The Active Manager Paradox: High-Conviction Overweight Positions
Is active management’s decade-long losing streak to passive management due to high fees, a lack of manager skill, or something else?
What’s required to answer this question is not rampant speculation but a fact-based assessment of manager decision making. As the saying goes, “You cannot manage what you cannot measure.”
Our research explored how active managers generate stock-selection alpha. We conducted a multi-year analysis that covered 114 US equity mutual funds from 57 fund families and evaluated more than 400,000 individual rolling one-year performance periods. Combined, our sample represented about $2 trillion in assets under management (AUM).
Our key focus? Manager conviction. How committed is the manager to the different subgroupings of equities within each fund? To find out, we measured the scale of overweight and underweight positions rather than the raw size of the holdings, which tends to be biased by the benchmark weightings.
Research Design and Objective
The primary categories of stock positions based on a manager’s active intent are
- High-Conviction Overweight
- Underweight
- Neutral Weight
We identified the constituents of these three categories by measuring real-time, daily mutual fund holdings and weights and rebalancing each group every 14 days. The fund holdings data came from Turing Technology Associates’ proprietary Hercules fund-replication system and corresponding Hercules Database.
Summary Results
The results, depicted below, feature two sets of data: the success rate of each category compared with the benchmark over rolling one-year periods and the average annual excess return of those rolling periods.
The Impact of High-Conviction Overweights, Gross of Fees
The Impact of High Conviction Overweights, Net of 85 bps Fees
The High-Conviction Overweights, composed of the managers’ best ideas, is the only category that delivers stock-selection alpha. High-Conviction Overweights achieved success rates of 84% gross of fees and 74% net of a theoretical 85 basis points (bps) fees. Underweights and Neutral Weights, by comparison, generated a success rate of 50% gross of fees — the equivalent of a pure beta portfolio — and materially inferior success rates after fees.
That High-Conviction Overweights are the sole category through which active managers could add alpha defies the long-held assumption that managers can improve performance throughout the entire stock-selection and portfolio construction process.
Active Manager Paradox
While our data shows that fund managers can exhibit persistent skill through their high-conviction best ideas, it also reveals a portfolio design paradox.
As the sole source of excess return, High-Conviction Overweights need to be the main emphasis of all actively managed portfolios. Any allocation to anything else will reduce returns.
Yet, according to our research, the average manager sabotaged their returns by shrinking the High-Conviction Overweight stocks to an overall portfolio weight of 55%. The corresponding portfolio allocation to Underweights and Neutral Weights thus acts as a “Beta Anchor” that severely dilutes the alpha generated by High-Conviction Overweight positions.
To use a sports analogy, this is like an NFL football team voluntarily removing its star quarterback from the game after the first half. It does not constitute a winning strategy.
To be sure, a “Beta Anchor” has a variety of justifications. Allocating to a market-neutral component reduces the portfolio’s tracking error versus the benchmark. It also decreases the likelihood of a relative performance failure compared with a more highly concentrated portfolio. Nevertheless, any risk-management benefit is offset by a significant performance penalty.
Implications for Investors
We held off claiming to have the solution to the Active Manager Paradox in this paper. And we didn’t address the risk-management considerations. But this topic is not trivial.
Active management is, by definition, a premium service. Its fees are higher because the expectation is that it will deliver higher returns.
But our research indicates that the current approach to actively managed fund design compromises the manager’s ability to outperform.
Outside research supports the cause-and-effect implications of reduced allocations to High-Conviction Overweight stocks. Morningstar currently classifies mutual funds as either active or passive and provides summary return data for the average actively managed mutual fund by asset class. The chart below compares the relative performance of actively managed large-blend funds with that of the S&P 500 Index over rolling calendar years since 1990.
The results are bleak.
Actively Managed Large-Blend Mutual Funds vs. the S&P 500
Large-blend active managers have outperformed the S&P 500 in only 5 of the 29 years analyzed. On average, active managers underperformed by –1.7% per calendar year.
The results are even worse for the most recent decade. Since 2010, active managers have failed to keep pace with the S&P 500 every year, lagging by –2.1% a year on average.
While it is industry convention to blame these outcomes on higher fees, our research suggests that fees are only a secondary contributor. Diluting the sole source of stock-selection alpha to a minority component of a portfolio has far greater structural impact than higher fees.
The decade-long failure of active managers to compete with their passive counterparts has not gone unnoticed. End investors have voted with their feet: In the last five years, approximately $1.3 trillion has been taken out of active funds, while $1.3 trillion has flowed into passive funds and exchange-traded funds (ETFs), according to Morningstar.
