Peter Dixon, Global Equities Economist, looks at advances in artificial intelligence and the effect these may have on finance.

Download the article as a pdf

Advances in artificial intelligence continue apace as computers compete to outdo humans at chess, driving, medical diagnoses and even writing. We can, however, console ourselves with the fact that we are still able to outperform in many of the service industries which require the human touch. Such as finance. Except even that is no longer true as wealth management algorithms have been developed which can provide bias-free advice at lower cost than a traditional asset manager. Indeed, the rise of robo advisors threatens to be as disruptive to the asset management industry as ETFs. That said we should not uncritically accept much of the hype surrounding digital advisers.

Is there anything they can’t do?

The extent to which computers can do jobs better than humans is rarely out of the headlines these days. Not content with designing machines which can beat humans at chess and verbal reasoning games such as the TV show Jeopardy!, the artificial intelligence community has raised the stakes still further by designing a machine capable of outperforming humans in the ancient Chinese strategy game Go. This is significant because Go involves significantly more calculations than chess and as a result is not so amenable to the brute force number crunching which has humbled chess grandmasters. It is reckoned that there are around 10,170 possible games which can be played on a Go board measuring 19x19 squares – almost 100 orders of magnitude larger than the number of atoms in the observable universe.

Chess programs solve problems by searching for the most promising looking moves and only then applying their number-crunching capacity to calculate all the possible combinations which can result. The AlphaGo program, which recently beat the world Go champion in a five game series, also uses these ideas but additionally applies deep learning techniques to facilitate the extraction of strategies from a huge mass of data. Since it is impossible to run through all strategies to the end of the game, the system is designed to look a few moves ahead to assess whether the pattern is developing in line with known winning strategies. This is akin to developing the kind of intuitive feel which human players develop for the game.

Many experts in the field reckon that deep learning is a key part of any future artificial intelligence system by allowing machines to apply trial and error strategies within a well- defined framework. This would bring them much closer to learning in the same way as humans. Such systems are still in their infancy but it is evident that advances in technology are going to impact on jobs, and many previously secure professions will feel the chill winds of competition. In the financial services industry, we are already seeing the application of technology – albeit at a less sophisticated level – in a financial advisory capacity which is changing the way in which many investors think. Welcome to the world of robo advising.

The new world of robo advising

A robo advisor is defined as a service which provides automated, algorithm-based portfolio management advice without the need for human financial planners. It is yet another aspect of the revolution which has swept across the financial services industry in recent years, breaking down many of the traditional barriers to entry and reducing the cost of financial services provision. It also promises to eliminate many of the inbuilt biases which characterise human investment decisions. But whilst robo advising captures the imagination, as efforts to supplant the human element are likely to do, the claims made for it should not be accepted uncritically.

The automation of financial planning decisions is the product of both technological advances and social change. But technical advances count for little if the acceptance threshold is low. Fortunately, the Millennial generation has taken to Fintech solutions like a duck to water. Millennials, defined as those born between 1982 and 2004, are the first generation to have grown up in the era of personal computing which has had a significant impact on the way in which they interact with data and technology to arrive at investment decisions. Up until around 30 years ago, financial information was hard to come by and many investors relied on their brokers to provide them with financial advice. The subsequent advent of the internet led to a democratisation of data, and the dramatic rise in the use of computer apps has further undercut the position of the financial information specialist. A survey conducted by the online broker E*TRADE Financial Corporation in 2014 suggested that people aged below 35 are more likely to use mobile devices to research investment strategies or monitor their portfolios than older age groups. In this environment, it is hardly surprising that a generation of investors has emerged which places its faith in technological investment planning.

There are essentially three types of robo advisor: fully- delegated, assisted and self-directed. The first of these is self-explanatory and implies that investors adopt a hands-off approach to their strategy by allowing the algorithm to make all decisions. For those not willing to go quite that far, assisted adviser platforms combine the automated elements of a robo advisor system with follow up from a human adviser. These services often offer additional packages such as social network investing which allow clients to benefit from the collective knowledge of other investors and thus replicate the more successful strategies of their peers. The self-directed model is described as a do-it-yourself strategy which aggregates funds data and constructs a platform which investors can then dip into as they choose.

Most robo advisors share a number of key characteristics:

  • Accessibility: The fact that such systems are web or app based means they are available 24/7
  • Low fees: Fee structures tend to be transparent and the absence of human advisers keeps costs down
  • Efficiency: Portfolio changes are free from the systematic biases which have traditionally characterised the fund management industry
  • Diversification: Systems generally adhere to modern portfolio theory which manages risk exposure through portfolio diversification

The future of the digital advisory business

Although these are valuable attributes, robo advice capabilities are still fairly basic. Indeed, whilst they represent a useful basket of services, they are still unable to meet the needs of investors with even moderately complex financial lives. At present, they use relatively simple surveys to generate client profiles in order to assess investment needs. But as deep learning techniques continue to evolve, it is likely that they will become an important component of the robo advising business by anticipating changing investment needs as client circumstances change. It is even possible to imagine that such systems will evolve towards a more interactive product, in which real time interaction between the client and the system becomes a reality.

We are still in the very early stages of the robo advisory business so we should beware of the hype which surrounds estimates of future market penetration. But in a recent survey, the consultancy AT Kearney reckoned that 48% of those polled had some form of interest in using robo advisory services at some point whilst only 31% ruled out their use to manage household taxable investment assets (see Charts 1 and 2). According to their estimates, the share of robo advisory services in the US market will increase 11-fold between 2015 and 2020 (see Chart 3) although even if this huge increase is realised, it will still leave market share only just above 5%. For those who believe such estimates are overdone, it is worth noting that the share of ETFs in US assets under management has shown spectacular growth over the past two decades (see Chart 4).

