基础率手册——销售增长：整合过去以更好预见未来

GLOBAL FINANCIAL STRATEGIES www.credit-suisse.com

The Base Rate Book – Sales Growth Integrating the Past to Better Anticipate the Future February 23, 2016

40 Authors 35 Base Rates Michael J. Mauboussin

Frequency (Percent)

30 [email protected] 25 Dan Callahan, CFA 20 [email protected]

3-Year Net Income CAGR (Percent)

Source: Credit Suisse HOLT® and FactSet Estimates.

“‘Pallid’ statistical information is routinely discarded when it is incompatible with one’s personal impressions of a case.”

Daniel Kahneman1

Successful active investing requires a forecast that is different than what the market is discounting.

Forecasts about outcomes relevant to us commonly suffer from biases of optimism and overconfidence.

Research reveals that consideration of the results for an appropriate reference class can enhance the quality of forecasts.

Sales growth is the most important value driver for most companies.

This report shows the base rate of sales growth rates for a large sample of companies over more than six decades. We sort the companies into deciles, allowing for easy identification of an appropriate reference class.

We provide a method to integrate individual views with base rates in order to improve forecasts.

Introduction

An investor’s primary task is to determine whether the expectations for future financial performance, as implied by the stock price, are too optimistic or pessimistic relative to how the company is likely to perform. In other words, the intelligent investor seeks gaps between expectations and fundamentals.2 This approach does not require forecasts of pinpoint accuracy, but rather only judgments as to whether the expectations embedded in the shares are too high or low.

Sales growth is the most important driver of corporate value.3 For companies that earn a return on invested capital in excess of the cost of capital, growth amplifies value creation. For those that earn a return below the opportunity cost, growth destroys value. Sales growth ripples through the income statement by determining operating leverage and the degree of economies of scale a company can realize.

Naturally, executives want their companies to generate healthy rates of sales growth, and more than a third of the companies in the S&P 500 Index provide guidance for sales growth.4 Researchers who study forecasts find two common biases: optimism and overconfidence.

Optimism about predictions that are personally important encourages perseverance in the face of adversity but also offers a distorted view of likely outcomes.5 Consider that only about one-half of new businesses survive five or more years. Notwithstanding that fact, a survey of thousands of entrepreneurs found that more than 8 of 10 of them rated their odds of success at 70 percent or higher, and fully one-third did not allow for any probability of failure at all.6 The bottom line on optimism: “People frequently believe that their preferred outcomes are more likely than is merited.”7

Overconfidence also distorts the ability to make sound predictions. This bias reveals itself when an individual’s confidence in his or her subjective judgments is higher than the objective outcomes warrant. For instance, more than five thousand people answered 50 true-false questions and provided a confidence level for each.

On average, they were 60 percent correct but were 70 percent confident in their answers. When these subjects were 100 percent confident in their response, they were correct only 77 percent of the time.8 Most people, including financial analysts, place too much weight on their own information.9

Overconfidence shows up in forecasts as ranges of outcomes that are too narrow.10 As a case in point, researchers asked chief financial officers to predict the results for the stock market, including high and low growth rates of return within which the executives were 80 percent sure the results would land. They were correct only one-third of the time.11

Exhibit 1 shows how this bias manifests in forecasts of revenue growth. The solid line is the distribution of sales growth rates, annualized over three years and adjusted for inflation, for the 1,000 largest companies in the world by market capitalization. These base rates reflect results from 1950 to 2014. The dashed line is the distribution of available analyst forecasts for the sales growth rates of the largest 1,000 companies today.

Exhibit 1: Overconfidence – Range of Sales Growth Rates Too Narrow 40 35 Base Rates Current Estimates

Frequency (Percent)

3-Year Sales CAGR (Percent)

® Source: Credit Suisse HOLT and FactSet Estimates.

Note: FactSet consensus estimates as of February 19, 2016.

