Gal Krubiner is clicking through a PowerPoint presentation on his laptop, pausing on a picture of a 1980s era New York Stock Exchange floor.
“Recognize that?” he asked.
“You have 5,500 people working there. You had people trading from the stomach. The back office was full of papers – everything was different,” said Krubiner (who, it should be noted, at 31 years old came of age well after the photo was taken.) “Ninety percent of those people became irrelevant.”
Krubiner, the chief executive of New York-based Pagaya Investments, is not giving a history lesson. He’s pitching the idea that technology firms are poised to do something similar in asset-backed securities and institutional asset management.
Earlier this year, Pagaya, a startup led by three Israeli-born principals, with tony offices in both New York and Tel Aviv, launched what it said was the securitization’s industry first robo-deal – a fully automated transaction in which collateral was compiled, priced and sold without any assist from human intervention.
In the privately placed, $100 million deal, Pagaya’s artificial intelligence platform pooled together individual, unsecured consumer loans directly from the managed portfolio of (at that time) an undisclosed lender. Typically, buyers of marketplace loans usually bid on static, preselected folders of loan collateral compiled by the issuer.
The portfolio compilation criteria on the deal – and three subsequent issuances that have involved loans acquired from Prosper and LendingClub – were based on criteria encoded into the firm’s proprietary A.I. platform. The benchmarks for choosing the loans involve more than 1,600 attributes that Pagaya assigned from across any number of data points: borrower credentials, loan market and investor trends, macroeconomic factors, etc.
That same technology is also being used to actively manage the portfolio going forward with a short-duration, high-yield strategy of buying and selling loan assets determined by the cold logic from data accumulated by Pagaya’s machine-learning tools.
“The concept is very simple,” said Krubiner. Pagaya is “taking the three layers of asset management – deployment, performance and operations – and asking the question if we need, and we do, to be an asset manager from scratch in data-driven asset classes, how does it look like?”
Rise of the bots
Machine-learning inroads into financial services has been well documented in recent years.
BNY Mellon has launched A.I.-based “bots” to automate many of bank’s functions that were previously manual tasks from employees (such as clearing U.S. Treasuries).
Banks like BMO Financial Group and Nordic bank Nordea are using A.I. to function as online chat advisors to consumers.
A.I. is also being deployed by lenders and institutions for credit-decisioning, consumer behavioral analysis, fraud detection and internal operations. New York-based AlphaSense has developed an A.I.-based search engine used by banks and investment firms to analyze public corporate filings, research and news stories through natural language processing to ferret out trends in financial markets.
In addition, A.I. is being tapped for economic analysis and market predictions. Cambridge, Mass.-based Kensho Technologies, for example, is a hedge-fund markets analyst software firm that got attention for projecting an extended drop in British currency following the 2016 Brexit vote in the UK, according to Fortune magazine.
(Kensho was acquired by S&P Global in 2018 for $550 million).
These first steps for big data analytics into financial services makes A.I. in portfolio and credit management a natural next step, said Brad Bailey, a research director in capital markets for fintech research firm Celent.
“When you look at fixed income broadly and credit broadly, it’s how people manage – not just down to the details of that construction, but how are you going to manage your hedges, your pricing and your, the different things that happen within the tranches,” said Bailey. “This can all be done in much more effective ways then it’s done.”
A.I. can help ABS investors in several ways. The tools can make greater use of reams of available data – such as faster comparisons and risk/return projections from Reg AB-II and disclosure data in asset classes like auto loans and mortgages – but can also gather and crunch more opaque information such as individual consumer behavioral data that might not be apparent in loan apps and credit bureau files.
“The big managers, they’re all going to be thinking about this as part of their portfolio,” including both in tactical strategies (i.e., trading and reinvestment) and improving efficiencies and transparency to investors, said Bailey.
Under the hood
In Pagaya’s case, the loans it acquires and securitizes through its platform – Pagaya AI Debt Selection Trust – are chosen and bid on based on algorithms that measure an array of data beyond borrower FICOs, payment history and job title.
Pagaya will apply macroeconomic data, for instance, to individual loans that may uncover hidden risks.
Krubiner points to examples such as an unsecured loans to a construction worker and a nurse. While both may have similar credit scores and income, the construction worker is in a field with four times the historical employment volatility of the healthcare worker.
So instead of taking on both loans with similar attributes with similar bids, Pagaya might avoid the higher-risk loan or seek a premium for it.
Pagaya may also consider paying a higher rate on a loan that is safer than its surface details reveal, if the algorithms determine the rate is right.
“The very basic type of [lending] models will ask the question, what is the [debt-to-income]. What is the monthly income and therefore what is the interest rate?” said Krubiner. “But when we are an asset manager, we care more than everything [about] performance. Not to lose money. We want to be in a situation where if the crisis comes, we are better off with our portfolio.”
Pagaya’s technology, using API connections with its MPL partners, will continue monitoring each loan it purchases for institutional clients like Citigroup – scoring the underlying borrower’s updated credit profile, the lender’s underwriting model and managed portfolio performance, among other factors.
The goal is to formulate projected returns for Pagaya’s clients based on the thousands of data points that might uncover attributes that mitigate or enhance the decisioning factor.
This will factor in the ongoing management of the asset within the portfolio that promises a greater level of operational savings and less “from-the-gut” thinking on buying and selling.
