An international analysis of earnings, stock prices and

bond yields

Alain Durré
Pierre Giot∗

February 7, 2006

ABSTRACT

This paper assesses the contemporaneous relationship between stock prices, earnings and
long-term government bond yields for a large number of countries. The time period of our data
spans several decades. In a time series framework our analysis first tests the presence of a long-
term contemporaneous relationship between these three variables (the so-called Fed model).
Next, we assess if government bond yields play a significant role in the long-run relationship.
Our empirical results question the validity of the Fed model in the sense that we show that long-
term market movements are mainly driven by the earnings yield and not the differential between
bond and earnings yields. As such, our analysis validates the results of Asness (2003) for a
much larger collection of countries while using a dynamic time series (cointegration) framework.
Finally, we also show that changes in long-term government bond yields have a short-term
impact on stock prices.

Key words: stock prices, earnings, long-run relationships, interest rates, inflation, market valuation.

∗ Alain Durré is from the European Central Bank, IESEG-Catholic University of Lille (France) and member of La-
bores (CNRS-U.R.A. 362). Phone:+49 (0) 6913445972. Email: alain.durre@ecb.int. Pierre Giot is Professor of finance
at the Department of Business Administration & CEREFIM, University of Namur, Rempart de la Vierge, 8, 5000 Namur,
Belgium and CORE at Université catholique de Louvain. Phone: +32 (0) 81 724887. Email: pierre.giot@fundp.ac.be.
Send all correspondence to pierre.giot@fundp.ac.be. We would like to thank discussants at the National Bank of Bel-
gium and the European Central Bank for insightful comments. The opinions expressed in this paper are those of the
authors and do not reflect the views of the European Central Bank or other institutions. We thank the editor and an
anonymous referee for very insightful comments which helped to improve the paper. The usual disclaimer applies.
JEL classification: C13, C22, F31, G14

2
1. Introduction

Although neither endorsed nor suggested by the U.S. Fed, the name ‘Fed model’ was coined by
market practitioners in the late 1990’s. According to this valuation model, there is an equilibrium
relationship between the earnings yield of a stock index and the 10-year government bond yield.
In a nutshell, when the earnings yield is below (above) the 10-year government bond yield, the
stock market is supposed to be overvalued (undervalued). Thus the ‘fair value’ for the stock index
should be equal to the earnings index divided by the prevailing 10-year government bond yield.
Strictly speaking, the expected earnings yield should be taken into account and equated to the pre-
vailing bond yield (we come back to this issue later). The main rationale of this model is the use
of a discounted cash-flow model. In a simplified setting, decreasing (increasing) government bond
yields imply a smaller (larger) discount factor, hence a smaller (larger) denominator in the valuation
formula. This should lead to higher (lower) stock prices. Note that this supposes that the other vari-
ables in the valuation formula are not affected by the modifications of the discount factor. A closely
connected model is the Stock Valuation Model of Yardeni (2003).

For the past five years, this simplified valuation model has been severely criticized. Critics argue
that the valuation argument is flawed, in the sense that an element is missing (the risk premium,
which is known to be time-varying) and that the concept of ‘inflation illusion’ is not taken into
account. Indeed, lower bond yields suggest lower anticipated inflation. Hence, firms should witness
smaller earnings growth rates because of a likely decrease in corporate pricing power. Therefore,
when the discount factor decreases, the EPS growth rate should also decrease. This implies that
higher stock prices are not necessarily warranted. As summarized in Lansing (2004): “Investors and
homebuyers appear to be adjusting their discount rates to match the prevailing nominal interest rate.
However, for some unexplained reason, they do not simultaneously adjust their forecasts of future
nominal cash flows, i.e., earnings distributions or imputed rents”. Thus, these critics stress that the
growth rate and discount factor variables are interrelated in the valuation formula. Consequently,
the impact of inflation on stock prices is difficult to quantify as it affects both the numerator and
the denominator of a discounted cash-flow model. Note that it also implies that rising interest rates
are not necessarily a negative factor for stock prices, which was already suggested by Modigliani
and Cohn (1979). Among others, Asness (2000), Ritter and Warr (2002), Asness (2003), Campbell
and Vuolteenaho (2004) and Cohen, Polk, and Vuolteenaho (2005) provide a treatment of these
issues. In the same vein, many academic papers stress the role of valuation ratios (such as the P/E

2
ratio) for fundamental stock market valuation, see e.g. Philips (1999), Campbell and Shiller (1998,
2001) or Asness (2003). In contrast to most papers published in this field however, Thomas (2005)
presents some recent evidence that earnings yield do vary with expected inflation. His analysis tends
to validate the rationale underlying the Fed model. To set the stage for our empirical analysis, we
provide a literature review in Section 2.

Given these recent developments, the goal of this paper is to empirically assess the contempora-
neous relationship between stock prices, earnings and long-term government bond yields for a large
number of countries (Australia, Austria, Belgium, Canada, Denmark, France, Germany, Italy, Japan,
Switzerland, The Netherlands, United Kingdom and the United States). Our analysis involves long-
term data on a quarterly basis. To assess the robustness of our results, we also run the analysis using
expected earnings instead of actual earnings and use a really long-term database for two countries
(the U.K. and the U.S.) for which this kind of data is available. While Asness (2003) already con-
ducted a related analysis for the U.S. stock market in an OLS framework, we use dynamic time series
models (cointegrated VAR models, also called VECM models) which allow a rigorous modeling of
the short-term and long-term dynamics. The latter is important for our study as the presence of a
valid long-term cointegrating relationship between stock prices, earnings and long-term government
bond yields for many countries would lend credence to the Fed model.1

Besides Asness (2003), our analysis is also close to Koivu, Pennanen, and Ziemba (2005) who
assess a restricted Fed model for three countries. In contrast to the latter study, we do not constrain
the weights in the long-term relationship (i.e. we do not claim that it is exactly the differential
between the earnings and bond yield that should be modeled) and indeed show that the long-term
parameters can deviate substantially from what is commonly assumed. In other words, we let the
empirical analysis choose the optimal long-term coefficients in the cointegrating relationship. We
can then test if these weights are equal to the Fed model weights (e.g., in the Fed model the coef-
ficient of the bond yield is exactly equal to -1) and if the long-term government bond yield plays a
significant economic role in the analysis.

While cointegration analysis is harder to implement than OLS, it also offers a much more rig-
orous testing framework. In particular, the long-term and short-term effects can be disentangled
and we can separately assess their influences. For several countries our empirical analysis shows
1 Note that we do not consider time-varying models (for the risk premium) as we focus on the contemporaneous
long-run relationship between stock prices, earnings and long-term bond yields. Hence, our econometric methodology
is closer to Harasty and Roulet (2000) and what some practitioners would like to test than the time-varying models of
Campbell and Shiller (1988 and 1989).

3
that there indeed exists a long-run relationship between earnings, stock prices and government bond
yields. Our analysis is also conclusive when we use expected earnings instead of actual earnings
and when we run the analysis with very long-term data for the U.K. and the U.S. In contrast to the
Fed model however, we show that the long-term bond yield does not really bear on stock markets
valuations, neither from a statistical point of view (the bond yield coefficient is not significant in the
long-run relationship) nor from an economic point of view (the bond yield coefficient is quite small).
In other words, while bond yield variations do impact stock prices in the short-run, the long-run re-
lationship is made up mostly of a linear combination of earnings and stock prices. This analysis
thus supports the view that earnings, and not bond yields, are the main determinants of stock prices
in the long run. Variations in bond yields help explain stock returns in the short term however.

The rest of the paper is structured as follows. Section 2 discusses the pros and cons of the
Fed model. We then present our dataset in Section 3. The cointegration econometric framework
is detailed in Section 4 and the empirical application is discussed in Section 5. Finally, Section 6
concludes.

2. The earnings yield, the bond yield and the Fed Model

This section describes the Fed model and its inputs. We also discuss present value models and
provide a discussion as to why the Fed model could be meaningful or meaningless.

2.1. From the dividend model to the Fed model

e ,
For an investor long one share in a given stock, the expected return from period t to t + 1, HPRt+1
e , and the expected change in the stock
can be expressed as the sum of the expected dividend, Dt+1
e −P :
price, Pt+1 t

e −P
Pt+1 De Pe + Dt+1
e
t
e
HPRt+1 ≡ + t+1 = t+1 − 1. (1)
Pt Pt Pt
e
Let us assume that the expected return is a constant h, i.e. HPRt+1 = h (we briefly discuss time-
varying models below). Rearranging the previous equation, we have:

4
 ·
e + De ¸
Pt+1 t+1
Pt = . (2)
1+h
Solving Equation (2) N periods forward, we get the usual specification for Pt :

" µ ¶i # "µ ¶N #
N
1 e 1 e
Pt = ∑ 1+h
Dt+i +
1+h
Pt+N . (3)
i=1

When N → ∞, the second term on the right hand side of Equation (3) tends to zero:

N µ ¶i
1 e
Pt = ∑ Dt+i . (4)
i=1 1+h

If dividends are expected to grow at a constant rate d, Equation (4) can be simplified into:

e
Dt+1 (1 + d)Dt
Pt = = , (5)
h−d h−d
which is valid if and only if h > d. This is the classical stock valuation model of Gordon (1962).
Equation (5) can also be written as:

δ(1 + d)Et
Pt = , (6)
h−d
where δ is the payout ratio (assumed constant here) and Et are the earnings of the firm at time t.
Finally, the required rate of return is usually expressed as r f + RP, where r f is for example the 10-
year government bond yield and RP is the risk premium demanded by investors (in excess of r f ) to
hold the stock. This finally yields:

δ(1 + d)Et
Pt = . (7)
r f + RP − d
Such a present value relationship sets the stage for the stock yield - bond yield relationship used in
the Fed model.2 In this simplified framework, decreasing bond yields lead to higher stock prices,
provided that the growth rate of earnings is not affected. In the same vein, an upward revision
in expected earnings (or their long-term growth rate) leads to higher stock prices, provided that
the discount rate does not increase when the growth rate of earnings increases. Besides the pure
‘mechanical’ relationship implied by Equation (7), market participants also constantly arbitrage the
2 Note that we formally characterize present value relationships and their econometric framework in Section 4.