Consistent with the overconfidence bias, the range of expected outcomes is narrower than what the results of the past suggest is reasonable. Specifically, the standard deviation of estimates is 12.2 percent versus a standard deviation of 18.4 percent for the past growth rates. Forecasts are commonly too optimistic and too narrow. The best explanations for the pattern of faulty forecasts include behavioral biases and distortions encouraged by incentives.12

Knowing that we are prone to optimism and overconfidence means we need to adopt techniques to manage those biases. One useful approach is to examine the experience of many companies over time, or base rates, and thoughtfully integrate those base rates with our own view. This is not our typical approach. As Daniel Kahneman, the eminent psychologist, notes bluntly, “People who have information about an individual case rarely feel the need to know the statistics of the class to which the case belongs.”13

The classic way to make a forecast is to gather information, combine it with our own view, and project an outcome. Left to our own devices, this is how we naturally go about forecasting. Here’s an example: In February 2015, Elon Musk, chairman, product architect, and chief executive officer of Tesla Motors, proposed that the company would be able to grow its sales 50 percent compounded annually for the next decade.14 How would you assess that proposition?

The natural way would be to open a spreadsheet and start counting. How big is the global automobile market? What share of total auto sales do electric cars have? Where will the market share of electric cars go? What share will Tesla have? And so forth. Combine this assessment with some knowledge of the company’s current and future offerings and you can come up with a forecast. You can do a similar exercise for the battery business. You would then compare the results of your analysis to Musk’s aspired growth rate.

Using Kahneman’s language, you have collected “information about an individual case.” But what you haven’t considered is “the statistics of the class to which the case belongs.” We don’t use base rates as much as we should for a couple of reasons. First, you trust and value the work you’ve done and hence are inclined to place

a lot of weight on it. Second, base rates are rarely at your fingertips. You have to find a suitable reference class and incorporate the information appropriately.

Though decision scientists have known for a long time that the proper integration of base rates improves the quality of forecasts, the technique remains remarkably underused.15 We believe this reflects the human desire for a narrative. Causality is clear in stories about the specifics, which makes those scenarios vivid. Base rates, on the other hand, are largely antiseptic and hence less appealing to the mind.

Base Rates of Sales Growth

We analyze the distribution of sales growth rates for the top 1,000 global companies by market capitalization since 1950.16 This sample represents roughly 60 percent of the global market capitalization and includes all sectors. The population includes companies that are now “dead.” The main reason public companies cease to exist is they merge or are acquired.17

We calculate the compound annual growth rates (CAGR) of sales for 1, 3, 5, and 10 years for each firm. We adjust all of the figures to remove the effects of inflation, which translates all of the numbers to 2014 dollars.

Exhibit 2 shows the results for the full sample. In the panel on the left, the rows show sales growth rates and the columns reflect time periods. Say you want to know what percent of the universe grew sales at a CAGR of 15-20 percent for three years. You start with the row marked “15-20” and slide to the right to find the column “3-Yr.” There, you’ll see that 6.8 percent of the companies achieved that rate of growth. The panel on the right shows the sample sizes for each growth rate and time period, allowing us to see where that percentage comes from: 3,261 instances out of the total of 48,136 (3,261/48,136 = 6.8 percent).

Exhibit 2: Base Rates of Sales Growth (1950-2014)

Source: Credit Suisse HOLT®.

Note: CAGR = compound annual growth rate.

Exhibit 3 is the distribution for the three-year sales growth rate. This represents, in a graph, the corresponding column in exhibit 2. The mean, or average, growth rate was 8.0 percent per year and the median growth rate was 5.5 percent. The median is a better indicator of the central location of the results because the distribution is skewed to the right. The standard deviation, 18.4 percent, gives an indication of the width of the bell curve.

Exhibit 3: Three-Year CAGR of Sales (1950-2014)

30

25

Frequency (Percent)

CAGR (Percent)

® Source: Credit Suisse HOLT .

