This post is available as a PDF download here.

Summary­

  • In order for active management to outperform passive investing, a manager needs both skill and opportunity.
  • Opportunities can come in the form of entering and exiting the market, security selection, and portfolio construction.
  • However, even if these opportunities can be quantified, developing a model that explains active management’s performance versus passive investing is difficult; developing a predictive model is even more challenging.
  • While focusing on systematic strategies with more direct links to these opportunity factors can make the exercise easier, predicting the out/underperformance of an active strategy is still tough.
  • By diversifying management styles and controlling the factors that are within your power – such as avoiding idiosyncratic risk, performing due diligence, and developing an investment plan that can be weathered – you can reap the benefits that both styles have to offer in different market environments.

The debate over the merits of active and passive investing can get heated at times, with the focus often falling on the fee and performance differences.

In our previous commentary entitled It’s Long/Short Portfolios All the Way Down, we discussed a framework to reduce overall fees for active management by barbelling high active share strategies with low-cost passive strategies.1

But the performance discrepancy between the two styles still stands.

Intuitively, we would expect ebbs and flows in performance over the short run. Why? If a particular method of active management always underperformed a passive strategy, then we could simply short the strategy (assuming that the costs of shorting were not prohibitive), giving us a strategy that now always outperforms!

Given this, deciding how much of each style, if any, to hold at a point in time is a difficult question. Do you take Jack Bogle’s approach and go purely passive? Or do you employ some of the Warren Buffett approach and look for attractive investing opportunities or pay someone with (supposed) skill to do it for you?

When Has Active Management Outperformed?

With the growth of low-cost passive investment vehicles, many articles, especially in the current bull market, tout that active investing is a thing of the past.

From a theoretical perspective, William Sharpe explained in his Arithmetic of Active Management2 that collectively, all active investors must equal “the market”. The consequence is that after fees, active managers must collectively underperform the market (i.e. passive investing).

Lasse Pedersen of AQR challenges this conclusion by pointing out that “the market” changes over time as shares are issued and repurchased, indices (e.g. the S&P 500) undergo reconstitutions, active managers can be a subset of all active investors (i.e. those who may trade based on liquidity needs rather than on information), and that few, if any, investors actually own the entire market portfolio.3

We tend to agree with Pedersen.

Focusing on large cap U.S. equity, it is true that active investing has lagged passive in 7 of the past 8 years. But the few decades of data is not enough to assert from a statistical perspective that either active or passive investing should outperform over the long run.

Even if we did have enough empirical evidence to make a call one way or the other, it can take a lot of years to prove that something was likely a false positive (we showed this in the case of factor investing in our recent commentary on the Factor Fimbulwinter4).

The chart below shows the difference between the average net performance of active large-cap U.S. equity funds (Morningstar Large-Blend Category of mutual funds that are not index funds) to the average net performance of the passive S&P 500 index funds. Positive bars indicate that the average return on the active funds was greater than the average return of the passive funds for the year.

Over the past 30 years, active and passive management have traded the lead many times, with an average performance differential of only 8 bps in favor of active. Common estimates of the frequency of the shift between the two styles is every 5 to 10 years.5

A simple frequency analysis of the annual data shows that 3, 4, 7, and 8 years are the dominant frequencies, indicating that our current 8-year period of passive trumping active is in line with the historical experience.

Source: Morningstar. Calculations by Newfound.

But what if there were a way to which predict style – active or passive – would outperform in a given year?

When Might Active Management Outperform?

According to Grignold’s and Kahn’s Fundamental Law of Active Management, excess returns require both investment skill and opportunity.

In the evaluation of fund managers, skill is often the primary focus. However, Mauboussin and Callahan at Credit Suisse highlighted in their 2015 paper that opportunity can be just as important in determining whether active management can deliver excess returns.

In their paper, they explore three areas of opportunity where active managers can potentially generate excess returns. If these areas have limited breadth, then even the most skillful manager’s performance may be dampened.

Without opportunity, all the skill in the world cannot add much value, especially when no investing skill is 100% foolproof.

The three areas of interest in evaluating opportunity are:

  1. Market timing
  2. Security selection
  3. Security weighting (portfolio construction)

For each of the three areas, they describe a metric to quantify the opportunity in a given year. Their main conclusion was to justify why 2014 was a bad year for active management since each of the three metrics was very low within its historical range.

In hindsight, it is generally simple to propose a reason why something happened. Whether that reason is the correct explanation for an observation requires good data, sound logic, and a degree of trust in the comprehensiveness of the analysis. But even the most accurate explanatory model may be of no use if it does not lead to a predictive model.

