PREDICTABILITY OF EQUITY REIT RETURNS: IMPLICATIONS FOR

PROPERTY TACTICAL ASSET ALLOCATION

by

JOHN OKUNEV

SCHOOL OF ACCOUNTING AND FINANCE

MACQUARIE UNIVERSITY, SYDNEY

PATRICK WILSON

SCHOOL OF FINANCE AND ECONOMICS

UNIVERSITY OF TECHNOLOGY, SYDNEY

August 2004
Contact: John Okunev
School of Accounting and Finance,
Macquarie University,
Sydney,
Australia 2019
Email Jokunev@efs.mq.edu.au
 PREDICTABILITY OF EQUITY REIT RETURNS: IMPLICATIONS FOR

PROPERTY TACTICAL ASSET ALLOCATION

This study presents further evidence on the predictability of excess Equity REIT returns.

Recent evidence on forecasting excess returns using fundamental variables has resulted in

diminishing returns from the 1990s onwards. Trading strategies based upon these forecasts have

not significantly outperformed the buy hold strategy in the 1990’s. We develop an alternative

strategy which is based upon the time variation of investors risk premium. Our results indicate

that a strategy based upon modeling this time variation of the risk premium is able to outperform

the buy hold strategy both in and out of sample. By modeling the dynamic behavior of the risk

premium we are implicitly capturing economic risk premiums which are not captured by

conventional multi beta asset pricing models.

2
1. Introduction

There has been considerable interest in recent times in modeling the

predictability of asset returns. This aspect of the finance literature has witnessed a

long history of this type of research by both practitioners and academics. Clearly,

practitioners are interested in being able to predict asset returns, as this will directly

influence their trading strategies. Academics are also interested in this area as it has

important implications in the development of asset pricing models.

Historically, there have been two approaches used in predicting asset returns,

these being the fundamental approach and secondly, modeling investor’s perceptions

of risk through the use of time varying risk premiums. The fundamental approach

relies upon the assumption that asset returns are influenced by some common factors

which reflect different states of the business cycle. Studies that have adopted this

approach in relation to stock markets are: Chen, Roll and Ross (1986), Conrad and

Kaul (1988), Fama and French (1990,1993), Ferson and Harvey (1991), Lo and

Mackinley (1992) and others. Similar methodologies have been applied to real estate

markets. For example, studies on the real estate market by Liu and Mei (1992), Liu

and Mei (1994), Ling and Naranjo (1997), Liu and Mei (1998), Karolyi and Sanders

(1998), Quan and Titman (1999) and Ling, Naranjo and Ryngaert (2000) have found

that economic factors conventionally considered important in predicting excess stock

returns were also important in predicting excess real estate returns. Variables that

were found to be important were: the yield on one month T/Bills, the yield spread

between AAA bond and T/bills, the dividend yield of the stock market and the cap

rate. For the period 1972-1989 Liu and Mei (1992) found the excess returns of the

3
Equity REITs over the T/bill rate were predictable with R-squares ranging from

14.6% to 16.6%. In a follow up paper Liu and Mei (1994) demonstrated that

significant profits can be generated by forecasting excess returns based upon the

variables described above. When the excess return was forecasted to be positive a

long position was taken in the Equity REITs, and when the forecasted risk premium

was negative a short position was taken. This long/short strategy produced significant

profits in and out of sample on both a non risk adjusted and risk adjusted basis.

Ling and Naranjo (1997) similarly found evidence to support a commonality of

some 'drivers' for both real estate and stock markets. They used nonlinear multivariate

techniques to jointly estimate the risk factor sensitivities and return premia that economic

variables such as T-bills, industrial production, per capita consumption, expected and

unexpected inflation etc. had on commercial real estate returns. Their analysis was

applied to both securitised real estate data as well as to 'unsmoothed' real estate data

from the National Council of Real Estate Investment Fiduciaries. Their study showed

that the growth rate in real per capita consumption and the real T-bill rate were important

economic variables (i.e. consistently priced across the four real estate portfolio groups

constructed), while the term structure and unexpected inflation did not carry statistically

significant risk premiums in their fixed-coefficient model. Karolyi and Sanders (1998)

examined the variation of economic risk premiums by employing a multiple beta asset

pricing model and found varying degrees of predictability among stocks, bonds and

REITs, and concluded that there are important economic risk premiums for REITs which

are not captured by conventional multiple beta asset pricing models.

