1. Reports of Value’s Death May Be Greatly Exaggerated
PRELIMINARY
NOT FOR DISTRIBUTION
Reports of Value’s Death May Be Greatly Exaggerated
Campbell R. Harvey
Duke University and NBER
(Joint work in progress with Rob Arnott, Vitali Kalesnik, and Juhani Linnainmaa)
November 2019

2. Why has value underperformed growth?
Did crowding reduce expected returns?
Different economic regime?
Different interest rate regime?
Less relative mean reversion?
Is value mismeasured?
Value has lagged because it has become cheaper?

3. Testable Implications
Crowded trade?
Permanently narrow valuation spread
Different economic regime?
Growth permanently more profitable vs. value
Different interest rate regime?
Less relative mean reversion?
Lower rate of price mean reversion
Is value mismeasured?
Potential to fix mismeasurement of intangibles
Value has lagged because it has become cheaper?
Relative valuations would explain the underperformance
Value is structurally
impaired
Value suffered from temporary setback

4. Diagnosing Value

5. Value Investing Is Not New
Graham and Dodd, Security Analysis (1934):
Derive intrinsic value of a company and compare it to the market price. Buy if cheap and sell if expensive.
“In general terms [intrinsic value] is understood to be that value which is justified by the facts, e.g., the assets, earnings, dividends, definite prospects, as distinct, let us say, from market quotations established by manipulation or distorted by psychological excesses. But it is a great mistake to imagine that intrinsic value is as definite and as determinable as is the market price. Some time ago intrinsic value (in the case of common stock) was thought to be the same as “book value,” i.e., it was equal to the net assets of the business, fairly priced. This view of intrinsic value was quite definite, but it proved almost worthless as a practical matter because neither the average earnings nor the average market price evinced any tendency to be governed by book value.”

6. Academic Origins of Value
Basu (1977) — First academic evidence of superior performance of value strategies
Stocks with low P/E (value) outperform stocks with high P/E (growth).
Fama and French (1992) — Risk-based theory of value
P/B becomes a standard academic definition of value.
Lakonishok, Shleifer, and Vishny (1994) — Mispricing theory of value

7. Value Is One of the Strongest Factors United States, Jul 1963–Dec 2018
Year of Discovery
Average
Return
CAPM
Alpha
Market
1964
4.20
Value
1990
4.15
***
5.24
Size
1975
2.53 *
1.32
Operating profitability
2013
3.70
4.74
Investment
2003
4.32
5.80
Momentum
1989
5.48
6.06
Low beta
1966
0.16
3.18
Asness, Moskowitz, and Pedersen (2013)
Value effect is pervasive across geographies and asset classes.
Beck, Hsu, Kalesnik, and Kostka (2016)
Value effect is robust to perturbation across definitions.
Source: Arnott, Harvey, Kalesnik and Linnainmaa (2019). Volatility is ex-post scaled to 10% annualized.
* -- significance at 10%, ** -- at 5%, *** -- at 1%.

8. Value Has Underperformed since 2006
Examine HML — Value vs. growth long/short performance (balanced by size)
In this computation with monthly rebalancing into HML, value most recently peaked at the end of December 2006.
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June HML is high (value) minus low (growth) long/short portfolio balanced by size around the median by the NYSE market capitalization.

9. Second Worst Drawdown
In terms of its depth, the most recent drawdown of value counts as the second deepest since July 1963.
But value is prone to drawdowns and prolonged periods of underperformance — How unusual was this?
Dates
Rank
Event
Start Date
Bottom
End
Length in Years
Drawdown 1
Tech Bubble
1998/9
2000/3
2001/3
2.4
-40.6% 2
Current
2007/1
2019/6
12.5
-36.5% 3
Iran Oil Crisis
1979/8
1980/12
1982/3
2.5
-28.2%
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

10. Estimated Probability of a Drawdown
Use the “Alice in Factorland” bootstrapping methodology to assess the likelihood of the drawdown that started in 2007.
Take the long/short return sample period up to December 2006.
Draw returns from this sample in six-month blocks.
Create a sample that matches the length of the actual total sample from July through June 2019.
For each simulated sample, record the size of the second-largest drawdown.
Draw 200,000 simulated samples.
We take the second-largest drawdown to be consistent with the actual data.
Drawdowns ranked by magnitude are order statistics.

11. Likelihood of Recent Drawdown Magnitude (Six-Month Bootstraps)
The second-largest drawdown in 4.1% of the simulated samples exceeds the actual drawdown of 36.5%.
4.1% is unusual but inconsistent with “broken”
Source: Research Affiliates, LLC, bootstrapping exercise using CRSP and Compustat data. United States, July 1963-June 2019.

