Lagos state university of science and

technology

Group6
Topic:Risk and Uncertainty: Trade Off
Return
Course: Business Finance
Group members are;
Oluwatayo Oluwaseun Victor 230201010051
Adeshina Mariam Oluwabukola 230201010052
Ayodele Wisdom Ayomide 230201010053
Adeyemi Qudus Adedeji 230201010054
Akinsanya Zainab Opeyemi 230201010055
Rasheed Muktar Olakunle 230201010056
Lateef Abigeal Oluwakemi 230201010057
Edukpe David Adebayo 230201010058
Longe Jubril Adeshina 230201010059
Akingbola omolola Christiana 230201010060

1|P a g e
 Risk and Uncertainty: Trade Off Return

Abstract

The Merton’s (1973) Intertemporal Capital Asset Pricing Model (ICAPM) implies that the conditional risk

should contain the unspanned Knightian uncertainty component which may be very important when interest rate is

not sufficient to describe the investment state, but the conventional risk–return tradeoff tests only consider the

spanned risk component and end up with inclusive findings. Borrowing a machine-learning based index of

economic policy uncertainty (EPU) from Baker, Bloom and Davis (2016) who design this index to capture the

Knightian uncertainty in macroeconomic, monetary, political, fiscal, national security, sovereign credit,

international trade, and currency areas, we detect a significant EPU–return tradeoff relation. Further empirical

analyses suggest that more than 60% of this EPU–return tradeoff should be attributed to the Knightian uncertainty

while less than 10% of it can be attributed to the spanned risk. More interestingly, we find that expected stock

return is significantly positively related to conditional risk as suggested by Merton when this EPU index is

employed as the proxy of the conditional risk, and this tradeoff is mainly attributed to the component of Knightian

uncertainty–return tradeoff rather than the component of spanned risk–return tradeoff. Using this EPU index, we

find that the risk–return tradeoff exists in corporate bond markets. Overall, our analyses suggest that the Knightian

uncertainty unspanned by capital market risks plays a very important role in detecting a positive risk–return

tradeoff relationship.

KeyWords: ICAPM, Conditional risk, Unspanned risk, Tradeoff, Knightian Uncertainty, Policy
Uncertainty

2.1 Introduction

A main tenet of modern finance is that investors demand compensation for investing in assets
whose payoffs are uncertain. The most challenging job in testing this positive risk-payoff
relationship is to properly measure the unobservable uncertainty. The traditional view is that

2|P a g e
 the tradeoff is inherently estimable by estimating the means and variances of equity returns, as
in Merton (1973, 1980). As such, extant studies test this relationship using market portfolio

1
return as the proxy of uncertainty but the findings are unfortunately inclusive. According to

Merton (1973), the risk-return tradeoff can be approximated by variance-return relationship if

and only if return variance is a sufficient stochastic of conditional risks, or equivalently, if and

2
only if interest rate is a sufficient statistic of the state. This sufficiency cannot be meet both

theoretically and empirically. According to Knight (1921) and Keynes (1937), uncertainties
beyond capital market are substantial and can significantly change investment state. The
literature defines the uncertainty that is measurable by investment portfolio return volatility as
risk, and the uncertainty which is not spanned by investment portfolio returns as Knightian
uncertainty. In this study, we call the later Knightian uncertainty and unspanned risks
interchangeably. Knightian uncertainty matters because it can have large consequences on
financial markets via changes in regulation, taxes, geo-political security, international treaties
(trade). Empirically, there is a large body of literature suggests that many unspanned economic
“uncertainties” can systematically impact stock prices. For example, there are enormous studies
show that political election, monetary policy, regulations, exchange rate, and others

3
significantly impact stock prices. Moreover, the Merton (1973) model suggests that such

uncertainty should be included in the conditional risk estimate if it matters in the generating
process of stock prices. Unlike the spanned risk (return variance), however, the Knightian
uncertainty is inherently hard to measure.