In a later comparative study on international real estate and stock markets Liu

and Mei (1998) found that own country economic state variables (short-interest rates,

4
spread between long and short rates, dividend yield), when exchange rate adjusted for

$US, account for a portion of the variation in the expected rates of return in some

countries, but not in others. In a further international study on the extent to which real

estate and stock prices move together Quan and Titman (1999) sampled 17 countries

and found strong evidence of a positive correlation between stock returns and real

estate price changes. Their results suggest that a large fraction of the observed positive

correlation was due to economic fundamentals, in particular changes in GDP, that affect

both real estate and stock prices

Recently, Ling, Naranjo and Ryngaert (2000) have shown the extent of

predictability of the risk premium is not nearly as great in the 1990s compared to the

1970s and 1980s. Using a rolling best fit regression model they are able to produce

fairly high in sample R-squares, but out of sample fits are quite low. With plausible

levels of transaction costs they show that trading strategies based upon forecasted risk

premiums do not out perform the buy hold Equity Reit strategy in the 1990s.

Recent evidence appears to indicate that forecasting the risk premium based

upon fundamental variables does not perform as well as their performance during the

70’s and 80’s. In this paper we adopt a different approach and model the dynamic

behavior of the ex ante risk premium. From a simple discounted dividend model we

extract the ex ante risk premium implied from the Equity REITs. Using this series we

observe when the normalized risk premium is high this is usually associated with

undervaluation of the Equity REITs. Conversely, when the normalized risk premium

is low this is usually associated with overvalued Equity REITs. Our results indicate

that a strategy based upon modeling the normalized risk premium is able to

5
outperform the buy hold strategy both in and out of sample. By modeling the

dynamic behavior of the ex ante risk premium we are implicitly capturing other

economic risk premiums which are not captured by conventional multi beta asset

pricing models ( see Karolyi and Sanders (1998)

2. METHODOLOGY AND EMPIRICAL RESULTS

2.1. Descriptive Statistics

Table 1 presents descriptive statistics indicating the general distributional traits

of the data utilized throughout this paper, consisting of total monthly returns (returns

include capital gain and dividends paid) from Equity REITs , the yield of Treasury Bills,

the yield of 20 year government bonds and the annualized realized risk premium for the

Equity REITs which is defined as the total monthly return minus the yield of T/Bills.

The period of the data is from January 1972 to December 2003. The returns of the

Equity REITs are negatively skewed and have a lower level of kurtosis than would be

expected under a normal distribution. This fact may explain why the Jarque-Bera test

indicates the Equity REITs series are not normally distributed. To also test for serial

correlation, Ljung-Box tests are reported for both the returns and squared returns for lags

up to half a year. Evidence of autocorrelation in squared returns can be a result of

conditional heteroskedasticity which is prevalent in financial times series data and for

which it is therefore worthwhile testing. However, there is no evidence of serial

correlation, most likely due to the low frequency data being examined.

INSERT TABLE 1

6
 Table 1 also contains a number of important test results that help characterize

the market. The Augmented Dickey-Fuller test results indicate returns are

stationary. Finally, a Zivot and Andrews (1992) test for structural breaks is also

conducted. This test benefits from the fact that it reports values for where potential

breaks occur when they are neither specified nor determined. The importance of

highlighting where a structural break may occur in the data is that it can have a

fundamental impact upon one of the major test statistics that will be examined in the

empirical section. Simply put, if no consideration is made for structural breaks, linear

regression may produce spurious results as parameter stability becomes questionable

over the whole sample period or periods straddling the structural break. As reported in

table 1, results of the Zivot and Andrews (1992) test indicate a significant structural

break for Equity REITs in August 1989. In a recent paper Glascock , Lu and So

(2000) found evidence of a structural break in 1993, arguing this may be due to

changes in tax legislation. We apply the Zivot and Andrews (1992) test to the data

from 1990 to 2003 and found no evidence of this additional structural break at that

time. We therefore proceed on the basis that only one break occurred and this being

August 1989. We therefore partition the data set into two sub samples, sub sample one

being from 1/72 to 8/89 and sub sample 2 from 9/89 to 12/03 is used as a hold out

sample1.

