Rochester Cahan Yu Bai 212 803-7973 212 803-7919

September 13, 2019 Stock Selection: Research and Results September 2019 Predicting Earnings Surprises: Some Useful Tools for a High-Stakes Game Do Earnings Surprises Even Matter?

 Do earnings surprises even matter? Unfortunately, the answer is yes, which is good for purveyors of caffeine but bad for the stock pickers who have to struggle through late-night model updates and interminable earnings calls four times a year. In the post-Crisis era the median stock has generated about a fifth of its annual relative return on the four days it announces earnings, a statistic that’s higher than it has been in past decades. The re- sult holds irrespective of whether the stock outperformed or underperformed over the year.

 At the same time, the portion of a stock’s annual return that’s produced in the week after earnings (excluding the announcement day itself) hasn’t changed much this decade and remains fairly insignificant. That means earnings reports are more consequential to a stock’s longer-term performance, but most of the action occurs on the day itself and there’s very little post-announcement drift. Today’s market is frightfully efficient and new information gets discounted almost immediately. To profit from earnings these days one either needs to be the first to react (i.e., probably be a robot) or one has to call the surprise correctly ahead of time. No Country for Old Surprises

 To make matters worse, predicting earnings surprises has become harder because there’s been a complete breakdown in the autocorrelation of surprises. Autocorrelation is just a statistic that measures the likelihood that an earnings beat will be followed by another beat next quarter (or a miss by another miss). Over the past five years the average autocorrelation in earnings surprises across all stocks has been exactly zero.

 However, there is one cohort of stocks that’s bucked the trend: serial beaters. In the 1990s serial beaters – which we define as stocks beating earnings for eight consecutive quarters – produced about 20% of the earn- ings beats in a given quarter. Now they account for close to 40% of all beats market-wide. Autocorrelation has vanished for most stocks, but an elite cohort of winners just keeps on winning. Unsurprisingly many of these firms are drawn from the technology sector, so what we may be capturing here is the ongoing strength of de- mand rather than a behavioral anomaly. Some Tools for Spotting Earnings Surprises

 Luckily, the collapse in autocorrelation doesn’t mean earnings surprises are completely unpredictable, it just means we need to inject additional information into the problem beyond what happened in previous quarters. We spent some time digging through the wide range of factors that we keep in our toolbox, to see if there are any that can help call the direction of earnings surprises. We found a few that can improve our win rate in predicting earnings surprises.

 However, win rates aren’t alpha and what ultimately matters is returns, which means it’s important to find stocks that surprise and move in the direction of the surprise. It turns out our Failure Model, which has been in live use for 15 years now, is hard to beat in that regard. It’s been good at identifying earnings blowups ahead of time and its Failure Candidates are almost 70% more likely to suffer a really big earnings blowup compared to a random stock picked from the market. Each week in earnings season we circulate a list of stocks reporting that week that screen as Failure Candidates. Let your salesperson know if you’d like to get that.

 Overall, our work in this report has reaffirmed our view that avoiding disasters is one of the best ways to use quantitative tools in a fundamental process. These days the market’s response to an earnings miss is swift and devastating so even sidestepping a few of them can make a big difference.

Sungsoo Yang (212) 803-7925 Nicole Price (212) 803-7935 Yi Liu (212) 803-7942 Iwona Scanzillo (212) 803-7915 © 2019, Empirical Research Partners LLC, 565 Fifth Avenue, New York, NY 10017. All rights reserved. The information contained in this report has been obtained from sources believed to be reliable, and its accuracy and completeness is not guaranteed. No representation or warranty, ex- press or implied, is made as to the fairness, accuracy, completeness or correctness of the information and opinions contained herein. The views and other information provided are subject to change without notice. This report is issued without regard to the specific investment objectives, fi- nancial situation or particular needs of any specific recipient and is not to be construed as a solicitation or an offer to buy or sell any securities or related financial instruments. Past performance is not necessarily a guide to future results. Stock Selection: Research and Results September 2019 Conclusions in Brief  This decade more of a stock’s annual return has come from  …And the increase has been larger for losers than winners: earnings announcement days…

Large-Capitalization Outperformers¹ Large-Capitalization Outperformers and Underperformers¹ Median Share of Annual Relative Returns that Occured on Earnings Median Share of Annual Relative Returns that Announcement Days² Occured Around an Earnings Announcement² % 2002 Through Late-August 2019 % 2003 Through 2007 and 2009 Through Late-August 2019 25 20

15 20 10

5 15 0

10 (5)

(10)

5 (15) Week Before Announcement Next Week Week Before Announcement Next Week Day Day Outperformers Underperformers 0 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 2003-07 2009-19 -to- Source: Empirical Research Partners Analysis. Date ¹ Stocks that outperformed or underperformed the equally-weighted market over the calendar year by more than ±1% Source: Empirical Research Partners Analysis. respectively. ¹ Stocks that outperformed the equally-weighted market over the calendar year by more than +1%. ² For earnings announcements outside of market hours the first trading day is used. Week before and next week exclude ² For earnings announcements outside of market hours the first trading day is used. the announcement-day return.  Autocorrelation in earnings surprises has collapsed…  …Except for an elite group of serial beaters:

Large-Capitalization Stocks Large-Capitalization Stocks Average Autocorrelation of Quarterly Earnings Surprises¹ that Beat Earnings 1996 Through Late-August 2019 Share by Number of Consecutive Beats % % 1996 Through Late-August 2019 10 40

35 8 30

6 25

20 4

15

2 10

0 5

0 (2) 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 Two Quarters in a Row Eight or More Quarters in a Row

Source: Empirical Research Partners Analysis. ¹ Autocorrelation based on earnings surprises over the prior 12 quarters. Source: Empirical Research Partners Analysis.