Generating viable solutions to the Active Manager Paradox is of paramount importance to both the end investor and the active management industry itself. We believe this research can contribute to finding those solutions.
The good news is that active managers are creating real value. The bad news is that value is too often lost before it can be delivered.
Research Design Methodology
This analysis is based on a proprietary database of daily fund positions and portfolio weights constructed and maintained by Turing Technology Associates Inc. The specific funds used in the research dataset include 114 unique US equity mutual funds, from 57 fund families, and represent $1.996 trillion in assets under management (AUM).
Fund Selection Process
The funds selected for use in the research came from the set of mutual funds included within a series of investment portfolios known as Ensemble Active Management (EAM) Portfolios. Turing licenses a series of proprietary technologies to clients to support their creation of such EAM Portfolios. Each EAM Portfolio is typically constructed from a set of 10 to 15 underlying mutual funds with a corresponding industry benchmark. As of early August 2019, Turing had 24 client-designed EAM Portfolios in live production.
All 114 funds used within the study were selected by clients or prospects of Turing related to the design of an EAM Portfolio. Because Turing’s clients selected the underlying funds and corresponding benchmark, the fund selection process maintained independence from the researchers.
Each paired fund and benchmark is a subject of the analysis. Benchmarks included the S&P 500, Russell 1000, Russell 2000, Russell 1000 Value, and Russell 1000 Growth. The time periods used were either January 2014 through July 2019, or January 2016 through July 2019, depending on available data.
Source of Daily Fund Positions
To access daily fund holdings, Turing applied its proprietary fund-replication technology known as the Hercules System. Hercules is a machine learning-based platform processing a multitude of publicly available data, with core concepts behind the approach in use and development for more than a decade. Hercules is not a regression-based approach. Daily estimated positions are generated by the Hercules System, with the out-of-sample portfolios rebalanced every 14 days.
For reference, the Hercules estimated fund holdings and weights for the funds used in this study typically generated a tracking error of less than 1%, and a correlation to the actual fund returns that was greater than 99.7%.
Isolating Manager Conviction
The focus of this research was to analyze the impact of manager conviction in security selection, and thus we embedded two critical design elements into the study. First, securities were categorized and evaluated based on portfolio weights relative to the benchmark. Rather than focus on actual portfolio weights, which are heavily influenced by benchmark weights, the emphasis was placed on a manager’s overweight and underweight decisions and the scale of the over or underweight positions. Second, we divided each fund into multiple, non-overlapping subportfolios determined by the level of Manager Conviction involved, and evaluated their performance separately. Each subportfolio was rebalanced every 14 days and treated as a distinct Model Portfolio. The three subportfolios analyzed were:
- High Conviction Overweights: A subportfolio consisting of the largest overweight positions for stocks in the fund. The subportfolio was selected to cumulatively represent 80% of aggregate portfolio overweights relative to the benchmark.
- Underweights: A subportfolio consisting of the largest underweight positions for stocks in the fund. The subportfolio was selected to cumulatively represent 80% of aggregate portfolio underweights relative to the benchmark.
- Neutral Weights: A subportfolio consisting of overweight securities that are not included in the Overweight subportfolio and underweight positions that are not included in the Underweight subportfolio.
All subportfolios capture distinct choices by a fund manager. The dynamic portfolio weights for each subportfolio are in proportion to the original fund weights, normalized to 100%. Securities outside of the benchmark were excluded as they cannot be properly evaluated in relation to a benchmark. All performance data was calculated both as gross of any fees and after factoring in a hypothetical 85 bps fee. Neither result reflected transaction costs.
The performance data presented represents rolling one-year data (daily step), which was evaluated to capture the percent of rolling periods where each subportfolio was able to outperform the corresponding benchmark (Success Rate), and the average excess (or negative) relative return.
A subportfolio consisting of securities included in the benchmark but not included in the mutual fund (i.e., Zero Weights) was built and analyzed. This fourth subgrouping was not included in the research results because the only way to capture any potential alpha would be through a 100% short portfolio, which is not allowed in a traditional mutual fund. For reference, the Zero Weight portfolio underperformed the benchmark by 78 bps, on average. Unfortunately, even a frictionless short portfolio of Zero Weight securities would not be able to earn the fees of even a standard long-only mutual fund.
If you liked this post, don’t forget to subscribe to the Enterprising Investor.
All posts are the opinion of the author. As such, they should not be construed as investment advice, nor do the opinions expressed necessarily reflect the views of CFA Institute or the author’s employer.