5
stock and bond markets. When new money has to be invested and interest rates are low, it is expected
that this money inflow will mostly find its way in the stock market (this is especially true if dividend
yields are high).3 The opposite should be true when interest rates are high. As such, there exists
a substitution effect between stocks and bonds which is strongly shaped by the relationship of the
dividend yield to the bond yield. Another example is the so-called ‘carry trade’, where market
participants take advantage of low interest rates to buy stocks on margin: stock markets indirectly
benefit from a low-rate environment as portfolio managers incur low borrowing costs when buying
shares. When interest rates rise, these portfolio managers sell their shares to put a cap on their rising
borrowing costs.

Besides practitioners’ discussions, many academic studies have also focused on these contem-
poraneous relationships. For example, the relationship between stock prices, dividends and govern-
ment bond yields has been keenly studied by British academics. As indicated in Mills (1991): “the
relationship between equity prices, dividends and gilt edged stocks was once felt by market practi-
tioners in the UK to be of primary importance for forecasting future movements in prices. . . ”. Mills
(1991) also puts forth a cointegration framework to model the stock price index (Pt ), the associated
dividend index (Dt ) and 20-year government bond yields (Rt ).4 Although not set in the cointegrating
framework, the so-called GEYR ratio is very similar. Indeed, the GEYR ratio, or gilt-equity yield
ratio, is defined as the ratio of the coupon yield on long-term government bonds to the dividend
yield on the stock index. Proponents of the GEYR ratio argue that it fluctuates around a central
value and that any deviation from this ‘equilibrium’ state indicates that the stock market is under- or
over-priced. Therefore the current GEYR ratio, i.e. GEY Rt , should have predictive power for fore-
casting future stock index returns. See Levin and Wright (1998), Harris and Sanchez-Valle (2000a)
or Harris and Sanchez-Valle (2000b) for some recent discussions and empirical applications.5

In the United States, the differential between bond yields and appropriately defined ‘equity
yields’ has recently been highlighted with the growing popularity of the so-called Fed and SVM
models. Widely popularized by market practitioners and finance journals (e.g. the Wall Street Jour-
3 As an illustration, the Wall Street Journal on Monday June 6, 2005, ran a story in its Money and Investing column
“Equities shrug off gloom”, quoting a practitioner as saying “To see a market that is this cheap compared to interest
rates, you have to go back 30 to 40 years . . . There’s never been a better moment to buy equities compared to bonds and
that’s a very strong support to the market”.
4 Using data from the U.K., Mills (1991) concludes that these three series expressed in logs, i.e. p = ln(P ), d =
t t t
ln(Dt ) and rt = ln(Rt ), are cointegrated (with 1 cointegrating vector). We come back to the issue of cointegration in
Section 4 as we detail our econometric methodology.
5 Switching to logs, we have that ln(GEY R) = ln(R) − ln(D) + ln(P), or ln(GEY R) = r − d + p. Readers familiar

with the cointegration framework will immediately recognize that proponents of the GEYR ratio indeed state that r, d
and p are cointegrated with ‘constrained’ weights for the long-term relationship set equal to (1, −1, 1).

6
nal, Barron’s,. . . ), the Fed model states that the ratio of the 10-year government bond yield to the
expected earnings yield for the S&P500 index should be relatively stable through time. When this
ratio is below (above) its long-term average, it is believed that the stock market is undervalued
(overvalued) as the earnings yield is particularly high (low). Recent modifications of the Fed model
include the SVM-1 and SVM-2 models introduced by Yardeni (2003). These models forecast a
‘fair value’ for the S&P500 based on the 10-year bond yield and earnings. They also motivate asset
allocation decisions based on the perceived degree of over and undervaluation of the S&P500 with
respect to its ‘fair value’.6 For example, Lander, Orphanides, and Douvogiannis (1997) present the
Fed model as:

Et
= a + Rt , (8)
Pt
where a is an intercept (or a constant risk premium) and Rt is a nominal bond yield. As underlined by
Vila-Wetherilt and Weeken (2002), Equations (6) and (8) are strongly related if we assume δ = 1 and
d = 0. This discussion shows that the Fed model approach is very similar to the GEYR framework,
with (anticipated) earnings instead of dividends and 10-year government bonds instead of gilts. Next
we look at the pros and cons of this simplified approach.

2.2. Why the Fed model could or could not be relevant?

The rationale underlying the Fed model has been discussed in the academic literature for the last five
years. For example, Lander, Orphanides, and Douvogiannis (1997), Asness (2003) or Campbell and
Vuolteenaho (2004) highlight the pros of this model, although they disagree regarding the way the
model should be interpreted. First, portfolio managers do arbitrage the equity and bond markets and
carry trades are much used. Secondly, this model is more or less in agreement with present value
models. Thirdly, the recent empirical evidence supports the rationale of the Fed model: the equity
yield has somewhat tracked the government bond yield over the last thirty years. As indicated
in Campbell and Vuolteenaho (2004), “the Fed model has been quite successful as an empirical
description of stock prices. Most notably, the model describes the rise in stock yields, along with
6 Note that the Yardeni (2003) SVM-1 model is exactly the model specified by Equation (7) when RP = d and δ = 1.
Yardeni (2003) recently introduced the so-called SVM-2 model to alleviate concerns regarding the RP = d and δ = 1
constraints and discuss the risk premium problem. This second model still hinges on the differential between the 10-year
government bond yield and the earnings yield.

7
inflation, during the 1970’s and early 1980’s, and the decline in stock yields during the past 20
years”.

Despite the recent empirical evidence the Fed model has also been severely criticized. Indeed,
there is some confusion regarding the role of inflation as the earnings yield (expressed in real terms)
is set equal to a nominal bond yield. This is neatly summarized in Lansing (2004) who echoes As-
ness (2000) and Modigliani and Cohn (1979) and shows that this leads to ‘expectational errors’. This
is also at odds with the empirical evidence that shows that equities could be a good hedge against
inflation, see e.g. Marshall (1992), Boudoukh and Richardson (1993), Anari and Kolari (2001) and
Spyrou (2004). More recently, the money illusion argument has also been studied by Campbell and
Vuolteenaho (2004) and Cohen, Polk, and Vuolteenaho (2005). They show that, although the Fed
model tends to fit the recent data quite well, this model is on very shaky grounds with few theoret-
ical justifications for its relevance. In the same vein, Ritter and Warr (2002) highlight two possible
problems regarding the Fed model. On the one hand, the discount rate is not adjusted for risk, which
yields capitalization rate errors (as defined by Ritter and Warr, 2002). On the other hand, when
focusing solely on the earnings growth without any adjustment for the firms’ wealth given specific
cases (e.g. due to changing inflation environment), a debt capital gain error is made. Therefore
the potential capital gain that may result from the reduction of the real value of the firm’s debt in
presence of inflation is not taken into account. More recently however, Thomas (2005) argues quite
convincingly that forward earnings yield should vary with expected inflation. His analysis is thus at
odds with most of the recent literature on this topic and is based on descriptive analysis and account-
ing simulations. In a nutshell, Thomas (2005) contends that accounting earnings include inflation
holding gains, which introduces a link between expected earnings and inflation.

Another potential problem stems from the fact that the risk premium is not really taken into
account in the simplified Fed model framework. More precisely, possibly time-varying risk premia
à la Campbell and Shiller (1988, 1989) are not taken into account as the relationship focuses solely
on the contemporaneous links between the variables. Finally, there is also some confusion regarding
the role of interest rates in the determination of stock prices and anticipated returns. For example,
Philips (1999), Campbell and Shiller (1998, 2001) or Jones, Wilson, and Lundstrum (2002) show
that valuation ratios are the main determinants of future stock price performance; prevailing bond
yields do not enter the relationships. In contrast, the Fed model takes as input the nominal bond
yield to set the ‘right’ stock index price (if the nominal bond rate would decrease to 1% for instance,
the ‘right’ P/E ratio would be at 100). Besides, in the recent 2001-2003 period, fears of deflation

8
have depressed the stock markets, while decreasing interest rates should have spurred the markets
according to the Fed model. Actually, there is a growing consensus that valuation ratios (such
as the price earnings ratio) strongly matter for the future long-term stock market outlook. In this
framework, the long-run equilibrium relationship should only involve earnings and stock prices,
interest rates should not be an input in the model. We focus on this research agenda in Sections 4
and 5.

3. The dataset

The empirical part of the paper focuses on thirteen countries: Australia, Austria, Belgium, Canada,
Denmark, France, Germany, Italy, Japan, Switzerland, The Netherlands, United Kingdom and
United States. As far as stock prices, earnings and long-term interest rates are concerned, reli-
able data has been available for this group of 13 countries since the early seventies. Therefore, the
sample period of our analysis ranges from January 1973 to December 2004 (quarterly data). Taking
into account the international and historical perspectives of this paper, we rely on data vendors that
ensure that the data is harmonized across countries. For the equity variables (stock indexes and
corresponding earnings), the primary source is Thomson Financial Datastream (TFD). More pre-
cisely, we use the stock and earnings harmonized indexes (as computed by Datastream) to facilitate
the comparison between countries. For example, the stock indexes are the so-called total market
indexes of the given country. The source for the long-term interest rate is the IMF International
Financial Statistics. The selected long-term interest rate is equivalent to the yield-to-maturity of
long-term government bonds, i.e. a 10-year yield. To switch from nominal stock prices and earn-
ings to real stock prices and real earnings, we first download the consumer price index (CPI) series
for each country from the harmonized OECD dataset.7 In a second step, the stock index and earn-
ings series are deflated accordingly. Therefore, for each country, we have six quarterly series: the
nominal stock index, the real stock index, the nominal earnings index, the real earnings index, the
long-term government bond yield and the inflation index normalized at 1 in 1973:01.

The global indexes supplied by TFD feature all the stocks of the given country. They are thus
more relevant than the more narrowly-defined (and better-known) S&P500, CAC-40 or DAX-30
indexes (henceforth called the standard indexes). The correlation of the global indexes with the
7 As these series are not seasonally adjusted, we compute seasonally adjusted CPI series using the Census X-12

ARIMA method run by the EViews 4.0 software.