Chart 1: Interest in robo advisory services


Chart 2: Likelihood of using robo advisory services to manage household assets


Chart 3: Share of robo advisory market forecast to in- crease sharply


Chart 4: ETFs may be a pointer to the future importance of robo advisors


To date, the industry has made most gains in the US but it has potential to grow in other developed markets. In the UK, for example, rules were introduced in 2013 to prohibit financial advisers from accepting commissions for investment recommendations in order to eliminate scope for conflicts of interest. As a result, between 2011 and 2013 the number of financial advisers fell by almost 25% which has opened up a gap in the market which the robo advisory industry is striving to fill. Other European markets are also making great advances, notably Germany, whilst tech savvy markets in Asia offer significant opportunities for expansion, particularly since the traditional adviser model is not so well entrenched.

Before we get too carried away…

…it is worth noting that since this is a new business, not everything will be plain sailing. In order to build upon the recent stellar growth rates, it will be important to broaden the appeal of the digital advisory business beyond the younger age groups which form its current client base. The key to continued growth will be to ensure that existing clients stay with the business, whilst continuing to sign up new clients. In this way, the robo advisor model can be expected to generate decent organic growth. However, this assumes that the existing client base will not be bid away by the more sophisticated offerings of the traditional advisory business, or indeed that they will not increasingly become susceptible to the charms of direct personal contact.

Technological advances may take care of the former issue if systems evolve to meet their more sophisticated investment needs later in life. Inertia may prevent the latter: having grown up with machine-based advising, existing clients may be less likely to switch to a more traditional service. But one of the clinching arguments in favour of retaining existing clients is that they will likely be reluctant to switch to platforms where costs are significantly higher. Currently, robo advisors provide basic financial advisory and asset allocation for a fee of between 40 and 60 basis points (bps), which is half of the 100bps charged by traditional advisers.

Would you miss the human touch?

In some areas – the production of written text, for example – it is already very difficult to differentiate between the output generated by robots and that of humans. On the one hand this is reassuring. If a robot can do a job better than a human, and you cannot tell the difference between the two, then we are less likely to worry about allowing tasks to be performed by a robot. But only when robo advisors are able to pass the Turing test1 will the industry have passed a critical threshold. We are clearly not there yet. One of the important aspects of the client-adviser relationship is to provide reassurance during difficult times. Imagine the conversations which a crash along the lines of 2008 would have generated.

Another of the selling points of the digital advisory business

is that it is free from many of the biases which characterise human decision-making. Many investors rely on heuristics – a practical approach to problem solving which is not guaranteed to lead to the right answer. This can result in biases such as loss aversion in which investors attach more importance to not losing money rather than making a gain, with the evidence suggesting that they assign a weight twice as high to the former as the latter. But in rational portfolio theory, in which the objective of the exercise is to maximise returns, this bias is not accounted for. In any case, rational investment decisions do not always produce superior returns to so-called irrational decisions, as demonstrated by the Game Theory paradox known as the Traveller’s Dilemma2. Moreover, unlike the algorithms used to solve problems in games such as chess or Go, there are no readily definable end-point conditions. Deep learning algorithms may thus be able to exclude sub-optimal strategies, but so long as there is any kind of randomness associated with the pricing of financial securities, they cannot define what the end result of any investment strategy will be.

Perhaps the real test for the robo advising sector will come when it has to deal with a shock like the one that hit financial markets in 2008. Calculated rational decision-making at a time when markets are demonstrating the highest degree of irrationality will test digital decision-making to its limits. If such strategies are at least able to minimise losses, investors will be forced to take the robo advisory sector a lot more seriously.

Last Word

We are on the verge of an exciting future in the world of AI which is going to change the face of many industries – and finance is no exception. Attention has focused recently on the impact of outsourcing portfolio investment decisions to machines rather than relying on traditional management approaches. Clearly the robo advisory sector offers considerable potential but it currently remains a small part of the asset management industry. If it is to expand more rapidly, it will have to break out of its traditional stronghold made up of millennial investors. It has considerable potential to do so by virtue of the fact that it offers much lower cost advice than the traditional model. However, this does not mean that there will not be a place for investment managers in future – after all, clients will always need some form of hand-holding. But the digitisation of financial advice may turn out to be as disruptive to the managed funds industry as ETFs have proven to be.

1 Devised by the mathematician Alan Turing in 1950 requiring that humans cannot distinguish whether they are in conversation with a machine or another person.

2 An airline loses two suitcases belonging to two different travellers but each suitcase contains an identical antique. The airline is willing to pay compensation, set at a minimum of USD2 and a maximum of USD100. In order to determine the appropriate amount of compensation, each traveller must – without conferring – write down the value of the lost luggage on a piece of paper. If both write the same number, this is treated as the appropriate level of compensation. However, if they write different numbers, the lower one will be treated as the true value which will be paid to both parties. Moreover, the lower bidder will receive a bonus of USD2 whereas the higher bidder will have USD2 docked from their compensation. Backward induction, of the type used by computer based algorithms in solving chess problems, would arrive at a rational Nash equilibrium solution of USD2. But pretty much any choice (≥5) will generate a higher outcome than the so-called rational choice.