Note: CAGR = compound annual growth rate.

While the data for the full sample are a start, you want to hone the reference class of base rates to make the results more relevant and applicable. One approach is to break the universe into deciles based on a company’s sales in the prior year. Within each size decile, we sort the observations of growth rates into bins in increments of five percentage points (except for the tails).

There is a modest survivorship bias because each sample includes only the firms that survived for that specified period. For example, a company in our 10-year sample would have had to have survived for 10 years.

About one-half of all public companies cease to exist within ten years of being listed.18

The heart of this analysis is exhibit 4, which shows each decile, the total population, and an additional analysis of mega companies (those with sales in excess of $50 billion). Here’s how you use the exhibit. Determine the base sales level for the company that you want to model. Then go to the appropriate decile based on that size.

You now have the proper reference class and the distribution of growth rates over the various horizons.

Let’s use Tesla as an example. Musk said he hoped to grow sales 50 percent per year for the next decade from a sales base of $6 billion (actual sales for 2015 were $4 billion). You first find the correct reference class, which is the decile that has a sales base of $4.5 - $7 billion. Next you examine the row of growth that is marked “>45,” representing sales growth of 45 percent or more. Going to the column “10-Yr,” you will see that no companies achieved this feat. Indeed, you have to go down to 30-35 percent growth to see any companies, and even there it is only one-fifth of 1 percent of the sample.

Exhibit 4: Base Rates by Decile (1950-2014)

Source: Credit Suisse HOLT®.

In total, exhibit 4 shows results for 44 reference classes (11 size ranges times 4 time horizons) that should cover the vast majority of possible outcomes for sales growth. The appendix contains the sample sizes for each of the reference classes. Bear in mind that these data are adjusted for inflation and that most forecasts reflect inflation expectations. We will show how to incorporate these base rates into your forecasts for sales growth in a moment. For now, it’s useful to acknowledge the utility of these data as an analytical guide and a valuable reality check.

Getting to the proper reference class is crucial, but there are some useful observations about the whole that are worth noting. To begin, as firm size increases the mean and median growth rates decline, as does the standard deviation of the growth rates. This point has been well established empirically.19 Exhibit 5 shows this pattern for annualized growth rates over three years. The lesson is to temper expectations about sales growth as companies get larger.

Exhibit 5: Growth Rates and Standard Deviations Decline with Size Mean Median Standard Deviation 45 40

Sales 3-Year CAGR (Percent)

Decile (Smallest to Largest by Sales) Universe Mega Source: Credit Suisse HOLT®.

Note: Growth rates are annualized over three years.

Exhibit 6 shows that sales growth follows gross domestic product (GDP) reasonably closely. U.S. GDP growth and the median sales growth in the same year have a correlation coefficient of 0.65. (Positive correlations fall in the range of 0 to 1.0, where 0 is random and 1.0 is a perfect correlation.) From 1950-2014, U.S. GDP grew at 3.2 percent per year, adjusted for inflation, with a standard deviation of 2.3 percent.

Corporate sales growth was higher than that of the broader economy for a few reasons. First, companies growing rapidly often need access to capital and hence choose to go public, likely creating a selection bias. Second, some companies, including contract manufacturers, generate growth that is not captured in the GDP figures. Finally, some companies grow outside the U.S., which shows up in sales growth but fails to be reflected in GDP.20

Exhibit 6: Median Sales Growth Is Correlated with GDP Growth 15 r = 0.65

Annual Real Sales Growth (Percent)

10

5

0 -5 0 5 10

-5

Annual Real GDP Growth (Percent)

Source: Credit Suisse HOLT® and Bureau of Economic Analysis.

Note: Sales growth is for the top 1,000 global companies by market capitalization in each year.