 Some quant factors, many of which capture sentiment, can  …B ut the F ailure M odel is hard to beat: help us predict surprises…

Large-Capitalization Failure Candidates Large-Capitalization Stocks Number of Big Earnings Misses Identified Ratio of Beats-to-Misses in the Best Quintile of Select Factors Relative to the Base Rate¹ Relative to that in the Worst Quintile: Top Ten x 1996 Through August 2019 x 1996 Through Early-September 2019 3.0 2.8

2.6 2.5 2.4

2.2 2.0 2.0

1.8 1.5 Parity 1.6

1.4 1.0 1.2

1.0 0.5 0.8 Earnings Earnings Market Nine- Media Free Growth ROE Short Failure Revisions Estimate Reaction Month Sentiment Score Pressure Model Dispersion Super Price (since Flow (Since Factor Trends 2002) Margin 2007) 0.0 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 Whole Period Since 2010 Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis. ¹ Big misses are defined as those inthe worst decile of earnings surprises each quarter.

2 Stock Selection: Research and Results September 2019

Predicting Earnings Surprises: Some Useful Tools for a High-Stakes Game Do Earnings Surprises Even Matter? For as long as most of us can remember the rhythm of our industry – and indeed professional life (some might have to drop the professional qualifier) – has marched to the beat of quarterly earnings. Four times a year things cre- scendo in a caffeinated blur of conference calls, obscure footnotes, late-night model updates and cold pizza. Then, just when you’ve caught your breath – and some sleep – it’s time to start prepping for the next one. Does any of this make any sense? After all, technology has vastly increased the rate at which information can be digested and it’s not clear one really needs to listen to every word of the earnings call when the robots are already eavesdropping on the line.1 Unfortunately, the importance of the earnings announcement day to the performance of individual stocks has been rising, not declining. One way to see this is to plot how much of a stock’s annual return is generated on those days. Exhibit 1 shows the results for stocks that outperformed the market over a calendar year. In the past five years the median outperformer generated almost a fifth of its annual return on the four days when it released earnings. That’s significantly higher than what was seen in the prior cycle. Some of this might be attributable to the leader- ship of the tech sector where earnings results matter, certainly more than in the commodity stock leadership of the 2000s cycle. Exhibit 2 shows the same chart for stocks that underperformed over a calendar year. For most of this cycle an increas- ing portion of a loser’s annual share price decline came on earnings announcement days. This year and last year have bucked that trend a bit, but overall this cycle has been challenging because a lot has hinged on just four trading days out of 252.

Exhibit 1: Large-Capitalization Outperformers1 Exhibit 2: Large-Capitalization Underperformers1 Median Share of Annual Relative Returns that Occurred Median Share of Annual Relative Returns that Occurred on Earnings Announcement Days2 on Earnings Announcement Days2 2002 Through Late-August 2019 2002 Through Late-August 2019 % % 25 0 Zero

20 (5)

15 (10)

10 (15)

5 (20)

0 (25) 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 -to- -to- Date Date Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

1 Stocks that outperformed the equally-weighted market over the calendar 1 Stocks that underperformed the equally-weighted market over the year by more than +1%. calendar year by more than (1)%. 2 For earnings announcements outside of market hours the first trading 2 For earnings announcements outside of market hours the first trading day is used. day is used.

Making things trickier is the fact that the week after earnings hasn’t really contributed more to annual performance (see Exhibits 3 and 4). The bars in these two charts show the portion of a stock’s annual relative return that came in the week after earnings, excluding the announcement day itself. This cycle only a 20th of the median stock’s annual return – for winners and losers – has come in the four discrete weeks after quarterly earnings. That suggests that while earnings days have been more decisive to annual performance this decade, all of the increase has been con- centrated on the announcement days. The discounting of the new information is close to immediate.

1 Stock Selection: Research and Results October 2017. “Earnings Calls: FOMO and MOMO.” 3 Stock Selection: Research and Results September 2019

Exhibit 3: Large-Capitalization Outperformers1 Exhibit 4: Large-Capitalization Underperformers1 Median Share of Annual Relative Returns that Occurred Median Share of Annual Relative Returns that Occurred in the Week After an Earnings Announcement2 in the Week After an Earnings Announcement2 2002 Through Late-August 2019 2002 Through Late-August 2019 % % 18 0

16 (2)

14 (4)

12 (6)

10 (8) 8

(10) 6

(12) 4

2 (14)