Image credit: ©Getty Images/sompong_tom
Interesting article. So how often and when do you play that star quarterback? Thats the question after considering risk.
Chuck T, according to our research, fund managers need to ensure that the ‘star quarterback’ is playing AT LEAST two-thirds of the ‘game’. And that is not the case today.
Risk is always a consideration, but usually it is in terms of ABSOLUTE risk. A large Beta Anchor does not reduce overall risk, just RELATIVE risk. The latter matters, and has large business implications for the fund firm, but assuming a fund has a market-level risk profile, then they had better reliably deliver performance that can beat the benchmark.
Keep in mind, the industry’s overall relative performance results show that the vast majority of active managers are underperforming. Flow data between passive and active funds/ETFs indicates that the marketplace has pretty much figured this out. Simply put, for quality managers, they need to maximize the percent of the portfolio allocated to High Conviction Overweights or their statistical ability to outperform is severely compromised.
Alexey
Great article!
I was very interested at how does the Hercules fund-replication works but could not find any material on the web, not even their website.
Do you have any sources you can share?
And congratulations on the great article!
Stephano, thank you for the kind words. As our database has been growing over the past year, we were anxiously waiting until it reached sufficient scale to justify this latest research. Hopefully it gains the attraction it deserves.
Regarding how our Hercules System works, we do not provide much detail given the highly proprietary nature of the technology. However, I can share that it is NOT regression based. That is a non-starter on many levels. We instead use Machine Learning-based technology. Also, the public data that would normally be used to support such an effort ranges in quality from highly accurate to quite weak. We therefore emphasize the most accurate and reliable data for our efforts (e.g., NAVs), and have built out a significant amount of QA/Cleanup stages to ensure that weaker quality data (e.g., EDGAR filings) are not compromising the overall accuracy.
Alexey
Hi Alexey,
Interesting study to better understand the general underperformance of active equity funds.
However, what is the use of stating differences in performance for periods less than one market cycle since 2010? Most active managers also do some market timing, which is thus not completely taken into account. I think pure calculus explains their worse performance of -40 bps since 2010, not their actual underperformance.
Norbert, I am glad you found the data interesting. While it can be argued from a purely statistical standpoint that full market cycles are most useful for analysis, practicality means that shorter periods are actually required. First, full market cycles typically take more than a decade to play out. While academically interesting, most investors have an investment decision horizon for active managers that runs 1-2 years. Also, if active managers underperform in rising markets (such as -150bp per year) and then add relative value in bear markets (such as +250bp per year), from a mathematical basis the client still loses.
We have looked at market timing events, and that research is being discussed for future publication. But we can state that on a risk-adjusted basis out of benchmark market timing bets in the sample set we evaluated had no statistical significance.
Finally, we use rolling one-year periods as a means of accessing a stable performance profile for the active managers.
Alexey
This is an absolutely fascinating paper. The takeaways are huge. It explains why most active managers are playing from behind all the time. It also means that they need to make changes. Fast.
Probably makes the Bogleheads out there happy. This research shows that the average fund is charging active fees for only one-half an active portfolio. Or alternately, that the ‘effective’ active portion has a fee hurdle of 2x before it can generate net return for clients. Using Alexey’s data, that is roughly 170bp. Tough hill to climb.
However, if this research ends up being the bread trail that allows active managers to get their house in order, then this research might end up being the key inflection point they need.
Thank you for the thought provoking study. When can we expect more revelations from your database??
Alexey,
Thank you for your reply. It is clear and widely known that clients mostly or always lose with active equity funds. It also makes sense, that if the time period, over which you compare the results of actice equity funds, does not matter statistically. Because their cyclicality is probably rather similar to each other due to their similar benchmark hugging, minimizing their risk of negative tracking error, but also eliminating their chance for any positive alpha for clients after fees. Thus, the time period of the comparison is only irrelevant if the cyclicalites of all compared funds are similar.
However, the cyclicalities of active and passive equity funds is usually rather different from each other as you stated: “…active managers underperform in rising markets (such as -150bp per year) and then add relative value in bear markets (such as +250bp per year)”.
Thus, if you compare active equity funds vs. their passive equity fund benchmark over less than the last market cycle, you systematically bias the result of the comparison against active funds. Because you omit the last bear market phase from 2007-2009, when active funds outperformed their benchmark on average, as shown in the third figure entitled “Actively Managed Large-Blend Mutual Funds vs. The S&P 500”. I.e., you make the average performance of active equity funds appear worse against the benchmark than it actually is.