Finally, notwithstanding our natural tendency to anticipate growth, 23 percent of the companies in the sample had negative sales growth rates for 3 years, after an adjustment for inflation, and 20 percent shrank for 5 years. Whereas a decline in sales need not be bad if it occurs for the right reasons, few analysts or corporate leaders project shrinking sales unless there is a clear strategy of divestiture.21

Using Base Rates to Model Growth

We have established that there are two ways of making a forecast. You can do bottom-up research, which is the most natural method, or you can turn to a base rate. The research in decision making shows that the bottom-up approach is subject to biases and that incorporating the base rate generally improves the quality of the forecast. Yet we don’t want to lean too much on either our own analysis or the base rate. We want to combine the two intelligently.

There is a technique to combine the two approaches, which we will apply to our sales growth data.22 Correlation is the key to the method. The correlation coefficient measures the degree of linear relationship between variables in a pair of distributions. The value of a correlation coefficient can fall between -1.0 (the rise in one variable perfectly correlates with the fall of the other) and 1.0 (both variables move in tandem). A zero correlation indicates randomness. We will examine a single variable, sales growth, measured over time, and all of the correlations are positive.

If the correlation between two distributions is high, then what happened before gives you a really good sense of what will follow. For example, the correlation for cash flow return on investment (CFROI®*) for companies in the consumer staples sector is about 0.90 from one year to the next.23 That means if you know Unilever’s CFROI from last year, you can forecast it this year with a great deal of accuracy. The bottom-up work is highly relevant.

® *CFROI is a registered trademark in the United States and other countries (excluding the United Kingdom) of Credit Suisse Group AG or its affiliates.

If the correlation is low, what happened before provides no inkling of what will happen next. Take the annual total shareholder returns for the S&P 500 as a case.24 The correlation from year to year, from 1928 through 2015, is essentially zero. Telling you last year’s return provides no help in forecasting the return for this year. Your best forecast is the average of the reference class.

The basic idea is that the correlation determines how you should weight the bottom-up analysis and the base rate. For Unilever, a sensible forecast is nine parts last year’s CFROI and one part last year’s average sector CFROI, the base rate. For your S&P 500 forecast, you should place minimal weight on what happened last year and rely largely on the average return since 1928, the base rate.

Studying base rates for sales growth is logical for two reasons. First, sales growth is the most important driver of value for most companies. Second, sales growth has a higher correlation from year to year than does earnings growth, which is the most commonly discussed item on the income statement.25 Sales growth is important and more predictable than profit growth.

Exhibit 7 shows that the correlation coefficient is 0.30 for the year-to-year sales growth rate.26 This includes the top 1,000 global companies by market capitalization from 1950 to 2014. Roughly 50,000 company years are in the data, and all of the figures are adjusted for inflation.

Exhibit 7: Correlation of One-Year Sales Growth Rates r = 0.30 75

60

Sales Growth Next Year (Percent)

Sales Growth (Percent)

Source: Credit Suisse HOLT® and Credit Suisse.

Note: Data winsorized at 2nd and 98th percentiles.

Not surprisingly, the correlations are lower for longer time periods. Exhibit 8 shows the correlations for one-, three-, and five-year horizons for the full population of companies. The base rate for the reference classes, the median growth rate, should receive the majority of the weight for forecasts of three years or longer. In fact, you might start with the base rate and seek reasons to move away from it.

Exhibit 8: Correlation of Sales Growth Rates for 1-, 3-, and 5-Year Horizons 0.40

0.30

Correlation 0.18

(r) 0.16

0.00 1-Year 3-Year 5-Year Period Source: Credit Suisse HOLT® and Credit Suisse.

This approach to modelling regression toward the mean does not say that some companies will not grow rapidly and others will not shrink. We know that companies will fill the tails of the distribution. What it does say is that the best forecast for a large sample of companies is something close to the median, and that companies that anticipate sales growth well in excess of the median are likely to be disappointed.

