The P/B-ROE Model Revisited

Jarrod Wilcox
Wilcox Investment Inc
&
Thomas Philips
Paradigm Asset Management
 Agenda

§ Characterizing a good equity model: Its virtues and uses

§ Static vs. dynamic models

§ The P/B-ROE model: Closed form & approximate solutions

§ Cross-sectional explanation using the P/B-ROE model

§ Cross-sectional prediction using the P/B-ROE model

§ Time-series explanation using the P/B-ROE model

§ Time-series prediction using the P/B-ROE model

2
 What Characterizes a Good Model?

§ Economic realism in its intellectual underpinnings

– Must be grounded in a realistic view of the firm
– Must allow the incorporation of economic constraints
• e.g. Earnings cannot grow faster than revenues in perpetuity

§ Parsimony and computability

– Should require relatively few inputs
– Inputs should be readily available or easily estimated from data

§ Widespread applicability
– Model prices should explain prevailing prices without significant bias
– Model residuals should predict future returns
– Should be applicable in cross-section and time-series

3
 Who Might Use a Good Model?

§ Corporate officers
– If the model can guide them on how best to increase firm value

§ Fundamental analysts
– If the model can help them better evaluate a firm and its management

§ Investment bankers and buyers and sellers of companies

– If the model can generate unbiased valuations

§ Investors
– If the model’s residuals are predictive of future returns

4
 Models in Widespread Use Today
∞
E[ FCFi ]
§ Dividend Discount Model (J.B. Williams, 1938): P0 = ∑
i =1 (1 + k )i
– Intellectual root of almost all models in use today

Free Cash Flow1

§ Gordon Growth model (1962): P =
k−g
– Free cash flows grow at a constant rate in perpetuity

∞
E[(ri − k ) × Bi −1 ]
§ Edward-Bell-Ohlson Equation (1961): P = B0 + ∑
i =1 (1 + k )i
– Apply clean surplus relationship to DDM and rearrange terms

§ Various multi-stage versions of the DDM

– 3 stages model growth, steady state and decline

5
 Static vs. Dynamic Models

§ A static model evaluates price at a point in time

– Estimate inputs at fixed points in time, discount back to get today’s price
– Examples: DDM, EBO

§ A dynamic model evolves some function of price over time

– Some evolve price, others evolve a valuation ratio
– Trajectory must be consistent with the model: a hint of continuous time
– Examples: Options (Black-Scholes), pricing a zero-coupon bond
• Bond price trajectory must be consistent with the yield curve

§ Both static and dynamic models can have the same intellectual roots
– Both ultimately give us a fix on today’s price
– Choice of one over the other is empirical – which works better in practice

6
 A Brief History of Dynamic Models
§ Jarrod Wilcox (FAJ 1984): P/B-ROE model.
– Two stage growth model ,with first phase ending at time T.
– Determine the trajectory of P/B subject to the constraint P/BT=1
– Obtain today’s P/B from trajectory & terminal condition: ln (P / B ) = ( r − k )T

§ Tony Estep (FAJ 1985, JPM 2003): T (or Total Return) model
– Follows P/B-ROE logic, but arbitrarily sets time horizon to 20 years
 r − g  ∆P / B
– Derives and tests a holding period return: T = g +  + (1 + g )
P/B P/B
§ Marty Leibowitz (FAJ 2000): P/E Forwards And Their Orbits
– P/E must evolve along certain paths (orbits) determined by k
– Has implications for current P/E
– Theoretical, no tests of explanatory or predictive power

7
 Our Two-Stage Dynamic Model
§ Firm has two stages – growth phase (t<T) and equilibrium phase (t>T)

§ Distinct growth rates, ROEs, and dividend yields in these two phases

§ Capital structure is time-invariant – firm is self financing

§ Exogenously determined expected return is time-invariant

GROWTH PHASE EQUILIBRIUM PHASE

Growth rate of book = g Growth rate of book = geq
Return on equity = r Return on equity = req
Dividend yield on book = d Dividend yield on book = deq
t
0 T
8
 Economic Intuition and More Notation
§ We evolve P / Bt = Price-to-Book ratio at time t

Trajectory of P/B in the growth

phase must be consistent with k

P/B P / BT

Growth Phase Equilibrium

0 T t

§ Dt = Cumulative dividend process at time t

§ r = Instantaneous ROE = growth + dividend yield on book = g+d

§ k = Required shareholder return, assumed constant for all t>0.