This is like comparing the average temperatures of certain cities only from Jan. until Sept. and not from Jan. until December. If you are only interested in the difference of the average temperatures, this would ONLY be OK, even though not perfect academically, IF both cities are located on the SAME hemisphere, north or south, with the same cyclicality or rather seasonality!
However, IF they lie on OPPOSITE hemispheres, i.e. north and south like London and Sidney, this is plainly WRONG due to different cyclicalities/seasonalities and would produce biased results. Namely that London has a relatively higher average temperature than Sidney as is the case when compared correctly over one whole year, taking into account all seasons in both cities completely.
As extreme investment example to make this more obvious, just take the (intended) extreme differences in cyclicality of a time series momentum managed futures fund or trendfollower (TF) and a passive equity fund. Let’s assume both of them achieved the same absolute return and risk values over the (assumed) last market cycle, i.e., from the market peak in 2007 until the last market peak in 2019.
If you compared them only over the last ten years from 2009 until today, the TF performance would appear to be much worse than the equity fund by several 100 bps p.a. Because you omitted the phase of the highest gains of TFs but included the phase of the highest gains of the equity funds. Because most TFs achieved them in 2008 due to their highly negative correlation to stocks during drawdowns, their main USP! In turn, this due to going short during pronounced drawdowns, exploiting the downward trend! As you correctly stated, real alpha can only be earned of going short, which TFs ideally do in bear markets.
Thus, such a comparison of funds with extremely different cyclicalities over less than one whole market cycle is just plainly wrong. But it can often be observed even with many professionals for whatever reason!? Not so seldomly this even leads to dumb bashing of well managed TFs, which intend these differences to diversify equity dominated portfolios in the first place! See: awealthofcommonsense.com/2017/04/managed-futures-dealing-with-uncorrelated-assets/
Thus, comparing active and passive equity funds – certainly differing much less in their cyclicality but usually still differing as you mentioned – only during a boom phase, as you have done since 2010, you also make the performance of active equity funds appear worse than they actually are by some ten bps p.a. vs. their passive benchmark.
Thus, I think that your statement “On average, active managers underperformed by -1.7% per calendar year. The results are even worse for the most recent decade. Since 2010, active managers have failed to keep pace with the S&P 500 every year, lagging by -2.1% a year on average.” is misleading. Because most probably the observed decrease in performance is mainly due to the biasing comparison over less than one whole market cycle and not due to further decreasing active fund manager performance or increase of fees. Or what else can explain the apparent decrease in performance of active equity fund managers during the current decade?
Actually, due to the immense pressure from low cost passive equity funds, you would rather expect that they really try hard to reduce their poor performance after fees mainly by lowering fees.
Sorry for the length of my post. I just felt it needs some more details to make my point.
Norbert, our research was limited to actively managed US equity funds, so our findings are all limited to such. Also, the industry data we provided (based on Morningstar data) covers annual performance of active and passive funds going back nearly 30 years. Over that time, there were two full market cycles (1990s leading up to the bear market of 2001 -2002; 2000s leading up to the 2008 bear market), and other pockets of significant market downturns. We believe that this data captures cyclicality of active managers completely.
For the information related to High Conviction Overweights (HCOs), we did not have sufficient data to go back through a full market cycle (i.e., back to 2007), but I will stick with my earlier comment that most investors have an investment decision horizon for active managers that runs 1-2 years, and thus the data we analyzed has significant value to the marketplace even over the time period covered.
Alexey, for sure investors do have such an irrationally short horizon of 1-2 years for their long-term investments. Thus, they may easily get trapped by presuming that changes in investment results during purely upward or downward market phases vs. results of complete market cycles, caused by simple math, are due to changes in actual manager performance.
Referring to your two statements:
““Large-blend active managers have outperformed the S&P 500 in only 5 of the 29 years analyzed. On average, active managers underperformed by –1.7% per calendar year.”
“The results are even worse for the most recent decade. Since 2010, active managers have failed to keep pace with the S&P 500 every year, lagging by –2.1% a year on average.”
These “worse” results are also due to simple math, imho. As this is an educational blog for investment professionals, I just thought it may be relevant for them to know that the actual reason behind these “worse” results is not that their performance is actually becoming worse overall.
These “worse” results can simply be explained by the fact, that during this recent decade we are only in a rather long upward market phase. During such phases, active managers mostly lag their correct index significantly. This is shown in the figure “Actively Managed Large-Blend Mutual Funds vs. the S&P 500” for “Calendar Year Periods 1990-2018” with approximately three complete market cycles.