12. Is the Value Engine Broken?

13. Value Engine Components
Value minus growth driven by revaluation as well as two structural components:
Migration
Profitability differences

14. How Value Works, on Average — Migration
P/B
Growth
New Growth
Value
New Value t
t+1
Before Rebalancing
t+1
After Rebalancing
Year
Fama and French (2007) — A large share of the value premium comes from:
Value stocks migrating to neutral and to growth.
Growth stocks migrating to neutral and to value.

15. How Value Works On Average — Profitability
P/B
Growth
New Growth
Portfolio
Profitability
Value
New Value t
t+1
Before rebalancing
t+1
After rebalancing
Year
Growth, on average, is more profitable than value, which contributes negatively to value’s return.

16. Migration Rates, Pre-2007 Current Quintile Quintile Next Year Growth 23 4
Value
70%
19% 6% 3% 2%
20%
42%
24% 9% 4%
21%
38%
26% 7%
27%
18%
72%
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

17. Migration Rates, Post-2007 Current Quintile Quintile Next Year Growth3 4
Value
71%
18% 6% 3% 2%
21%
45%
22% 8% 4% 5%
20%
41%
25% 9%
46%
19%
72%
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

18. Migration Rates
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

19. Migration Rates
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

20. Migration Rates Migration rates are virtually indistinguishable.
Pre-2007
Post-2007
Current Quintile
Quintile Next Year
Growth 2 3 4
Value
70%
19% 6% 3% 2%
71%
18%
20%
42%
24% 9% 4%
21%
45%
22% 8%
38%
26% 5%
41%
25% 7%
27%
46%
72%
Migration rates are virtually indistinguishable.
This time is not different.
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

21. Historical Profitability Differences Pre-2007
Growth, on average, is more profitable than value, which contributes negatively to value’s return.
Investors, on average, overpay for earnings.
ROE
Earnings Yield
Growth
18% 5%
Neutral
11% 8%
Value 7% 9%
HML
-11% 4%
H/L
0.37
1.70
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-December 2006.

22. Profitability Differences
Profitability differences pre- and post-2007 are very close.
We paid a little more premium for a little more growth.
This time is not different!
ROE
Earnings Yield
Pre-2007
Post-2007
Growth
18%
16% 5% 3%
Neutral
11% 9% 8%
Value 7% 4%
HML
-11%
-12% 2%
H/L
0.37
0.26
1.70
1.50
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

23. How Big Was Revaluation Alpha Post-2007?

24. Valuation Cycle for the Market
Fama and French (2002) and Arnott and Bernstein (2002)
HML Portfolio
Return ≈ +
Return Due to Change in Valuation
“Revaluation Return”
“Structural Return”
Valuation- Adjusted Return
Source: Arnott, Robert D., and Peter L. Bernstein “What Risk Premium is ‘Normal’?” Financial Analysts Journal, vol. 58, no. 2 (March/April):64–85 and Fama, Eugene F., and Kenneth R. French “The Equity Premium.” Journal of Finance, vol. 57, no. 2. (April):637–659.

25. Portfolio Alpha Decomposition
Generalizing Fama and French (2002) and Arnott and Bernstein (2002)
Alpha due to a change in relative valuation
Averaging roughly zero in the long run given that relative valuation is likely stationary.
Further, relative valuation tends to be mean reverting — revaluation alpha mean reverts.
Contributes significantly to strategy performance in the “short run.”
“Short run” can mean decades!
Structural Alpha: Migration + Profitability differences
Portfolio
Alpha ≈ +
Return Due to Change in Relative Valuation
“Revaluation Alpha”
“Structural Alpha”
Valuation- Adjusted Alpha
Source: Arnott, Robert D., and Peter L. Bernstein “What Risk Premium is ‘Normal’?” Financial Analysts Journal, vol. 58, no. 2 (March/April):64–85 and Fama, Eugene F., and Kenneth R. French “The Equity Premium.” Journal of Finance, vol. 57, no. 2. (April):637–659.