9
 Exact Solution - I

§ Total Return = Price Return + Dividend yield

§ If all parameters are time invariant:

∂Pt ∂Dt
+
∂ t ∂t 1 ∂Pt 1 ∂Dt
§ k = Total Return = = + (1)
Pt Pt ∂t Pt ∂t

§ In addition, we always have Pt = B t × P / B t (2)

§ Differentiate (2) w.r.t. time and divide by price to get

1 ∂Pt 1 ∂P / Bt 1 ∂Bt
§ = + (3)
Pt ∂t P / Bt ∂t Bt ∂t

10
 Exact Solution - II
∂P / Bt
§ Substitute and rearrange to get P / Bt × (k − g ) = +d
∂t
§ Solve this differential equation to give

§ P / B0 = P / BT × e( g − k ) T +
d
(
* 1 − e ( g − k )T )
k−g

§ d / k-g = P/B* = P/B if the initial conditions prevail in perpetuity.

 
 
 P / B0 − P / B * 
§ ln  r − k r − k  = ( g − k )T
 eq − 
 k−g k − g 
 eq 

11
Approximation: All Profits Are Reinvested In Growth Phase

§ Then d=0, r = g, and P / B0 = P / BT × e ( r −k ) T

Impact of Dividends on P/B-ROE Model

16
8
Price/Book
2 1
.5 4

-10% 0% 10% 20% 30% 40%

Expected Return on Equity

d = 0% d = 6%

12
 Approximate Solution: The P/B-ROE Model
§ Take natural log on both sides to get
ln (P / B0 ) = ln (P / BT ) + (r − k )T = [ln (P / BT ) − kT ] + rT
§ In Jarrod’s 1984 paper, P / BT = 1, but this is unrealistic today

§ The P/B-ROE model can be estimated from data via OLS regression
– Can proxy r with ROE, as profitability tends to be stable and mean-reverting.
– Can use analysts’ estimates to further enhance our estimate of r.
– Hard to extract information from constant, so focus on estimating T

§ Run cross-sectional (U.S. stocks) and time-series (S&P 500) regressions

§ Determine fit of regression (cross-sectional & time-series explanation)

§ Use residuals to forecast returns (cross-sectional and time-series prediction)

13
 Cross Sectional Explanation

§ How much should CEO’s expect stock price to increase for each 1%
in additional ROE?

§ Sample: ValueLine Datafile 1988-2002, companies with fiscal year-

ends in December, positive book value, and ROE between -10% and
+40%. Over 20,000 observations.

§ Run panel regression of ln(P/B) against ROE

§ Slope of regression line depends on past volatility of ROE.

§ We interpret this slope as a measure of the investment horizon T.

14
 Long-term Panel Results

1 2 3 4
QUARTILES: Highest 5-year Lowest 5-year
ROE Volatility ROE Volatility

T (years) 2.37 3.87 5.00 6.47

§ The pooled slope within each year of the 1988-2002 period is 3.66
years. For very stable companies it rises to about 9 years.

§ A stable ROE allows projecting recent values further into the future.

§ Independent of risk premium, ROE stability can either help or hurt

through its impact on investment horizon T.

15
 Example Drawn From 1988-2002 Averages

§ Consider a stock with ROE = 15%, and in the 4 th quartile of

ROE stability (5-year standard deviation of ROE < 2.5%).

§ Regression slope (our estimate of T): 6.47 years

§ Question: Other things equal, how much higher would its

price be if its ROE were 20%?

§ Answer: Its stock price would have been 38% higher, not
counting any increase in book value B.

16
How Predictable is the Investment Horizon?

5 4 Annual Pooled Investment Horizon Over Time

ln(PB) vs ROE Slope (years)
2 3

Question for research: Is investment

horizon also a predictable function of
market-wide variables such as the state
1

of the economy?
0

1988 1990 1992 1994 1996 1998 2000 2002

17
 Interpretation of Explanatory Models
§ Across the full sample, R2 = 26%. It approaches 50% for more stable
companies. R2 biased upward by random B, and downward by
pooling across company types and time.

§ Statistical models involving valuation ratios should be translated into

standard errors in log price to judge their merits.

§ Apparent degrees of freedom are inflated because of clustering of

observations by industry. However, they are still very large.

§ Though useful in practice, interpreting slope as T may also incorporate

an errors-in-variables bias from using ROE as a proxy for r.

§ In an efficient market, even a very good explanatory model for prices

may not forecast returns.
18
 Cross Section Prediction

Valuation Residual Return Prediction

For Successive Months after December
.05
0
Correlation
-.05

Questions for Research: How much P, B

and ROE information was available
before December? How late does the
-.1

surprise portion become known? How

would a risk premium look?
-.15

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

19
 Regression Coefficients
1988-2002 December Residuals vs. Future Returns.

Period OLS Coefficient t-statistic Adjusted R2

January – March -.055 -21.2 2.22%

April – June -.022 -8.7 0.38%

July – September -.011 -4.5 0.10%

20
Hypothetical Cross-sectional Return Forecast Success

-12 Ability to Forecast April-June Return Differences

Does P/B-ROE capture periods of

reversion to fundamentals?
-8
t-Statistic
-4
0
4

1989 1991 1993 1995 1997 1999 2001 2003

21
 Cross-Sectional Summary

§ P/B-ROE gives both the company and the market a helpful tool
to calibrate the impact of financial plans on shareholder value.