The question naturally arises as to what economic event(s) has led to the

structural break in August 1989 and suggest two possible reasons. Goetzmann and

Wachter(1996), Quan and Titman (1999) and Case, Goetzmann and Rouwenhorst

(1999) have provided compelling evidence that real estate returns are highly

correlated with changes in global and domestic GDP, and that the negative returns of

1
This partitioning is also similar to periods used by Ling, Naranjo and Ryngaert (2000) . They partition their data
into 1/80-12/89 and 1/90-12/96.

7
real estate markets in the early nineties was due to a global recession in 1990. This

being especially the case for the US, where both global and local economic factors are

important in explaining movements in real estate returns (see Case, Goetzmann and

Rouwenhorst, 1999). This event may also have been exacerbated by the S&L crisis

which occurred in 1989 that resulted in a dramatic shift in the financing of real assets.

From the Gordon Shapiro discounted dividend model we extract the implied

ex ante risk premium from the following equation.

D (t )(1 + g (t ))
P (t ) = (1)
( R (t ) − g (t ))

Where D(t) are dividends paid at time t

g(t) is the annualized growth rate at time t;

R(t) is the cost of capital at time t;

P(t) is the Equity REIT price index at time t.

The ex ante risk premium is defined as:

EARP(t) = R(t)- TB(t)

The ex post risk premium is defined as EPRP(t) = R12(t) – TB(t), where R12(t) is the

previous twelve months total return of the Equity REITs and TB(t) is the T/Bill rate

at time t.

Using equation 1 and information available at time t, the ex ante risk premium

is calculated. Figure 1 displays the ex ante risk premium from 1/74 to 12/03.2 It is

2
The ex ante risk premium is calculated from 1/74 as we require two years of data to determine the growth rate
of dividends.

8
apparent that the ex ante is mostly positive and is not nearly as volatile as the ex post

risk premium. It is also interesting to note that the ex ante risk premium is negative

in 1976 , 1977 and 1981. These results lend some support to the findings of

Boudoukh, Richardson and Smith (1993) who develop tests of inequality restrictions

implied by conditional asset pricing models. As an application, they test whether the ex

ante risk premium is always positive. Using annual data on aggregate US stock returns,

inflation, long and short rates of interest and dividend yields over three time periods

(1802-1990, 1802-1896, 1897-1990), they report reliable evidence that the ex ante risk

premium is negative in some states of the world; these states being related to periods of

high expected inflation and especially to downward sloping term structure. For the

period April to August 1981, the yield curve was inverted and inflation varied between

8 % to 13%. For 1976 and 1977 the yield curve was not inverted but inflation varied

between 3% and 8%, which is above the average inflation rate.

INSERT FIGURE 1

Liu and Mei (1992), Mei and Liu (1994), Karolyi and Sanders (1998) and

Ling, Naranjo and Ryngaert (2000) have shown that variables such as the T/Bill rate,

the spread between the yields on long term bonds and T/bills, the dividend yield of the

equity market and cap rate at time t are able to explain between 20%-30% of the

variation in the realized risk premium from t to t+1. The realized risk premium is defined

as the monthly return of the Equity REITs from t to t+1 minus the T/bill rate at time t. It

therefore seems sensible to determine whether these same variables can forecast the ex

ante risk premium. We adopt an approach which is similar to Liu and Mei (1992) where

the ex ante risk premium at time t+1 is regressed against the following variables:

January dummy, T/bill rate, spread between the yield on a long bond and the T/bill, the

9
dividend and earnings yields of the SP500 at time t 3. The sample is divided into two.

The first subsample is the in sample period from 1/72 to 8/89, and the second sample is

the out of sample from 9/89 to 12/03. The results of the regressions are presented in

Table 2. We use generalized least squares regression as there was some evidence of

serial correlation in the residuals. It is evident that macroeconomic factors can explain

little of the future movements in the ex ante risk premium both in and out of sample.

These results are in stark contrast to prior research which has found that macrofactors

offer significant explanatory power in movements of the realized risk premium. It

appears that movements of the ex ante risk premium are not driven by macrofactors, but

are capturing some other factors which are influencing Equity REIT price movements.

INSERT TABLE 2

We now investigate whether the dynamic behavior of the ex ante risk premium

contains information about likely price movements of the Equity REITs in the future.

