An analytics approach to debiasing asset- management decisions

Wealth & Asset Management December 2017

Nick Hoffman Martin Huber Magdalena Smith An analytics approach to debiasing asset-management decisions

Asset managers can find a competitive edge in debiasing techniques— accelerated with advanced analytics.

Investment managers are pursuing analytics- Richard H. Thaler, one of the leaders in this field aided improvements in many different areas of (see sidebar, “The debiasing nudge”). the business, from customer and asset acquisition to operations and overhead costs. The change Asset management: An industry facing area we focus on in this discussion is investment challenges performance improvement, specifically the Investment managers are facing significant debiasing of investment decisions. With the challenges to profitability. Demonstrating the help of more advanced analytics than they are value of active management has become more already using, funds have been able to measure difficult in a market where returns are narrowing. the role played by bias in suboptimal trading Dissatisfaction with active performance is causing decisions, connecting particular biases customers to migrate toward cheaper passive to particular decisions. Such discoveries funds (Exhibit 1). Actions active funds are taking in provide the necessary foundation for effective response include reducing fees, which has created countermeasures—the debiasing methods that can a competitive cycle that is compressing margins. bring significant performance improvements to a Some funds (such as Allianz Global Investors pressured industry. and Fidelity) are changing their fee and pricing structures to make them more dependent on Business leaders are increasingly recognizing the outperformance through the use of fulcrum-type risk of bias in business decision making. Insights fee structures in their funds. from the fields of behavioral economics and cognitive psychology continually emerge to reveal To push back on these trends, some investment that individuals and institutions do not base managers are broaching the topic of bias in financial and other decisions on purely rational investment decisions. They are seeking a considerations. The contours of the irrational competitive edge and—perhaps more important— biases have become increasingly known, and the to improve the value proposition of active ways bias operates in our thought processes can management. Evidence from within and outside often be predicted. Most important of all, the the industry strongly suggests that even the leading methods and means to counteract biases are asset managers in top-performing funds could becoming more sophisticated and effective, at least improve their investment performance by applying for those executives willing to inquire into their debiasing techniques. Our recent experience debiasing needs. working with investment managers has shown that enhancing these techniques with analytics can Advances in the use of debiasing techniques to improve performance significantly. improve decision making have been inspired by the research of many pioneering scholars. Bias and debiasing in action Their innovations have had numerous practical Bias is a risk in all business decision making— confirmations in the public sector and business the more significant the decision, the greater the settings as well as in the decisions of individuals. risk. Particular biases have been behind many As if in recognition of the deepening relevance costly missteps by companies and institutions in of debiasing in economic life, the Nobel Prize every sector. Consequently, behavioral scientists in Economic Sciences was awarded this year to and business leaders have developed methods for

2 An analytics approach to debiasing asset-management decisions The debiasing nudge

In October 2017, the Nobel Committee awarded its decision making is undermined by biases. Among prize in Economic Sciences to Richard H. Thaler of the biases the authors analyzed are the stability the University of Chicago, who has thought deeply biases to which investment decision making is highly about rational and irrational behaviors in economic susceptible: anchoring, for example—the tying decision making. He and his colleagues in the field of of actions to an initial value and failure to adjust to behavioral economics—including Dan Ariely, Amos take into account new information. Loss aversion Tversky, and Daniel Kahneman (a previous Nobel is another such bias—the familiar fear that makes winner)—analyze the psychological dimension of us more risk averse than logic would allow. Where economics, partly leaning on insights from the field of human decision making is more prone to bias than cognitive psychology. Their research reveals patterns to reasonable deliberation, Thaler and Sunstein that did not align with the rational assumptions recommend the debiasing “nudge”—a benign, often embedded in prevailing descriptions of economic small adjustment that counters irrational impulses. systems and individual financial actions. They The authors discuss many examples of successful found that especially under conditions of risk and nudges; reviews of Nudge most often cited the opt- uncertainty, individuals as well as institutions fail to in default for employee retirement savings plans. In behave as might be expected when making financial this example, many more employees will save for decisions. Assumed universal principles, such as retirement when they are automatically enrolled in actual self-interest and a sure grasp of dynamic a plan. That is, the nudge of requiring employees inputs including time and probability, often give way to to save unless they opt out of the plan gets better irrational and unpredictable actions, based on narrow results than an approach that requires them to opt in. or flawed data and personal experiences. Everyone recognizes the need to save for retirement, but not everyone acts on it. The opt-in nudge is in fact In their book, Nudge: Improving Decisions About a form of debiasing for those irrationally ignoring their Health, Wealth, and Happiness (2008), Thaler and own future financial well-being. coauthor Cass Sunstein reveal ways that rational

