16

Active Versus Passive and Something About the “Pharma Bro”

We can’t measure directly if the market is efficient from a capital allocation perspective. But indirectly, we can study whether abnormal profits are possible.

—JPP

NOW THAT WE HAVE DISCUSSED RISK FACTORS AND THE role of alternative assets, we must discuss one more important question on portfolio construction: Should we allocate to active or passive building blocks?

In recent years, index strategies have gained market share over active managers. The common narrative is that on average, active managers don’t outperform index-tracking strategies. In my view, this narrative is flawed. Not every active manager is “average.” Skilled active managers—with a repeatable and disciplined process, significant resources, the right culture, reasonable fees, and a long time horizon—have delivered excess returns to investors over decades and, ultimately, have contributed to better retirement outcomes. There are many examples of skilled active managers, so let’s use a hypothetical example (which will please my legal department because it simplifies the compliance requirements for this book), albeit one that’s not divorced from reality.

Suppose that through active security selection (via fundamental research), tactical asset allocation (as described earlier in this book), and structural portfolio design advantages (also based on many of the concepts we reviewed in this book), an actively managed target-date fund has outperformed a target- date fund that allocates to index-based strategies by 82 bps per year since inception, after fees. Is that a meaningful difference?

Let us assume that twin sisters Mary and Lucy have the exact same savings patterns but at two different employers. Mary’s employer uses the active target-date fund as the default investment option, while Lucy’s employer uses the index-based option. Suppose the sisters each contribute $12,000 per year toward their retirement. Let’s assume that the index-based strategy returns 6% per year (after fees), compared with 6.82% (after fees) for the actively managed strategy, for a difference of 82 bps.

There are many ways to run this simulation. For example, we can include stochastic asset returns, salary growth, employer match, inflation, etc. But for illustration purposes, we can get a reasonable estimate with the “FV” function in Excel. This function calculates the future value of a stream of payment for a given rate of return and time horizon. We find that over her entire career (40 years, from age 25 to age 65) Mary accumulates a retirement balance of $2,287,248 with the active strategy (isn’t compounding great?), compared with $1,857,243 for Lucy’s index-based target-date fund.

In this example, the total active management advantage is $430,104. If both employers have 5,000 employees, or “plan participants,” the decision of the plan sponsor at Mary’s company to rely on a skilled active manager has created, over time, a staggering total of over $2.15 billion (5,000 × $430,104) in extra retirement security for the employees at Mary’s company compared with the choice of a passive strategy. That’s meaningful.

Though inspired by actual results, this example is “cherry-picked,” as participants would be invested in different vintages and experience different entry and exit dates. Also, most individuals contribute less than $12,000 per year. Selecting an active manager involves the risk of underperformance. For every buyer of a security, there’s a seller. For every seller, there’s a buyer. In that context, it’s hard to imagine how average active performance could be much higher than . . . average. But again, not every active manager is average. Like almost everything in investment management, the active versus passive decision boils down to a risk versus return decision. Investors who want to take active risk can be rewarded for it.

In the end, though, financial markets have room for both active and passive investment management. Investors who prefer to minimize fees, even at the cost of possibly higher after-fee returns, should choose passive strategies.

The Revenge of the Stock Pickers

A recent trend popular with asset owners globally has been to invest in blend strategies, which optimize the asset mix between passive and active building blocks. Blend portfolios give investors exposure to active management within a specified fee budget. With massive, bubble-like inflows into index strategies in recent years, there’s an argument to be made that even fee- sensitive investors can benefit from some exposure to active management. In the 2018 article “The Revenge of the Stock Pickers,” my colleagues Hailey Lynch, Rob Panariello, Jim Tzitzouris, David Giroux, and I provide empirical evidence that the more assets flow into index strategies, the easier it gets for active stock pickers to outperform the index.