0 (16) 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 -to- -to- Date Date Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis. 1 Stocks that outperformed the equally-weighted market over the calendar year 1 Stocks that underperformed the equally-weighted market over the by more than +1%. calendar year by more than (1)%. 2 Excluding the earnings announcement day. 2 Excluding the earnings announcement day. It’s also interesting to note that the earnings day contribution has increased by more for losers than winners (see Exhibit 5). Reacting to bad numbers after they’ve been delivered has become less effective; you either have to cor- rectly call the miss ahead of time or you need to be a robot so you can be the first one out, reacting before the lowly humans have had time to click on the press release. No single type of stock has been driving these changes. We can see that in Exhibit 6, that plots the contribution that earnings days made to the annual returns of stocks that underperformed over a calendar year. In the latest decade the whole line is shifted down, meaning the greater impact of earnings days has been pervasive across all stocks. In the 2003-07 expansion 56% of annual losers had a net negative contribution from their four earnings days; this cycle the number has risen to 64%. That means 64% percent of stocks that underperformed on their four earnings days were also down for the whole year. An earnings miss is much harder to overcome when growth isn’t strong enough to bail out the perpetrator.

Exhibit 5: Large-Capitalization Outperformers and Underperformers1 Exhibit 6: Large-Capitalization Underperformers1 Median Share of Annual Relative Returns that Occurred Distribution of the Share of Annual Relative Returns Around an Earnings Announcement2 that Occurred on Earnings Announcement Days2 2003 Through 2007 and 2009 Through Late-August 2019 2003 Through 2007 and 2009 Through Late-August 2019 % % 20 200 Stock Underperformed Over the Year but Earnings Days Helped Mitigate that Underperformance

15 150 General Motors 2018 Relative Return: (7.3)% 100 Earnings Days Only: +7.1% 10

50 5 56%

0 0 64% (50)

(5) (100) ExxonMobil (10) 2018 Relative Return: (7.5)% (150) Earnings Days Only: (7.2)%

Stock Underperformed Over the Year and Earnings Days

(15) Days on Earnings Occured that Returns Relative of Annual Share Contributed to that Underperformance Week Before Announcement Next Week Week Before Announcement Next Week (200) Day Day 0 102030405060708090100 Outperformers Underperformers Percentiles (100 = Highest Share) 2003-07 2009-19 2003-07 2009-19 Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

1 Stocks that outperformed or underperformed the equally-weighted 1 Stocks that underperformed the equally-weighted market over the market over the calendar year by more than ±1% respectively. calendar year by more than (1)%. 2 For earnings announcements outside of market hours the first trading day is 2 For earnings announcements outside of market hours the first trading used. Week before and next week exclude the announcement-day return. day is used.

4 Stock Selection: Research and Results September 2019

For example, last year ExxonMobil underperformed by a cumulative (7.2)% on its four announcement days, which accounted for almost all of its annual loss. In contrast, General Motors outperformed by +7.1% on its earnings days, but the stock was still down for the year because it lagged badly on non-earnings days. The chart is telling us that there are more ExxonMobil-type situations these days, where the stock crashes on earnings and nothing the firm does in the rest of the year can paper over the carnage. It’s worth noting that the number of earnings misses has been fairly steady over time, at about 10-15% of all the re- ports (see Exhibit 7). So it’s not the frequency of misses that’s driving their rising impact. Rather, the magnitude of the negative response to misses has gotten larger over time as the nominal rate of economic growth came down (see Exhibit 8). The idiosyncratic characteristics of companies matters more when the tailwind is weaker.

Exhibit 7: Large-Capitalization Stocks Exhibit 8: Large-Capitalization Stocks Share of Earnings Announcements that Beat or Miss Average Relative Returns on Earnings Announcement Estimates by Year1 Days 2002 Through Late-August 2019 by Direction of Surprise and Year1 2002 Through Late-August 2019 % % 70 3

60 2

50 1

40 0

30 (1)

20 (2)

10 (3)

0 (4) 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 -to- -to- Date Date Beat Inline Miss Beat Inline Miss Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis. 1 Beats and misses are defined as actual EPS being more than two cents away 1 For earnings announcements outside of market hours the first trading from the pre-announcement consensus and at least 3% different from consensus. day is used. All others are considered inline with expectations. All of this is consistent with our view that avoiding losses is a big part of the game. owners, burnt badly in the Crisis, crave consistent return streams and a good night’s sleep, even if it means giving away some potential up- side. There’s very little stomach for losses of any kind and that shows up in a sell-first-ask-questions-later response to earnings disappointments (see Exhibit 9). Technology and new tools like ETFs have been enablers, making it eas- ier to duck for cover at the first sign of trouble.

Exhibit 9: Large-Capitalization Stocks Exhibit 10: Large-Capitalization Stocks Average Relative Returns Around Earnings Average Autocorrelation of Quarterly Earnings Announcement Days Surprises1 by Direction of Earnings Surprise1 1996 Through Late-August 2019 2003 Through 2007 and 2009 Through Late-August 2019 % % 2.0 10

1.5

1.0 8

0.5

0.0 6

(0.5)

(1.0) 4

(1.5)

(2.0) 2

(2.5)

(3.0) 0 Week Announcement Next Memo: Week Announcement Next Memo: Before Day Week Next Before Day Week Next Year Year 2003-07 2009-Present (2) Beat Miss 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis. 1 For earnings announcements outside of market hours the first trading day 1 Autocorrelation based on earnings surprises over the prior 12 quarters. is used. Week before, next week and next year exclude the announcement- day return. 5 Stock Selection: Research and Results September 2019