debiasing decision making. Many of the companies top performers were those that least anchored that have adopted these methods can attest to their their capital-allocation decisions to decisions effectiveness in improving outcomes: made the previous year. “Dynamic reallocators” achieved a median return that was nearly ƒƒ McKinsey cross-sector research has suggested four percentage points higher than that of that one of the most prevalent irrational biases, “Dormant reallocators.” the form of stability bias known as anchoring, undermines optimal capital reallocation. The ƒƒ An executive at the German electric utility research compared capital allocation and total RWE recently discussed his experience leading shareholder returns over a 20-year period a debiasing effort.1 Like most of its peers, the (1990 to 2010) at more than 1,500 publicly company had until recently based capital- traded companies in the United States. The investment decisions on ever-rising commodity

An analytics approach to debiasing asset-management decisions 3 CDP 2017 An analytics approach to debiasing asset-management decisions Exhibit 1 of 4

Exhibit 1 Customers are migrating to passive equity funds, which have lower margins.

North America net ow growth and revenue margin by asset class1

Passive Active Alternatives

Bubble size 2016 AUM

150 145 Public market alternatives 60 Active specialty 55 equity Active specialty 50 Multi-asset fixed income basis points 45 40 35 Active core 30 Active core equity fixed income 25 Passive 20 Passive fixed 15 equity Passive Passive income 10 Money other

2016 revenue margin, multi-asset 5 market 0 –13 –12 –11 –10 –9 –8 –7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 2016 net ows,

1 Active core equity includes US large cap and yieldincome equity; active core fixed income includes core, core plus, and municipal bonds; active specialty equity includes foreign, global, EM and US smallmid-cap; active specialty fixed income includes global, EM, high yield, TIPS, and unconstrained. Source: 2016 McKinsey Performance Lens Global Asset Management Survey and Global Growth Cube

and power prices. RWE leaders realized that ƒƒ A global investment bank reviewed its traders’ debiasing efforts could have challenged their decisions on equity position weighting. The earlier assumptions, enabling them to hedge bank recognized overcommitment to positions potential adverse effects. For future decisions, and suboptimal execution due to endowment they implemented a farsighted cultural change effects and confirmation bias. The endowment program to identify and counter cognitive effect, discussed in the research of Richard biases throughout the organization. Care was Thaler and other behavioral economists, refers taken to ensure that dissenting and outside to the psychological effects that prejudice analyses were fully articulated. For important owners in favor of retaining their assets despite moves, a “devil’s advocate” was appointed; for changing conditions. Confirmation bias is a the company’s most pressing strategic problem, form of pattern-recognition bias, causing us separate internal and external teams—a to see nonexistent patterns in information: “red team–blue team” approach—were assigned evidence supporting a favored belief is to develop solutions independently of each overvalued, while evidence to the contrary is other. (The devil’s advocate and the red team– discounted. To counteract these biases, the blue team approach are defined, along with bank implemented “premortem” evaluations other debiasing techniques, toward the end of before each trading stage, with checklists to this article.) ensure that the stages were investigated fully.2