In this study, we test this risk-return relationship using a novel approximation of the

conditional risks which contains both conventional spanned risks and unspanned Knightian

uncertainty. We first

verify the importance of Knightian uncertainty in the generating process of stock prices by
introducing an additional state variable to the model and ending up with a two-risk-factor
model, out of which the first approximates the conventional spanned risks and the second
represents the Knightian uncertainty. The rationale in our two-factor model is similar to
that in Maio and Santa-Clara (2012), who extend the single-factor ICAPM model to a
multi-factor model by introducing an explicit state variable to capture the “state” in capital
market beyond interest rates to justify the Fama-French 3-factor and other multi-factor
models. In our two-factor model, however, which risk contributes more to the risk-return

3|P a g e
 relationship seems an empirical question. We use the “Economic Policy Uncertainty”
(EPU henceforth) Index proposed by Baker, Bloom and Davis (BBD, 2016) as the proxy
for the unspanned Knightian uncertainty. This index is appropriate because it is novelly
constructed by extracting “policy uncertainty” in capital market, political, monetary, fiscal,
national security, taxing, international trade, currency, government spending, healthcare
and other areas from 10 leading newspapers in the United States using text analysis

4
approach.

Specifically, we test the risk-return tradeoff relationship using the comprehensive new risk

proxies by Baker, Bloom and Davis, and compare our test with the conventional tests which

focus only on the spanned risk proxies, including the variance, skewness and kurtosis of market

portfolio returns, the range and GARCH volatility of market portfolio returns. We test whether

each individual proxy is a sufficient stochastic of investment risk and focus on the EPU index

since it is constructed to capture both spanned and unspanned risks from the most areas beyond

stock markets. Consistent with previous studies, monthly stock return is negatively related to

the variance of past-month daily returns over the period from 1900 to 2016 while the

corresponding Newey-West statistic is as small as 0.10. Monthly stock returns are also

negatively (and insignificantly) related to the skewness or the range volatility of past-month

daily returns. Monthly stock returns are positively related to the GARCH variance but this

relationship becomes negative after controlling macroeconomic variables. Monthly market

returns is positively related to the kurtosis of daily returns in past month while this relationship

is pretty insignificant. The Economic Policy Uncertainty (EPU) index, however, is significantly

positively related to market returns and this relationship remains after controlling conventional

macroeconomic variables. We further find that the positive EPU-return tradeoff remains

when the expected returns are approximated by cumulative market returns over
subsequent 3 to 12 months.
In short, we detect a significant EPU-return tradeoff relation. We are very interested in

whether this tradeoff is mainly driven by the Knightian uncertainty component in the EPU

index. In addition to the Knightian uncertainty, the literature, however, suggests three

alternative explanations for the existence of this tradeoff, including spanned risks, the

dynamics of investor’s risk aversion and macroeconomic investment opportunity state. Since

4|P a g e
 all the four explanations are not mutually ex-clusive, we identify the dominant one by linearly

decomposing the EPU index into four components: the component explained by conventional

risk proxies, the component by investor’s time-varying risk aversion, the component by

macroeconomic state variables, and a residual part as a proxy of the unobserved Knightian

uncertainty. To be solid, we employ all most conventional spanned risk proxies, including the

conventional variance, range variance, GARCH variance, skewness and kurtosis of the market

portfolio returns, and find that all of these proxies together have less than 8% power to explain

the EPU-return tradeoff. The investor’s time-varying risk aversion, approximated by the

standardized Sharpe ratio of CRSP value-weighted stock market portfolio, has around 2%

5
explanatory power of the EPU-return tradeoff. The market investment opportunity state

variables, including the dividend yield of the S&P 500 index, inflation rate, default spread, and

stock market portfolio return, altogether have around 28% explanatory power. In other words,

the total explanatory power of the three alternative sources contained in the EPU-return

tradeoff is less than 40%, and more than 60% of the tradeoff should be attributed to the

Knightian uncertainty contained in the EPU index. In sum, our analyses suggest that unspanned

Knightian uncertainty in monetary policy, fiscal policy, regulation, national security, currency,

taxes and other areas have significant impacts on the generating process of stock returns and

cannot be ignored in testing the risk-return relationship.