Current Expectations

Exhibit 1 shows the current expectations for sales growth over three years for a thousand public companies around the world. The median expected growth rate is 1.7 percent. Exhibit 9 represents the three-year sales growth rates, adjusted for inflation, which analysts expect for ten companies with sales in excess of $50 billion. We superimposed the expected growth rates on the distribution of historical sales growth rates for the reference class of mega companies.

Exhibit 9: Three-Year Expected Sales Growth Rates for Ten Mega Companies 30 Unilever Wells Fargo GE 25 Apple

Frequency (Percent)

Wal-Mart 20 Alphabet

CAGR (Percent)

® Source: Credit Suisse HOLT and FactSet Estimates.

Note: FactSet consensus estimates as of February 19, 2016; Growth rates are annualized; P&G = Procter & Gamble, IBM = International Business Machines, and GE = General Electric.

Analysts expect negative sales growth for four of the ten, which corporate actions or commodity prices can largely explain. The standard deviation of growth rates for this small sample is 7.0 percent.

Summary

Active investing requires having a point of view that is different than that of the stock market. Implicit in such a variant perception is a forecast of outcomes that is at odds with what the market price implies.

Research shows that optimism and overconfidence can creep into our forecasts and distort them. This is especially pronounced when the outcomes have personal relevance. Research also shows that incorporating a base rate can improve the quality of our forecasts. Notwithstanding the utility of this method, it remains substantially underutilized.

In this piece we provide the base rates for sales growth rates for a large sample of global companies over a span of more than six decades. We start with sales growth because it is the most important value driver. We then provide a method to integrate our views with base rates to sharpen the quality of our forecasts.

Appendix: Observations for Each Base Rate by Decile (1950-2014)

Source: Credit Suisse HOLT®.

Endnotes 1 Daniel Kahneman, Thinking, Fast and Slow (New York: Farrar, Straus and Giroux, 2011), 249.

2 Alfred Rappaport and Michael J. Mauboussin, Expectations Investing: Reading Stock Prices for Better Returns (Boston, MA: Harvard Business School Press, 2001).

3 Ibid. Growth only creates value when a company earns in excess of the cost of capital. Growth at a negative spread destroys value.

4 Benjamin Lansford, Baruch Lev, Jennifer Wu Tucker, “Causes and Consequences of Disaggregating Earnings Guidance,” Journal of Business Finance & Accounting, Vol. 40, No. 1-2, January/February 2013, 26-54. 5 Kahneman, 257.

6 See Small Business Association, Office of Advocacy, “Frequently Asked Questions,” January 2011 (https://www.sba.gov/sites/default/files/sbfaq.pdf) and Arnold C. Cooper, Carolyn Y. Woo, and William C.

Dunkelberg, “Entrepreneurs’ Perceived Chances for Success,” Journal of Business Venturing, Vol. 3, No. 2, Spring 1988, 97-108.

7 Cade Massey, Joseph P. Simmons, and David A. Armor, “Hope Over Experience: Desirability and the Persistence of Optimism,” Psychological Science, Vol. 22, No. 2, February 2011, 274-281. Also, David A.

Armor, Cade Massey, and Aaron M. Sackett, “Prescribed Optimism: Is It Right to Be Wrong About the Future?” Psychological Science, Vol. 19, No. 4, April 2008, 329-331. For a more detailed discussion of optimism, see Tali Sharot, The Optimism Bias: A Tour of the Irrationally Positive Brain (New York: Pantheon Books, 2011).

8 Michael J. Mauboussin and Dan Callahan, “IQ versus RQ: Differentiating Smarts from Decision-Making Skills,” Credit Suisse Global Financial Strategies, May 12, 2015.

9 Geoffrey Friesen and Paul A. Weller, “Quantifying Cognitive Biases in Analyst Earnings Forecasts,” Journal of Financial Markets, Vol. 9, No. 4, November 2006, 333-365.

10 Jack B. Soll and Joshua Klayman, “Overconfidence in Interval Estimates,” Journal of Experimental Psychology: Learning, Memory, and Cognition, Vol. 30, No. 2, March 2004, 299-314.