4 An analytics approach to debiasing asset-management decisions ƒƒ A global asset management company likewise investment managers could now select and identified confirmation bias in the decisions of apply debiasing methods to greater effect in their its investment managers. The company used investment decisions. devil’s advocates to develop opposing views on investment decisions.3 The analytics edge The approach requires the creation of an ƒƒ A leading pharmaceutical company tackled integrated data set covering the investment overconfidence bias in the proposals of its portfolio. Included are all stock decisions since researchers by using premortems to challenge a fund’s inception. Security selection, security research proposals and renewals before weighting, and selling timing are captured, as investing in any project.4 are the activities and actors that developed and maintained the portfolio. Data is thus compiled for ƒƒ In the healthcare industry in the United States, the processes and decisions made by the individual research revealed that anchoring and stability investment managers and their teams. This history biases significantly undermined diagnostic includes team communications and, insofar as accuracy and effective treatment in hospitals. possible, the behavior, reasoning, and emotions According to one widely cited study, errors associated with individual decisions. occurred in 6.5 percent of hospital admissions in 2000–10. NIH and other research further The funds deploy machine learning, guided by suggests that diagnostic and other errors were hypotheses developed jointly by fund managers causing as many as 200,000 avoidable deaths and experts about the biases that might have annually.5 In response, leading hospitals negatively affected their investment decisions. implemented a strong debiasing culture, Typical biases afflicting asset management including requisite consideration of “must not performance are overconfidence, loss aversion, miss” diagnoses, checklists for ICU, surgery, or the false analogy—the logical fallacy in which and diagnostic procedures, the presentation of inductive reasoning is simulated through invalid evidence in support of alternative diagnoses, comparisons. The hypothetical biases are tested and group discussions to encourage others’ by building an exploratory model to understand opinions in complex situations. emotions and processes associated with trading decisions. Emotions behind biased decisions are Debiasing in asset management diagnosed in the data through K-means cluster In the asset management sector, investment analysis—an iterative vector analysis enabled by decisions are being analyzed in light of debiasing machine-learning algorithms, which can isolate experiences in other industries. A few leading patterns in highly complex data sets (Exhibit 2). funds have employed analytics in this effort, to improve effectiveness in diagnosing bias and The use of analytics in this way, to discover biases its drivers. Working with analytics experts and and their sources, is new. The biases themselves behavioral scientists, they applied machine- are familiar and susceptible to the debiasing learning algorithms to their historical investment methods elaborated in the scholarly literature and data. They discovered clusters of suboptimal practical approaches. investment decisions that showed potential biases. By looking more closely at these suboptimal Initial performance decomposition analysis decisions, the funds identified consistent bias in The approach starts with performance the processes by which the decisions were reached decomposition analysis, the relatively simple and the accompanying emotions experienced diagnostic tool long used in the industry. This by the decision makers. Having exposed the initial step is followed by further analyses, aimed patterns of bias with the help of analytics, these at discovering the processes and emotions surrounding specific investment decisions as well

An analytics approach to debiasing asset-management decisions 5 CDP 2017 An analytics approach to debiasing asset-management decisions Exhibit 2 of 4

Exhibit 2 To detect and correct for biases in investment decisions, some funds have successfully applied analytics to large sets of historical data.

ngoing data and analyses

ortfolio ● All security decisions since inception of fund ● Decomposition of performance attributed to security selection, weighting, and selling timing ututs

eam communications ● Diagnoses of and rocesses Analytics: aly machine individual fund learning by deeloing managers’ decision- ● Decision-making processes hyotheses making biases and of the fund managers and underlying drivers, ● Find patterns of bias their analysts consistently associated and their drivers with specific security ● Security selection, ● Find clusters where characteristics weighting, and suboptimal security selling processes decisions and drivers Plans and methods ● Individual behavior and traits of bias are consistent for debiasing future decisions ndiidual fundmanager behaior ● Decision-making processes by stock ● Individual behavior and traits at point of decisionnondecision

as the periods when fund managers did not make At one high-performing equities fund, performance such decisions. decomposition analysis revealed that over an eight- year period, superior stock-selection decisions For the portfolio as a whole and for each were driving most of the returns. While both stock constituent security, decomposition analysis weighting and stock-selling timing were acting shows the performance contribution of three as drags on performance, in the years since the types of decisions: security selection (including financial crisis the main negative performance purchase timing), security weighting, and the factor was selling timing. Consequently, this timing of security selling. From this approach, factor was prioritized for more advanced analysis simple patterns of suboptimal decision making (Exhibit 3). can be generally discerned. Fund managers can then prioritize the patterns, making them the Suboptimal timing and the contributing biases basis for more complex analyses of decision Through analysis of historic selling timing, bias. For example, portfolio segments might be asset managers discovered that approximately identified for which skills in the three types of three out of ten stocks were sold too early or too decisions have significant impact on the total late, according to the fund’s own standards. To return of a fund. identify the biases that likely influenced these decisions, a detailed questionnaire was developed,