9
standard indexes is very high (typically larger than 0.9). In contrast to Lander, Orphanides, and
Douvogiannis (1997), we use the current earnings and not the expected earnings because the latter
are not available for so many countries and for such a large time frame. Indeed, expected earnings
(such as provided by the I/B/E/S database) have only been available for a much shorter time period
(this is particularly relevant for the European countries). As our paper features data spanning three
decades for a very large number of countries, we thus cannot use the expected earnings. Note
that Koivu, Pennanen, and Ziemba (2005) use the same kind of dataset, i.e. they also focus on
current earnings and not expected earnings. In the last part of the paper, we nevertheless assess
the robustness of our results using a proxy for anticipated earnings. More precisely, we use the
actual 1-year ahead earnings as the anticipated earnings in the models.8 Finally, we further check
the long-term robustness of our results by looking at two countries for which really long-term data
is available, i.e. the U.K. and the U.S. This however necessitates a change of database. For these
two countries, we use the information provided by Global Financial Data, a data vendor of long-
term time series. For the U.K. and the U.S., we thus have all series starting from 1958 and 1871
respectively.9

4. The Fed model in the cointegration framework

4.1. Why cointegration instead of OLS?

In the existing empirical literature, the Fed model is usually tested using ordinary least squares
methodology (OLS). For example, Asness (2003) estimates the following model:

∆Pt = α + β1 (EPt − Rt ) + εt (9)

where ∆Pt is the ten-year real returns for the Standard&Poors 500 index, α is an intercept and
EPt − Rt is the contemporaneous difference between the earnings yield (EPt ) and the ten-year US
Treasury bond yield (Rt ). Over the 1881-2001 time period, Asness (2003) shows that the weak
statistical significance of the β1 coefficient (its t-statistic is equal to 1.41) is due to the inclusion of the
earnings yield in the expression. When the earnings yield and the ten-year bond yield are included
8 We thank the anonymous referee for this suggestion.
9 Incontrast to dividends, long-term data for earnings are not available for the other countries in our dataset. Strictly
speaking, the earnings series is available from Global Financial Data since 1933, however the 10-year government bond
yield has only been available since 1958. Hence we set the starting date at 1958 for the U.K.

10
separately, EPt is strongly significant (with a t-statistic of 4.13) while Rt is not significant (with a
t-statistic of 0.06). Besides, the switch from Equation (9) to the new model improves substantially
the explanatory power of the regression as the adjusted R2 increases from 11.9% to 30.1%. In a
related OLS framework, other authors predefine the weights for the variables and then assess the
forecasting properties of the combination of variables (e.g. the forecasting performance of the P/E
ratio, as in Campbell and Shiller, 1998, 2001, or the GEYR ratio as in Harris and Sanchez-Valle,
2000).

From an econometric point of view, an OLS analysis of the Fed model is ‘restrictive’ in the sense
that it has a hard time combining the long-term and short-term dynamics if these are present in the
analysis. Indeed, OLS regressions that do not include dynamic time series effects cannot disentangle
and properly assess the short-run and long-run dynamics. From a finance point of view, it is crucial
that both effects be separated. Indeed, even if bond yields do not impact stock prices in the long
run (i.e. what is argued by those who criticize the Fed model), we cannot rule out the fact that
maybe variations in long term bond yields affect stock prices in the short run. This could explain
the popularity of the Fed model in the financial press.

In this paper, we jointly analyze these short-term and long-term effects within the cointegra-
tion framework (dynamic time series models - VECM models). Originally developed by Engle and
Granger (1987), the cointegration methodology allows an assessment of possible long-term relation-
ships between economic or financial variables while explicitly modeling the short-term dynamics.
Put simply, a cointegrated model lets the future time path of a variable be governed by a long-term
equilibrium and a short-term dynamics. To our knowledge, with the exception of Mills (1991) in
his study on the GEYR ratio and recently Koivu, Pennanen, and Ziemba (2005), there have been
no cointegration studies of the Fed model. Even Koivu, Pennanen, and Ziemba (2005) do not run a
complete cointegration analysis as they constrain the coefficients of the long-term relationship (see
below).10
10 Regarding the relationship between earnings and stock prices, tests of present value relationships (as pioneered by

Campbell and Shiller, 1987, 1988) have become very popular. While the early paper of Campbell and Shiller (1987) did
not get meaningful cointegration results (using stock prices and dividends as input variables), MacDonald and Power
(1995) validate the present value relationship between earnings and stock prices for the U.S. market. They suggest that
earnings, and not dividends, should be included in the analysis (they argue that both dividends and retained earnings,
which sum to earnings, should be taken into account). More recently, the international analysis conducted by Harasty
and Roulet (2000) also supports the cointegration hypothesis.

11
4.2. The cointegrated model

The Fed model in the cointegration framework Let us illustrate the methodology by focusing
directly on the variables involved in the Fed model. For each country, the variables are et = ln(Et ),
the log earnings index, pt = ln(Pt ), the log stock index and rt = ln(Rt ), the log government bond
yield.11 Provided that there is one cointegration relationship among the three variables, the cointe-
gration model can be written as:12

∆et = γe + αe (et−1 + β p pt−1 + βr rt−1 ) + Short-run dynamics for e + εe,t , (10)

∆pt = γ p + α p (et−1 + β p pt−1 + βr rt−1 ) + Short-run dynamics for p + ε p,t , (11)

∆rt = γr + αr (et−1 + β p pt−1 + βr rt−1 ) + Short-run dynamics for r + εr,t . (12)

These equations illustrate the rationale of a cointegration model. For example, the change in stock
prices (∆pt ) is driven both by the past disequilibrium in the long-run relationship et−1 + β p pt−1 +
βr rt−1 and a short-run dynamics.

Interpretation of the coefficients What should β p and βr be if the Fed model is correct? Taking
the log of the supposed Fed model relationship Et−1 /Pt−1 = Rt−1 gives et−1 − pt−1 − rt−1 . Hence,
if the Fed model is valid, we should have β p = −1 and βr = −1. If the Fed model is only partially
valid in the sense that long-term government bond yields do not really matter while the bulk of the
adjustment comes from the earnings and stocks prices, then β p should be significantly negative (but
not necessarily equal to -1) and βr should not be significant.

In the empirical application, we test these hypotheses (in contrast, Koivu, Pennanen, and Ziemba,
2005, impose these constraints, i.e. they suppose that the Fed model is true). Regarding the short-
∗ ∗ ∗
term dynamics, it can be expressed as ∑kj=1
−1
δe, j ∆et− j + ∑kj=1
−1
δ p, j ∆pt− j + ∑kj=1
−1
δr, j ∆rt− j , i.e. a
weighted sum of past changes in the three variable (k∗ − 1 being the optimal number of lags to
include).
11 An alternative specification would include pt = ln(Pt ), et = ln(Et ) and Rt , and not rt = ln(Rt ), as inputs. We
however prefer to work with the log government bond yield as taking the log of the supposed Fed model relationship
Et /Pt = Rt gives et − pt − rt . As detailed in the equations given below, this is thus the supposed long-run relationship
if the Fed model is valid. The log specification is also suggested in many papers, see e.g. Campbell and Shiller (1989),
Timmermann (1995), Campbell, Lo, and MacKinlay (1997) or Koivu, Pennanen, and Ziemba (2005).
12 In theory, there could be up to 2 cointegration relationships. Anticipating on the empirical results, we always have

0 or 1 cointegration relationship, hence we do not detail the specification which features 2 cointegration relationships.

12
 In the cointegration literature, the αe , α p and αr coefficients are called the adjustment speeds, as
they determine how each variable is affected by the disequilibrium in the lagged long-run relation-
ship et−1 + β p pt−1 + βr rt−1 . For example, if α p was equal to 0, the dynamics of pt would solely be
governed by the short-run effects. Economic good sense suggests a positive α p if β p is negative: if
stock prices increase more than warranted by earnings increases, et−1 + β p pt−1 + βr rt−1 becomes
negative and the system should ‘correct’ by having stock prices decrease.

Given this discussion and the literature review of Section 2, we suggest three testable hypotheses
regarding the validity or partial validity of the Fed model:

Hypothesis 1: There is a cointegration relationship between earnings, stock prices and government
bond yields.

Hypothesis 2: The cointegrating relationship of hypothesis 1 implies that a deviation from the
long-run equilibrium impacts stock prices such that the equilibrium is restored.

Hypothesis 3: Although there is a cointegration relationship, government bond yields do not play a
significant role in the long-term relationship.

Note that hypothesis 2 is key to the adjustment process. For example, it predicts that, if stock
prices are too high with respect to the equilibrium level fixed by the earnings and bond yields, they
should decrease in the near future. Hypothesis 3 modifies hypothesis 2 in the sense that the bond
yield no longer influences the ‘return to equilibrium’ of stock prices. It was also the main focus of
Asness (2003) but we test it here within the error-correction framework. Hypothesis 3 also looks at
the validity of constraining (as in the strictest version of the Fed model and as assumed in Koivu,
Pennanen, and Ziemba, 2005) the long-run relationship to be exactly et−1 − pt−1 − rt−1 .

Unit root tests and the cointegration test Prior to estimating this model, we must first test that the
log variables are integrated of order 1. This is done via the usual augmented Dickey-Fuller (ADF)
unit root tests with constant and/or trend included in the specification. Then we must of course test
for the presence of cointegration. Indeed, if the et , pt and rt variables are not cointegrated, then it
makes no sense to estimate such a VECM model. We use the Johansen trace and Max Eigenvalues
tests.13 Additional details are presented in Enders (1995), Brooks (2002) or Harris and Sollis (2003).
13 The number of optimal lags (k∗ ) included in the multivariate model is set such that the last included k∗ + 1 lagged
variables in the VAR specification are jointly non significant. We also check for autocorrelation and absence of normality
in the residuals and look at the AIC criteria. Because cointegration tests are known to have relatively low power in small
samples and can depend on the chosen k∗ , we also perform the cointegration tests for k = k∗ + 1 and k = k∗ − 1.

13
Assessing the possible contemporaneous influence of bond yields on stocks prices To assess
this influence, we estimate the single-equation ECM model for the stock price adjustments. This
model can be written as:

∆pt = γ p + α p rest−1 + δe,0 ∆et + δr,0 ∆rt + Short-run dynamics for p + εt , (13)

where rest are the residuals from the previous (and separate) estimation of the long-run relationship
∗ ∗ ∗
and the short term dynamics is ∑kj=1
−1
δe, j ∆et− j + ∑kj=1
−1
δ p, j ∆pt− j + ∑kj=1
−1
δr, j ∆rt− j . Note that we
use the same Greek letters for the coefficients as in the VECM model, but of course they will take
different numerical values.