26. Understanding Relative Valuations
P/B
Growth
Value
Year
t+1
t+2
t+3
t+4
t+5
t+6
Expanding Relative Valuations:
Value Underperforms
Narrowing Relative Valuations:
Value Outperforms t

27. Valuation Cycle for the Value Factor
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

28. Realized year-over-year changes in valuation
The Path of Valuations
We measure how HML’s returns can be explained by the path of valuations.
Dependent variable: What was HML’s return in year t?
Explanatory variable 1: What was the value spread between growth and value at the beginning of year t?
Explanatory variable 2: What was the change in the value spread over year t?
where the bars denote the averages up to 2007, and so 𝐻𝑀𝐿 = a = 5.6%.
This formulation assumes that HML’s structural alpha is equal to its long- term historical mean, 5.6%.
𝐻𝑀𝐿 𝑡 = 𝑎 𝑏 1 × 𝑏𝑚𝑠 𝑡−1 0 − 𝑏𝑚𝑠 𝑏 1 × ∆ 𝑏𝑚𝑠 𝑡 − ∆𝑏𝑚𝑠 + 𝑒 𝑡
Structural
alpha
Effect of starting
valuation level
Realized year-over-year changes in valuation
HML
return

29. The Path of Valuations: Pre-2007 Sample
Independent Variable
Estimates
Intercept
0.056
(7.98)
𝑏𝑚 𝑡−1 0
-0.191
(-3.89)
∆bmt
-0.903
(-12.84) N 44
Adjusted R2
80.9%
In the years growth has become more expensive relative to value, value has underperformed.
Changes in relative valuations and starting relative valuations explain more than 80% of value performance.
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-December 2006.

30. Estimating Valuation–Implied Value Alpha
𝐻𝑀𝐿 𝑡 =0.056−0.191 × 𝑏𝑚𝑠 𝑡−1 0 − 𝑏𝑚𝑠 0 −0.903 × ∆ 𝑏𝑚𝑠 𝑡 − ∆𝑏𝑚𝑠
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

31. Structural Alpha – The Value Engine
Use pre-2007 to compute values for the post-2007 sample.
𝐻𝑀𝐿 — Realized annualized value alpha = -2.7%
𝐻𝑀𝐿 — Fitted value annualized alpha implied by valuation path = -3.2%
𝐻𝑀𝐿 − 𝐻𝑀𝐿 — Realized vs. Fitted annualized alpha = 0.4% (t-value = 0.17)
For value minus neutral (pre-2007 alpha = 2.25%)
𝐻𝑀𝐿 — Realized annualized value alpha = -1.3%
𝐻𝑀𝐿 — Fitted value annualized alpha implied by valuation path = -0.6%
𝐻𝑀𝐿 − 𝐻𝑀𝐿 — Realized vs. Fitted annualized alpha = -0.7% (t-value = -0.40)
For neutral minus growth (pre-2007 alpha = 3.31%)
𝐻𝑀𝐿 — Realized annualized value alpha = -1.5%
𝐻𝑀𝐿 — Fitted value annualized alpha implied by valuation path = -1.3%
𝐻𝑀𝐿 − 𝐻𝑀𝐿 — Realized vs. Fitted annualized alpha = -0.2% (t-value = -0.23)
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

32. Structural Alpha – The Value Engine
Use pre-2007 to compute values for the post-2007 sample
𝐻𝑀𝐿 — Realized annualized value alpha = -2.7%
𝐻𝑀𝐿 — Fitted value annualized alpha implied by valuation path = -3.2%
𝐻𝑀𝐿 − 𝐻𝑀𝐿 — Realized vs. Fitted annualized alpha = 0.4% (t-value = 0.17)
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

33. Structural Alpha – The Value Engine
Starting valuation: one of the narrowest in history (19.5th percentile)
Ending valuation: one of the widest in history (95.8th percentile)
It was wider only during the dot-com bubble (2000) and GFC (2009).
The association between changes in valuations and HML in the pre sample substantially explains why value has underperformed post-2007.
Post-2007 return can be fully attributed to revaluation.
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

34. What If Starting In 2007… Mean Reversion Followed Historical Patterns?
Predicted path of valuations and returns starting 2007
Initial return would have been somewhat lower due to narrow valuations.
We compute expected relative valuations by measuring the rate of mean reversion in pre-2007 data.
Year
Expected
Realized
Relative Valuation
Return
2007
-0.039
2008
-0.023
4.9%
0.124
-14.3%
2009
-0.014
5.2%
0.143
-8.3%
2010
-0.008
5.3%
0.091
12.1%
2011
-0.005
5.4%
0.103
-11.1%
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

35. What If Starting 2019… Mean Reversion Followed Historical Patterns?
Predicted path of valuations and returns starting 2019
Returns above the historical average—the current valuations are attractive.
A lot of value locked up in value: If valuations normalized tomorrow, HML would earn a return of 35.3%.
Even conservative halfway mean reversion would imply 16.4% outperformance.
Year
Expected Relative Valuations
Expected
Return
2019
0.290
2020
0.172
12.9%
2021
0.102
9.9%
2022
0.088
8.2%
2023
0.052
7.1%
Source: Research Affiliates, LLC, Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June Estimated information provided for illustrative purposes only.