In that context, before we review our portfolios, let us take a short detour into the world of stock picking. This detour will help us better understand the active versus passive decision. My coauthors and I show that index investors distort security-level pricing, which creates opportunities for active managers, especially stock pickers. We focus on the stock-picking opportunities created by the popularity of exchange-traded funds (ETFs).

ETFs now account for 30% of all trading volume on US exchanges, up from less than 2% in 2000.¹ This trend may have created opportunities for stock pickers. When an ETF trades heavily around a theme, correlations among its constituents increase significantly. Even some securities that have little or negative exposure to the theme itself begin to trade in lockstep with other ETF constituents. In other words, because ETF investors are blind to security-level information, they often “throw the baby out with the bathwater.”

The events of 2020 have provided additional evidence of this effect. As the prices of individual stocks get dragged up or down with ETFs, mispricings can become significant, and the profits realized by taking advantage of them may represent one of the hidden costs of ETF investing. In a 2017 editorial, Financial Analysts Journal coeditor Daniel Giamouridis calls for more research on this topic. He refers to “higher trade commonality in ETF constituent stocks (in down markets), increased commonality in their liquidity/market impact, and less idiosyncratic risk compared with nonconstituent stocks.” He emphasizes that future research should clarify how volatility and correlations change as well as the likelihood of price deviations from fundamentals (and reversions).

To answer this call for research and, importantly, to estimate the size of this opportunity for stock pickers, we have designed a simple, contrarian trading strategy that buys oversold constituents when an ETF sells off in a high-volume panic. We focus on the downside because, as we discussed in Chapter 9, investors are less rational when faced with losses than when faced with gains; i.e., diversification fails much more often on the downside than on the upside. Hence, extreme downside correlations are almost always higher than upside correlations. As Rob Panariello and I (2018) argued, in financial markets, fear is more contagious than optimism.

We identify oversold constituents by their beta to their ETF. We use nine sector ETFs because they are more susceptible to speculative, retail-oriented trading than broad index ETFs. We also use an S&P 500 ETF and a small cap ETF.

Giamouridis specifically calls for research to cover “not only stocks in broad market indexes that are ETF constituents but also specific segments of the equity market,” such as sectors.

We don’t suggest that anyone implement this strategy without fundamental oversight, but our results are striking: when high-volume sell-offs occur, ETF investors may be leaving as much as 200–300 bps of alpha on the table for stock pickers to capture over the following 40 days. Across ETFs, such events occur an average of 30 times per year, for a total of 240 events throughout our study period (January 4, 2010 to December 29, 2017).

This strategy doesn’t require any stock-picking skills other than the ability to measure a stock’s beta to its ETF. We suspect stock pickers can capture even more alpha from ETF investors. They can carefully analyze why the ETF is selling off and whether certain constituents are simply being dragged down with it for no good fundamental reason. Our goal is merely to estimate the size of the opportunity, because it’s impossible to backtest a discretionary, fundamental approach.

Cahan, Bai, and Yang (2018) suggest that most ETF investors ignore the fundamentals of the underlying constituents.² The authors refer to the “arbitrage opportunity” that arises when “the short-term trading activity in an ETF is inconsistent with the real-world fundamentals of the underlying stocks.” They use the term “arbitrage” in an informal way, not in the academic sense of riskless profit. But they show that investors can generate alpha if they select ETFs based on constituent fundamentals.

They find that sector ETFs are the most disconnected from fundamentals, but the effect is also present for broad market and smart beta ETFs.

Although we reach similar conclusions, our approach is different in that we pick stocks (i.e., we look for mispricings within ETF constituents), whereas Cahan et al. pick ETFs (they look for mispricing across ETFs, based on stock-level analysis).

Similarly, others have documented evidence of the effects of indexing on security-level comovements. When a stock is added to an index, its correlation with its peer index constituents immediately increases.³

As a corollary, Xiong and Sullivan (2012) argue that in general, index investing contributes to systematic equity market risk. Regarding ETFs specifically, Da and Shive (2017) show that the higher the turnover on an ETF, the higher the correlation between its constituents. They conclude that these comovements are excessive—that is, not driven entirely by fundamentals.