“The concept to understand is there is no ‘good’ or ‘bad’ loan,” said Krubiner. “There is a loan that is priced well or not.”
Alternative data
Dan Petrozzo, a veteran technology venture capitalist, likens the A.I. evolution to an ironic return to classic investing and lending: making more intrinsic decisions based on data that bankers and issuers may only otherwise glean from deep, one-on-one interaction.
“It’s very interesting. What the technology ultimately does, it’s making a personalized decision based upon a whole bunch of data,” said Petrozzo, who once served as global head of technology for investment management at Goldman Sachs, and now leads fintech investments for venture capital firm Oak HC/FT (a key investor in Pagaya).
“This is what would have happened before FICO existed. [Bankers] would call you up. They would call your friends, do you work a steady job. They would look you in the eye, are you a good person.
“Effectively, that’s what the [A.I.] algorithm is doing at scale.”
The core, fundamental skillset for lenders, investors and fixed-income managers is derived from a simple concept.
What is a good investment – and what is a bad bet?
For decades, firms have increasingly poured millions into their intelligence-gathering mission to gain an edge to raise the odds of finding the former and avoiding the latter.
But in the new era of big data, one in which artificial intelligence and machine-learning tools are taking center stage in firms’ back offices, a new dynamic is taking shape.
What if the best way to choose investable assets is no longer a binary “yes-or-no” question?
Analytics have been a long been staple of investor decisioning and risk management, but the new A.I. promises a deeper trove of alternative data, said Celent’s Bailey.
“If you think about what [lenders and issuers] have done with thinking about how they look at credit, how they look at credit rating, how they look at the consumer or other types of credit,” securitization investors can also take the deeper dive different types of machine learning algorithm to gain the same insight, said Bailey.
“You might have found something that, [for example], there’s a 3 percent correlation between certain assets. Has that been something investors worry and think about?”
Krubiner said that while big-data analytics is nothing new, what institutional clients have lacked in portfolio management is collateral context.
The views into underlying assets may be too opaque – or in the case of assets backed by online consumer loans, too new – to properly assess the “value proposition” from the buy-side perspective that Pagaya’s platform and its technology seek to attain.
Pagaya’s platform buildout was led by chief technology officer Avital Pardo, who was a data scientist and analyst for small-biz lender Fundbox. (Pardo deployed A.I. in building a platform that created ability-to-pay projections for small- and medium enterprises, said Krubiner.)
Pardo and Krubiner launched Pagaya in 2016 with another Israeli tech industry entrepreneur, Yahav Yulzari. It was seeded with early investments from the $2.5-billion asset firm Viola Ventures in Israel. The company received a later investment from retired American Express chief executive Harvey Golub during the firm’s Series B funding in August 2018 (“He literally gave us the first million dollars,” said Krubiner). Golub became a member of Pagaya’s board, which is chaired by Viola Group partner Avi Zeevi.
The faith from early believers in A.I. jumpstarted Pagaya to raise more than $950 million in capital used to acquire its loans, of which more than $515 million have been securitized in private deals via Cantor Fitzgerald. The four deals have made Pagaya a top-10 issuer in the MPL ABS space in 2019. (It’s most recent deal priced in October, topping $200 million in loan-backed securities).
Pagaya had its eye on securitization early by recruiting an asset-backed specialist as its first employee. Benjamin Blatt, a Capital One veteran as well as the former capital markets manager for student-loan refinancing firm CommonBond, joined the firm in a similar capital markets post for Pagaya in 2017. (He worked out of a shared WeWork space in the same New York building where Pagaya now has its co-headquarters location).
In 2018 Pagaya made its biggest hire to date when it recruited former BlackRock managing director Ed Mallon as chief investment officer, with plans to expand Pagaya’s investment platform into new asset classes including mortgages, real estate and auto loans, according to Krubiner.
Adoption grows
Petrozzo said AI’s applicability to asset management was first pioneered in equity trading by firms like Two Sigma Investments, the New York-based hedge fund founded in 2001 which adapted A.I., machine learning and distributed computing into its trading strategies.
“These types of companies sort of set the groundwork for how you can use computers to an advantage… in the buying of securities. But they tend to still be very much on the short-term trade, not on a buy and hold strategy,” said Petrozzo.
“That’s not like how we would how we would think as investors about asset management.”
A.I. is gaining traction in financial services because of the results that financial services firms are reaping from investments in the technology.
In a report earlier this year, Deloitte surveyed 206 global financial institutions on the impact of institutions’ use of A.I. technology (such as advanced analytics, process automation, “robo” advisors and self-learning programs”),
Deloitte found that “frontrunner” institutions in A.I. investments gained 19% returns on their machine-learning outlays compared to 12% for firms that have not adopted these tools as widely.
Deloitte pointed to Nordic bank Nordea as a firm using AI in multiple ways across its organization. The company developed an internal chatbot (“Nova”) which used natural-language processing for responding to online customer queries, and looking at means to automate claims handling, fraud detection and personalizing recommendations for clients.
Industry adoption is also easier because the barriers for entry have dissolved for upstart tech firms. The massive investments in computing power and storage has shrunk with the availability of affordable cloud services, noted Bailey.
“All the limitations that existed even five years ago are gone,” said Bailey. “If you have a good data set, you can run.”