Thus, over several complete market cycles results of active managers must be “better” than over just one upward market phase since 2010. Simply because they also include a number of downward market phases, corresponding to the number of upward market phases. During downward market phases, active managers usually achieve “better” results vs. their benchmark, thus, partly compensating their usually “worse” performance during upward market phases.
Your further comments to Tony, including the dependency of a quantitative performance breakdown of active managers on exposing their convictions to the market forces and on fees and costs, are very valuable! They help to explain the higher probability of HF managers, particularly CTAs, to achieve outperformance over complete market cycles. Because they operate at a much lower ratio of fees and costs to their market exposure.
Thank you for sharing your insights. I have few remarks/questions:
– the beta anchor portion of all mutual fund should mathematically eliminate the alpha generated by the high conviction portion, in the context of active underperforming passive. That does not appear obvious when we look at the different probabilities of success you provided?
– There is a widespread tendency among PMs to put bigger weights on more defensive stocks ( so called low beta), implying that higher weight does not equal necessarily high conviction but just a risk mitigation behaviour.
– As the low volatility style has outperformed substantially over more than 10 years, this could simply explain the outperformance of the high conviction stocks. That could just mean that there is no sustainable skill out there, just exposure to the right style combined with risk control.
A “55% overweight” to high conviction positions seems like it should be enough to push the active managers total portfolio performance well above any benchmark. What causes the offsetting drag to produce overall underperformance? How does this approach compare to all the “Active Share” work out there?
Tony, let me answer both questions. Regarding the 55% overweight position, some simple math lends insight. Assume that the manager generates 200bp in excess return from the high conviction overweights. At a 55% allocation, that means that the effective excess return is 110bp. Factor in transaction costs (often estimated at 50-65bp, but let’s use 50bp) and total fund fees (assume 85bp), and you get an average annual excess return of -25bp. This manager will win sometimes because this is an average value. But over longer periods the negative profile will result in longer-term underperformance. The manager can always try to improve returns from their best ideas, but candidly they should have been doing that all along. They can also increase the allocation to their best ideas. A 70% allocation translates to an average 5bp positive return. But in most circumstances, our research shows that a 55% allocation to high conviction overweights is a losing hand.
Regarding Active Share, there are a number of similarities. The biggest difference is in the calculation. Active Share treats all deviation from the benchmark as adding to the overall score. And higher Active Share scores are proven to be a prerequisite for outperformance. However, our research shows that the contributions from underweighted securities and from benchmark holdings that are not in the fund do not add to alpha production. We would exclude them. Thus our interpretation of Active Share would just be the portion coming from overweight positions.
Very good. That all sounds very reasonable. I hope the academic community follows up on this and does more research; also, the active management community to see if they can tilt the ledger in their favor going forward.
Very interesting research. Thank you. The underperformance of high-conviction positions may be a function of the behavioral bias to sell one’s winners and lock in gains. There may also be a personal seasoning effect. How many portfolio managers early in their career have the confidence to really ride (i.e. keeping them significantly oversized) winners for a long period of time? At what point does a great company become a bad stock (i.e. overvalued)?
It seems like it pays to keep your winners regardless of their weight in the portfolio, and sell your losers. Quite often many PMs reduce their weights in their best perfoming stocks, and redeploy the cash to their smaller holdings. Part of the reason is that there are constraints in how much one can hold in any particular stock and/or sector. How do you get around that?
Interesting research and the conclusion certainly makes sense.
I would add:
When 5 names comprise over 20% of one’s benchmark, it is hard to overweight those 5 names even if the portfolio manager believes that they will outperform.
Active portfolio managers tend to construct portfolios that roughly “mirror” the underlying index and then make relatively small overweight/underweight bets that do not “move the needle”.
Active portfolio managers often have diversified portfolios with too many names to allow for meaningful outperformance, if there is any outperformance at all. A former colleague used to refer to the practice as “diworsification”. (mispelling is intentional)
I keep hearing active managers proclaim that the time for active management outperformance has arrived.
It appears as if that train is late again.
Re comparisons
If the average investor did 7.13% v the s&p market 10.65% a loss of 3.52% in the 30 yr period to Dec 31, 2021. (I don’t know 30-year result for the average manager )but as per the article the average managed fund failed in a 10-year period compared to the index by about 2.10% does it still not speak well for active managers as a means to help the average investor. Though they underperformed the index they outperformed the average investor by about 40% ( -3.52 v -2.10) Though the index beats the average active manager 60% of the time it also beats the average investor, due to cognitive bias and investment behavior glitch it seems the average investor can’t achieve better than the average active so he should still consider active as part of the portfolio. Another way of saying it is -other than the index what percentage of active managers outperformed the average investor? thanks