Note, however, that such research does not mean indexing is bad for markets per se. Wurgler (2010), for example, says that “for the sake of balance, it is important to start by acknowledging the many considerable benefits that indexes and index-linked investment products provide.” Similarly, Hill (2016) explains that the natural tension between macro investors, who trade ETFs and other index products in response to dynamic market conditions, and fundamental investors, who take the long-term view, is healthy for financial markets: “Each type of investor depends on the presence of the other to provide liquidity and to drive prices to appropriate levels.”

Brown, Davies, and Ringgenberg (2018) approach the issue from a different angle. Unlike Cahan et al.’s loose definition of “arbitrage,” they study the true arbitrage between an ETF’s price and its net asset value. [To take advantage of the discount or premium, arbitrageurs simultaneously sell (buy) the ETF and buy (sell) the underlying securities.] Their dataset provides a unique and transparent view of arbitrage activities. They show that an increase in ETF arbitrage activity signals nonfundamental demand shocks (perhaps because of sentiment, or “thematic,” trading). In turn, these shocks appear to predict subsequent return reversals at the one-month horizon for both ETFs and their constituents.

This wide body of research all points to the same conclusion: index/passive investing may cause mispricings and abnormal correlations (or “correlation bubbles”).⁴ Surprisingly, Madhavan and Morillo (2018) arrive at the opposite conclusion. They use a factor model to analyze what drives correlations over time and find that macro factors are more important than the increase in ETF assets in driving cross-stock correlations higher. One of their key arguments is that “although cross-stock correlations rose in the period when ETF assets increased, they are not at unprecedented levels relative to the past, well before the rise of passive indexing.” But as our research shows, averages can be misleading. If we isolate days with high ETF volume, the picture is quite different and supports the mainstream conclusion that indexing causes correlation abnormalities.

Also in the skeptical camp is an earlier study that supports Madhavan and Morillo’s (2018) critique. Glosten, Nallareddy, and Zou (2016) suggest that jumps in cross-constituent correlations could be explained by macro shocks or, more generally, systematic fundamental information. In this case, some illiquid ETF constituents may even benefit from ETF trading volume because they become more efficiently priced (i.e., they react more promptly to macro fundamental news). But the authors reach mixed conclusions. They find that systematic price discovery only partially explains ETF activity.⁵

If ETF volumes improve the pricing of systematic shocks but don’t distort the pricing of nonsystematic information, we shouldn’t observe predictable reversals, such as those reported by Brown et al. (2018). Moreover, Ben-David, Franzoni, and Moussawi (2018) observe that ETFs attract “high-frequency demand” and, based on observed reversals, confirm that “demand shocks in the ETF market translate into non-fundamental price changes for the underlying securities.”

To build on this body of research, we posit that the main reason for the distortions and reversals is that some ETF constituents aren’t exposed to macro shocks in the same way—or to the same extent—as their peers. We call these constituents “outsiders.” We recognize that the list of outsiders can change as a function of the nature of the macro shock. Ultimately, the more different constituents are from one another, the more opportunities there are for distortions.

We show that these abnormalities present an alpha opportunity for stock pickers who can distinguish between systematic shocks and ETF-driven price distortions. We suggest a practical shortcut: focus on the behavior of outsider constituents around significant jumps in ETF volumes. This approach is different from everything else we have found in the literature. For example, Brown et al. (2018) sort stocks based on ETF-driven volume, without considering whether a given stock is an outsider or not.

Ultimately, while we recognize the role of index products in financial markets, we conclude that stock pickers may be able to “pick off” the rising number of ETF investors if they can answer two simple questions: Why is the ETF selling off, and should this constituent be selling off with it? The implication for asset allocators is that they should allocate at least part of their portfolios to skilled active managers if they want to hedge against bubble-like situations in index-based products. This research supports the idea of active-passive diversification. It also provides a warning sign to avoid a recent trend (a fad?) toward niche ETFs.