Figure 2 displays the 36 month rolling normalized ex ante risk premium (ZRP) from

1/1974 to 12/2003.4 The ex ante risk premium can be calculated either by subtracting

the T/bill rate from the implied cost of capital or alternatively subtracting the yield on

10 year government bonds from the cost of capital. We adopt the latter approach in

the following analysis as it is more in keeping the long term nature of the discounted

dividend formula.5 We suggest that when the risk premium is high relative to the

historic average investors will demand a greater return in holding a risky asset. This is

3
Liu and Mei (1992) did not use the earnings yield on the stock market, but used the cap rate. We found that the
dividend and earnings yield were significant in many of the regressions. These results are also supported by the
findings of Ling, Naranjo and Ryngaert (2000).
4
We examined a number of rolling windows and found the 36 month window to be long enough to capture timely variation in
the risk premium.
5
We calculated both versions and found the ensuing results to be similar.

10
usually associated with falling prices. Similarly, when the normalized risk premium is

low relative to the historic average investors are requiring a lower return for holding

risky assets. This is usually associated with rising prices. Observation of Figure 2

would suggest that in 1980, 1985, 1987, 1992, 1995,1999 and 2001 the Equity REITs

were oversold (prices depressed) and that the market was overbought in 1976, 1990,

2002 and 2003 (prices inflated). Figure 2 also displays that the normalized risk

premium is mean reverting and generally oscillates between +- 2 standard deviations.

Large ZRP values are usually associated with reversals in the next 3-6 months.

INSERT FIGURE 2

To obtain some insight into the likely future performance of the Equity REITs

we examine the returns when ZRP is greater and less than zero. Table 3 displays the

average return and standard deviations of these strategies. It is evident from Table 3

when ZRP was greater zero, the ensuing average return was 19.1% pa over the entire

sample.6 The results are similar in each of the sub periods. On the other hand, when

ZRP was less zero, the ensuing average return was -0.8% pa over the entire sample.

These results confirm our initial statement that high risk premiums are usually

associated with higher subsequent returns, and lower risk premiums are usually

associated with lower subsequent returns. The question then arises how should one

invest when ZRP is less than zero?

To address this issue we devise a strategy which goes long in the Equity

REITs when ZRP>=0 and convert to cash or bonds otherwise. The alternative strategy

is if ZRP<0, we long in the Equity REITs and invest in cash or bonds otherwise.

6
The strategy was, if ZRP>=0 at time t-1 go long Equity REITs at time t, else zero. A similar approach was adopted when

11
For example, the strategy ZRP>0 +cash is: if ZRP>=0 invest in the Equity REITs

otherwise invest in cash. Similarly, for the strategy ZRP>=0 + Bonds is : if ZRP>=0

invest in the Equity REITs otherwise invest in long term government bonds. The

reason we go long in the Equity REITs when ZRP<0 is to illustrate that returns of this

strategy are indicative that one should be out of the Equity REITs during this time.

We also compare these results to a forecasting model and show that the performance

of the forecasting model has produced lower returns than the buy/hold strategies since

1990. Using an approach similar to other researchers, we perform a rolling 60 month

regression of the realized excess return of the Equity REITs from t to t+1 on lagged a

variable at time t. The lagged variable we adopt is the dividend yield of the Equity

REITs minus the yield on the 10 year government bond.7

Table 4 presents the returns and standard deviations of the following

strategies, ZRP>=0 + cash, ZRP <0 + cash, ZRP >=0 + bonds, ZRP <0 + bonds and

ZRP long/short strategy. Also displayed in Table 4 are results of the forecast based

strategy, FRP>=0 + cash, FRP<0 + cash, FRP>=0 + bonds, FRP<0 + bonds and

FRP long/short strategy. For example, the strategy FRP>=0 + cash is: if the FRP

(forecasted excess return) is positive go long in the Equity REITs and cash otherwise.

For the period 1/78 – 8/89 it is apparent that the strategies ZRP >=0 + cash,

ZRP >=0 + bonds, FRP>=0 + cash and FRP>=0 + bonds all outperform the buy hold

strategy 8. Each of these strategies have a greater return with similar standard

ZRP<0.
7
Alternatively we could have used the Tbill rate, spread and cap rates as lagged variables, but using the dividend yield minus
the yield on the 10 year bond produced similar results.
8
We evaluate the strategies from 1/78 owing to start up requirements. Estimation of the ex ante risk premium
required two years of data to estimate the growth rate of dividends plus another 3 years of data to obtain values of
the normalized risk premium.