Regarding the short-term dynamics of the model, coefficient δr,0 is important as it shows how
contemporaneous changes in the bond yield affect the stock prices (we suspect that this coefficient
will turn out to be significantly negative). In the short-run dynamics, coefficient δ p,1 highlights the
momentum effect in stock prices as it summarizes the influence of the past returns (it should not be
statistically different from zero if the stock market is weakly efficient).

A key feature of the single-equation ECM is that it features contemporaneous terms on the right-
hand side. As such, this type of model is not much used in forecasting applications as scenarios for
the right-hand side variables must be made prior to computing the forecasts. In contrast, the VECM
only features lagged variables on the right-hand side. The single-equation approach is often used by
financial institutions which use that kind of model in conjunction with a scenario analysis.

5. Empirical analysis

5.1. A long-run equilibrium?

We report the cointegration results for all countries in Table II (nominal data) and Table III (real
data).14 Prior to the cointegration analysis, we also ran augmented Dickey-Fuller unit root tests on
the nominal and real series (and on their first differences). These results are displayed in Table I.
The unit root results are similar to those previously documented in the literature (e.g. Harasty and
Roulet, 2000; Koivu, Pennanen, and Ziemba, 2005). Indeed, all series exhibit a unit root. When a
14 We report the cointegration trace tests. The Max Eigenvalues tests deliver the same results and are not reported to
save some space in the tables.

14
constant and a trend are both included, a few series fail the test at the 5% level. Nevertheless a visual
inspection of those cases does not invalidate the analysis and we therefore proceed similarly for all
countries.15

Among the group of 13 countries (nominal data), eight countries are characterized by exactly
one cointegration relationship, while 5 countries (Austria, Belgium, Germany, Italy and Japan) do
not exhibit any cointegration. Regarding the real stock prices, real earnings and long-term bonds,
seven countries feature exactly one cointegration relationship. In this case, there are thus 6 countries
(Australia, Austria, Belgium, Germany, Italy and Japan) that do not exhibit any cointegration.16 For
the countries that do exhibit cointegration, we report the coefficients (β p , βr ) that enter the long-run
relationship et−1 + β p pt−1 + βr rt−1 . Besides, we also give the adjustment speeds (αe , α p and αr
for the nominal data; αer , α pr and αr for the real data). These results tend to support hypothesis
1 of Section 4. They are also consistent with previous results, e.g. MacDonald and Power (1995)
for U.S. data, Harasty and Roulet (2000) or more recently Koivu, Pennanen, and Ziemba (2005) for
three countries.17

To illustrate, for two countries that exhibit one cointegration relationship (the U.K. and the U.S.),
we plot the long-run equilibrium relationship (called cr) vs time in the bottom of Figure 1. For each
of these figures, the top figure shows the earnings yield, while the middle figure presents the ratio
of the earnings yield to the long-term government bond yield. A look at the long-run relationships
visually confirms that the cointegrating vector is stationary. Note that the cycles are quite long,
which supports the view that a meaningful cointegration analysis needs a large time sample. These
graphs also show that the long-run relationship is not dissimilar to either the ratio of the earnings
yield to the long-term government bond yield (middle figure) or the earnings yield (top figure), but
at the same time it is distinctively different.18
15 Unit root tests on the first differences of the series tend to confirm the unit root hypothesis for the original series.
16 The significance level of the cointegration is reported in the P column, while the number of lags in the VECM
system is given in the Lags column.
17 Regarding Belgium and Italy, they exhibited cointegration on a smaller sample. Note also that Japan is a very

difficult market to model given 15 years of bull market followed by 15 years of bear market, with some deflation.
Indeed, on October 9th, 2005, the Financial Times in its “Arguments persist over assessing valuations” story discussed
the Fed model and also noted that Japan seemed at odds with other countries.
18 To save space, we present these graphs for only two countries. Graphs for the other countries are nevertheless

similar and are available on request.

15
5.2. Deviations from the equilibrium

As far as hypothesis 2 is concerned (does a deviation from the equilibrium in the long-run rela-
tionship impact stock prices in such a way that the system reverts to equilibrium?), the evidence
is mostly conclusive, although some coefficients are not significant. Indeed, coefficient β p is neg-
ative and α p is positive, although not significant in some cases. To illustrate, this implies that, if
stock prices rise more than warranted by the rise in earnings, the system gets out of equilibrium
(et−1 + β p pt−1 + βr rt−1 becomes negative) and the positive adjustment speed (α p ) channels this
effect into a negative price adjustment at the next period. For example, the U.K. is characterized
by β p = −0.64 and α p = 0.17, while these coefficients are respectively equal to -0.74 and 0.17 for
the U.S. If β p was exactly equal to -1, then the log earnings yield would exactly enter the coin-
tegration relationship. This would indicate that high (low) P/E ratios would lead to poor (good)
future stock market performance. Although we do not have β p = −1, the combination of a negative
β p and positive α p indicates that high stock prices with respect to earnings do lead to poor future
stock market performance. This supports hypothesis 2 and the conventional wisdom prevailing for
stock market performance and high/low stock prices to earnings ratios. It should be stressed that
the set of long-term coefficients is very far from (1 -1 -1), i.e. the set of weights posited by the Fed
model and constrained as such in Koivu, Pennanen, and Ziemba (2005) (see below for an additional
discussion).

To further highlight the possible stock index adjustment to the level of the cointegration relation-
ship, we plot scatter plots of 3-, 12-, 24- and 60-month forward-looking returns vs the value of the
cointegration relationship (at the time the return is computed). Campbell and Shiller (1998, 2001)
present similar graphs for forward-looking returns vs P/E ratios. If the valuation argument is correct,
we expect that low (high) values for the cointegration relationship indicate overvalued (undervalued)
markets. Thus these values should lead to negative (positive) forward-looking returns and hence the
scatter plots should trend upwards. We plot these scatter plots for two selected countries (United
Kingdom and United States) in Figures 2. The shape of the scatter plot is as expected. Note however
that these graphs are only illustrative as it is known that overlapping returns tend to mechanically
lead to such patterns.

16
5.3. Are bond yields relevant?

Let us now focus on hypothesis 3, i.e. whether long-term government bond yields are economically
and/or statistically relevant in the equilibrium relationship. The P-value for this test (χ2 (1) LR test)
is reported in the H0:βr = 0 column. To better understand the role of the bond yield, we also report
the estimation results from a constrained cointegration analysis, i.e. the estimation results from
the VECM where the βr coefficient is constrained to be equal to zero. Indeed, if the bond yield
does not really matter in the cointegration analysis, we expect that the exclusion of the bond yield
from the long-run analysis cannot really impact the other coefficients. More formally, this yields
a new cointegrating vector et−1 + β p pt−1 , which only takes the log earnings (or real earnings) and
stock prices as inputs. In this constrained framework, the bond yield is absent from the long-run
relationship, which is thus shaped by an appropriately defined linear combination of the stock price
and earnings index. Note of course that bond yields still enter the system through the short-term
dynamics.

For all countries that exhibit cointegration, the long-term bond yield coefficient is not signifi-
cantly different from zero in the cointegrating space (according to the LR test). The U.S. is the only
country for which the long-term bond yield somewhat matters (at 9%) in the cointegration relation-
ship. For the real data, the government bond yield is never significant in the long-run relationship,
with large P-values for the H0:βr = 0 test. The results also indicate that the bond yield does not
matter much in an economic sense: the coefficients are not significant and they take low values for
all countries (the United States seems to be the exception, with a coefficient equal to -0.4). This
discussion leads us to accept hypothesis 3. It also invalidates the ‘second’ part of the Fed model, i.e.
the bond yield should not enter the long-run equilibrium relationship as posited in Equation (8). In
contrast to Et /Pt = a + Rt , we thus have that the linear combination of ln(Et ) and ln(Pt ) is stationary
(with weights appropriately defined by the cointegration analysis), and the Rt term is not needed.
Regarding the literature discussed in Section 2, our estimation results are similar to Asness (2003).
His results do not however hinge on the cointegration framework and only focus on U.S. data. Our
results are also in agreement with Siegel (2002) (the equity yield and/or dividend yield is a strong
determinant of future long-run stock market performance). Besides, this also shows that (1 -1 -1)
should not be imposed as the set of long-term weights as this is clearly rejected by the model.

To illustrate a practical application of the model, let us rewrite the cointegration relationship
such that P becomes the left-side variable and is thus the ‘fair value predicted by the model’. Indeed,

17
when the system is at equilibrium, the cointegration relationship should be equal to 0 (e + β p p +
βr r + constant =0) and we can solve for the ‘right’ p given the prevailing e and r. This can be
done with both the unconstrained and constrained long-run relationships. Let us illustrate with the
United States. For this country, the results in the table indicate that the unconstrained equilibrium
relationship is ln(E) − 0.74ln(P) − 0.4ln(R) + 2.03. Rewriting this relationship with respect to P
gives P∗ = exp((ln(E)−0.4ln(R)+2.03)/0.74). This P∗ can thus be interpreted as the stock market
value given the prevailing earnings (E) and long-term interest rate (R).

When the bond yield is explicitly left out of the cointegration relationship, the new (constrained)
relationship is ln(E) − 0.62ln(P) + 0.57. When the system is at equilibrium, this should again be
equal to 0, which yields P∗∗ = exp((ln(E) + 0.57)/0.62). We plot these P∗ and P∗∗ , along with the
actual P, for the United States in Figure 4 and United Kingdom in Figure 3. An assessment of these
figures shows that the fit is quite good. Moreover, P∗ and P∗∗ are quite close, which is in agreement
with our third hypothesis.19 A look at the R2 (given in the last column of the two tables) shows that
any stock market forecasting exercise will have a hard time at being economically (or financially)
significant, at least on a quarterly basis. Indeed, the R2 is between 5% and 13%, with Denmark
being the exception with a higher R2 of 16% (nominal data). These R2 levels are consistent with
results previously given in the literature, taking into account the fact that the left-hand side variable
of the VECM is a stock return.