Unfortunately, the speed and accuracy of the initial response to an earnings release these days means there isn’t much post-announcement drift to play any more. Where it does still exist is mostly on the downside, see the right- hand black bar. Stocks that missed earnings in this decade went on to underperform by another (180) basis points on average over the following year, excluding the announcement-day return. The same doesn’t apply to earnings beats, where all of the outperformance has come on the announcement day in the post-Crisis period. No Country for Old Surprises Part of the reason there’s no long-term, post-announcement drift any more is because of a steady decline in the au- tocorrelation of earnings surprises (see Exhibit 10 overleaf). Autocorrelation, as it relates to earnings surprises, is just a statistic that tells us how often a surprise is in the same direction as last quarter’s surprise. For example, a company that consistently beats the analysts’ expectations quarter after quarter has a high autocorrelation. But keep in mind that a company that misses quarter after quarter would also have high autocorrelation. Autocorrelation doesn’t tell us anything about the direction of the surprise, only whether the surprises were consistently in the same direction. The collapse in autocorrelation has been pervasive across the market. Of the ten sectors experiencing the highest autocorrelation in surprises during this decade, eight have autocorrelations that are lower than what they had in the 1990s (see Exhibit 11). Many of the top ten sectors are cyclical in nature, which makes sense because economically- sensitive businesses often beat consistently in an economic expansion and then miss consistently in a contraction. For example, semiconductors have had the highest autocorrelation of any sector over time (see Exhibit 12). Here the past is something of a guide to the future despite what the disclaimers will tell you: if a semi stock beat last quarter it’s a better-than-average bet that it will do so again this quarter (of course, the same applies to a miss).

Exhibit 11: Large-Capitalization Stocks Exhibit 12: Large-Capitalization Semiconductor Stocks Average Autocorrelation of Quarterly Earnings Surprises Average Autocorrelation of Quarterly Earnings by Sector: Top Ten1 Surprises1 1996 Through 1999 and 2009 Through Late-August 2019 1996 Through Late-August 2019 % % 16 35 26%

14 30

12 25

10 20

8 15

6 10

4 5

2 0

0 Semi- Tech Capital Biotech- Health Pharma- Commercial Insurance Retail & Consumer (5) conductors Hardware Equipment nology Care ceuticals Services & Other Durables Equipment Supplies Consumer Cyclicals (10) 2009-19 Memo: 1996-99 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis. 1 Autocorrelation based on earnings surprises over the prior 12 quarters. 1 Autocorrelation based on earnings surprises over the prior 12 quarters. Exhibit 13 shows the ten sectors with the lowest autocorrelations this decade. Commodities, energy, and financials haven’t been able to catch a break this cycle. Every time they showed some promise with a beat, the next quarter was likely to undo all the good work with a miss. Some of the travails of value investing in the post-Crisis world are encapsulated in this chart. Even the largest stocks in the market, which have generally done well during this expansion, haven’t been able to sidestep the declining autocorrelation (see Exhibit 14). In the past five years the autocorrelation for the 50 largest- cap stocks in the market has averaged close to zero, just like that for the wider market. This broad-based decline in autocorrelation doesn’t mean that earnings surprises are unpredictable; it just means they can’t be predicted by simply extrapolating past outcomes. Instead, one needs to inject additional information into the problem. With so much of a stock’s return hinging on four earnings days each year the potential reward for an accurate forecasts is up. But so too is the degree of difficulty because of the breakdown in autocorrelation. In the next section we present some research we did trying to root out the stocks most likely to beat or miss expectations. 6 Stock Selection: Research and Results September 2019

Exhibit 13: Large-Capitalization Stocks Exhibit 14: Large-Capitalization Stocks Average Autocorrelation of Quarterly Earnings Surprises Average Autocorrelation of Quarterly Earnings by Sector: Bottom Ten1 Surprises1 1996 Through 1999 and 2009 Through Late-August 2019 1996 Through Late-August 2019 % % 10 12

8 10

6 8 4

6 2

0 4

(2) 2

(4) 0 (6)

(2) (8) (15)%

(10) (4) Media Consumer Banks Real Indutrial Telecom- Integrateds, Trans- Capital Utilities Staples & Estate Commod- munications Oil ports Markets Consumer ities Services Finance & (6) Refiners 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19

2009-19 Memo: 1996-99 Top 50 by Market Cap All Stocks Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

1 Autocorrelation based on earnings surprises over the prior 12 quarters. 1 Autocorrelation based on earnings surprises over the prior 12 quarters.

Winners Gonna Win, Win, Win The first observation that can help us is the fact that that a disproportionate share of beats each quarter come from a shortlist of serial beaters: firms that have beaten estimates for eight or more consecutive quarters (see Exhibit 15). In the mid-1990s about a fifth of the earnings beats in a given quarter were delivered by these serial beaters, now their share has almost doubled. In contrast, the number of earnings misses that come from serial losers hasn’t changed much over time (see Exhibit 16). Note that we define serial losers as companies missing estimates for four or more consecutive quarters because it’s quite rare to have eight or more misses in a row. In listed companies you tend to get delisted, drop out of the in- vestable universe, get taken over or go bankrupt well before you build up a long losing streak.