12
deviation to the Equity REITs. The results also show that the returns are generally

similar except for ZRP >=0 + bonds which has the highest average return of 20.1%

pa. The results for the forecasting model are consistent with other studies using a

macro based forecasting equation. For example Ling, Naranjo and Ryngaert (2000)

using a number of macro factors attained an average return of 17.6% pa (with zero

transaction costs), whereas the FRP>=0 + cash strategy attained an average return of

16.7% pa for that period for the period 1/80 to 12/89.

INSERT TABLE 4

The long/short strategies produced mixed results. The ZRP strategy marginally

outperformed the buy hold strategy, whereas the forecast strategy underperformed the

buy hold strategy.

We now turn to the out of sample period 9/89 – 12/03. The strategies, ZRP

>=0 + cash, ZRP >=0 + bonds and FRP>=0 + bonds outperformed the buy hold

strategy. Each of these strategies dominated the buy hold strategy on both a risk and

non risk adjusted basis. Similar to the findings of Ling, Naranjo and Ryngaert (2000),

we find that the forecasting strategy that invests in cash and Equity REITs has

underperformed the buy hold strategy in the nineties. With regards to the long/short

strategies it is evident that the forecasting strategy has performed poorly with an

average return of 6.7% pa, whilst the ZRP long/short strategy has produced an

average return of 11.7% pa which is similar to the Equity REIT return.

Table 5 presents similar results to those of Table 4, but in this case returns are

measured relative to a buy hold Equity REIT benchmark. The table presents average

excess returns (return of the strategy – the Equity REIT return) and the standard

deviation of the excess return. Also displayed is the information ratio which is defined

13
as the ratio of the excess return divided by the standard deviation of the excess return.

This measure is similar to the Sharpe ratio. The information ratio is a risk adjusted ratio

which measures reward to risk relative to a benchmark. Strategies with higher

information ratios are preferred to strategies with low information ratios . The fourth row

of the table are the t values. The results from Table 5 indicate that none of the forecasting

strategies (both in and out of sample) significantly outperformed the Equity REIT

benchmark. With regards to the ZRP >= 0 strategies both the cash and bond strategies

produced significant excess returns at the 10% level in sample, and also produced

significant excess returns at the 1% level over the whole sample period. The results to

date have been quoted without the inclusion of transaction costs, however transaction

costs do not play a significant role in relation to ZRP strategies. The strategies are

generally slow moving because if one examines Figure 2 it is apparent that since 1990

there are only 16 trades. This occurs every time the normalized ex ante risk premium

crosses the time axis.

INSERT TABLE 5

Figure 3 displays the rolling 12 month return of the strategy ZRP >=0 + bonds.

The average excess return was 3.6% pa with a standard deviation of 8.4%pa. It is evident

from figure 3 there may be prolonged periods when the strategy is long in the Equity

REITs and consequently this produces zero excess returns. There are also periods of

underperformance, this occurred in 1992, 2001 and 2003.

INSERT Figure 3

The strategies outlined in this paper may not be suitable for an Equity REIT fund

manager who holds only equities. However if the portfolio manager holds bonds in the

portfolio this strategy may be appropriate in determining tactical shifts between equity

14
REITs and bonds. Since changes in exposure may be infrequent, as there are periods

when positions are not changed for two to three years, the strategy is probably better

suited to a large balanced portfolio with regards to the asset allocation decision. Because

the strategy is slow moving an asset allocator could increase or decrease exposure to

Equity REITs away from the strategic benchmark allocation to property. For example,

when ZRP >=0, exposure could be increased to Equity REITs by directing incoming

funds ( or reducing exposure to an unattractive asset class) to Equity REITs. Similarly

when ZRP<0, the exposure to Equity REITs could be decreased from the benchmark

allocation to property.

3. CONCLUSION

As pointed out earlier there has been considerable interest in recent times in

modeling the predictability of asset returns. Historically two approaches have been used

in predicting asset returns, these being the fundamental approach and secondly,

modeling investor’s perceptions of risk through the use of time varying risk premiums.