In this commentary, we will explore the metrics used in the paper to assess the feasibility of predicting the performance trends of active and passive investing.

The Metrics

The metrics used in the Credit Suisse paper are not unique. Others could be chosen, but they are simple enough to understand and distill each of the three areas of active portfolio management down to a single number that can be compared across time frames.

Market Timing

Investors are notoriously bad at timing the market. This also holds for professional investors. However, in a strongly trending market with little volatility, timing the market using a systematic approach like momentum should be more effective than when the market is moving sideways or trending but volatile.

We quantify the opportunity for market timing using the average directional movement index (ADX).6 The metric describes the strength of a trend using a number between 0 and 100, with numbers below 20 indicating no trend and those above 40 indicating a strong trend, either positive or negative.

Security Selection

Security selection can add more value when the range of returns over the investment universe is wide. Having returns all clustered around a single value leaves little room for adding value by selecting securities that outperform (or shorting securities that underperform).

We measure the opportunity for security selection with dispersion. It is calculated by taking the average returns of the top half and bottom half of securities and looking at the differences between these two averages.

Portfolio Construction

Once the securities are selected, how to allocate capital among them is the remaining component. For this, we will look at the dispersion of dispersion, which is essentially just the dispersion calculated for the top performing half of securities and the bottom performing half of securities. These stats assess the long-side and short-side portfolio construction opportunities, respectively.

The Data

To calculate some concrete numbers for these statistics we will utilize the market returns and 49 industry group returns from the Kenneth French Data Library. The start date for the sectors is July 1969 when we have returns for all 49 industry groups. For the market returns, we will utilize the full data set back to 1926.

Market Timing – Trendiness

Below we plot the ADX using the Kenneth French market returns from 1927 to 2017 along with the critical values that indicate whether the trend is strong, weak, or absent.

If you are familiar with the ADX indicator, you may notice that this chart looks less smooth than is typical. We have left off the final step in smoothing in order to have a timelier measure of momentum. Using monthly data, smoothing over 12 months twice adds significant lag into the data.

Source: Kenneth French Data Library. Data from 1927 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

We can see the qualitative aspects a bit better by coloring each year based on which trend region it fell into.

Trendiness by Year in U.S. Equities

Source: Kenneth French Data Library. Data from 1927 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

Years with prolonged drawdowns and run-ups (e.g. recently 2008, 2013, and 2017), have all exhibited high values of trendiness.

But while it may seem like there is possibly a predictive pattern that emerges, from a statistical point of view, the one-year lagged correlation of the ADX is only -0.12. Grouping the ADX into the three regions (divided by the values 20 and 40), and denoting each trend regime according to the signals 1 for strong, 0 for weak, and -1 for no trend leads to essentially zero correlation from year to year.

This lack of strong relationship is also evident if we look at the conditional probabilities of each state. Regardless of the current trend – or absence of one – it was most likely followed by a strong trend. The unconditional probabilities of seeing the different trends were 59% for strong, 21% for weak, and 20% for no trend. Because the unconditional probabilities of seeing the trends were very close to all the conditional probabilities, there does not appear to be much predictive power from year to year.

Transition Table of Year to Year Trend Strength in U.S. Equities 

Source: Kenneth French Data Library. Data from 1927 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

Perhaps including the other variables will be beneficial.

Security Selection – Dispersion

Using the data for the 49 industries in the Kenneth French data library, we can see how the yearly dispersion changes over time.

From 1970 to 2017 the difference between the top half and bottom half of industry returns has ranged from 17% to 51% with a median value of 25%. The Tech Bubble years and coming out of the Financial Crisis in 2009 were the recent years with the largest dispersion, which should have led to opportunities for active managers to differentiate themselves through their security selection.

2017 was a lower year at 19.7%, and 8 out of the last 10 have been below the average over the whole period.

Source: Kenneth French Data Library. Data from 1970 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

The lagged correlation in the dispersion numbers is only 0.17, so there is not a strong predictive relationship of dispersion alone. However, since we care most about predicting how active management will fare, we will look at the final piece of the model that proxies portfolio construction.

Portfolio Construction – Dispersion of Dispersion

Zeroing in on the individual halves by performance of the industries in each year, we can see the potential for adding value through over or underweighting within a portfolio.

Source: Kenneth French Data Library. Data from 1970 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

The dispersion of dispersion for the top performing half of industries ranges from 10% to 45% with a median value of 17% while the dispersion of dispersion for the bottom performing half ranges from 8% to 30% with a median of 15%. We would expect the bottom half of industries to exhibit lower numbers since losses are capped at 100%.