5.4. Short-term dynamics

The discussion above has focused on the long-run relationship and how the system reacts when it
gets out of balance. For market participants, short-term effects are, however, important and we can-
not disregard the fact that bond yields variations maybe do matter on a short term basis. Fortunately,
there are a number of time series tools that allow an exploration of the short-term dynamics in coin-
tegration models. In this section, we first focus on the variance decomposition of the log stock index
to ascertain if the bond yield could partially explain the dynamics of ln(P) in the short-run. In a sec-
ond step, we estimate the single-equation ECM (as given by Equation (13)) to look at the possible
contemporaneous influence of changes in the bond yield on ln(P).
19 In
an earlier version of the paper still available as a discussion paper, we conducted that exercise for other countries
and we concluded similarly.

18
 The results from the variance decomposition are presented in Tables IV and V. In both tables,
the left panel is for the nominal data, while the right panel is for the real data.20 The variance of the
stock price n quarters ahead (in the table we present results for n being successively equal to 1, 4 and
20 quarters) is mainly explained by its own innovations; earnings innovations do not matter much.
Regarding the long-term bond yields, albeit their influence was weak in the long-run relationships,
they do impact the variance of the stock prices in the short-run, whatever the variable ordering may
be. This result could be consistent with arbitrage effects and/or carry trades that could take place
in the short run. Moreover, even if bond yields do not really matter for long-term stock market
valuation, increasing bond yields tend to raise the cost of borrowing (for example for investors who
bought stocks on margin), which could lead some investors to unwind speculative positions.

Since most of the long-term bond yield fluctuations are known to be explained by the inflation
rate, the size of the bond yield’s impact on the variance of the stock price might be related to the
country’s history in terms of inflation stabilization. Taking into account the results of Tables IV
and V, the countries can be rearranged into three groups regarding the impact of the bond yield on
stock prices: very low but stable impact, very large but stable impact and a intermediary category
with moderately large impact. Switzerland is in the first group. This is consistent with its stable
history in terms of monetary policy objectives.21 Not surprisingly, the United States, the United
Kingdom and Canada are in the second group. Indeed, these countries were affected by monetary
policy ‘instabilities’ over the sample. This is particularly true for the United States, as suggested by
Favero and Mosca (2001) and Clarida, Gali, and Gertler (2000). By anchoring its exchange rate to
the U.S. dollar, Canada imported the same instability. For the third category of remaining countries,
the impact is moderately large.

The estimation results given in Table VI also contribute to this discussion. This table pertains
to the estimation of the single-equation ECM as expressed in Equation (13). For the short-term
analysis, the ∆rt column is particulary interesting, as it gives the impact of the contemporaneous
change in the bond yield on the change in the stock price. For Switzerland and the Netherlands,
the contemporaneous effect is weak and not significant. For the other countries, the impact is sta-
tistically significant, although the range of the ∆rt coefficient is quite large. Canada and the United
Kingdom feature the largest effect, while Australia and France rank second. The impact is lowest for
Denmark and the United States. This table also shows that there is no ‘momentum’ effect (save for
20 Because the results are similar for the nominal and the real data, we focus on the discussion of the nominal data.
21 Note also that Switzerland is the country with the lowest (among our sample of 13 countries) real long-term interest
rate over the sample period.

19
Denmark, in the terminology of Harasty and Roulet, 2000) for the stock market, i.e. no significant
AR(1) effect for the stock returns, which is consistent with the weak form of the Efficient Market
Hypothesis (EMH).

5.5. Expected earnings

The Fed model specifies that the current long-term bond yield should be compared to the anticipated
earnings yield, i.e. the expected earnings yield. This version of the model is however hard to assess
from an empirical point of view as long-term databases of anticipated earnings do not exist for many
countries. This explains why many tests of present value models (as discussed in the introduction)
or the Fed model study of Koivu, Pennanen, and Ziemba (2005) rely on current earnings. An anony-
mous referee for this paper however suggested an approximation which may allow us to tackle this
problem. Indeed, for the stock market index, it is not unreasonable to suppose that actual earnings
n-quarter ahead are not systematically out of line with expected earnings n quarters before. In other
words, knowing the actual earnings, one can equal these on average to the previously anticipated
Et∗
earnings. With this caveat, the Fed model can be written as Pt = a + Rt , where Et∗ are the expected
earnings and are set equal to Et+n where n is the time shift. We implement this variable change
in our analysis and thus test the cointegration hypothesis between anticipated earnings (as equal
to future earnings with n = 4), current stock prices and current long-term government bond yield.
The cointegration results are given in Table VII. All countries (with the exception of Australia) that
exhibited cointegration still do so in this case, while we now accept the cointegration hypothesis at
10% for Germany and Italy. Note that the long-term bond yield remains statistically not relevant in
the cointegration relationship. Adjustment weights can be interpreted as indicated above. Hence the
switch from actual to expected earnings actually confirms the hypotheses put forth in the paper and
our analysis is still conclusive.

5.6. A long-term analysis for two countries

What cointegration results can we expect if we run a really long-term analysis? The literature on the
Fed model has indeed pointed out that the supposed long-run relationship between earnings yield
and bond yield has only been valid for the last 30 years. Hence it makes sense to run a cointegration
analysis on longer term data. Unfortunately really long-term data for earnings, stock prices and bond

20
yields are hard to come by (especially for earnings). Using the new long-term database of Global
Financial Data described in Section 3, we can nevertheless retrieve long-term earnings and 10-year
government bond yield data since 1958 for the U.K. and since 1871 for the U.S. We thereafter
proceed similarly, i.e. we assess the cointegration and estimate the VECM models in both nominal
and real terms. The results are presented in Table VIII. For the U.S., the long-term evidence strongly
supports the view that a cointegration relationship exists and that, as noted above, the long-term bond
yield is not statistically significant in this relationship. Hence, for more than one century, earnings
and stock prices have been cointegrated; it would however be misleading to include long-term bond
yields in the relationship with a -1 weight. For the U.K., the cointegration relationship is still present.
In this case the long-term bond yield plays a role in the long-run relationship in nominal terms. In
real terms, we can again accept the fact that the long-term bond yield can be left out of the picture.
To save some space, we do not report the long-term results with expected earnings (as defined in the
preceding sub-section). We nevertheless conclude similarly, both in nominal and real terms.

6. Conclusion

For thirteen countries and a time period spanning more than three decades, this paper looks at the
long-run relationship between earnings, stock prices and long-term government bond yields. The
starting point of our analysis is the popular Fed model which links the equity yield of a stock index
to the prevailing 10-year government bond yield. The Fed model argues that the ‘fair value’ equity
yield for the index should be equal to the 10-year government bond yield. In the first part of the
paper, we provide a literature review of the pros and cons of the Fed model. Critics argue that
the rationale of the Fed model is flawed from a theoretical point of view. Indeed, the Fed model
relates a real quantity (the stock index earnings yield) to a nominal bond yield. In the same vein,
inflation illusion is not really taken into account as the Fed model would mechanically drive down
stock prices when inflation increases. Correspondingly, very low inflation would warrant very low
earnings yields, hence extremely high P/E ratios. Recently, Thomas (2005) presents a conflicting
view according to which earnings yield are shaped by expected inflation. In this latter framework,
there is however some support for the Fed model.

In the second part of the paper, we address these issues from an empirical perspective. More
precisely, we estimate dynamic time series models (cointegration models) for the thirteen coun-
tries in our dataset and ascertain if there exists a long-run relationship between the earnings index,

21
the stock index and the long-term government bond yield. Our empirical results show that such a
long-run relationship indeed exists for many countries (including the United States and the United
Kingdom). However, in contrast to the rationale of the Fed model, the long-term bond yield is
mostly not statistically significant in this relationship. Put simply, the long-term bond yield does not
really affect the ‘equilibrium’ stock market valuation. As such, we validate and extend the analysis
of Asness (2003) who ran a related analysis (OLS framework, no dynamic models) on U.S. data.
From a methodological point of view, the dynamic time series approach used in this paper offers
a compelling and rigorous testing environment. It also allows us to disentangle the short-term and
long-term effects of the variables involved in the model. Regarding the possible short-term effects,
we show that rising/decreasing bond yields do impact contemporaneous stock market returns and
thus have an important short-term impact on the stock market. To assess the robustness of our re-
sults, we conduct the analysis in both nominal and real terms. Furthermore, we also estimate the
models using a proxy for the expected earnings and run a really long-term analysis for the U.K. and
the U.S., two countries for which longer term data exist. These additional estimations corroborate
our earlier results.

22
References
Anari, A., and J. Kolari, 2001, Stock prices and inflation, Journal of Financial Research 24, 587–
602.

Asness, C., 2000, Stocks vs. bonds: explaining the equity risk premium, Financial Analysts Journal
56, 96–113.

Asness, C., 2003, Fight the Fed model, Journal of Portfolio Management 30, 11–24.

Boudoukh, J., and M. Richardson, 1993, Stock returns and inflation: a long-horizon perspective,
American Economic Review 83, 1346–1355.

Brooks, C., 2002, Introductory econometrics for finance. (Cambridge University Press).

Campbell, J.Y., A.W. Lo, and A.C. MacKinlay, 1997, The Econometrics of Financial Markets.
(Princeton University Press Princeton).

Campbell, J., and R. Shiller, 1987, Cointegration and tests of present value models, Journal of
Political Economy 95, 1062–1088.

Campbell, J., and R. Shiller, 1988, Stock prices, earnings and expected dividends, Journal of Fi-
nance 43, 661–676.

Campbell, J., and R. Shiller, 1989, The dividend-price ratio and expectations of future dividends
and discount factors, Review of Financial Studies 1, 195–228.

Campbell, J., and R. Shiller, 1998, Valuation ratios and the long-run stock market outlook, Journal
of Portfolio Management 24, 11–26.

Campbell, J.Y., and R.J. Shiller, 2001, Valuation ratios and the long-run stock market outlook: an
update, Cowles Foundation DP1295.

Campbell, J.Y., and T. Vuolteenaho, 2004, Inflation illusion and stock prices, Forthcoming in Amer-
ican Economic Review Papers and Proceedings.

Clarida, R., J. Gali, and M. Gertler, 2000, Monetary policy rules and macroeconomic stability:
evidence and some theory, Quarterly Journal of Economics 115, 147–180.

Cohen, R.B., C. Polk, and T. Vuolteenaho, 2005, Money illusion in the stock market: the Modigliani-
Cohn hypothesis, NBER Working Paper 11018.