The fundamental approach relies upon the assumption that asset returns are influenced

by some common factors which reflect different states of the business cycle. Strategies

based upon forecasts of excess returns from macroeconomic variables have produced

positive excess returns in the 70’s an 80’s, however the performance of these models in

the 90’s has been disappointing. We adopt the alternative approach and model the time

variation of the ex ante risk premium. Variation of the ex ante risk premium appears to

capture other factors that influence Equity REIT price movements, and we concur with

Karolyi and Sanders (1998) that there are important economic risk premiums for REITs

which are not captured by conventional multiple beta asset pricing models.

15
 Returns based upon the strategies employing the time variation of the risk

premium generate positive excess returns in and out of sample. However the

strategies are probably better suited to tactical shifts of property asset allocation

owing to the slow movement of the indicators of the model. The results in and out of

sample suggest that, in the long term, strategies based upon modeling the time variation

of the risk premium can generate significant positive returns.

16
BIBLIOGRAPHY

Case, B, W. Goetzmann and K. Rouwenhorst, (1999), “Global Real Estate Markets –

Cycles and Fundamentals”, working paper Yale University.

Chen, N.F., R.Roll and S.Ross (1986) “Economic Forces and the Stock Market” Journal

of Business, Vol.59, No.3 pp.383-404.

Conrad, J. and G. Kaul, (1988), “Time Variation in Expected Returns,” Journal of

Business,61, 409-425.

Dickey, D.A. and Fuller,W.A. (1981) "Likelihood Ratio Statistics for Autoregressive

Time Series with a Unit Root", Econometrica, Vol.49.

Fama, E. and K.French, (1990), “Business Conditions and Expected Returns on Stocks

and Bonds”, Journal of Financial Economics, 25, 23-49.

Fama, E. and K.French, (1993), “Common Risk Factors in Returns on Stocks and

Bonds”, Journal of Financial Economics, 33, 23-49.

Ferson, W. and C. Harvey, (1991), “The Variation of Economic Risk Premiums”,

Journal of Political Economy, 99,385-415.

Ferson, W. and C. Harvey, (1993), “The Risk and Predictability of International Equity

Returns”, Review of Financial Studies,6,527-566.

Glascock, J, Lu C. and R. So, (2000), “ Further Evidence on the Integration of Reit,

Bond and Stock Returns”, Journal of Real Estate Finance and Economics, 20, 177-194.

Goetzmann, W. and S. Wachter, (1996), “The Global Real Estate Crash – Evidence

from an International Data Base”, working paper Yale University.

Karolyi, G. and A.B. Sanders, (1998), “The Variation of Economic Risk Premiums in

Real Estate”, Journal of Real Estate Finance and Economics, 17, 245-262.

17
Ling, D. and A. Naranjo, (1997), “Economic Risk Factors and Commercial Real Estate

Returns”, Journal of Real Estate Finance and Economics, 14, 283-307.

Ling, D., A. Naranjo and M. Ryngaert, (2000), “The Predictability of Equity Reit

Returns: Time Variation and Economic Significance”, Journal of Real Estate Finance

and Economics, 20, 117-136.

Liu, C. and J. Mei, (1992), "The Predictability of Returns on Equity REITs and Their

Co-movements with other Assets," Journal of Real Estate Finance and Economics, 5,

401-418.

Liu, C. and J. Mei, (1994), "The Predictability of Real Estate Returns and Market

Timing," Journal of Real Estate Finance and Economics, 8, 115-135.

Quan, D. and S. Titman, (1999), “Do Real Estate Prices and Stock Prices Move

Together? An International Analysis”, Real Estate Economics, Vol. 27, 2, 183-207.

Zivot, E. and D. Andrews, (1992), "Further Evidence on the Great Crash, the Oil-

Price Shock, and the Unit-Root Hypothesis", Journal of Business and Economic

Statistics, 10(3), 251-70.