To better understand how the EPU index effectively capture the unspanned Knightian

uncertainty and whether the captured uncertainty in this index are the main underlying drivers

of positive risk-return tradeoff, we decompose the EPU index into two orthogonal components,

that is, the spanned risks by market return moments and the unspanned uncertainty, by

regressing this index

on conventional spanned risk proxies. The fitted EPU index captures the
uncertainty contained in conventional risk proxies and the EPU residual contains
unspanned Knightian uncertainty beyond conventional risk proxy. The fitted EPU
index (spanned risk) out of all linear regression models fails to describe the
positive risk-return tradeoff relationship but the unspanned uncertainty part
strongly supports this relationship.

5|P a g e
 In addition to investigating the risk-return relationship in stock markets, we also test
whether the Knightian uncertainty matters in corporate bond markets. We use the long-
term corporate bond issue index defined by Goyal and Welch (2008) as the approximation

6
of corporate bond markets. Consistent with the findings in stock market, we do not find a
significant spanned risk-return tradeoff but detect a significant EPU-return tradeoff in
corporate bond market. The three alternative explanations of spanned risk, time-varying
risk aversion and macroeconomic state variables together explain around 35% of the bond
market EPU-return tradeoff, implying that around two thirds of the tradeoff should be
attributed to the Knightian uncertainty.

Since the EPU index is constructed by extracting uncertainty texts from 12

categories,including economic policy, monetary policy, fiscal policy, taxes, government
spending, healthcare, national security, government entitled programs, regulations, trade,
and sovereign credit and currency. we further test the EPU-return relationship using
categorical EPU indexes, We find that the EPU-return tradeoff exists in 11 categories and
significantly exists in 9 categories, and that Knightian uncertainty contributes 60% to 96%
of the categorical EPU-return tradeoffs. We also examine the risk-return relationship for
18 international markets using similar Economic Policy Uncertainty index constructed for
each market based on local leading English newspaper(s). The results are mixed and
suggest the complication in building an effective unspanned index for other markets.

The main contribution of this study is to show the importance of Knightian uncertainty

unspanned by capital market risks on stock market and the risk-return relationship in particular.

The paper by Joslin, Priebsch and Singelton (2014) show that unspanned macro risks play an

important role in pricing US treasury securities. Our study shows that unspanned risks are also

very influential in the

pricing process of corporate bonds and stocks. There are enormous studies show
that unspanned risks from political, monetary, fiscal, national security, taxing and
currency systematically impact stock prices. We further show that incorporating
these uncertainties is critical for both corporate bond and stock investors to
estimate their investment risks.

6|P a g e
 The paper which is most close to our study is Anderson, Ghysels and Guergens (2009) who

test whether uncertainty beyond stock return variance is priced in stock prices. They use a

similar framework but assume that investors are averse to model misspecification caused by

Knightian uncertainty. We assume that investors acknowledge the pricing power of Knightian

uncertainty. They use the disagreement of analysts forecasts as a measure of uncertainty or

deviations from the benchmark. We are interested in the economic sources of the deviations.

Hence, we have to find proxies for political economic uncertainty that go beyond those

measured by market variances and covariances. We use the EPU index of Baker, Bloom and

Davis, and show that EPU does have an impact on asset prices. Our study is also related to

Brogaard and Detzel (2015) who test whether current EPU can predict subsequent stock

returns. In addition to that we provide a simple model to justify their test, we use a longer

version of the EPU. They use only one from 1985-2014. It is hard to measure political

uncertainty within a short time span. We use the EPU that starts from 1926. This reasonably

long sample allows us to not only have more power in the tests but also to correctly identify

political uncertainty variations. We further decompose EPU into spanned and unspanned parts

and focus on the Knightian effect. We also extend these analyses to corporate bond markets.

The remainder of this paper is organized as follows. Section 2 describes the

motivation and methodologies. Section 3 reports the main empirical results. Section 4
concludes the paper.