Exhibit 15: Large-Capitalization Stocks Exhibit 16: Large-Capitalization Stocks that Beat Earnings that Miss Earnings Share by Number of Consecutive Beats Share by Number of Consecutive Misses 1996 Through Late-August 2019 1996 Through Late-August 2019 % % 40 35

35 30

30 25

25

20

20

15 15

10 10

5 5

0 0 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19

Two Quarters in a Row Eight or More Quarters in a Row Two Quarters in a Row Four or More Quarters in a Row

Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

The elite cohort of serial beaters has also been outperforming their less-consistent peers by a wider margin in recent years (see Exhibit 17). Since the mid-1990s stocks in the highest quintile of the number of consecutive beats outper- formed by a modest +1 percentage point per annum; this decade their alpha is up five-fold. At the same time the turnover of stocks in the highest quintile of consecutive beats has been slowing over time (see Exhibit 18). Once again, the winners-keep-winning dynamic that’s characterized the post-Crisis era is on full display.

7 Stock Selection: Research and Results September 2019

Exhibit 17: Large-Capitalization Stocks Exhibit 18: Large-Capitalization Stocks Relative Returns to the Highest and Lowest Quintiles in the Highest Quintile of Consecutive Earnings Beats of Consecutive Earnings Beats Average Monthly Turnover Monthly Data Compounded to Annual Periods 1996 Through August 2019 1996 Through Late-August 2019 % % 5 6

4 5 3

2 4

1 3

0

2 (1)

(2) 1

(3) Most Consecutive Beats Least Consecutive Beats Memo: Misses Quintiles of Earnings Beats Streak 0 1990s 2000s 2010s Memo: Last Five Years Whole Period 2010s Last Five Years Decades Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

The current cutoff to make the top quintile is 14 consecutive quarterly beats and it’s telling that only six industries are currently overrepresented in the top quintile of serial beaters, with three of those drawn from the technology sector (see Exhibit 19). We’ve documented many times the reasons for tech’s persistent dominance – the buildout of the Cloud, globalization of supply chains on the hardware side, lower tax rates, and so forth – so we won’t dwell on them here. Suffice it to say that a lot of the persistence in serial beaters boils down to the same story. Meanwhile, for stocks that miss earnings the market takes a dim view of even the first miss, and by the second con- secutive miss investors have generally had enough (see Exhibit 20). There are a lot more sectors that are overrepre- sented in the bucket of stocks that have missed earnings recently, compared to what we saw for the serial beaters (see Exhibit 21). That just hammers home how concentrated the market’s earnings leadership has been this cycle: a narrow group of mostly tech stocks have enjoyed robust autocorrelation in earnings surprises while most of the rest of the market has seen their earnings surprise persistence collapse.

Exhibit 19: Large-Capitalization Stocks Exhibit 20: Large-Capitalization Stocks Sectors Overrepresented in the Highest Quintile Relative Returns by Number of Consecutive Earnings of Consecutive Earnings Beats Misses Relative to their Benchmark Representation Monthly Data Compounded to Annual Periods As of Early-September 2019 1996 Through August 2019 x % 4.0 0.5

3.5 0.0

3.0 (0.5)

(1.0) 2.5

(1.5) Benchmark 2.0 Representation (2.0) 1.5 (2.5)

1.0 (3.0)

0.5 (3.5)

0.0 (4.0) Semi- Technology Health Care Technology Pharma- Retail & Memo: conductors Software & Equipment Hardware ceuticals Other Big Growers Services & Services Consumer (4.5) Cyclicals First Miss Two Three Four or More Overrepresented Sectors Number of Consecutive Misses Highest Quintile of Consecutive Earnings Beats Memo: Earnings Misses Whole Period 2010s Last Five Years Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

All of this means we need to be a little circumspect about literally extrapolating the success of serial beaters going forward, because it’s effectively a bet that the market’s tech/growth leadership will continue to win. The aftermath of the New Economy era is a sobering reminder of what can happen to serial beaters when the mood changes and the cycle rolls over (see Exhibit 22). 8 Stock Selection: Research and Results September 2019

Exhibit 21: Large-Capitalization Stocks Exhibit 22: Large-Capitalization Stocks Sectors Overrepresented in the Cohort of Stocks Relative Returns to the Highest Quintile of Missing Earnings Consecutive Earnings Beats1 Relative to their Benchmark Representation Measured Over One-Month Holding Periods As of Early-September 2019 1996 Through August 2019

x % 2.5 3

2 2.0

1 1.5 Benchmark Representation

0

1.0

(1)

0.5

(2)

Zero Zero Zero Zero 0.0 Utilities Commercial Banks & Telecom- Trans- Industrial Insur- Integrateds, Media Consumer Capita Tech Memo: Services & Cons- muni- ports Commod- ance Oil Services Durables Equip- Hard- Q1 of (3) Supplies umer cations ities & Refiners ment ware Valuation Finance Overrepresented Sectors (4) Earnings Misses Memo: Highest Quintile of Consecutive Earnings Beats 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

1 Data smoothed on a trailing six-month basis.

Some Tools for Spotting Earnings Surprises With that in mind, we spent some time looking for other metrics that might help us call the direction of earnings surprises, beyond the history of a stock’s past surprises. Exhibit 23 shows factors that have historically done a good job of identifying earnings beats ahead of time. The way to read the chart is each bar shows the number of beats identified by the top quintile of each metric, relative to what we’d expect randomly. For example, stocks in the best quintile of consecutive earnings beats have been about 22% more likely to beat earnings in the next quarter com- pared to stock randomly picked from the market at large. The magnitude of the numbers illustrate how difficult it is to systematically play the earnings game. We need to take a lot of swings at the plate to get the benefits of these small advantages. Exhibit 24 repeats the exercise for companies that miss earnings, using the worst quintiles of factors. There’s an overlap between the metrics that appear in the two charts, indicating that many factors do a good job of simultane- ously flagging beats in the best quintile and misses in the worst quintile.

