Can an AI asset manager beat active managers and passive index trackers?

Active managers have lagged the index for years, often to the irritation of family office clients, who can now access an AI system which claims to have beaten both.

Pave Finance’s system was first developed in 1997. It makes no promises on performance, but its US stock picks beat the S&P 500 index by an annualised 3% between May 2010 and December 2021. Its sample portfolios have also beaten the indices during a volatile 2022.

AI has not, as yet, come close to disrupting the industry. But Pave could be the shape of things to come

Its system was first developed by Steve Evans, former investment professional at Bank of America’s Merrill Lynch and, latterly, US Trust. It went on to manage $7 billion at wealthy clients at US Trust.

Evans left US Trust in 2017 to manage capital for two family offices, leaving US Trust to fill his shoes. His system is being married with software devised by Peter Corey, a trader at Steve Cohen’s top performing SAC Capital hedge fund between 1995 and 2004. Corey joined hedge fund Timescale Capital, which closed in 2012 due to partnership issues, and now runs a macro consultancy which dovetails with Pave. They joined forces with Pascal Cevaer, a former McKinsey consultant.

Special legal counsel, Jay Gould has backed the firm from the start, arguing it is an “exciting” new adviser based on an algorithm which has helped billionaires for more than ten years. Tiger 21, a family office networking group, agreed to host its new service, Pave Pro, in January.

Chief executive Chris Ainsworth – previously wealth adviser at Deutsche Bank and US Trust – said Pave wanted to target investors losing out, due to poor investment advice.  

Around 70% of Robinhood traders contrived to lose money during the bull markets of 2020 and 2021. Standard & Poor’s says 97% of US active funds have lagged the index over twenty years to their fees and behaviour flaws. Research in 2022 by Alex Chinco and Marco Sammon suggests 38% of US stocks are now indexed, including the 15% in funds.

The number of active managers and hedge funds who beat the index, net of fees, is vanishingly small, and they only take on individuals, sovereign funds and institutions who place big bets. 

AI has not, as yet, come close to disrupting the industry. But Pave could be the shape of things to come.

According to Ainsworth: “We wanted to create a product that was available to everybody, giving them the opportunity to have their money managed on a personalised basis, based on their values and beliefs, and give them the opportunity – while not guaranteed – to perform in line with – or better than – the market.”

Pave’s software can be applied to small sums of money at the outset to give clients the chance to try it out without disrupting their portfolios: “We think that’s a huge advantage because it allows people to get comfortable with us, without hiring and firing managers.” That said, consultants would argue that a client needs a 10% portfolio exposure to really impact performance.

Pave Pro system is made of two components. One is an optimisation process, based on a benchmark, which adjusts risk positions to compensate for a client’s preferred stock selection. So, if a client wants to own, and retain, a volatile stock like Tesla, its portfolio is optimised through software which invests in lower-risk stocks to share the ride. Family offices owning assets they do not want to sell can also rely on the software to adjust its portfolio.

Pave can benchmark on a variety of indices. It will be able to build portfolios around a hybrid bond/equity index by early June, when all of Corey’s software will be completely integrated with Evans’s system. Gold, bonds, commodities and ETFs can be part of the mix.

Pave’s second system component is an alpha-scoring engine, which rates, and ranks, every stock in a portfolio. It operates on the assumption that market trends tend to stay in place for longer than you might think,

Ainsworth says: “We have identified 87 factors that have an ability to predict outperformance over a significant period of time.” Value, momentum, earnings trends, surprises are just the most obvious factors taken into account. 

Ainsworth says: “The system scores are stock every week, but it doesn’t automatically buy or sell – it looks at the marginal utility of a stock and likelihoods and probabilities, plus tax efficiency.” Pave regularly carries out manual checks on the efficiency of its software to ensure its efficiency.

In a normal environment, portfolios get updated once a quarter, but the frequency has increased of late, due to disruption caused by market volatility. 

Even so, Pave would not expect to make changes more than once a week. It also updates its clients with a macro news reporting service, to inform their view of the market. 

Because Pave’s system uses machine learning, Ainsworth expects greater efficiency over time: “As we integrate more asset classes, more data, we may be able to interpret more quickly. Factors, correlations and the covariance of securities are always changing, and we hope to get better and better.”

Fees vary. A small investor running over $10,000 through a single consumer account would pay upwards of $10 monthly. A consumer account for less than $10,000 costs nothing. A $100 million family office requiring several licences, would cost $1,000 to £1,500 a month. 

However, Pave would collect the spread between buying and selling for each of its trades because, after all, there’s no such thing as an entirely free lunch.