2.2 Knightian Uncertainty and ICAPM

In this section, we propose a theoretical model to show that Kinghtian uncertainty is priced in

stock prices but not reflected in stock return volatility by introducing an explicit state variable

of macroeconomic uncertainty to the Merton’s (1973) ICAPM model. Although Merton’s

(1973, equation 13) model allows the generating process of stock returns is a function of a

multi-dimension state variable, it is well known that the illustrative ICAPM model does not

explicitly identify the state variable in the covariance matrix of market and individual returns.

Some papers introduce an explicit state variable in Merton’s model. For example, Anderson,

Ghysels and Juergens (2009) introduce a state variable proxy into the Merton’s model and

derive a two-factor model. In this model, the explicit state variable does not directly but

7|P a g e
 indirectly impact stock prices through model ambiguity. Maio and Santa-Clara (2012) extend

the Meteron’s single-factor ICAPM model to multi-factor models by introducing an undefined

state variable. As pointed by Merton, and Maio and Santa-Clara, any state variable which is

included in Merton’s model should be able to predict market portfolio returns.

th
The dynamics of the k investor’s wealth are given by:
N N
dW = w (m r )W dt + (r W c )d + w W s dB ;
t å i;t i;t f ;t t f ;t t t t å i;t t i;t i;t (3)
i=1 i=1

where wi;t denotes the portfolio weight for stock i in month t, and ct is the
th
consumption. The k investor lifetime utility function is given by:

Z¥
E0 U(c(s); s)ds ; (4)
0

subject to the intertemporal budget constraint in Equation (3). The investor’s

dynamic optimization becomes:

w;c N
( (m
i=1

r )W
0 = max U(c;t) + Jt + Jw å i;t f ;t t ct + Jxax
i=1 j=1
N N Ni=1

2
+(1=2)Jww åwi å w jsi; jWt + Jxwbx åwisiqx;iWt + (1=2)bxxJxx ; (5)

where qx;i denotes the instantaneous correlation between dBt and dBi;t . The first
order conditions (FOC) are:
0 = Uc(c;t) Jw(W; x;t) (6)

and for i = 1; 2; :::; n:

N

8|P a g e
 The market demand becomes:

N+1 K K
dM = å DidPi=Pi + å(Wi ci)dt = ådWi (10)
i=1 i=1 i=1

Replace wi by Di=M and assume that there exists a stock whose return is perfectly related to the
Knightian uncertainty (we denoted its standard deviation as sx), then Equation (8) can be simplified as:
N

mi r f = (M=P) å w jsi j + (Qbx=Psx)six: (11)

j=1

The market portfolio return process can be simplified as:

2
mM r f = (M=P)sM + (Qbx=Psx)sMx: (12)

Equation (12) says that the expected returns of market portfolio is a linear function of market portfolio
(spanned) risk and macroeconomic volatility unspanned by market portfolio returns. Following
Cochrance (2005, Chapter 9), we approximate Equation (12) in discrete time as:

2
E (R R )=g s +g s : (13)
t M;t+1 f ;t+1 M M K Mx

Empirically, we test the following tradeoff relationship:

Et [RM;t+1] = a + bM Riskt + bK U ncertaintyt ; (14)

where bM is the relative risk aversion of the representative agent to spanned risks and b K is the risk
aversion to unspanned Knightian uncertainty.
The literature suggests that spanned risks can be approximated by stock return volatility or
higher moments. Following the convention, we approximate spanned risks using singleton or
various sets of market return volatility, range volatility, GARCH volatility, skewness, and kurtosis.
To partition the unspanned risks and spanned risks captured in the EPU index, we decompose it
using the following linear regression:

EPUt = k + b Risk(Rt ) + et ; (15)

where Risk(Rt ) can be either a singleton or a vector of spanned risk proxies. According to our

argument, the fitted EPU index from the equation captures conventional spanned risks and e t
captures unspanned uncertainty.

3 Empirical Analysis

In this section, we first describe the data used in empirical analyses. Then we test whether risk
return and uncertainty return relationships empirically exist with conventional risk proxies of (i.e.,
stock return volatilities) and the EPU index as a proxy of Knightian uncertainty. As we will show
soon that a significant EPU return tradeoff exists in capital market but a significant risk return
tradeoff does not. We conduct further analyses to test whether the EPU return tradeoff should be
attributed to Knightian uncertainty by excluding all alternative explanations for the EPU return
relation. We then turn to test the efficacy of the Knightian uncertainty return tradeoff.