A Case Study: Pharmaceuticals, Hillary’s Tweet, and the Valeant Subpoena

As an illustration, consider the behavior of US healthcare and pharmaceutical stocks in September 2015. Between September 18 and 28, the Health Care Select Sector SPDR ETF (XLV) plummeted by –10.7%, compared with –5.4% for the S&P 500. Volume on the ETF over these seven trading days jumped to its 99th percentile, whereas volume on the S&P 500 remained in its 33rd percentile.⁶

Two important events appear to have driven most of the sell-off in healthcare stocks. First, on September 21, Hillary Clinton tweeted that she would unveil a plan to curtail “price gouging” by pharmaceutical companies.⁷ (The day before, the New York Times had published an article on how Turing Pharmaceuticals had just increased the price of a lifesaving drug from $13.50 to $750.00.⁸ The media nicknamed Martin Shkreli, Turing’s young and brash CEO, the “pharma bro.”) Second, on September 28, Democrats in the US House of Representatives asked to subpoena Valeant Pharmaceuticals for documents on drug price increases.⁹ XLV volume on that day reached an all-time high.

Both events threatened to put pressure on pharmaceuticals-sector revenues, but not necessarily those of other healthcare stocks. Although some companies were directly in the line of fire, it’s hard to imagine how regulation aimed at human drug pricing would affect companies that make animal medicines and vaccines, such as Zoetis, or medical equipment, such as Baxter International.¹⁰ Yet all XLV constituents—without exception—sold off over these seven trading days.

Pharmaceuticals contribute a significant percentage to XLV’s total volatility. Such high-beta stocks tend to be at the center of most high-volume thematic sell-offs in this ETF. In contrast, stocks with a low beta to XLV are often unaffected by the theme behind the sell-off, at least from a fundamental perspective. Nevertheless, they get dragged along, like the baby thrown out with the bathwater. Hence, an easy way to identify outsiders within a list of ETF constituents is to look for stocks that have a low beta to their ETF.

We would have identified five “outsiders” stocks on September 28, 2015, if we had ranked XLV’s constituents by their ETF beta and selected the bottom 10%. These stocks should not have been affected by the drug pricing controversy. These companies sell such products and services as dental equipment, pet supplies, and lab tests. Yet they sold off on the political posturing around drug pricing. Because of an increase in constituent correlations—which is common when ETF volumes spike—they sold off more than expected based on their ETF betas. An equally weighted portfolio of these five stocks returned –8.3% during the seven-day sell-off compared with an ETF beta-implied return of –6.1%.

This overreaction created an opportunity for stock pickers. Suppose an investor had bought the five outsider stocks (equal weights) at the end of the sell-off and levered the portfolio to an ETF beta of 1.0 (we levered the portfolio to calculate alpha versus the ETF). Over the next 40 days, the investor would have outperformed XLV by +4.2% after transaction and borrowing costs.¹¹

Another Case Study: Financials, the Impact of Interest Rates, and REITs

On February 11, 2016, then US Federal Reserve Chair Janet Yellen concluded her semiannual testimony to Congress with an indication that the Fed was not in a rush to raise rates. “Financial conditions in the United States have recently become less supportive of growth,”¹² she said, adding that negative rates were “not off the table.”¹³ These comments hurt the Financial Select Sector SPDR ETF (XLF) because financials tend to benefit from rising rates. For example, when rates rise, banks can lend at a rate that is higher than their overnight borrowing costs and thereby increase net interest revenues.

From February 4 to 11, 2016, XLF returned –6.6%. Although higher trading volumes on this ETF had been recorded around the financial crisis, its volume for those six trading days in February 2016 was in the 91st percentile for all six-day periods over the previous five years. Volume on the S&P 500 was also elevated relative to the previous five years. It was in the 94th percentile, which reflected the systemic importance of monetary policy and, presumably, Yellen’s comments on weaker economic growth.