11 Itzhak Ben-David, John R. Graham, and Campbell R. Harvey, “Managerial Miscalibration,” Quarterly Journal of Economics, Vol. 128, No. 4, August 2013, 1547-1584.

12 Bent Flyvbjerg, Massimo Garbuio, Dan Lovallo, “Better Forecasting for Large Capital Projects,” McKinsey on Finance, Autumn 2014, 7-13. Also, Bent Flyvbjerg, “Truth and Lies about Megaprojects,” Speech at Delft University of Technology, September 26, 2007.

13 Kahneman, 249.

14 Tesla Motors, Inc. Q4 2014 Earnings Call, February 11, 2015. See FactSet: callstreet Transcript, page 7. 15 Berkeley J. Dietvorst, Joseph P. Simmons, and Cade Massey, “Algorithm Aversion: People Erroneously Avoid Algorithms After Seeing Them Err,” Journal of Experimental Psychology: General, Vol. 144, No. 1, February 2015, 114-126.

16 The sample size is somewhat smaller than 1,000 in the early years but reaches 1,000 by the late 1960s. 17 Most public companies “die” as the result of mergers and acquisitions. See Michael J. Mauboussin and Dan Callahan, “Why Corporate Longevity Matters: What Index Turnover Tells Us about Corporate Results,” Credit Suisse Global Financial Strategies, April 16, 2014.

18 Madeleine I. G. Daepp, Marcus J. Hamilton, Geoffrey B. West, and Luís M. A. Bettencourt, “The mortality of companies,” The Royal Society Publishing, Vol. 12, No. 106, April 1, 2015.

19 Michael H. R. Stanley, Luís A. N. Amaral, Sergey V. Buldyrev, Shlomo Havlin, Heiko Leschhorn, Philipp Maass, Michael A. Salinger, and H. Eugene Stanley, “Scaling Behaviour in the Growth of Companies,” Nature, Vol. 379, February 29, 1996, 804-806. Also, Rich Perline, Robert Axtell, and Daniel Teitelbaum, “Volatility and Asymmetry of Small Firm Growth Rates Over Increasing Time Frames,” Small Business Research Summary, No. 285, December 2006.

20 Tim Koller, Marc Goedhart, and David Wessels, Valuation: Measuring and Managing the Value of Companies, 6th Edition (Hoboken, NJ: John Wiley & Sons, 2015), 126-127.

21 Sheridan Titman, K. C. John Wei, and Feixue Xie, “Capital Investments and Stock Returns,” The Journal of Financial and Quantitative Analysis, Vol. 39, No. 4, December 2004, 677-700.

22 William M. K. Trochim and James P. Donnelly, The Research Methods Knowledge Base, Third Edition (Mason, OH: Atomic Dog, 2008), 166. See http://www.socialresearchmethods.net/kb/regrmean.php. 23 Michael J. Mauboussin, Dan Callahan, Bryant Matthews, and David A. Holland, “How to Model Reversion to the Mean: Determining How Fast, and to What Mean, Results Revert,” Credit Suisse Global Financial Strategies, September 17, 2013.

24 “Credit Suisse Global Investment Returns Yearbook 2016,” Credit Suisse Research Institute, February 2016. 25 Michael J. Mauboussin, Dan Callahan, and Darius Majd, “The Base Rate Book – Earnings Growth,” Credit Suisse Global Financial Strategies, December 16, 2015. Louis K.C. Chan, Jason Karceski, and Josef Lakonishok, “The Level and Persistence of Growth Rates,” Journal of Finance, Vol. 58, No. 2, April 2003, 643-684. Also, Michael J. Mauboussin, “The True Measures of Success,” Harvard Business Review, October 2012, 46-56.

26 We winsorize the top and bottom two percent of the growth rates. Companies with growth rates in the top two percent are generally extremely small firms or firms that engaged in a significant merger and acquisition activity.