6 An analytics approach to debiasing asset-management decisions probing what might be called the structural such biases as anchoring, loss aversion, or the and emotional environment surrounding each endowment effect. In a number of instances, trade. The structural environment includes such positions in strong-performing stocks were sold too factors as stock returns, valuation, M&A and early. More often than not, the fund learned that capital discipline, ESG (environment, social, and these sales took place in an emotional environment corporate governance), trading conditions, the defined by pride and optimism. The usual source of investment case, portfolio construction, and this mood was a fund manager’s conviction in the alternative investment opportunities. The original investment case and valuation. Convictions structural values spark emotions associated with of this kind are connected to the bias known as individual trades and groups of trades. These can anchoring—the tendency to allow one’s actions to beCDP positive, 2017 negative, or neutral, ranging from be governed by a fixed logic, regardless of changing optimism and confidence to fear or impatience. conditions. Sometimes, however, managers were An analytics approach to debiasing asset-management decisions reluctant to hold on to strongly performing stocks Exhibit 3 of 4 An analysis of the emotions that led to particular for a different reason: aversion to losing profits types of repetitive suboptimal decisions identified made so far. Here the operative bias is known as

Exhibit 3 At one fund, the data revealed that success was due to superior stock selection, while selling timing had become the main negative factor.

ontribution to fund erformance ersus market inde total returns to shareholders S ercentage oints A

he nds oerall otperormance o the maret inde oth eighting and selling-timing sills as drien by eceptional stoc selection contribted negatiely to perormance

Stock selection by year Stock eighting by year

Avg

iming as prioritied or rther analysis based on Avg impact and manager hypotheses Stockselling timing by year

Avg

An analytics approach to debiasing asset-management decisions 7 loss aversion—referring to the strong preference to ƒƒ Checklists have been proven to be very efficient avoid loss despite favorable odds. in slowing down decision making. They have been used widely in medicine and law. For The biases underlying the inaction that leads fund managers, a checklist of factors beyond to stocks being sold too late were identified as valuation is reviewed before the final decision anchoring, the endowment effect, and regret is made. These factors could include strong aversion (a fear that in hindsight any choice will external buy recommendations or the relative appear to be suboptimal). Overconfidence and loss strength of the stock in comparison to peers. aversion were the biases associated with selling too early. The entire analytics-based approach, ƒƒ Clean sheet redesign. In this approach, the including K-means cluster analysis, is sketched in manager takes a fresh look at the investment, Exhibit 4. revising the strategy as much as needed in light of current conditions and rebasing contemplated Tailoring debiasing methods moves on the new analysis. Essentially this Once the funds in question could see the biases technique encourages decision makers to treat behind the clusters of suboptimal trades in the past, a decision as if it is a new investment. steps were taken to ensure that future decisions would undergo debiasing when certain conditions ƒƒ Devil’s advocate. This is a formal role assigned were present. Metrics were established, called to an individual before a final decision is made. “triggers,” that would signal the presence of those The job of the devil’s advocate is to challenge the conditions. The triggers included such events as current view of the fund manager, marshaling a 25 percent movement of the stock price within as much pertinent contrary evidence as possible. a three-month period, an investment reaching its Accordingly, the person in this role establishes fair value, a fund manager thinking of selling a the deciding factors in such a way that most position, the failure to add to a position during a convincingly supports the opposite outcome. 60-day period when the stock is drifting down, or a negative attribution to fund performance of over 50 ƒƒ Premortem analysis is a method for basis points within a year. understanding the potential causes of failure. The approach encourages people to express Debiasing techniques that will most effectively the doubts, criticisms, and second thoughts change the decision-making process to reduce the that might otherwise be suppressed due to presence and impact of bias can then be selected organizational biases. In medicine, for example, by the fund manager. The techniques themselves an assumption is made that a patient has died, are not specific to the asset management sector but and the team then seeks to discover how it have been used to help companies and institutions happened. A premortem approach has been make more effective and profitable decisions used in many academic, professional, and throughout the private and public sectors. What business settings, since it has been practically is specific to the industry is the way the methods proven to reduce failure. are selected and applied, taking into account the investment process, the fund’s mandate, its ƒƒ Red team–blue team. In this widely used culture, and the personality traits of those involved debiasing approach, two independent groups (or in the decisions. individuals) are assigned to represent opposing positions, for and against, on contemplated The following list describes some of the debiasing decisions. The fund manager does not participate techniques that can be tailored for the needs of in the discussions but only observes as each side funds and the specific circumstances in which they challenges the other’s analysis and arguments. will be applied:

8 An analytics approach to debiasing asset-management decisions CDP 2017 An analytics approach to debiasing asset-management decisions Exhibit 4 of 4

Exhibit 4 Repetitive patterns of selling timing and behavioral bias were uncovered with machine learning.