However, marketwide selling was not as intense as in financials: the S&P 500 returned –4.4%. What happened to the outsiders within XLF? Among the eight stocks with the lowest beta to XLF (the bottom 10%), seven were REITs and the eighth was American Express. Unlike banks, REITs tend to trade as positive duration assets. Real estate assets are almost always valued based on discounted cash flow models. Cash flows (i.e., rents) are fairly predictable. When rates go down, the value of real estate assets goes up; when rates go up, their value goes down (i.e., these assets behave like bonds).

As for American Express, the company’s 2015 annual report explains that its revenues have positive duration: “Amex is negatively exposed to interest rates.”¹⁴ According to the American Express Company (2015), “The detrimental effect on our annual net interest income of a hypothetical, immediate 100 basis point increase in interest rates would be approximately $216 million.”

Therefore, as the market suddenly had to digest the possibility of lower rates, REITs and American Express should have performed better than other financials. In fact, because the growth shock was downplayed (Yellen said that despite weaker expectations, it would not “be fair to jump to any conclusion about the state of the economy”),¹⁵ perhaps they should have rallied. Treasuries were up, for example.

But an equally weighted portfolio of the eight outsiders returned –8.5% during the six days leading up to and including the end of Yellen’s testimony on February 11. We surmised that REITs and American Express were oversold because of the spike in ETF trading volume, which led to indiscriminate selling across financials. As in our first case study on healthcare stocks, if a stock picker had bought the outsiders (equally weighted portfolio) after the sell-off, levered them to an ETF beta of 1.0, and held the portfolio for 40 days, she would have outperformed the ETF significantly—in this case, by +20.0% after transaction and borrowing costs. It is worth noting that later that year, REITs were spun off from financials and reclassified as a separate sector.

Such ETF-driven stock-picking opportunities appear to be pervasive. Beyond our two case studies, there are correlation bubbles everywhere. For example, suppose a company’s earnings disappoint. Investors may use an ETF to sell exposure to the entire sector, even though from a fundamental perspective, several competitors should not be affected (and perhaps some should benefit from a gain in market share). Macro factors also seem to matter. For example, a drop in oil prices may lead to a sell-off in an energy-sector ETF, dragging down companies that may have little or negative exposure to oil. Emerging markets ETFs may also sell off with oil prices, even though some markets and companies within the emerging markets index are net importers. And so on.

The challenge for stock pickers is twofold. First, they must look for situations when an ETF sells off with very high volume, based on a specific theme. Second, they must identify the outsiders—the oversold companies that should not be affected by the theme from a fundamental perspective. The good news is that simple filters may work quite well: we found that most spikes in ETF trading volume lead to abnormal correlations, and low ETF betas appear to be a good way to identify outsiders.

These correlation abnormalities can create a plethora of buying opportunities at the security level. To illustrate, we backtest a simple systematic strategy. For each volume spike accompanied by a negative return, we systematically buy the outsiders and hold them for 40 days. We do so across all 11 ETFs in our sample and across time. We identify outsiders the same way we did in our pharmaceutical and financials case studies: we rank constituents by ETF beta and build an equally weighted portfolio of the bottom 10%. Then, to calculate alpha versus the ETF, we lever the portfolio to an ETF beta of 1.0. Essentially, we replicate our case studies but on a much bigger scale, across a total of 240 volume spikes. All our data are out-of-sample, based on what would have been available at the time.

We calculate the average cumulative alpha (the return for the levered outsider portfolio minus the ETF) across all events, from 1 to 40 days after the volume spike and before and after trading costs.¹⁶ Average alpha on the first post-spike day is slightly negative, which indicates that even if we lagged the implementation time by one day, the strategy would still work. Then, as the time window expands, average alpha cumulates positively and consistently—all the way to 40 days.