Enders, W., 1995, Applied econometric time series. (Wiley).

23
Engle, R.F., and C.W.J. Granger, 1987, Co-integration, and error correction: representation, estima-
tion and testing, Econometrica 55, 251–276.

Favero, C., and F. Mosca, 2001, Uncertainty on monetary policy and the expectations model of the
term structure of interest rates, Economics Letters 71, 369–375.

Gordon, M.J., 1962, The investment, financing and valuation of a corporation. (Irwin).

Harasty, H., and J. Roulet, 2000, Modeling stock market returns, Journal of Portfolio Management
26, 33–46.

Harris, R.D.F., and R. Sanchez-Valle, 2000a, The gilt-equity yield ratio and the predictability of UK
and US equity returns, Journal of Business Finance and Accounting 27, 333–357.

Harris, R.D.F., and R. Sanchez-Valle, 2000b, The information content of lagged equity and bond
yields, Economics Letters 68, 179–184.

Harris, R., and R. Sollis, 2003, Applied time series modelling and forecasting. (Wiley).

Jones, C.P., J.W. Wilson, and L.L. Lundstrum, 2002, Estimating stock returns, Journal of Portfolio
Management 29, 40–50.

Koivu, M., T. Pennanen, and W.T. Ziemba, 2005, Cointegration analysis of the Fed model, Forth-
coming in Finance Research Letters.

Lander, J., A. Orphanides, and M. Douvogiannis, 1997, Earnings forecasts and the predictability of
stock returns: evidence from trading the S&P, Journal of Portfolio Management 23, 24–35.

Lansing, K.J., 2004, Inflation-induced valuation errors in the stock market, FRBSF Economic Letter
2004-30.

Levin, E.J., and R.E. Wright, 1998, The information content of the gilt-equity yield ratio, The
Manchester School Supplement 25, 89–101.

MacDonald, R., and D. Power, 1995, Stock prices, dividends and retention: long-run relationships
and short-run dynamics, Journal of Empirical Finance 2, 135–151.

Marshall, D.A., 1992, Inflation and asset returns in a monetary economy, Journal of Finance 47,
1315–1342.

Mills, T.C., 1991, Equity prices, dividends and gilt yields in the UK: cointegration, error correction
and ‘confidence’, Scottish Journal of Political Economy 38, 242–255.

Modigliani, F., and R. Cohn, 1979, Inflation, rational valuation and the market, Financial Analysts
Journal 35, 24–44.

24
Philips, T.K., 1999, Why do valuation ratios forecast long-run equity returns?, Journal of Portfolio
Management 25, 39–44.

Ritter, J.R., and R.S. Warr, 2002, The decline of inflation and the bull market of 1982-1999, Journal
of Financial and Quantitative Analysis 37, 29–61.

Siegel, J.J., 2002, Stocks for the long run. (McGraw-Hill).

Spyrou, S.I., 2004, Are stocks a good hedge against inflation ? Evidence from emerging markets,
Applied Economics 36, 41–48.

Thomas, J., 2005, Understanding the Fed model and the implications of this remarkable empirical
regularity, Mimeo, Yale University School of Management.

Timmermann, A., 1995, Cointegration tests of present value models with a time-varying discount
factor, Journal of Applied Econometrics 10, 17–31.

Vila-Wetherilt, A., and O. Weeken, 2002, Equity valuation measures: what can they tell us?, Bank
of England Quarterly Bulletin 42, 391–403.

Yardeni, E., 2003, Stock valuation models 4.1, Mimeo, Prudential Financial Research.

25
 Table I
Unit root tests.

Country e = ln(E) er = ln(Er ) p = ln(P) pr = ln(Pr ) r = ln(R)

c c+t c c+t c c+t c c+t c
Australia 0.58 0.12 0.23 0.09 0.90 0.06 0.81 0 0.76
Austria 0.96 0.59 0.86 0.60 0.97 0.40 0.90 0.38 0.77
Belgium 0.23 0.04 0.49 0.27 0.95 0.30 0.92 0.13 0.94
Canada 0.61 0 0.04 0.08 0.96 0.07 0.93 0.15 0.85
Denmark 0.42 0.04 0.59 0.02 0.92 0.04 0.95 0.01 0.94
France 0.36 0.30 0.74 0.31 0.89 0.27 0.88 0.07 0.90
Germany 0.82 0.30 0.78 0.48 0.86 0.23 0.78 0.15 0.86
Italy 0.77 0.89 0.83 0.51 0.81 0.80 0.70 0.28 0.87
Japan 0.55 0.59 0.14 0.49 0.59 0.95 0.70 0.85 0.76
Switzerland 0.95 0.05 0.88 0.13 0.95 0.11 0.94 0.07 0.44
The Netherlands 0.77 0 0.74 0 0.91 0.37 0.90 0.15 0.89
United Kingdom 0.27 0.25 0.15 0 0.81 0.65 0.79 0.33 0.93
United States 0.69 0.04 0.75 0.32 0.96 0.10 0.94 0.08 0.68
P-values for the ADF unit root tests for the log earnings index, log real earnings index,
log stock index, log real stock index and log government bond yield. The P-values
reported in the table refer to the null hypothesis of a unit root in the given series.
The time period is 1973:01 - 2004:04 (quarterly data) for all countries. The c column
indicates that a constant was included in the unit root test, while the c + t column
indicates that both a constant and time trend were included in the ADF test.

26
 Coint. Coint. vector Adjustment speeds H0:βr = 0
Country Time period Lags level (1 β p βr ) αe αp αr (P-value) R2
Australia 1973:03 - 2004:04 1 0.13 (1 −0.69 −0.14) -0.10 (0.03) 0.16 (0.06) -0.04 (0.04) 0.41 0.06
Australia 1973:03 - 2004:04 1 0.13 (1 −0.65 0) -0.09 (0.03) 0.13 (0.06) -0.05 (0.03)
Austria 1973:03 - 2004:04 - - - - - - - -
Belgium 1973:03 - 2004:04 - - - - - - - -
Canada 1973:04 - 2004:04 2 0.03 (1 −0.74 −0.05) -0.15 (0.03) 0.03 (0.03) 0.03 (0.03) 0.85 0.05
Canada 1973:04 - 2004:04 2 0.04 (1 −0.72 0) -0.15 (0.03) 0.03 (0.03) 0.02 (0.02)
Denmark 1973:04 - 2004:04 2 0.01 (1 −0.72 0.13) -0.17 (0.04) 0.08 (0.02) -0.02 (0.03) 0.74 0.16
Denmark 1973:04 - 2004:04 2 0.01 (1 −0.77 0) -0.17 (0.04) 0.08 (0.02) -0.02 (0.03)
France 1973:03 - 2004:04 1 0.06 (1 −0.77 −0.02) -0.12 (0.03) 0.07 (0.04) -0.04 (0.02) 0.94 0.09
France 1973:03 - 2004:04 1 0.06 (1 −0.77 0) -0.12 (0.03) 0.07 (0.04) -0.04 (0.02)
Germany 1973:03 - 2004:04 - - - - - - - -
Italy 1973:03 - 2004:04 - - - - - - - -
Japan 1973:03 - 2004:04 - - - - - - - -

27
Table II

Switzerland 1973:03 - 2004:04 1 0.08 (1 −0.68 −0.01) -0.23 (0.05) 0.05 (0.08) -0.11 (0.06) 0.96 0.03
Switzerland 1973:03 - 2004:04 1 0.08 (1 −0.68 0) -0.23 (0.05) 0.05 (0.08) -0.12 (0.06)
The Netherlands 1973:03 - 2004:04 1 0.16 (1 −0.50 0.27) -0.12 (0.04) 0.05 (0.04) -0.09 (0.03) 0.41 0.08
The Netherlands 1973:03 - 2004:04 1 0.16 (1 −0.57 0) -0.13 (0.04) 0.06 (0.05) -0.08 (0.03)
United Kingdom 1973:01 - 2004:04 1 0 (1 −0.64 0.01) -0.11 (0.02) 0.17 (0.05) -0.07 (0.03) 0.96 0.12
United Kingdom 1973:01 - 2004:04 1 0 (1 −0.64 0) -0.11 (0.02) 0.17 (0.05) -0.07 (0.03)
United States 1973:03 - 2004:04 1 0.02 (1 −0.74 −0.40) -0.05 (0.02) 0.17 (0.05) 0.05 (0.05) 0.09 0.13
United States 1973:03 - 2004:04 1 0.02 (1 −0.62 0) -0.05 (0.02) 0.13 (0.04) -0.02 (0.04)
Cointegration analysis for the log earnings index e = ln(E), log stock index p = ln(P) and the log government bond yield r = ln(R). The
long-run relationship is e + β p p + βr r, while the adjustment speeds are for ∆et , ∆pt and ∆rt respectively. As H0:βr = 0 is not rejected, we
report the long-run relationship e + β p p + 0r, i.e. when βr = 0 is enforced in the estimation (second line of results for each country). The
last column gives the R2 of the regression for the stock price adjustments in the model. Cointegration tests were done using the Johansen
methodology. A - indicates that there is no cointegration for that country, while the ‘Coint. level’ column gives the percentage level at
which we reject the absence of cointegration. ‘Lags’ gives the number of lags included in the short-term dynamics.
Cointegration analysis (VECM): earnings, stock prices and government bond yields.
 Coint. Coint. vector Adjustment speeds H0:βr = 0
Country Time period Lags level (1 β p βr ) αer α pr αr (P-value) R2
Australia 1973:03 - 2004:04 - - - - - - - -
Austria 1973:03 - 2004:04 - - - - - - - -
Belgium 1973:03 - 2004:04 - - - - - - - -
Canada 1973:04 - 2004:04 2 0.04 (1 −0.70 −0.12) -0.12 (0.03) -0.003 (0.03) 0.04 (0.02) 0.81 0.06
Canada 1973:04 - 2004:04 2 0.04 (1 −0.62 0) -0.12 (0.03) -0.01 (0.03) 0.04 (0.02)
Denmark 1973:04 - 2004:04 2 0.01 (1 −0.50 0.20) -0.20 (0.05) 0.09 (0.03) -0.03 (0.03) 0.65 0.15
Denmark 1973:04 - 2004:04 2 0.01 (1 −0.64 0) -0.19 (0.05) 0.09 (0.03) -0.02 (0.03)
France 1973:03 - 2004:04 1 0.05 (1 −0.61 0.15) -0.15 (0.03) 0.04 (0.05) -0.04 (0.03) 0.59 0.09
France 1973:03 - 2004:04 1 0.05 (1 −0.68 0) -0.15 (0.03) 0.06 (0.05) -0.03 (0.03)
Germany 1973:03 - 2004:04 - - - - - - - -
Italy 1973:03 - 2004:04 - - - - - - - -
Japan 1973:03 - 2004:04 - - - - - - - -