18
 Table 1. Descriptive Statistics 1/72-12/03

384 Equity T/bill Long Bond Risk Premium

Observations REITs Yield Yield Return

Mean 12.7 6.3 9.6 7.4

Std. Dev. 13.5 2.9 3.1 10.4
Skewness -0.28129
Kurtosis 2.1569
Jarque-Bera 13.82a
Qx(6) 10.1
Qxx(6) 7.94
ADF(prices) -2.61
ADF(returns) -10.30a
Zivot & -5.15a
Andrews (8/89)
Qx(6) and Qxx(6) are Ljung-Box test statistics for the null hypothesis of
no serial correlation of up to the 6th order in returns and squared returns
respectively. 6 lags are incorporated as this will measure serial
dependence up to a full half year. The Jarque-Bera statistic, under the null
hypothesis, tests whether a series is normally distributed by examining the
last two moments. Its statistic is distributed as χ2 with two degrees of
freedom. The ADF statistics are Augmented Dickey-Fuller (1981) tests
for the null of non-stationarity, all incorporating 12 lags. The Zivot and
Andrews test statistics follow a t-distribution with the null hypothesis of
no structural break. There is only one period, 8/89, when the t-stat. is
significantly different from zero.
a
Indicates if there is a rejection of the null hypothesis at the 5% level.

Table 2

Results of the Ex Ante Risk Premium Against Macrofactors using Generalized Least
Squares

EXRP(t + 1) = a + bJandum(t ) + cTB(t ) + dSP(t ) + eDY (t ) + fEY (t ) + e(t )

Dependent Period Constant Jan dum TB SP DY EY adj R2

Variable
EXRP 1/73-8/89 -062 -0.001 -0.0014 -.001 -.006 0.03 0.0006
(-1.58) (-0.48) (-0.78) (-0.58) (-0.86) (2.10)
EXRP 9/89-12/03 .0.065 -0.001 0.001 .002 -.0002 -0.001 0.0003
(0.45) (-0.12) (0.46) (1.33) (-0.04) (-0.39)

19
 Table 3

Normalized Risk Premium Returns

ZRP >= 0 1/78- 8/89 9/89 – 12/03 1/78 – 12/03

Average Return 18.9 19.4 19.1

Standard Deviation 12.8 12.7 12.7

ZRP < 0

Average Return -4.5 0.8 -0.8

Standard Deviation 15.3 11.6 12.3

20
 TABLE 4

RETURNS AND STANDARD DEVIATIONS FOR DIFFERING STRATEGIES

STRATEGY 1/78 – 8/89 9/89 – 12/03 1/78 – 12/03

ZRP>=0 + CASH* 17.9 13.5 15.5
(12.0) (10.0) (11.0)
ZRP <0 + CASH 7.1 2.7 4.6
(5.4) (7.3) (6.5)
ZRP >=0 + BONDS 20.1 15.0 17.2
(13.6) (11.0) (12.2)
ZRP <0 + BONDS 6.6 4.2 5.3
(13.0) (10.1) (11.4)
ZRP LONG/SHORT 17.1 11.7 14.1
(13.1) (12.5) (12.8)
FRP>=0 + CASH 16.7 10.2 13.1
(12.4) (10.9) (11.6)
FRP < 0 + CASH 8.0 5.9 6.9
(4.6) (6.3) (5.6)
FRP>=0 + BONDS 18.9 13.1 15.7
(14.3) (11.7) (12.9)
FRP < 0 + BONDS 7.9 7.3 7.8
(12.2) (9.8) (10.9)
FRP LONG/SHORT 11.6 6.7 8.9
(13.6) (12.8) (13.2)
BUY HOLD 16.1 11.7 13.7
(13.2) (12.5) (12.8)
• The first row of each strategy is the average return. The second row is the
standard deviation of the return.

21
 TABLE 5

EXCESS RETURNS AND STANDARD DEVIATIONS FOR DIFFERING

STRATEGIES

STRATEGY 1/78 – 8/89 9/89 – 12/03 1/78 – 12/03

ZRP >=0 + CASH* 1.8 1.8 1.8
(5.4) (7.4) (6.6)
0.19 0.25 0.27
(1.2) (0.96) (1.41)
ZRP <0 + CASH -9.0 -9.1 -9.2
(12.1) (10.1) (11.0)
-0.75 -0.90 -0.82
(-2.6)** (-3.4)** (-4.3)**
ZRP >=0 + BONDS 3.9 3.2 3.6
(8.7) (8.0) (8.4)
0.45 0.40 0.42
(1.55)* (1.5)* (2.2)**
ZRP <0 + BONDS -9.5 -7.5 -8.5
(14.3) (11.3) (12.6)
-0.66 -0.68 -0.66
(-2.3)** (-2.5)** (-3.4)**
FRP>=0 + CASH 0.6 -1.6 -0.6
(4.7) (6.3) (5.6)
0.12 -0.25 -0.11
(0.45) (-0.95) (0.55)
FRP < 0 + CASH -7.3 -6.4 -6.8
(12.4) (11.1) (11.7)
-0.59 -0.57 -0.59
(-2.0)** (-2.1)** (-3.0)**
FRP>=0 + BONDS 2.7 1.4 2.0
(6.8) (6.6) (6.7)
0.40 0.22 0.30
(1.40) (0.80) (1.52)*
FRP < 0 + BONDS -7.4 -4.4 -5.8
(15.3) (13.1) (14.1)
-0.49 -0.34 -0.41
(-1.70)* (-1.26) (-2.1)**
• The first row of each strategy is the average excess return, the second row is the
standard deviation of the excess return. The third term is the information ratio
which is defined as the excess return divided by the standard deviation of the
excess return. The fourth row is the t value.
• * Significant at the 10% level
• ** Significant at the 1% level

