Funds that implement machine learning algorithms in end-to-end portfolio design are a subset of quantitative funds, and the larger quant space represents well under 3% of the allocation to global wealth assets. Yet, the message from the media is that there is an explosion in the adoption of machine learning tools by investment firms. We dispel several myths, such as widespread investing in quant funds, and consider reasons for the slow rate of investor adoption in the quant space. We examine investor behavior in the context of the innovation diffusion literature.

Will the next-generation quants using artificial intelligence with big data face the same obstacles to AUM growth as their quant predecessors? Credible advisors who clearly communicate the potential advantages and risks can lower investor resistance to quantitative investing. Each point made in this note is expanded in Rick’s article “Quant’s Quandary: Crossing |the Chasm,” published in The Journal of Investment Consulting.¹

Survey Results

Quantitative investment strategies may come in the form of hedge funds, mutual funds, separately managed accounts, and ETFs. Across forty-five conferences, consistently asking the question, “Have you made an allocation to a quantitative investment fund?” Little Harbor Advisors expected that about 15% of attendees would answer in the affirmative. Yet only a mere handful of Certified Financial Planners, Chartered Financial Analysts, and Chartered Alternative Investment Analysts did so. While perhaps not scientifically rigorous, the informal survey results were enough to prompt further study into reasons for such low allocations to a strategy with many benefits.

Quant Fund Refresher

On average, the number of quantitative funds’ holdings are six times larger than discretionary or fundamental funds. Quant funds are therefore highly diversified, and |the large number of positions held by quant funds also means they can have a low hit ratio and still generate a profit.

Quant fund managers take a systematic, disciplined approach that mitigates cognitive biases. Many studies show that well-known biases, such as confirmation bias, anchoring, and loss aversion, reduce profits. Systematic investment models and algorithmic strategies can reduce human weaknesses in speed, attention, fatigue, and biases.

Several research articles indicate that the performance of quantitative models is superior to discretionary managers in down markets and when cutting losses short and realizing gains.² Quant funds tend to take more short positions than discretionary managers and have a low correlation to traditional asset classes, offering the opportunity to increase a portfolio’s risk-return profile.

Quant Funds are Sold not Bought

The quantitative approach to fund management represents an innovation in asset management, as does the use of machine learning to make investment decisions. While quant funds have been around for decades, they are still new relative to the funds run by traditional managers. Traditional managers primarily perform fundamental analysis using their discretion to choose stocks in which they take mostly long positions. And while quant funds offer many benefits relative to conventional funds, the rate of adoption is low. Research on consumer resistance to positive changes attributes resistance to functional barriers and psychological barriers.³ For quant and machine learning funds, the information barrier is even more potent than those in the functional and psychological framework.

Among the functional barriers to the diffusion of beneficial innovation, the value and risk hurdles are most relevant to the case of quant funds.

The value barrier

Consumers must feel the added value of a new product outweighs the effort required to understand and use a new product. For investment products, fee structures can strengthen or weaken the value barrier. More than half of quant funds are within hedge fund vehicles, which charge much higher fees than mutual funds, raising the value barrier. Within hedge funds, average quant hedge fund fees are lower than qualitative hedge funds⁴, which suggests that the quant managers are aware of the value barrier.

The risk barrier

In the context of innovation adoption, risk is the perceived risk of trying or buying a new product or service. Proven ways to overcome the risk barrier when selling new products and services is to use client endorsements or testimonials. Another obvious way to overcome resistance to adoption is to offer free samples or product trials. The use of client testimonials is strictly prohibited by the SEC and “free samples” are a non-starter in asset management. However, allowing investors to visualize the interaction between various asset managers and investment strategies can illustrate the benefits of diversification by quantitative vs. qualitative security selection.

The information barrier

One universal truth of adoption intention is the lower the communicability of an innovation such as quantitative investing, the higher the resistance and reluctance to adopt. The biggest myth about quantitative investment is that it’s a black box.

Quantitative investment is rules-based, systematic, repeatable, and sustainable. (Although, models are subject to alpha decay.) Once quantitative managers explain their process, they arguably can be more transparent than their discretionary peers. It is the human brain, the brain of a discretionary manager or trader, that is the real black box.

Algorithmic Ascendance

An ever-increasing number of organizations have embraced algorithms to make what were traditionally human based decisions. An algorithm is a sequence of instructions that are carried out to perform a specific task. In investment management, an algorithm is a mathematical recipe that harnesses models, data, computers, and telecommunications to buy or sell securities. Many studies show that algorithms perform better than humans. For example, studies show that algorithms frequently outperform human experts in predicting the survival of cancer patients, predicting heart attacks, and assessing different kinds of pathologies.⁵

Much like certain unconscious human cognitive biases, algorithms can be biased. Unfortunate examples have appeared in many areas, including credit scoring and facial recognition.⁶ In response, a new science of socially aware algorithmic design is emerging.⁷ Algorithms can reflect the biases of programmers and datasets, from the data selected, collected, and omitted, to training the model. We are not suggesting the algorithms should be blindly or indiscriminately accepted or followed. On balance, however, algorithms have multiple built-in advantages of increased decision-making capacity, lower costs, minimization of errors, consistency, and when required, anonymity. When it comes to investing, in most instances, the quantitative or algorithmic method is superior to a discretionary manager’s subjective judgment.