We see years like 1999 when the dispersion of dispersion in the top half was significantly higher. There are also a few times – like in 2013 – when the bottom half shows more dispersion than the top.

2017 was a low reading for both numbers, solidifying what we anecdotally knew already: even though the market was trending, many things in the market were trending together.

The correlations of 0.26 for the top half dispersion of dispersion and 0.08 for the bottom half again point to weak relationships in readings from year to year.

The Explanatory Model

Now that we have the framework, let’s see how well, these factors can explain active management’s average performance relative to passive for the large-cap U.S. equity universe from 1985-2017.

Explanatory Model for Active Management Outperformance – Large-Cap U.S. Equity

Source: Kenneth French Data Library, Morningstar. Data from 1985 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

Unfortunately, this model stinks. None of the coefficients of the trendiness, dispersion, or dispersion of dispersion are significant at most commonly accepted levels. The adjusted R-squared for the model is a paltry 0.02.

The Predictive Model

When trendiness, dispersion, and dispersion of dispersion – factors that should affect how well active managers can perform – can’t even explain performance in hindsight, we do not have much faith in their ability to predict performance in the future. For this to be the case, there would likely have to be a significant relationship between the factors themselves (i.e. autocorrelation), and we saw in previous sections that this was not the case.

Indeed, the regression results on the relative performance of active management lagged by one period confirm that making a prediction from a model like this would essentially be a flip of the coin.

Predictive Model for Active Management Outperformance – Large-Cap U.S. Equity

Source: Kenneth French Data Library, Morningstar. Data from 1985 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

So what’s going on?

Well, it could be a number of things:

  1. We may not be quantifying the metrics in the correct way.
  2. We may be missing metrics that have more explanatory power for how active management will perform.
  3. Our data for the metrics may not be representative enough of the managers’ investment universes.
  4. Managers may not actually be tapping into the sources of return that they are supposed to be (or that we think they should be) benefitting from.

It is likely a combination of these.

The term “active” is very broad when it comes to investing. It can encompass everything from rebalancing to systematic allocation strategies to idiosyncratic stock picking.

If we set the broad, aggregate active management data aside and constrain our analysis to a more systematic strategy, perhaps the model will have some use.

Modeling a Systematic Active Strategy

For our systematic active strategy, we will utilize momentum and trend following. Other factors such as quality, value, and volatility could certainly be used.

In effort to avoid overfitting our results to any one particular model or parameterization of trend following, we have constructed our signals employing a model-of-models approach.7

We will utilize the same data covering 49 industries from before, and we will actually construct two strategies: a fully invested equity strategy that looks at the relative momentum of the industries and an equity strategy that takes the first one and applies a trend following overlay to tilt the portfolio between 50% and 150% market exposure.

For relative momentum, we use four different definitions for a given N-period lookback:

  • Time Series Momentum: Simply the return over the past N-periods.
  • Risk-adjusted Time Series Momentum: The return over the last N-periods divided by the realized volatility over the time period.
  • EWMA Time Series Momentum: The (N/2)-length exponentially-weighted moving average return.
  • Ordinary Least-Squares (OLS) Momentum: The slope of the best-fit line to the price series over the N-periods.

For each of these four models, we also run a number of parameterizations covering 3-to-18-month lookbacks. In grand total, there are 4 models with 16 parameterizations each, giving us 64 variations of momentum signals.

For each model, we rank the industries by their momentum. We then take to the top 12 (one-quarter) industries and equally weight them for the portfolio. We tranche the portfolio over a 21-day period to account for timing luck.

The trend-based variant of this strategy is constructed using a similar philosophy.

We use four different definitions for trend for a given N-period lookback:

  • Time Series Momentum: When the return over the prior N-periods is positive, invest. Otherwise, divest.
  • Price-Minus-Moving-Average: When price is above its N-period simple moving average, invest. Otherwise, divest.
  • EWMA Cross-Over: When the (N/4)-length exponentially-weighted moving average is above the (N/2)-length exponentially-weighted moving average, invest. Otherwise, divest.
  • Ordinary Least-Squares (OLS) Trend: When the slope of the best-fit line to the price series over the prior N-periods is positively sloped, invest. Otherwise, divest.

The average trend signals across all 64 models/parameterizations are tranched over a 21-day period. The exposure to the relative momentum portfolio is scaled between 50% and 150% as the average trend signal ranges from 0 to 1. Cash returns and borrowing costs are both assumed to be the risk-free rate.

We can run the same model analysis as before over the same time period using the market return as the passive benchmark.