§ Model residuals have predictive power, and are likely to be a

useful addition to the investor’s toolbox, even before
disaggregating by time and industry.

§ P/B-ROE allows a value approach for growth stocks, and is less

biased against high quality growth than are traditional ratios
like P/E, P/B, P/S, and P/CF.

22
Why Improve Explanatory Models for the S&P 500?

§ To increase market stability by showing relevance of

fundamentals and identifying bubbles.

§ To better show forecasters the impact of changes in

fundamentals.

§ If the market departs from forecastable fundamentals, to

help forecast returns using valuation residuals.

23
 Relevant Structure

§ If ln(P/B) = ln(P/BT) + T * (r - k), comparisons of E/P to interest

rates (so-called Fed Model) are badly mis-specified.
• See also “Fight The Fed Model” by Cliff Asness (JPM, 2003)

§ Changing monetary inflation complicates this picture further

§ Higher rates of inflation both:

• Raise nominal k
• Lower replacement cost profitability and thus r from
reported ROE.

§ We therefore model ln(P/B) as a linear function of ROE,

inflation, and real interest rates.

24
 Model Inputs (updated)
§ ROE: S&P 500’s S E/ S P
§ Inflation: 12 Month CPI % Change
§ Real Interest: Moody’s AAA Yield – 12 Month CPI % Chg.
P/B-ROE S&P500 Model Inputs
.25
.2 .15
Annual Rate
.05 .10

1976 1980 1984 1988 1992 1996 2000 2004

gdate

ROE Real_Interest
Inflation

25
 Full Sample S&P500 Index Model

§ Because of omitted variables, the model errors are highly

autocorrelated. R-squared of the fit is highly inflated.

§ ln(P/B)S&P500 = 1.1 + 6.3 * ROE – 15.9 * Inflation – 8.0 * Real Interest

§ When appraising the model’s hypothetical use as an

prediction tool, it is important to avoid look-ahead bias.
– Use expanding window regression after 5-year warm-up period.

26
 What Does P/B-ROE Tell US? (updated)

S&P500 Actual P/B and Expanding Window Explanation

6
5 4
Price-to-Book Ratio
2 1 3

1976 1980 1984 1988 1992 1996 2000 2004

Actual No_LookAhead

27
Using A Regression Model With Unstable Missing Variables

§ We know that the regression model is not fully satisfied

– The process is not stable

§ Residuals are highly autocorrelated due to missing variables

– Changes in risk preference?
– Changing ROE cross-sectional dispersion?
– Changing taxation?

§ Consequently, we do not assume that correlation automatically

translates into a successful investment decision,

§ But...

28
Correlation: P/B-ROE Residuals vs. 1 month S&P 500 Returns

0.20 Do P/B-ROE Residuals Predict S&P 500 Returns?

0.20
0.10

0.10
0.00

0.00
-0.10

-0.10
-0.20

-0.20
-15 -10 -5 0 5 10 15
Lag

29
Predicting S&P 500 Returns with P/B-ROE Residuals

Std.
Coefficient t P>|t| 95% Confidence Interval
Error
1 month return -0.0418 0.017 -2.41 0.016 -0.0759 -0.0077
3 month return -0.1199 0.050 -2.39 0.018 -0.2188 -0.0210
6 month return -0.2243 0.093 -2.41 0.017 -0.4077 -0.0409

§ All t-statistics are corrected for correlation (Newey-West)

§ Predictions are both economically and statistically significant

30
 P/B-ROE Time-Series Confirms Cross-section
§ Implied investment horizon T against ROE for the S&P500 is similar
to that found in cross-section for stocks in the most stable quartile.

§ When supplemented by allowance for time-varying inflation and

interest rates, P/B-ROE:
– Identifies key fundamentals controlling valuation, useful for planning
– Is structurally different from E/P versus interest rate comparisons
• See “Fight The Fed Model”, Cliff Asness (JPM 2003)

– Provides useful short-term coincident explanation

§ In addition, its residuals also show potential for use as an ingredient

in tactical asset allocation (TAA)

31
 Summary

§ P/B-ROE is both simple and effective for a wide range of problems

§ Some investment managers have used P/B-ROE for many years as

an additional valuation factor...

§ But it not used as widely as it could be:

– By CEO’s, CFO’s and analysts
– And for identifying undervalued growth stocks
– And to better identify bubbles
– And as an ingredient in tactical asset allocation

§ And generally to enhance the importance of fundamentals as

opposed to momentum in investing and pricing.

32