22
 NORMALIZED RISK PREMIUM Risk Premium %
31 31
/0 /0
1 1
31 /1 31 /19
/0 97 /0 75
5

-4
-3
-2
-1
0
1
2
3
4
-40
-30
-20
-10
0
10
20
30
40
50
60

1 1
31 /1 31 /19
/0 97 /0 76
1 6 1/
31 /1 31 19
/0 97 /0 77
1 7 1
31 /1 31 /19
/0 97 /0 78
1 8 1/
31 /1 31 19
/0 97 /0 79
1 9 1
31 /1 31 /19
/0 80
/0 98
1/ 0
1
31 /19
31 1 /0 81
/0 98
1 1
1 31 /19
31 /1 /0 82
/0 98
1 2 1/
31 19
31 /1 /0 83
/0 98 1
Ex Post Risk Premium
Ex Ante Risk Premium

1/ 3 31 /19
31 1 /0 84
/0 98 1
1 4 31 /19
31 /1 /0 85
/0 98 1/
1/ 5 31 19
31 1 /0 86
/0 98 1
1/ 6 31 /19
31 1 /0 87
/0 98 1
1 7 31 /19
31 /1 /0 88
1

Figure 2
/0 98 31 /19
1/ 8
31 1 /0 89
1/
/0 98
1 9 31 19
31 /1 /0 90
/0 99 1
0
Date
31 /19
1

DATE
31 /1 /0 91
1
Figure 1

/0 99
1 1 31 /19
/0 92
31 /1 1
/0 99
1 2 31 /19
/0 93
31 /1 1/
/0 99 31 19
1/ 3 /0 94
31 1 1
/0 99 31 /19
1 4 /0 95
31 /1 1/
/0 99 31 19
1 5 /0 96
31 /1 1
/0 99 31 /19
1 6 /0 97
31 /1 1
Ex Post and Ex ante Risk Premium over Tbills

/0 99 31 /19
1 7 /0 98
31 /1 1
/0 99 31 /19
1/ 8 /0 99
1
NORMALIZED RISK PREMIUM RELATIVE TO LONG BONDS 36 MONTH WINDOW

31 1 31 /20
/0 99
1 9 /0 00
31 /2 1
/0 00 31 /20
1 0 /0 01
31 /2 1
/0 00
31 /20
1 1 /0 02
31 /2 1/
20
/0 00

23
03
1/ 2
20
03
 EXCESS RETURN %
31
/1
2/
31 197

-20.00
-10.00
0.00
10.00
20.00
30.00
40.00
/1 8
2/
31 197
/1 9
2/
31 198
/1 0
2/
31 198
/1 1
2/
31 198
/1 2
2/
31 198
/1 3
2/
31 198
/1 4
2/
31 198
/1 5
2/
31 198
/1 6
2/
31 198
/1 7
2/
31 198
/1 8
2/
31 198
/1 9
2/
31 199
/1 0
2/
31 199
/1 1

DATE
2/
31 199
/1 2
2/
Figure 3 - ZRP >= 0 + BONDS

31 199
/1 3
2/
31 199
4
ROLLING 12 MONTH EXCESS RETURN

/1
2/
Average excess return = 3.6% , Stdev = 8.4%

31 199
/1 5
2/
31 199
/1 6
2/
31 199
/1 7
2/
31 199
/1 8
2/
31 199
/1 9
2/
31 200
/1 0
2/
31 200
/1 1
2/
31 200
/1 2
2/
20
03

24