Exhibit 23: Large-Capitalization Stocks Exhibit 24: Large-Capitalization Stocks Number of Earnings Beats Identified Number of Earnings Misses Identified Relative to the Base Rate: Top Ten by the Worst Quintile of Select Factors 1996 Through Early-September 2019 Relative to the Base Rate: Top Ten 1996 Through Early-September 2019

x x 1.25 1.5

1.20 1.4

1.15 1.3 Parity

1.10 Parity 1.2

1.05 1.1

1.00 1.0

0.95 0.9

0.90 0.8 Consecutive Earnings Media Market Low Free Nine- Tight Growth High Quant Earnings Wide Nine- Market Free ROE Free Failure Media Growth Earnings Revisions Sentiment Reaction Short Cash Month Earnings Score Hedge Fund Revisions Earnings Month Reaction Cash Cash Candidates Sentiment Score Beats (since 2002) Super Pressure Flow Price Estimate Ownership Estimate Price Super Flow Flow (since Factor (Since 2007) Margin Trends Dispersion (Since 2000) Dispersion Trends Factor Yield Margin 2002) Whole Period Since 2010 Whole Period Since 2010 Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

9 Stock Selection: Research and Results September 2019

For example, consider a factor like media sentiment, a metric we added to our library last year off the back of our Big Data initiative.2 If we plot its win rate in identifying beats and misses by quintile, we see that the best quintile does a good job of finding beats and avoiding misses, whereas the worst quintile is opposite, finding lots of misses and few beats (see Exhibit 25). Taking the ratio of the two bars gives us a favorable, monotonic pattern: the best quintile of media sentiment finds about 1.5 times more beats than misses while the worst quintile only contains about 0.8 beats for every miss (see Exhibit 26). The right-hand black bar shows the ratio of beats-to-misses for the best quintile relative to the worst quintile. The higher this number is, the better the factor has been at calling beats and misses.

Exhibit 25: Large-Capitalization Stocks Exhibit 26: Large-Capitalization Stocks Number of Earnings Beats and Misses Identified Beats-to-Misses Ratio by Quintile of Media Sentiment by Quintile of Media Sentiment Relative to the Base Rate Relative to the Base Rate 2002 Through August 2019 2002 Through August 2019 x x 1.2 2.0

1.1 1.8 Parity

1.6 1.0

1.4 0.9

1.2 0.8 Parity

1.0

0.7

0.8

0.6 Best Second Third Fourth Worst Quintiles of Media Sentiment 0.6 Best Second Third Fourth Worst Best/Worst Beats Misses Quintiles of Media Sentiment Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

Exhibit 27 shows the top ten factors ranked by their beats-to-misses ratio in their best quintiles relative to their worst quintiles. This is a starting point for thinking about what metrics might be helpful in trying to predict future earn- ings surprises. However, it’s only a starting point because of Exhibit 28, which shows the relative returns to each of the factors depicted in the previous chart. The grey bars are the returns only for stocks announcing earnings and the black bars are for all stocks. Some metrics, like our Failure Model, have a good win rate in predicting beats and misses and generate alpha from a return perspective.3 Other factors, like nine-month price momentum, have a good win rate but are poor at generating alpha despite their good batting averages. What’s going on? The answer lies in the fact that the market anticipates the direction of some surprises. Remember, a surprise is de- fined relative to analyst consensus, but that doesn’t mean the market actually believed the consensus. We can see what’s going on if we take price momentum as an example. Exhibit 29 shows that the factor has a good history of predicting beats and misses, which is why it ranked highly back in Exhibit 27. So far so good. The complication arises in Exhibit 30. Stocks in the worst quintile of price momentum that beat earnings – the left- most grey bar – tend to do very well, for an obvious reasons: the market action up to that point suggested that in- vestors believed things would keep getting worse so the positive surprise is genuine. Meanwhile, stocks in the worst quintile of price momentum that miss earnings – the right-most black bar – don’t underperform by that much; that’s because even though the stock missed analyst expectations the market already had a fairly unfavorable view of the stock so the miss came as less of a shock. Even though there are a lot more misses in the worst quintile of the momentum, the few beats that show up do so well they overwhelm the favorable win rate, meaning momentum is actually a perverse signal in the subset of stocks that are announcing earnings (see Exhibit 31). The net result is that calling earnings surprises is a balancing act: we want a high win rate but we also want to make sure we’re focused on calling surprises that will lead to a big price reaction in the same direction. There’s no point in leaning heavily on price momentum because even though there are lots of misses in the worst quintile, the mar- ket already discounts that to some degree. The same goes for beats in the best quintile. Given that, we can rule out

2 Stock Selection: Research and Results August 2018. “Big Data: Harnessing News and Social Media Sentiment to Improve Our Timing.”