Investors’ Algorithm Aversion

Despite the many benefits of algorithms, people often prefer to use human predictions. A Wharton study showed that participants did not lose confidence in human forecasters even when they produced twice as many errors as the prediction algorithm, and often demanded infallibility from algorithms.⁸

Two Sigma is a leading-edge quantitative investment strategy firm guided by the scientific method when building its algorithmic models. David Siegel (2015), Co chairman and founder of Two Sigma, summarizes our thoughts on algorithm aversion: The sooner we learn to place our faith in algorithms to perform tasks at which they demonstrably excel, the better off we humans will be. If the fear of the unknown really is driving skeptics’ irrational bias against algorithms, then it is the task of the practitioners who do understand their power (and limitations) to make the case in their favor.

Code Dependency

Although simple algorithms usually outperform humans, that does not mean there is no place for human input. Successful algorithms in the financial services are heavily dependent on domain experts who coach and counsel coders who write the programs. Data scientists and programmers rely on the acquired knowledge of domain experts and end users. Quant models can find false positive “discoveries” by overfitting data that may perform well on training data but will not perform out of sample. Quant models can also uncover spurious correlations that have no underlying causation. Domain expertise is required to distinguish between models generating sustainable alpha from those with a pattern that is not profitable. Humans are also required to monitor models that decay over time.

Quant’s Quandary & A Successful Chasm Transit

Change agents are needed to overcome barriers to quantitative investment adoption. The biggest barrier is information. Geoffrey A. Moore (author of Crossing The Chasm) augments the pioneering work of Everett M. Rogers (author of Diffusion Of Innovations) by noting there are chasms between Roger’s stages of adoption (innovators, early adopters, early majority, late majority, and laggards.) To cross the chasm, the message must change to suit the next group of adopters. Additionally, innovation diffusion studies show that repetition – frequency of contact – is the most important determinant of success.

Some industry experts believe that quants have a public relations problem of their own making:⁹ They willingly or unwittingly reinforced the black-box stereotype and are unwilling or unable to articulate their investment concepts in common language. If quant funds are to transform from misunderstood ugly ducklings into gracious swans, what must advisors and analysts do? Taking a page from Roger’s Diffusion of Innovations¹⁰ playbook, they must take on the role of change agent to persuade institutional and affluent investors to consider the potential benefits of quantitative investment.

References

Abis, S. 2017. Man vs Machine: Quantitative and Discretionary Equity Management. Columbia Business School, working paper (July 31). https://www8.gsb.columbia.edu/research archive/articles/25656.

Agrawal, A., J. Gans, and A. Goldfarb. 2018. Prediction Machines: The Simple Economics of Artificial Intelligence. HBR Press.

Chincarini, L. 2010. A Comparison of Quantitative and Qualitative Hedge Funds. (January 7). https://ssrn.com/abstract=1532992.

DiBartolomeo, D. 2013. The Near-Death Experience of Quant Asset Management. Northfield Information Services, Inc. (July 8).

Dietvorst, B. J. Simmons, and C. Massey. 2014. Algorithm Aversion: People Erroneously Avoid Algorithms After Seeing Them Err. Journal of Experimental Psychology 144, no. 1: 114-146.

Kearns, M. and A. Roth. 2019. The Ethical Algorithm: The Science of Socially Aware Algorithm Design. New York: Oxford University Press.

Logg, J. M. 2017. Theory of Machine: When Do People Rely on Algorithms? Harvard Business School Working Paper, No. 17-086 (March). http:// nrs.harvard.edu/urn-3:HUL. InstRepos:31677474.

Lohr, S. 2018. Facial Recognition is Accurate, If You’re a White Guy. New York Times (February 9). https://www..nytimes.com/2018/02/09/ technology/facial-recognition-race-artificialintelligence.html

Moore, G.A. 3013. Crossing the Chasm: Marketing and Selling High-Tech Products to Mainstream Customers (3rd editions). New York: Harper Business Essentials.

Roche, Richard P. (2019) Quant’s Quandary: Crossing the Chasm. The Journal of Investment Consulting, Vol 19, No. 1 pp. 53-65.

Rogers, E.M. 1995. Diffusion off Innovation (4th Edition). New York: The Free Press.

Sheth, J. N., and S. Ram. 1987. Bringing Innovation to Market: How to Break Corporate and Customer Barriers. New York: John Wiley & Sons, Inc.

Siegel, D. 2015. Human Error Is Unforgivable When We Shun Infallible Algorithms. Financial Times (June 4). https://www.twosigma.com/wp-content/ uploads/FT_OP_ED_PDF.pdf.

1. Roche, Richard P. (2019) “Quant’s Quandary: Crossing the Chasm,” published in The Journal of Investment Consulting, Vol 19, No. 1 pp. 53-65.
2. See Agrawal, et. al. (2018) and Chincarini (2010).
3. See Sheth and Ram (1987).
4. See Abis (2017).
5. See Logg (2017).
6. See Lohr (2018).
7. See Kearns, M. and A. Roth (2020).
8. See Dietvorst et al. (2014).
9. See DiBartolomeo (2013).
10. See Rogers (1995).