenly the intense media interest about AI and FinTech then? The mass media uses the analogy that AI and FinTech are driving the so-called Fourth Industrial Revolution. Like new technologies in previous industrial revolutions, we expect complaints from those whose skills are being phased out and whose jobs may be put at risk. However, new jobs requiring perhaps different skills will also be created. If the world suddenly finds itself in a situation where all major productive activities are automated, there will be a massive dislocation of the global job market, leading to civil unrest and perhaps even open conflicts, especially when countries cannot agree on a “fair” distribution of new jobs and new economic opportunities. Thankfully, we are not in that situation today: once again, what is changed today is not the technology per se but its social acceptance.

Hence, we should first try to understand how AI and FinTech will impact society, and then work out how best to manage the potential outcomes. Setting up more entities to regulate AI and FinTech is one way to manage the potential outcomes but not the only way. Other obvious and potentially more effective steps may include educating the younger generation on new and relevant skills so that they are ready for new opportunities that may come up, and incentivizing successful technology companies to reinvest earnings in R&D and to allow their technology to be deployed for pro bono causes, with one policy goal for them to give back to society.

1.2Unhealthy Myths

Before any debates on how best to manage the potential outcomes, it may be helpful to set the facts straight and dispute certain unhealthy myths about AI and FinTech:

1.AI, Big Data, and FinTech are not some form of magic. The current state of the art can automate certain repetitive tasks, and analyze a massive amount of data, but AI is still a long way from displacing many financial industry jobs. An article on The Wall Street Journal correctly pointed out the excessive hype over AI and FinTech prevalent in today’s mainstream media, even though some of its descriptions can be made more technically precise.²

2.Despite the AI-driven FinTech hype, finance is still one of the most “backward” industries in adopting modern technology. One plausible explanation is that financial data formats and standards remain highly fragmented even in developed markets today. The amount of integration required by adopting any “newer” technology in finance will create immense headcounts. In addition, because no new technology is perfect, a skillful combination of traditional techniques and new technologies can still tackle unsolved or hard-to-solve problems: for example, AI can be used to recognize a high-dimensional pattern from a massive volume of financial data that human brains cannot visualize and are, therefore, less likely to design traditional rule-based algorithms to work with such data. The market’s rule of thumb is that AI should produce a clear improvement before being deployed to replace a more traditional modeling technique or decision paradigm.

3.If everyone uses the same techniques, it can be shown by simulation that the chances for system-wide instability will increase dramatically.³ Diverse parameterization and error feedback mechanisms can be critical to allow a machine to learn and diversify its solutions instead of producing a highly similar solution for every investor. AI can also bring an unintended benefit: If every investor in the market relies on low-fee indexing (therefore free riding on someone else’s research), eventually, society may no longer be allocating capital in the most efficient manner.⁴ AI/FinTech can address that specific problem as a still cost-effective alternative to ETFs today. Fees can still be lowered, but assets can be chosen based on each investor’s specific criteria and preferences to encourage market diversity.

1.3New Regulatory Framework

A new regulatory body as suggested by Elon Musk may not be the right solution. As a practical thought experiment, let’s ask whether and how such a new regulatory body can work:

1.Who will staff such a regulatory body? A new agency as such tends to attract compliance officers and lawyers who are unlikely to have deep technical training and first-hand experience working with the underlying technology. There is a well-known challenge among those who talk about rapidly advancing technology to come up with effective policies. The handful of experts with both technical and regulatory backgrounds can expect to fill the (usually) better-paying jobs in the industry instead of taking public policy roles.

2.Given today’s internet speed, if one country decides to put in place more stringent regulations, researchers and suppliers can simply deploy the same technology and service using servers hosted at a different country. A classic example is that, after the US restricted the export of certain drone technology, American military allies simply shopped elsewhere. Today’s global framework in finance is that most firms are regulated from the jurisdictions where services are hosted and customers are billed. Moving jurisdiction is a relatively straightforward and inexpensive investment. If businesses located in a specific country “miss the boat” because of stringent restrictions unique to one jurisdiction, eventually the regulators there may be forced to give in due to intense business lobbying.

3.Ultimately, even if there is such an agency staffed by competent staff, and we can overcome the enormous challenge to coordinate the many jurisdictions involved globally, exactly what is such a regulator expected to accomplish, as in what rules it should set and when it should intervene?

1.4Defining a Road Map

While it is tempting to hypothesize on the potential impact of technology and automation on finance by reading today’s mass media, the reality is often more about working through the more precise details instead of prophesying on abstract disruption scenarios. AI and FinTech do not need to mean only the destruction of jobs. They can also mean more employment opportunities. The policy focus should be about to share those opportunities so that the net outcome is a general improvement to most buyers and sellers of financial services. More regulations are no “silver bullet”, and they may create more problems than they solve.

In this section, we aim to identify the most relevant topics by making an educated guess on which technology may matter the most to a post-AI financial industry.

1.4.1Nature of Financial Services

Let’s begin with a deeper understanding of financial services because we aim to use AI and FinTech to automate finance. The following is an excerpt from an e-mail providing invited feedback to a government program:

In order to give a clear explanation of how to apply artificial intelligence (AI) to investing: to the technology-trained crowd, you have to start by explaining how investments are made in the real world, for instance, the difference between retail, high net worth and institutional investments. To the finance-trained crowd, you have to explain what AI does and does not do, by giving some simple examples such as how AI engines can tell the difference between a dog and a wolf. Only then, we can go into a deeper explanation of why AI investing is not just mechanical pattern recognition, or the computer may simply recommend buying only Government bonds for any risk-averse investor.

It can be ineffective and even misleading to give an oversimplified explanation instead. To give you an analogy, you may have seen recent online postings on how “Facebook killed their AI engine that was inventing a new language”.^a As someone who has been working on machine learning algorithms over the last 30 years, I went back to the original post and read that Facebook engineers merely tried to solve a standard train and test problem in machine learning. There was no surprise. After two or three rounds of rewriting by journalists who did not have the necessary technical knowledge to comprehend the problem at hand, the story was rewritten to read like how “Skynet” from the Terminator movies was about to conspire against the human race by inventing its own secret language!

^a15 June 2017, The Atlantic, https://www.theatlantic.com/technology/archive/2017/06/artificial-intelligence-develops-its-own-non-human-language/530436/.

All financial markets aim to solve matching problems for buyers and sellers of financial assets. The entire financial market cannot outperform itself. Unless your next door neighbor is someone who has exceptionally better insights on public equities or bonds, like a Mr. Warren Buffet, to most investors investing is primarily coming up a collection of investments with suitable statistical properties to match future assets against estimated liabilities or future spending. According to the Berk-Green argument in [Berk 2004],⁵ there are limits on the ability to add value by selecting superior assets because:

1.First, performance on larger institutional portfolios tends to be heavily driven by allocating to the “right” region, country and/or sector;

2.Then, by finding misunderstood companies (e.g., over-hyped companies that are expected to fall in value, or vice versa); and

3.Finally, by finding “home run” investments that require domain expertise to pick let’s say the next Google or Facebook, which is few and far in between statistically.

By comparison, the following elements of institutional investing that are likely to take a long time to, or may never, be automated:

1.Face-to-face relationsh...

Table of contents

• Cover
• Halftitle
• Title
• Copyright
• Contents
• Preface
• 1 Introduction
• 2 Navigation and Vocabulary
• I CONSTRUCT PORTFOLIOS
• II SELECT ASSETS
• III DECIDE AND EXECUTE
• IV DELIVER REPORTS
• V DEPLOY
• Bibliography
• Index