3 Portfolio Strategy August 2019. “The Failure Model: Loaded for Bear in Earnings Season.” 10 Stock Selection: Research and Results September 2019 a number of metrics with good win rates but poor returns: earnings revisions, earnings dispersion, and the afore- mentioned price momentum for example. Stocks with lofty all-around growth credentials, as measured by our growth score, also tend to beat earnings more than we’d expect randomly, but the disappointment effect in the handful of big expectation stocks that don’t deliver is so large it swamps the favorable win rate.

Exhibit 27: Large-Capitalization Stocks Exhibit 28: Large-Capitalization Stocks Ratio of Beats-to-Misses in the Best Quintile Monthly Return Spread Between the Best and Worst of Select Factors Quintiles of Select Factors Relative to that in the Worst Quintile: Top Ten Contingent Upon Whether the Company Reported 1996 Through Early-September 2019 Earnings that Month Measured Over One-Month Holding Periods 1996 Through August 2019 x % 2.8 1.4

2.6 1.2

1.0 2.4 0.8 2.2 0.6 2.0 0.4 1.8 Parity 0.2 1.6 0.0

1.4 (0.2)

1.2 (0.4)

1.0 (0.6)

0.8 (0.8) Earnings Earnings Market Nine- Media Free Growth ROE Short Failure Earnings Earnings Market Nine- Media Free Growth ROE Short Failure Revisions Estimate Reaction Month Sentiment Cash Score Pressure Model Revisions Estimate Reaction Month Sentiment Cash Score Pressure Model Dispersion Super Price (since 2002) Flow (Since 2007) Dispersion Super Price (since 2002) Flow (Since 2007) Factor Trends Margin Factor Trends Margin Whole Period Since 2010 Stocks Reporting Earnings All Stocks Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

Exhibit 29: Large-Capitalization Stocks Exhibit 30: Large-Capitalization Stocks Number of Earnings Beats and Misses Identified Relative Announcement-Month Returns to Stocks by Quintile of Nine-Month Price Trends that Beat or Miss Earnings Relative to the Base Rate by Quintile of Nine-Month Price Trends 1996 Through August 2019 Measured Over One-Month Holding Periods 1996 Through August 2019 x % 1.3 3

1.2 2

1.1 1 Parity

1.0 0

0.9 (1)

0.8 (2)

0.7 (3)

0.6 (4) Best Second Third Fourth Worst Best Second Third Fourth Worst Quintiles of Nine-Month Price Trends Quintiles of Nine-Month Price Trends Beats Misses Beats Misses Memo: All Stocks Reporting Earnings Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

At the same time, other factors that have a lower win rate but do better in calling returns can be helpful, some of which are presented in Exhibit 32. Putting everything together, we assembled a rough prototype of an earnings surprise indicator, comprised of 13 equally-weighted factors that strike a good balance between predicting the di- rection of the surprise and the likely market response to that surprise (see Exhibit 33). To focus the indicator on stocks likely to have a large response to earnings we applied the indicator to the subset of stocks that screen in the highest quintile of arbitrage risk, a metric we’ve long used to assess the degree of controversy enveloping a stock. When a stock is embroiled in dispute prior to its reporting day even a small edge in calling the direction of the sur- prise can help because the stakes are so high. On face value the indicator does help shade the odds a bit in our favor (see Exhibit 34). Historically, high arbitrage risk stocks with favorable earnings surprise characteristics have been about +14% more likely to beat earnings than a random stock drawn from the market, see the left-hand bars in the chart. Meanwhile, stocks in the worst quintile have been about +54% more likely to miss earnings than a randomly-selected stock, see the right-hand bars. 11 Stock Selection: Research and Results September 2019

Exhibit 31: Large-Capitalization Stocks Exhibit 32: Large-Capitalization Stocks Relative Monthly Returns to the Quintiles Monthly Return Spread Between the of Nine-Month Price Trends Best and Worst Quintiles of Select Factors Contingent Upon Whether the Company Contingent Upon Whether the Company Reported Reported Earnings that Month Earnings that Month Measured Over One-Month Holding Periods Measured Over One-Month Holding Periods 1996 Through August 2019 1996 Through August 2019 % % x 0.9 1.4 1.5

0.8 1.4 1.2

0.7 1.3 1.0 0.6 1.2

0.5 0.8 1.1

0.4 0.6 1.0

0.3 0.9

Best and Worst Quintiles 0.4 0.2 0.8 Quinile Worst to the Relative Monthly SpreadMonthly Return the Between 0.1 0.2

0.7 Quintile Bestin of the Beats-to-MissesRatio

0.0 0.0 0.6 Capital Sector Valuation Free Earnings 10-K/Q Quant (0.1) Deployment ETF Flows & Super Cash Quality Disclosure Hedge Fund Super Equivalent Factor Flow Super Model Ownership Factor Volume Yield Factor (Since 2000) (0.2) (Since 2010) Best Second Third Fourth Worst Quintiles of Nine-Month Price Trends Stocks Reporting Earnings All Stocks Reported Earnings All Stocks Memo: Ratio of Beats-to-Misses in Best Quintile Relative to Worst Quintile Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