28
Switzerland 1973:04 - 2004:04 1 0.12 (1 −0.58 0.12) -0.20 (0.05) -0.06 (0.07) -0.09 (0.05) 0.42 0.03
Table III

Switzerland 1973:04 - 2004:04 1 0.12 (1 −0.61 0) -0.21 (0.05) -0.05 (0.07) -0.07 (0.06)
The Netherlands 1973:04 - 2004:04 1 0.17 (1 −0.38 0.19) -0.18 (0.05) 0.02 (0.06) -0.09 (0.04) 0.46 0.07
The Netherlands 1973:04 - 2004:04 1 0.17 (1 −0.44 0) -0.19 (0.05) 0.03 (0.06) -0.08 (0.04)
United Kingdom 1973:01 - 2004:04 2 0 (1 −0.32 0.05) -0.24 (0.03) 0.13 (0.08) -0.02 (0.05) 0.68 0.07
United Kingdom 1973:01 - 2004:04 2 0 (1 −0.35 0) -0.24 (0.03) 0.13 (0.08) -0.02 (0.05)
United States 1973:03 - 2004:04 2 0.02 (1 −0.45 −0.06) -0.10 (0.03) 0.13 (0.06) 0.06 (0.07) 0.76 0.11
United States 1973:03 - 2004:04 2 0.02 (1 −0.42 0) -0.10 (0.03) 0.12 (0.06) 0.04 (0.07)
Cointegration analysis for the log real earnings index er = ln(Er ), log real stock index pr = ln(Pr ) and the log government bond yield
r = ln(R). The long-run relationship is er + β p pr + βr r, while the adjustment speeds are for ∆er,t , ∆pr,t and ∆rt respectively. As H0:βr = 0
is not rejected, we report the long-run relationship er + β p pr + 0r, i.e. when βr = 0 is enforced in the estimation (second line of results for
each country). The last column gives the R2 of the regression for the stock price adjustments in the model. Cointegration tests were done
using the Johansen methodology. A - indicates that there is no cointegration for that country, while the ‘Coint. level’ column gives the
percentage level at which we reject the absence of cointegration. ‘Lags’ gives the number of lags included in the short-term dynamics.
Cointegration analysis (VECM): real earnings, real stock prices and bond yields.
 Table IV
Variance decompositions for the log stock index and log real stock index (I).

Australia
ln(P), explained by innovations in ln(Pr ), explained by innovations in
ln(E) ln(P) ln(R)) ln(Er ) ln(Pr ) ln(R))
Quarters ahead I II I II I II I II I II I II
1 1.8 0 95.8 97.6 2.4 2.4
4 6.9 2.4 87.4 91.9 5.7 5.7
20 32.5 22.6 61.7 71.6 5.8 5.8
Canada
ln(P), explained by innovations in ln(Pr ), explained by innovations in
ln(E) ln(P) ln(R)) ln(Er ) ln(Pr ) ln(R))
Quarters ahead I II I II I II I II I II I II
1 1.6 0 84.5 86.1 13.9 13.9 1 0 83.8 84.8 15.2 15.2
4 1.7 0 73.7 75.4 24.6 24.6 0.6 0.1 73.1 73.6 26.3 26.3
20 4.8 1.3 69.4 72.9 25.8 25.8 0.3 1.5 69.5 68.3 30.2 30.2
Denmark
ln(P), explained by innovations in ln(Pr ), explained by innovations in
ln(E) ln(P) ln(R)) ln(Er ) ln(Pr ) ln(R))
Quarters ahead I II I II I II I II I II I II
1 1 0 96.6 97.6 2.4 2.4 1.2 0 95.8 97 3 3
4 0.6 1.7 92.4 91.3 7 7 0.5 1.5 90.7 89.7 8.8 8.8
20 24.4 32.1 68.8 61.1 6.8 6.8 20.8 28.9 71.0 62.9 8.2 8.2
France
ln(P), explained by innovations in ln(Pr ), explained by innovations in
ln(E) ln(P) ln(R)) ln(Er ) ln(Pr ) ln(R))
Quarters ahead I II I II I II I II I II I II
1 3.1 0 94.9 98 2 2 2.5 0 94.4 96.9 3.1 3.1
4 2.2 0.4 88.4 90.2 9.4 9.4 1.2 0.5 85.5 86.2 13.3 13.3
20 13.9 5.5 79 87.4 7.1 7.1 6.8 1.8 81.9 86.9 11.3 11.3
Variance decompositions (1, 4 and 20 quarters ahead) for the log stock index (left panel) and for the
log real stock index (right panel) in the VECM models. Each figure gives the proportion of variance
explained by the highlighted variables. There are two variable orderings: I, for ln(R), ln(E) and
ln(P); II, for ln(R), ln(P) and ln(E) (and correspondingly for the right panel: I, for ln(R), ln(Er )
and ln(Pr ); II, for ln(R), ln(Pr ) and ln(Er )).

29
 Table V
Variance decompositions for the log stock index and log real stock index (II).

Switzerland
ln(P), explained by innovations in ln(Pr ), explained by innovations in
ln(E) ln(P) ln(R)) ln(Er ) ln(Pr ) ln(R))
Quarters ahead I II I II I II I II I II I II
1 5.8 0 93.2 99 1 1 4.2 0 94 98.2 1.8 1.8
4 3.7 0.4 93.7 97 2.6 2.6 1.3 1.4 93.1 93.0 5.6 5.6
20 10.3 1.5 87.6 96.4 2.1 2.1 0.6 1.8 91.8 90.6 7.6 7.6
The Netherlands
ln(P), explained by innovations in ln(Pr ), explained by innovations in
ln(E) ln(P) ln(R)) ln(Er ) ln(Pr ) ln(R))
Quarters ahead I II I II I II I II I II I II
1 0 0 99.4 99.4 0.6 0.6 0 0 98.8 98.8 1.2 1.2
4 0.5 0.3 92.8 93 6.7 6.7 0.1 0.1 90.3 90.3 9.6 9.6
20 13.1 12.2 80 80.9 6.9 6.9 3.4 3.1 84.7 85 11.9 11.9
United Kingdom
ln(P), explained by innovations in ln(Pr ), explained by innovations in
ln(E) ln(P) ln(R)) ln(Er ) ln(Pr ) ln(R))
Quarters ahead I II I II I II I II I II I II
1 0.4 0 80.5 80.9 19.1 19.1 0.5 0 74.7 75.1 24.8 24.9
4 0.7 1.1 77.8 77.4 21.5 21.5 0.8 0.1 64.1 64.8 35.1 35.1
20 18.4 22.5 63.8 59.7 17.8 17.8 2.7 4.4 62.4 60.7 34.9 34.9
United States
ln(P), explained by innovations in ln(Pr ), explained by innovations in
ln(E) ln(P) ln(R)) ln(Er ) ln(Pr ) ln(R))
Quarters ahead I II I II I II I II I II I II
1 1.6 0 95.8 97.4 2.6 2.6 3.1 0 93 96.1 3.9 3.9
4 1.7 0.3 85.1 86.5 13.2 13.2 3.5 0.6 78.7 81.6 17.8 17.8
20 33.9 26.5 46.8 54.2 19.3 19.3 19.1 9 65.2 75.3 15.7 15.7
Variance decompositions (1, 4 and 20 quarters ahead) for the log stock index (left panel) and for the
log real stock index (right panel) in the VECM models. Each figure gives the proportion of variance
explained by the highlighted variables. There are two variable orderings: I, for ln(R), ln(E) and
ln(P); II, for ln(R), ln(P) and ln(E) (and correspondingly for the right panel: I, for ln(R), ln(Er )
and ln(Pr ); II, for ln(R), ln(Pr ) and ln(Er )).

30
 Nominal data Real data
Country Time period rest−1 ∆rt ∆pt−1 Time period resr,t−1 ∆rt ∆pr,t−1
Australia 1973:03 - 2004:04 0.20 (0.07) -0.27 (0.14) 0.06 (0.09) - - - -
Canada 1973:04 - 2004:04 0.04 (0.03) -0.45 (0.10) 0.15 (0.09) 1973:04 - 2004:04 0.02 (0.03) -0.48 (0.11) 0.16 (0.09)
Denmark 1973:04 - 2004:04 0.05 (0.02) -0.16 (0.08) 0.25 (0.09) 1973:04 - 2004:04 0.04 (0.03) -0.18 (0.08) 0.23 (0.09)
France 1973:03 - 2004:04 0.13 (0.05) -0.29 (0.16) 0.07 (0.09) 1973:03 - 2004:04 0.10 (0.06) -0.33 (0.16) 0.06 (0.09)
Switzerland 1973:03 - 2004:04 0.13 (0.09) -0.14 (0.11) 0.03 (0.10) 1973:03 - 2004:04 0.02 (0.08) -0.07 (0.11) -0.01 (0.10)

31
The Netherlands 1973:03 - 2004:04 0.06 (0.05) -0.10 (0.13) 0.06 (0.09) 1973:03 - 2004:04 0.01 (0.06) -0.15 (0.13) 0.07 (0.09)
Table VI