3.1 Data

The data used for our analysis span from January 1926 to December 2016, and are from various sources. We use
the value-weighted index portfolio from the Center for Research in Security Prices (CRSP) as a proxy of the U.S.
stock market. We use the long-term corporate bond issues, which spans from January 1926 to December 2016,

10
defined by Goyal and Welch (2008) as a proxy of the U.S. corporate bond market. Our main proxy of unspanned

uncertainty is the so-called “Economic Policy Uncertainty” (EPU) constructed by Baker, Bloom and Davis (2016),
a text analysis based measure of policy uncertainty news out of six (between 1926 and 1984) to ten (from 1985 to
2016) leading newspapers in USA. Uncertainty is measured by the change in the scaled counts of news articles
containing terms of economic and policy uncertainty, such as uncertainty, uncertain, economic or economy,
regulation, deficit, legislation, Congress, White House, Federal Reserve, the Fed, regulations, regulatory, deficits,

11
congressional, legislative, and legislature. To make the magnitude of the index are comparable to stock returns

and variance, in our analysis we take log o

12
this index and then scaled by 100. We also simply scale the index by 1000 and
13
the results are almost unchanged.
In addition, we follow the literature and construct the spanned Knightian risk proxies
using market portfolio returns. Specifically, we construct conventional volatility and range
volatility of market portfolio returns based on daily returns within the most recent month,
and GARCH volatility using monthly market portfolio returns throughout the whole
sample period. We also compute the skewness and kurtosis of daily market returns within
each month and use them as alternative proxies of spanned conditional risk. The
macroeconomic variables, including the dividend yield of the S&P 500 index, the inflation
rate based on Consumer Price Index (all urban consumers), and the default spread defined
as the yield spread between Moody’s rated BAA and AAA corporate bonds, as well as the
risk-free rate defined as the Treasury-bill rate are from Amit Goyal’s website.

Figure 1 plots the time series of all risk and uncertainty proxies over the entire sample

period and suggests that each proxy evolves differently from others. The market portfolio

return variance, the conventional conditional risk spanned risk proxy, jumps during the great

recession period in 1930s, the WWII period, the oil crisis period in 1980s, and the recent

housing crisis period. The Economic Policy Uncertainty Index, which contains both spanned

and unspanned risks by construction, jumps during the 1930s but does not during other times.

Table 1 reports the summary statistics of the market portfolio returns and each uncertainty

index over the sample period and the pairwise correlations of these variables. Over the whole

sample period from January 1926 to December 2016, Panel A in Table 1 shows that the market

portfolio delivers an average monthly return around 1% with a standard deviation of 5%. The

market portfolio return is significantly negatively autocorrelated, evidence of an AR(1)

process. The mean, median and standard deviation of each conventional risk proxy, including

market volatility (%), range volatility (%), GARCH volatility (%), skewness and kurtosis are

reported.

nd th 14
from the 2 to 6 rows. Column 5 suggests that most of these series (except the
skewness) are AR(1) processes and the GARCH volatility series is highly
autocorrelated with an AR(1) coefficient of 0.98. The summary statistics of the EPU
th
index, are in the 7 row and suggest that it is a AR(1) process. Panel B in Table 1
shows that conventional risk measures of market return volatilities, including the
conventional volatility, range volatility and GARCH volatility, are highly correlated
with each other. The correlation between conventional volatility and range volatility is
as high as 0.95. The conventional risk proxies are significantly related to the EPU
index. However, the small magnitudes of these correlations suggest that the unspanned
risk proxies may capture risks beyond uncertainty in capital market.
3.2 Does EPU–Return Tradeoff Exist?