The strategy does not work perfectly for all ETFs or at all time horizons, but on average, it generates significant after-cost alpha.¹⁷ Because we force the outsider portfolios’ ETF betas to 1.0, the strategy is not expected to take on any systemic factor exposure—such as market beta, value, or momentum exposures—relative to the ETF. Because we measure performance relative to the ETF, we expect these alphas to be “idiosyncratic” (i.e., stock-picking alphas).¹⁸

Notably, the strategy does not work well for the Materials Select Sector SPDR ETF (XLB), and although it works in the short term, it ends in negative territory for the Technology Select Sector SPDR ETF (XLK). These outcomes highlight the risk of systematic, simple trading rules. In these cases, the trading rules lead to large positions in low-ETF-beta stocks that underperformed their ETFs after the high-volume sell-offs.

Perhaps fundamental analysis would have helped. A stock picker would have analyzed the theme behind each sell-off. She would have taken into consideration whether the low-ETF-beta outsiders were truly outsiders to the theme and, if so, whether these companies presented a risk of short-term underperformance for other reasons. Then she would have scaled the positions relative to the theme according to a risk-return analysis. Once the long positions were established, she would have applied discipline to determine when to sell them, considering market developments and the health of the sector and the companies involved.

Last, although the strategy identifies ETF volume spikes and conditions them on down days, it does not condition on the size or duration of the sell-off. Focusing on the largest sell-offs, with a flexible time horizon, might have enhanced performance. Volumes across ETFs and index funds also need to be monitored, of course, because several index products may trade in the same sector. Ultimately, a lot more can be done when the strategy incorporates fundamental analysis. Hence, our goal is to indicate the potential size of the opportunity, not to design a purely systematic approach.

Takeaways

Are ETF investors increasingly at risk of getting “picked off”? Because of the growing popularity—as well as the liquidity and tax benefits—of passive investing, the percentage of trading volume on US exchanges from ETFs has increased significantly. Some ETF investors focus on top-down market views or themes, whereas others believe that markets are efficient and simply want broad index exposures. In all cases, when they trade, most ETF investors—and index investors in general—ignore security-level fundamentals. They simply buy or sell all securities in the index in proportions determined by the index provider (typically, market capitalization weights).

As a result, we find that when ETF volumes spike, correlations among constituents increase to levels that are not justified by company-level fundamentals. Our study of 240 events between 2010 and 2017, compiled across 11 ETFs, suggested that these correlation bubbles may create opportunities for stock pickers. Investors who buy oversold constituents after high-ETF-volume days and hold them as they mean-revert over the next 5 to 40 days may generate alpha at the expense of index investors. Are we witnessing the revenge of the stock pickers?

My view is that there’s a place for both passive and active management in financial markets. Passive investors and stock pickers can happily coexist. We report gains for stock pickers that are of practical significance, but these results don’t mean ETFs are “bad.” They simply mean that different investors can make markets more liquid and efficient together. Market efficiency remains a paradox: profit opportunities, such as the one we have identified (and which indicates inefficiencies), are necessary to make markets more efficient. Such are the ebbs and flows of financial market equilibrium.

Notes

1.   Robin Wigglesworth, “ETFs Are Eating the U.S. Stock Market,” Financial Times (January 24, 2017). Volume data are from Credit Suisse, as of 2016. In 2016, seven of the ten most traded securities were ETFs, not stocks. The Wall Street Journal reports that the ETF industry has grown to $3.5 trillion in size: Asjylyn Loder, “Investors Win from ETF Price War,” Wall Street Journal (July 12, 2018), www.wsj.com/articles/etf-fees-tumble-as-price-war-heats-up-among-big-fund-firms-1531396800.

2.   In a related article, Chao, Shah, Finelli, et al. (2018) showed that a contrarian strategy that buys stocks with high ETF outflows and sells stocks with high ETF inflows generates substantial profits.

3.   See Wurgler (2010), Barberis, Shleifer, and Wurgler (2005), and Greenwood and Sosner (2007).

4.   For an extreme example, the case of the VanEck Vectors Junior Gold Miners ETF is interesting. See Asjylyn Loder and Chris Dieterich, “How a $1.4 Billion ETF Gold Rush Rattled Mining Stocks Around the World,” Wall Street Journal, April 23, 2017. The authors say that “money rushing into exchange-traded funds investing in gold mining stocks sparked wild trading in the stocks while the price of gold was largely flat.”