The results for the explanatory model are much more intuitive for both strategies.

Explanatory Model for Relative Momentum Model Outperformance vs. Passive 

Source: Kenneth French Data Library. Data from 1985 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

For the relative momentum strategy, the loading on dispersion and dispersion of dispersion (DoD) for the top performers is significant. The loading for DoD Top is likely negative since we are simply equally weighting the holdings. The important thing for this strategy was getting them into the portfolio.

Explanatory Model for Relative Momentum Model w/ Trend Overlay Outperformance vs. Passive

Source: Kenneth French Data Library. Data from 1985 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

Adding on the trend component boosts the significance of the loading on the trendiness factor. The loadings on dispersion and DoD Top are not as significant as with the relative momentum model, but they are still much better than with the aggregate active management data.

So where does that leave us for predicting the efficacy of this strategy versus passive going forward?

Unfortunately, not much better off.

Predictive Model for Relative Momentum Model Outperformance vs. PassiveSource: Kenneth French Data Library. Data from 1985 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

Predictive Model for Relative Momentum Model w/ Trend Overlay Outperformance vs. Passive

Source: Kenneth French Data Library. Data from 1985 – 2017. Calculations by Newfound Research. Returns are gross of all fees, including transaction fees, taxes, and any management fees. Returns assume the reinvestment of all distributions.  This document does not reflect the actual performance results of any Newfound investment strategy or index.  All returns are backtested and hypothetical.  Past performance is not a guarantee of future results.

The significance of the loadings in the predictive models generally worsened, and the parameters gauging the model fit declined.

One interesting thing to note is that the signs of the dispersion and DoD Top loadings flipped, indicating that high current dispersion tends to lead to lower active outperformance of these strategies, or vice versa. This is likely an indication of mean reversionary effects when dispersion is high and accelerating momentum effects when dispersion is low.

Conclusion

Reports written over the past 3 years have “predicted” that active management would make a comeback. While it has not lagged terribly over that period, the hoped-for outperformance, like that seen after the Tech Bubble or Financial Crisis, has not been repeated.

The metrics we examined in this commentary may makes sense from an intuitive level to explain why active investing has beaten passive investing in certain years. But with low statistical explanatory power – let alone predictive power – for these factors, it is difficult to predict which style will prevail going forward.

If you want to rely on intuitively appealing, quantitative factors to assess or predict active management’s outperformance, then you had better make sure that the active management is actually striving to align with your model. Using a model to predict the outperformance of even a systematic strategy that explicitly aims to benefit from trendiness and dispersion may not be a wise endeavor.

Our recommendation is to diversify and control the factors that are within your power.

Avoid idiosyncratic risks by focusing on systematic active investing (e.g. using well-documented factors). Do thorough due diligence on the process for selecting securities and on the portfolio construction. Above all, develop a plan you can adhere to. The grass will always be greener at some point along the journey, and switching from active to passive or vice versa based on a whim can lead to performance that is worse than either on its own.

Which style will be the best over the next 1, 3, 5, or 10 years?

Only time will tell.

  1. https://blog.thinknewfound.com/2017/11/longshort-portfolios-all-the-way-down/
  2. https://www.cfapubs.org/doi/pdf/10.2469/faj.v47.n1.7
  3. Sharpening the Arithmetic of Active Management: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2849071
  4. https://blog.thinknewfound.com/2018/06/factor-fimbulwinter/
  5. https://www.institutionalinvestor.com/article/b157ccvh5rs4s2/what-the-active-vs-passive-debate-is-really-about
  6. In a previous research piece, we used the MDI (market divergence indicator) to measure trend strength and analyze the performance of trend-following strategies based on the value of this indicator. See https://blog.thinknewfound.com/2016/06/tactical-trend-following-core-alternative/
  7. Nothing in this commentary reflects an actual investment strategy or model managed by Newfound and any investment strategies or investment approaches reflected herein are constructed solely for purposes of analyzing and evaluating the topics herein.

Nathan is a Vice President at Newfound Research, a quantitative asset manager offering a suite of separately managed accounts and mutual funds. At Newfound, Nathan is responsible for investment research, strategy development, and supporting the portfolio management team.

Prior to joining Newfound, he was a chemical engineer at URS, a global engineering firm in the oil, natural gas, and biofuels industry where he was responsible for process simulation development, project economic analysis, and the creation of in-house software.

Nathan holds a Master of Science in Computational Finance from Carnegie Mellon University and graduated summa cum laude from Case Western Reserve University with a Bachelor of Science in Chemical Engineering and a minor in Mathematics.