In this section, we test whether the spanned risk-return tradeoff and EPU-return tradeoff
exist empirically. According to Merton’s (1973) ICAPM model, we expect to detect
significant positive tradeoffs if the proxies of spanned risk are effective. We first conduct
simple regressions of risk-return tradeoff test for each singleton spanned risk proxy and the
EPU index and conduct further tests controlling the main macroeconomic predictors used
in the literature. The empirical test results are illustrated in Table 2 and do not suggest any
spanned risk-return tradeoff but suggest a significant EPU-return tradeoff. Consistent with
previous studies, the tradeoff relationship in Panel A (simple regressions) is slightly
negative and statistically insignificant when conditional risk is approximated by the

15
conventional volatility or range volatility of market portfolio returns. The tradeoff
relationship becomes positive but insignificant when conditional risk is approximated by

16
GARCH volatility or kurtosis of the market portfolio returns. However, the EPU-return
relationship is positive (0.33) and statistically significant (t=2.15). The results are
consistent with Baker, Bloom and Davis (2016) that the EPU index can effectively
measure risks broadly spanned from capital markets to areas of macroeconomy, monetary
policy, fiscal policy, national security, international.
trade, sovereign credit and so on. Panel B in Table 2 (multiple regressions) shows that the findings in Panel

A remain after controlling other return predictors. In fact, the EPU-return relationship is more positive (0.70)

and significant (t=3.19) after controlling additional return predictors.

3.3 Knightian Uncertainty and the EPU–Return Tradeoff

In this section, we investigate whether the positive EPU-return relationship documented in last section can be
attributed to the Knightian uncertainty contained in the EPU index. The extant studies on risk-return
relationship provide four potential explanations, which may not be mutually exclusive. First, the
conventional risk-return tradeoff studies suggest that the existence of EPU-return tradeoff may be due to that
EPU can be spanned by conventional risk proxies. Recent studies (e.g. Campbell and Cochrane, 1999;
Mezly, Santos and Veronesi, 2004; Wachter, 2006; Guiso, Sapienza and Zingales, 2018) suggest that the
existence of EPU-return may be because EPU can capture investor’s time-varying risk-aversion. Moreover,
the conventional risk-return tradeoff tests also implies that EPU may be an effective proxy of
macroeconomic investment opportunities. Regardless, the EPU index by construction is designed to capture
uncertainties of monetary policy, political events, international relationship and others beyond capital
markets, which we refer to as Knightian uncertainty in the sense of unspanned by investment return
moments. Because the four explanations are not mutually exclusive, we examine the relative importance of
each one in explaining the variation of the EPU index. The most important explanation should have the
largest explanatory power for the variation of the EPU index.

We first test whether the EPU index is a sufficient proxy of spanned risks by regressing EPU on the

conventional risk proxies, including market return variance, range variance, GARCH variance, skewness and

kurtosis. The empirical results are in Table 3 and do not support the spanning relationship between EPU and the

conventional risk proxies. Table 3 provides two interesting observations. First, the EPU index is positively and

significantly related to each conventional spanned risk proxy except the market return kurtosis, evidence that it

does effectively capture conventional spanned risks. Columns 1 to 4 show that the coefficients of these individual

conventional risk proxies from simple regressions are around 0.15 and statistically significant at 1% level. The

significant positive relationship between the EPU index and the market return volatility or skewness remains in

the multiple regression (column 6) while such regression may suffer from collinearity
because the conventional spanned risk proxies are significantly highly correlated to
each other as shown in Table 1. Second, the EPU index also captures uncertainties
2
unspanned by conventional risk measures. The R s of all regressions in Table 3 span
 2
from 0.04% (column 5) to 4.42% (column 1). The low R s suggest that more than 95%
of the variation of the EPU index cannot be explained by conventional spanned risk
measures, and consistent with Baker, Bloom and Davis (2016) that the EPU index is
designed to capture all risks and uncertainties.