36. How Much Mean Reversion is Necessary for Value to Outperform Growth?
A modest mean reversion from the 95.8th percentile to the 93.6th percentile implies value breakeven with growth.
Source: Research Affiliates, LLC, Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June Estimated information provided for illustrative purposes only.

37. How This Time Was Different

38. Profitability Concentration
ROE
Earnings Yield
Pre-2007
Post-2007
Growth
18%
16% 5% 3%
Neutral
11% 9% 8%
Value 7% 4%
Small-Growth 6% 1%
Small-Neutral
10%
Small-Value
Big-Growth
19%
25%
Big-Neutral
12%
Big-Value
HML
-11%
-12% 2%
Big-Growth has been unusually profitable.
Growth in profits was fueled by
Globalization
Low interest rates
Higher monopolization and concentration of profits
Are these changes going to mean revert or continue to change in the future?
Is book value the right denominator???
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June 2019.

39. Recent Market Performance United States, July 1963–June 2018
Market performance was unusually high.
% of Samples that Were Worse than Actual
Average Return
Simulation Assuming Normality
Bootstrap
(Resampling Returns
from Realized History)
Factor
1963:7–2006:12
2007:1–2019:6
1m Blocks
12m Blocks
Market
3.8%
5.6%
92.6%
92.7%
93.1%
Source: Research Affiliates, LLC, using CRSP and Compustat data with Arnott, Harvey, Kalesnik and Linnainmaa (2019) bootstrapping methodology. Market excess return is scaled ex-post to 10% annualized volatility.

40. Conclusions Value engine was quite healthy.
Rates of migration on par with history.
Differences in profitability on par with history.
Caveat: large-cap growth has been unusually profitable.
Is book value the right denominator for value???
Post-2007 return can be largely attributed to revaluation!
Starting valuation — one of the narrowest in history (19.5th percentile)
Ending valuation — one of the widest in history (95.8th percentile)
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June Estimated information provided for illustrative purposes only.

41. Conclusions Do we have a growth bubble?
If valuations were to mean-revert today, value will outperform growth by 35.3%.
Even halfway mean reversion would imply 16.4% outperformance!
A modest mean reversion from 95.8th to 93.6th percentile implies value breakeven with growth
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June Estimated information provided for illustrative purposes only.

42. Supplementary Material

43. Value Mismeasurement
Investments in intangible assets such as research and development, unpatented technology, trademarks, software and client relationships are increasingly important as the U.S. evolves from a manufacturing economy to a service-oriented economy.
Accounting systems developed over 100 years ago treat intangibles as expenses rather than valuable long-term investments.
This is the reason that the notion of book value is increasingly irrelevant (intangible expenses are deducted from book value).

44. Value Mismeasurement Implications:
Many “growth” stocks are misclassified and should be neutral or value
Similarly, some “value” stocks are misclassified

45. Value Mismeasurement Repairs:
Capitalize the intangibles and adjust the book value*
Use alternative measures of value
*See Lev and Srivastava (2019)

46. Economic Conditions Value minus growth has a negative beta
It makes sense that in a bull equity market that the value “underperforms”
However, it is important to consider the hedging value
Currently, we are in an inverted yield curve environment – how does value perform over the last seven inversions?

47. Economic Conditions
Market excess returns perform poorly after inversions (which historically have preceded recessions)
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June Estimated information provided for illustrative purposes only.

48. Economic Conditions Value outperforms growth after inversions
Source: Research Affiliates, LLC, using CRSP and Compustat data. United States, July 1963-June Estimated information provided for illustrative purposes only.

49. Economic Conditions
Over 50% of U.S. CFOs believe that low long-term interest rates are bad for the economy according to the Duke-CFO Survey
CFOs see historic low interest rates favoring firms with market power
FANMAG stocks have market power and have accounted for the bulk of the S&P 500 performance.

50. Economic Conditions Are the duopolies sustainable?
Google and Facebook in Advertising
AWS and Azure in the Cloud
Android and iOS in mobile OS
Google and Amazon in consumer AI (personal assistants)
Walmart and Amazon in retail
Amazon and Apple as ecosystems of bundled consumer subscription lifestyles.
See