Advanced analytics can isolate biases underlying suboptimal trades

INVESTMENT-FUND EAMPLES

K-means cluster analysis Structural analysis Emotional patterns Comparison K-means cluster analysis, The structural factors for each Particular emotions can The emotions associated an advanced-analytics cluster can then be analyzed, be related to clusters of with each cluster are technique, can be used to including the decision-making suboptimal decisions compared to those that identify clusters of trading processes, prevailing conditions in and consistent trading behavioral scientists decisions with similar the industry and region, the basket patterns. expect for certain biases. emotional profiles. of stocks, and stock weights.

The clusters identi ed Original Cluster analysis The biases noticed Some 30 of trades 6 6 Within the clusters 6 were identified as groups were identified suboptimal, of which 4 4 with consistent bias: 80 were found to + Cluster ● anchoring: sold too early be the outcome of center biased decision x 2 2 ● anchoring: sold too late making, resulting in + ● loss aversion: losing gain sales made too late + 0 0 ● loss aversion: fear of or too early. further loss

–2 –2 ● regret aversion –2 0 2 4 6 –2 0 2 4 6 y y ● endowment effect

How K-means cluster analysis is used to identify clusters

Compute distance Generate For each stock, Assign each Are all the K-cluster Final cluster between each pair K-initial cluster calculate its observation to centers steady memberships of stocks, according centers using distance to the closest cluster are determined to the similarity of random or each cluster and recalculate for all stocks the underlying known seeds1 center centers emotions: the more alike the emotions, Evaluate model No Yes the shorter the distance

1“Seeds” are the values used as starting points to generate numbers for simulations; the starting points can be random or known.

ƒƒ Visual nudging. In keeping with the principles analysts’ upgrades, price performance relative discussed in Richard Thaler and Cass to other stocks in the sector or region, or Sunstein’s book, Nudge (see sidebar, “The changes to the risk model. debiasing nudge”), this debiasing technique presents fund managers with alternative As the foregoing discussion indicates, suboptimal metrics to consider before making a decision. decisions can have a significant impact on Typically, these metrics reveal the structural performance. Our initial analysis suggests that, environment in greater detail—for example, even under conservative assumptions, debiasing

An analytics approach to debiasing asset-management decisions 9 will allow funds to undo this negative impact and reap the performance rewards. In looking at actual 1 See “A case study in combating bias,” McKinsey Quarterly, funds, we saw potential improvements of between May 2017, McKinsey.com. 100 and 300 basis points. 2 See Iris Bohnet, What Works: Gender Equality by Design, Cambridge, Mass.: Harvard University Press, 2016. Like In addition to the debiasing exercises, asset devil’s advocate and red team-blue team, premortem analysis is one of the debiasing techniques defined toward the end of managers are also applying change programs to this article. reinforce informed and well-reasoned decision 3 See Chuck Widger and Daniel Crosby, Personal Benchmark: making. These programs include role modeling, Integrating Behavioral Finance and Investment Management, awareness building around bias, and capability Hoboken, NJ: John Wiley & Sons, 2014. building to support improved decision making, as 4 See Michael D. Mumford and Michael Frese, editors, The well as more formal procedures. As with debiasing, Psychology of Planning in Organizations: Research and programs to address mind-sets and behaviors Applications, New York: Routledge, 2015. have been widely used in the public and private 5 See Charles Andel, Stephen L. Davidow, Mark Hollander, and sectors. The experience with such cultural-change David A. Moreno, “The economics of health care quality and programs demonstrates that, while they may not medical errors,” Journal of Health Care Finance, Fall 2012, ensure a successful transformation, success is Volume 39, Number 1, pp. 39–50. much less likely without them. Nick Hoffman is a partner in McKinsey’s London office, where Magdalena Smith is an expert; Martin Huber is a senior partner in the Düsseldorf office. Real investment managers have used these debiasing techniques to create real performance Copyright © 2017 McKinsey & Company. All rights improvements. These leaders are gaining a reserved. competitive edge over their peers, while generally improving the value proposition for active funds. The advantages of debiasing are compelling and inexpensive. Even a devil’s advocate would find it hard to argue against starting your debiasing program now.

10 An analytics approach to debiasing asset-management decisions Contact for distribution: Magdalena Smith Phone: +44 (20) 7961 7245 Email: [email protected]

December 2017 Designed by Global Editorial Services Copyright © McKinsey & Company