Second, we test whether the EPU index is spanned by time-varying risk aversion by
controlling a time-varying risk-aversion variable in previous spanning tests. Existing
studies propose several approaches to proxy investor’s time-varying risk aversion.
Campbell and Cochrane (1999) argue that time-varying risk aversion is implied in the
process of Sharpe ratio maximization. Following this argument, we use the squared Sharpe
ratio of the daily returns of the CRSP value-weighted index portfolio within each month as
a proxy of time-varying risk aversion. The squared Sharpe ratio ensures non-negative
results and captures the non-linear relationship between risk-aversion and investment
states. Based on the framework of Bad Environment-Good Environment, Bekaert,
Engstrom and Xu (2017) construct a proxy of time-varying risk-aversion using
macroeconomic fundamentals, asset prices and option prices. Moreover, Baker and
Wurgler (2006) propose an investor sentiment index to approximate the dynamics of
aggregated risk preference. In this study, we use the squared Sharpe ratio of market returns

17
as our main proxy because the sample periods of other two proxies are relatively short.

The results are in Table 4. First, the coefficient of the squared Sharpe ratio is between -
0.19 and -0.15 and statistically significant at 1% level, suggesting that the EPU index is
significantly related to the time-varying risk aversion. Second, the R-squared is still small,
between 3% and 10%, suggesting that the majority part of the variation of the EPU index
cannot be explained by the variations of the conventional risk proxies and investor’s risk-
aversion dynamics.

Lastly, we test whether the EPU index can be spanned by macroeconomic investment

opportunities by further controlling the main macroeconomic state variables, which are the

dividend yield of the

S&P 500 index, inflation rate, default spread, and stock market portfolio returns, proposed by Ghysels,
Santa-Clara and Valkanov (2005). The results are in Table 5. The coefficients of the macroeconomic
state variables except the stock market return are significant. The R-squareds increase to between 35%
and 37%. Both suggest that the EPU is related to macroeconomic states. However, the R-squareds also
suggest that the majority (more than 60%) of the variation of the EPU cannot be explained by
conventional risk proxies, risk-aversion dynamics and macroeconomic states. Following the argument in
Section 2, we argue that the EPU-return relationship should be driven by the Knightian uncertainty out
of capital markets.

3.4 Knightian Uncertainty–Return Tradeoff

In last section, we show that the significant EPU-return relationship should be attributed to the Knightian
uncertainty which is not spanned by risk proxies. In this section, we investigate how important it is to
incorporate unspanned risks (uncertainty) into the conditional risk in detecting the risk-return relationship.
We re-conduct the conditional risk-return tradeoff tests after including both spanned and unspanned risk
measures. We use the unexpected EPU index (residuals from the decompositions in Table 3 or Equation 15)
as the proxy of unspanned uncertainties and two types of proxies of spanned risks, that is, the conventional
conditional risk measures based on market portfolio returns and the fitted EPU index. By construction, we
can perceive the relative importance of spanned and unspanned risks in the conditional risk-return
relationship. The empirical results are reported in Table 6. In Panel A, we include the unspanned proxy
(unexpected EPU Index) and various sets of the conventional spanned risk proxies in the risk-return tradeoff
tests. The most impressive observation from Panel A is that the coefficients on all uncertainty proxies
(models) are positive and statistically significant at 1% level. The coefficients on market return volatilities
are positive but only two out of four cases are marginally significant (10%level). The coefficients on
GARCH volatility are positive but small and insignificant. The coefficients on market return skewness or
kurtosis are negative. In Panel B, we conduct conditional risk-return tests using the fitted EPU indexes from
Table 3 as spanned risk proxies. Similar to Panel A, the coefficients on the unspanned risk proxies are around
0.70 and statistically significant at 1% level across all models. The coefficients on all spanned risk proxies
are insignificant while some of them are positive.

To sum up, Table 6 suggests that unspanned uncertainty is the dominant component of

investor’s total risk, and that the inconclusive findings on risk-return tradeoff relationship in
previous studies may be caused by the failure of incorporating unspanned risk into the

conditional risk estimation.

3.5 Uncertainty–Return Tradeoff: Long Horizon

In previous sections, we find significant risk-return tradeoff relationship after

incorporating unspanned risks into conditional risk. In this section, we test how robust the

findings are. We first test whether whether the tradeoff relationship exists at long horizon,

that is, we test risk-return relationship between conditional risk and cumulative excess

returns over future three to 12 months.

k T T
Pi =1[1 + (RM;t+1 R f ;t+1)] 1 = a + b Riskt + d Controlt + et+1; (16)

k
where Pi =1[1 + (RM;t+1 R f ;t+1)] 1 is the cumulative excess market return over months t + 1 to
t +k, Riskt denotes the vector of spanned and unspanned risk measures in month t, and Control t is the
vector of controlled macroeconomic variables in month t, including short-term interest

rate, default spread, the dividend yield on the S&P 500 index, and one-month lagged

market portfolio return.