Exhibit 33: A Prototype Earnings Surprise Indicator Exhibit 34: Large-Capitalization Stocks Factors Deployed in the Highest Quintile of Arbitrage Risk As of Early-September 2019 Share of Earnings Beats and Misses Identified by the Best and Worst Quintiles of a Prototype Earnings Surprise Indicator Relative to the Base Rate 1996 Through August 2019 x 1.6 Number of Consecutive Earnings Beats 1.5

1.4 Existing Fundamentals & Frameworks 1.3 Valuation Market Reaction Free Cash Flow Margin Capital Deployment ROE 1.2 Free Cash Flow Yield The Failure Model 1.1 Parity

1.0

0.9

0.8

Market Structure 0.7 Quant Hedge Fund Big Data Ownership Media Sentiment 0.6 Sector ETF Flows & Equivalent Beats Misses Beats Misses Volume 10-K/Q Disclosure Model Short Pressure Best Quintile Worst Quintile Whole Period 2010s Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis.

Unfortunately, a good win rate alone doesn’t put bread on the table. Ultimately, returns determine the success or failure of an investment strategy. Our prototype earnings surprise screen does manage to generate some alpha, but it’s noteworthy that the performance of the worst quintile of the screen is almost exactly in-line with that of our Failure Model, that’s been in live use for 15 years now (see Exhibit 35). That suggests there’s no silver bullet when trying to predict earnings surprises. Our existing Failure framework has been quite effective at rooting out big misses, and it has the advantage of being an all-purpose model that we can deploy all the time rather than just four times per year (see Exhibit 36). It seems unnecessary to completely reinvent the wheel here when our existing tech- nology has proven itself in a wide range of market conditions over the past decade and a half. Conclusion: Systematic Tools Can Help When Playing High-Stakes Poker Earnings days now account for a greater share of a stock’s annual returns, raising the ante for calling them correct- ly. Unfortunately, almost all of the new information released is discounted immediately, so to generate alpha we need to call the direction of earnings ahead of time (or we can be very patient, a topic for another day). That game is hard because competition for alpha is acute and hedge funds congregate in the high arbitrage risk sandbox, which often overlaps with stocks set to report earnings (see Exhibits 37 and 38). 12 Stock Selection: Research and Results September 2019

Exhibit 35: Large-Capitalization Stocks Exhibit 36: Large-Capitalization Failure Candidates in the Highest Quintile of Arbitrage Risk Number of Big Earnings Misses Identified Relative Returns to the Quintiles of a Prototype Relative to the Base Rate¹ Earnings Surprise Indicator 1996 Through August 2019 Measured Over One-Month Holding Periods 1996 Through August 2019 % x 1.0 3.0

2.5 0.5

2.0 0.0

1.5

(0.5)

1.0

(1.0)

0.5

(1.5) Best Second Third Fourth Worst Memo: 0.0 Failure Quintiles of the Prototype Earnings Surprise Indicator 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 Candidates¹ Source: Empirical Research Partners Analysis. Source: Empirical Research Partners Analysis. ¹ Failure Candidates are drawn from the entire large-cap universe, ¹ Big misses are defined as those in the worst decile of earnings not just the highest quintile of arbitrage risk. All returns are relative to surprises each quarter. the entire large-cap universe.

Exhibit 37: Fundamental Equity Research Analysts Exhibit 38: Large-Capitalization Stocks By Firm Type in the Highest Quintile of Fundamental Hedge Fund 2019 Ownership Share of Stocks in the Highest Quintile of Arbitrage Risk 2000 Through August 2019 % 4,000 50

3,500 45

3,000 40

35 2,500

30 2,000 25

1,500 20

1,000 Random Draw 15

500 10

0 5 Long Managers Hedge Funds Sell-Side Analysts 0 Next 10 Next 80 All Others Sellside 00 02 04 06 08 10 12 14 16 18 Source: eVestment Alliance, Bigdough.com, Greenwich Associates, Source: Empirical Research Partners Analysis. Empirical Research Partners Analysis.

Luckily, if one chooses to play the high-stakes game of earnings prediction there are some quantitative tools that can help. Our longstanding Failure Model has done a good job of identifying potential earnings disasters in real-time use over 15 years and after crunching through lots of other permutations of things that might predict earnings we couldn’t come up with anything that significantly improves upon it. The fact the Failure Model is a general purpose model that we can use outside of earnings season is a plus, because it avoids having to turn to a specialized model four times per year. Overall, our work on the topic has reaffirmed our view that avoiding torpedoes is one of the best ways to use sys- tematic tools in a fundamental process. We’re often asked what Big Quant is up to, and from this analysis one thing they appear to be attuned to is downside risk in earnings. It’s telling that stocks in their long books have historically been involved in fewer earnings debacles than other stocks in the market; collectively the quants have been better- than-average at sidestepping earnings blowups. We think most investors can benefit from a disciplined approach to earnings surprise detection, particularly on the downside where the market’s response to an earnings disappointing is swift and devastating these days. Every week during earnings season we send out the list of Failure Candidates that are set to report earnings that week. If you’d like to get on that distribution please contact your salesperson. 13