5.   The authors didn’t find this same increase in informational efficiency “for big firms, stocks with high analyst following, and for stocks with perfectly competitive equity markets.” With the exception of IJR (the small cap ETF), all 10 of our other ETFs were made up of firms in the S&P 500, which are generally big firms with high analyst following.

6.   Percentiles calculated for all seven-day periods from December 22, 1998 to September 28, 2015.

7.   https://twitter.com/hillaryclinton/status/645974772275408896?lang=en.

8.   Andrew Pollack, “Drug Goes from $13.50 a Tablet to $750, Overnight,” New York Times, September 20, 2015, http://www.nytimes.com/2015/09/21/business/a-huge-overnightincrease-in-a-drugs-price-raises-protests.html.

9.   Valeant actually began trading under the name Bausch Health Companies Inc. on Monday, July 16, 2018.

10.   It could be argued that drug-pricing pressures could ultimately affect the entire medical system and thereby impact medical equipment providers. However, these stocks’ reaction still seems exaggerated.

11.   Throughout this research project, including in the 2010–2017 backtest, transaction costs were estimated at 10 bps, or 17 bps considering leverage (on average). Borrowing costs were based on LIBOR + 50 bps and depend on how long the position was held. They cumulate to about 10 bps on average after 40 days. Hence, a rough estimate of total costs (transaction and borrowing) for 40 days would be 27 bps.

12.   Zacks Equity Research, “Stock Market News for February 11, 2016,” NASDAQ, http://www.nasdaq.com/article/stock-market-news-for-february-11-2016-cm578585.

13.   Larry Elliott and Jill Treanor, “Stock Markets Hit by Global Rout Raising Fears for Financial Sector,” Guardian, February 11, 2016, http://www.theguardian.com/business/2016/feb/11/stock-markets-hit-by-global-rout-raising-fears-for-financial-sector.

14.   Ben Levisohn, “American Express: No, Higher Interest Rates Won’t Help,” Barron’s, December 22, 2016, http://www.barrons.com/articles/american-express-no-higher-interest-rates-wont-help-1482422718.

15.   Zacks Equity Research, “Stock Market News for February 11, 2016.”

16.   Our choice of the 1- to 40-day windows was motivated by prior studies, as well as the need to avoid too many overlapping events. Ben-David et al. (2018) found that “most of the contemporaneous stock-price effect of ETF flows reverts over the next 40 days, in line with the view that the demand shocks in the ETF market translate into nonfundamental price changes for the underlying securities.” Brown et al. (2018) used a one-month horizon. Though extending the window beyond 40 days was possible (and we found that alpha continues to accumulate after 40 days), it creates too many overlapping events and makes it more difficult to attribute alpha. The median number of days between ETF spike dates in our sample was 38.

17.   See the online supplemental material, available at http://www.tandfonline.com/doi/suppl/10.1080/0015198X.2019.1572358, for details of the statistical test.

18.   Brown et al.’s (2018) analysis shows that ETF-driven reversals generate significant alphas after controlling for the Fama-French three factors plus momentum. Using betas calculated over the 252 pre-event days, the expected beta to the S&P 500 was slightly higher for the levered outsider portfolio than it was for the ETF. Consequently, after adjusting for the exposure to the market in the 40 post-event days, we could see a reduction in the alpha generated by our strategy of about 20 bps—a fraction of our 300 bps of precost alpha. We left it to the reader to interpret whether this slight excess beta constitutes a systematic bias, but if so, the impact remains small relative to the magnitude of the net alphas. Regarding liquidity, our outsiders have a similar liquidity profile, on average, to that of their peer constituents. The distribution is symmetrical: roughly half the low-ETF-beta stocks have above-average liquidity, and half have below-average liquidity.