United Kingdom 1973:01 - 2004:04 0.11 (0.04) -0.71 (0.12) 0.11 (0.09) 1973:01 - 2004:04 0.02 (0.07) -0.75 (0.12) 0.04 (0.09)
United States 1973:03 - 2004:04 0.19 (0.05) -0.17 (0.08) 0.11 (0.09) 1973:03 - 2004:04 0.17 (0.05) -0.18 (0.09) 0.12 (0.09)
Results from the single-equation ECM analysis (quarterly nominal and real data). We report relevant results for the single-equation ECM: ∆pt =
γ p + α p rest−1 + δe,0 ∆et + δr,0 ∆rt + Short-run dynamics for p + εt . The rest−1 column gives the adjustment speed to the long-run disequilibrium; the
∆rt column reports the contemporaneous influence of the log government bond yield; the ∆pt−1 column gives the momentum effect (i.e. impact of
lagged stock price changes). All numbers in parenthesis are standard errors.
Cointegration analysis (single-equation ECM): nominal and real data.
 Coint. Coint. vector Adjustment speeds H0:βr = 0
Country Time period Lags level (1 β p βr ) αe∗ αp αr (P-value) R2
Australia 1973:03 - 2003:04 - - - - - - - -
Austria 1973:03 - 2003:04 - - - - - - - -
Belgium 1973:03 - 2003:04 - - - - - - - -
Canada 1973:04 - 2003:04 2 0.04 (1 −0.75 −0.02) -0.12 (0.03) 0.06 (0.03) 0.05 (0.03) 0.94 0.12
Canada 1973:04 - 2003:04 2 0.04 (1 −0.74 0) -0.12 (0.03) 0.06 (0.03) 0.05 (0.03)
Denmark 1973:04 - 2003:04 2 0.01 (1 −0.58 0.26) -0.06 (0.05) 0.10 (0.03) -0.06 (0.03) 0.46 0.21
Denmark 1973:04 - 2003:04 2 0.01 (1 −0.69 0) -0.07 (0.06) 0.10 (0.03) -0.04 (0.04)
France 1973:03 - 2003:04 1 0 (1 −0.88 −0.36) -0.01 (0.04) 0.28 (0.05) -0.01 (0.03) 0.07 0.25
France 1973:03 - 2003:04 1 0 (1 −0.77 0) -0.04 (0.04) 0.25 (0.05) -0.03 (0.03)
Germany 1973:03 - 2003:04 1 0.10 (1 −0.63 0.65) -0.16 (0.05) 0.14 (0.06) -0.05 (0.04) 0.07 0.08
Germany 1973:03 - 2003:04 1 0.10 (1 −0.83 0) -0.09 (0.04) 0.15 (0.05) 0.01 (0.03)
Italy 1973:04 - 2003:04 2 0.10 (1 −0.79 0.27) -0.04 (0.04) 0.12 (0.05) -0.09 (0.03) 0.18 0.14
Italy 1973:03 - 2003:04 2 0.10 (1 −0.86 0) -0.06 (0.04) 0.14 (0.05) -0.07 (0.03)
Japan 1973:03 - 2003:04 - - - - - - - -

32
Switzerland 1973:04 - 2003:04 1 0 (1 −0.69 −0.15) -0.10 (0.05) 0.39 (0.07) 0.03 (0.06) 0.13 0.22
Table VII

Switzerland 1973:04 - 2003:04 1 0 (1 −0.67 0) -0.11 (0.05) 0.37 (0.07) -0.01 (0.06)
The Netherlands 1974:01 - 2003:04 3 0.06 (1 −0.24 1.55) -0.09 (0.03) 0.04 (0.03) -0.07 (0.02) 0.04 0.12
The Netherlands 1974:01 - 2003:04 3 0.06 (1 −0.59 0) -0.16 (0.05) 0.12 (0.06) -0.05 (0.04)
United Kingdom 1973:01 - 2003:04 1 0.01 (1 −0.59 0.13) -0.09 (0.03) 0.20 (0.06) -0.05 (0.04) 0.56 0.10
United Kingdom 1973:01 - 2003:04 1 0.01 (1 −0.62 0) -0.10 (0.03) 0.20 (0.06) -0.04 (0.04)
United States 1973:03 - 2003:04 1 0.01 (1 −0.64 −0.10) -0.06 (0.02) 0.14 (0.05) 0.02 (0.05) 0.71 0.10
United States 1973:03 - 2003:04 1 0.01 (1 −0.61 0) -0.06 (0.02) 0.13 (0.04) 0.004 (0.05)
Cointegration analysis for the log expected earnings index e∗ = ln(E ∗ ), log stock index p = ln(P) and the log government bond yield
r = ln(R). The long-run relationship is e∗ + β p p + βr r, while the adjustment speeds are for ∆et∗ , ∆pt and ∆rt respectively. As H0:βr = 0
is not rejected, we report the long-run relationship e∗ + β p p + 0r, i.e. when βr = 0 is enforced in the estimation (second line of results
for each country). The last column gives the R2 of the regression for the stock price adjustments in the model. Cointegration tests were
done using the Johansen methodology. A - indicates that there is no cointegration for that country, while the ‘Coint. level’ column
Cointegration analysis (VECM): expected earnings, stock prices and bond yields.

gives the percentage level at which we reject the absence of cointegration. ‘Lags’ gives the number of lags included in the short-term
dynamics.
 Coint. Coint. vector Adjustment speeds H0:βr = 0
Country Time period Lags level (1 β p βr ) αe αp αr (P-value) R2
U.K. Nominal 1958:04 - 2004:04 2 0.10 (1 −0.93 −0.60) -0.04 (0.02) 0.07 (0.03) -0.01 (0.01) 0.04
U.K. Nominal 1958:04 - 2004:04 2 0.10 (1 −0.87 0) -0.02 (0.01) 0.07 (0.02) -0.03 (0.02)
U.K. Real 1958:04 - 2004:04 2 0.08 (1 −0.52 −0.17) -0.07 (0.02) 0.06 (0.03) -0.004 (0.02) 0.52
U.K. Real 1958:04 - 2004:04 2 0.08 (1 −0.40 0) -0.08 (0.02) 0.05 (0.04) -0.01 (0.03)
U.S. Nominal 1871:04 - 2004:04 2 0.02 (1 −0.88 −0.20) -0.05 (0.02) 0.04 (0.01) 0.02 (0.01) 0.30 0.10
U.S. Nominal 1871:04 - 2004:04 2 0.02 (1 −0.92 0) -0.05 (0.01) 0.04 (0.01) 0.01 (0.01)
U.S. Real 1871:04 - 2004:04 2 0.02 (1 −0.70 −0.21) -0.07 (0.02) 0.02 (0.014) 0.02 (0.01) 0.21 0.10

33
U.S. Real 1871:04 - 2004:04 2 0.02 (1 −0.77 0) -0.06 (0.02) 0.02 (0.01) 0.02 (0.01)
Table VIII

Cointegration analysis (long-term data) for the log earnings index e = ln(E), log stock index p = ln(P) and the log government bond
yield r = ln(R). The long-run relationship is e + β p p + βr r, while the adjustment speeds are for ∆et , ∆pt and ∆rt respectively. As
H0:βr = 0 is not rejected, we report the long-run relationship e + β p p + 0r, i.e. when βr = 0 is enforced in the estimation (second line of
results for each country). The last column gives the R2 of the regression for the stock price adjustments in the model. Cointegration tests
were done using the Johansen methodology. A - indicates that there is no cointegration for that country, while the ‘Coint. level’ column
gives the percentage level at which we reject the absence of cointegration. ‘Lags’ gives the number of lags included in the short-term
data): earnings, stock prices and bond yields.

dynamics.
Cointegration analysis (VECM) for the United Kingdom and the United Sates (long-term
 earnyield earnyield
0.14

0.30

0.12
0.25

0.10
0.20

0.08
0.15

0.06
0.10

0.04
0.05

1970 1975 1980 1985 1990 1995 2000 2005 1970 1975 1980 1985 1990 1995 2000 2005

1.6 earnyieldGOVT
earnyieldGOVT
1.8

1.4
1.6

1.4 1.2

1.2
1.0

1.0
0.8

0.8

0.6
0.6

1970 1975 1980 1985 1990 1995 2000 2005 1970 1975 1980 1985 1990 1995 2000 2005

0.50 cr cr
−1.8

0.25 −1.9

−2.0
0.00
−2.1

−0.25 −2.2

−2.3
−0.50

−2.4

−0.75
−2.5

−1.00 −2.6

1970 1975 1980 1985 1990 1995 2000 2005 1970 1975 1980 1985 1990 1995 2000 2005

Figure 1. United Kingdom and United States. From top to bottom: earnings yield, earnings
yield/government bond yield and cointegration relationship (nominal data). Left figures are for
United Kingdom, right figures for United States.

34
 y3m × cr y12m × cr
0.50
0.5
0.25

0.0
0.00

−0.5
−0.25

−1.25 −1.00 −0.75 −0.50 −0.25 0.00 −1.25 −1.00 −0.75 −0.50 −0.25 0.00

y24m × cr y60m × cr

0.5 1.0

0.0 0.5

−0.5 0.0

−1.0
−1.25 −1.00 −0.75 −0.50 −0.25 0.00 −1.25 −1.00 −0.75 −0.50 −0.25 0.00

0.2 y3m × cr y12m × cr

0.25

0.0 0.00

−0.25
−0.2

−0.50

−2.8 −2.6 −2.4 −2.2 −2.0 −2.8 −2.6 −2.4 −2.2 −2.0

y24m × cr y60m × cr
0.50
1.0
0.25

0.00 0.5

−0.25
0.0
−0.50

−2.8 −2.6 −2.4 −2.2 −2.0 −2.8 −2.6 −2.4 −2.2 −2.0

Figure 2. United Kingdom and United States (forward-looking returns vs cointegration re-
lationship). From top left to bottom right: 3-month, 12-month, 24-month and 60-month forward-
looking returns on the stock index vs the estimated cointegration relationship (cr). The top four
graphs are for the U.K., the bottom four graphs for the U.S.

35
 2005
2000
1995
1990
1985
FAIR

1980
TOTMKUKPI

1975
FAIR2

1970
5000

4000

3000

2000

1000

Figure 3. United Kingdom (stock index and fair values), 1973:03 - 2004:04. Actual stock index
(TOTMKUKPI) and fair values as forecasted by the VECM model: FAIR is the forecast based
on the original VECM model, FAIR2 is the forecast based on the constrained VECM model (the
coefficient of the government bond yield is constrained at zero in the long-run relationship).

36
 2005
2000
1995
1990
1985
1980
FAIR
TOTMKUSPI

1975
FAIR2

1970
1500

1250

1000

750

500

250

Figure 4. United States (stock index and fair values), 1973:03 - 2004:04. Actual stock index
(TOTMKUSPI) and fair values as forecasted by the VECM model: FAIR is the forecast based
on the original VECM model, FAIR2 is the forecast based on the constrained VECM model (the
coefficient of the government bond yield is constrained at zero in the long-run relationship).

37