Same to Panel A Table 6, we use various versions of unexpected EPU index from Table 3
as proxies of unspanned risks and the corresponding conventional spanned risk measures in our

18
tests. The empirical results of risk-return tradeoff tests between spanned and unspanned risks

and cumulative excessive market portfolio returns over future three, six and 12 months are
reported in Panels A, B and C in Table 7, respectively. Table 7 illustrates two interesting
findings. First, the coefficient of unspanned risk proxies are positive and statistically significant
at 1%level in all specifications and all panels. This coefficient is around 2 for tradeoff tests of
risk and three-month cumulative returns, around 3.6 for tests with six-month cumulative
returns, and around 6.3 for tests with 12-month cumulative returns. These are significant
evidence that risk-return tradeoff exists in long-horizon. Second, the coefficients on spanned
risk proxies, including range and GARCH volatilities, skewness, and kurtosis of market
portfolio returns, are either negative or positive but insignificant across all panels (horizons).
The coefficient of the conventional market-return volatility
is positive and insignificant. In the tradeoff test of risk and 6-month returns, however, this

coefficient becomes positive and significant when all spanned risk measures are included. This

inconsistency arises because of the collinearity issue in the specification as we already see in

Table 1 that the correlation between the conventional and range volatility of market returns is

as high as 0.95, and the correlation between the conventional and GARCH volatilities is 0.5

and significant. To sum up, Table 7 suggests that risk-return tradeoff relation exists in long

horizons and ignorance of unspanned risks in the conditional risk estimation may fail to detect

such a relationship.

3.6 Knightian Uncertainty and Corporate Bond Markets

3.6.1 Does EPU-Return Exist in Bond Market?

In this section, we test whether the risk-return tradeoff and EPU-return tradeoff exist in

corporate bond markets. According to Section 2, we expect to detect a significant positive

EPU-return tradeoff. We use the average returns of the long-term corporate bonds issued by

19
U.S. firms as the main approximation of corporate bond market. The sample spans from

January 1926 to December 2016. Since the daily returns are not available, we use GACRH

variance as a proxy of Knightian risks in bond markets. We conduct both simple regressions of

return-risk tradeoff test for the GARCH variance and the EPU index and regressions of risk-

return tradeoff tests controlling the main macroeconomic predictors used in the literature. The

empirical test results are illustrated in Table 8. The first three columns are the results of simple

regressions and the second three columns are results of multiple regressions. Neither case,

however, suggests a significant risk-return tradeoff but both cases suggest a significant EPU-

return tradeoff in bond market. Similar to studies on risk-return tradeoff equity markets (e.g.

Bali, 2008; Nyberg, 2012), the coefficient of the GARCH variance is positive but insignificant

in both simple and multiple regressions. However, the EPU-return relationship is positive and

very significant. The coefficient of the EPU index is 0.29 (t=4.43) in the simple regression and

0.26 (t=3.17) in the multiple regression. The coefficients of the EPU index are unchanged

when the GARCH variance is added in the regressions (Columns 3 and 6).
3.6.2 EPU–Return Tradeoff and Knightian Uncertainty

Given the significant EPU-return tradeoff, we investigate whether this tradeoff is a proxy of Knightian
uncertainty-return tradeoff by excluding the three alternative explanations suggested in Section 3.3. We
conduct similar regressions, that is regressing the EPU index on the bond market GARCH variance, the
squared Sharpe ratio, the four macroeconomic variables, or various combinations of these variables. The
results are in Table 9. The first column reports the results of whether the EPU-return relationship can be
explained by return variance of bond market portfolio. The coefficient of the GARCH variance is 0.31
(t=5.21) and the R-squared is only 9.45%, suggesting that the spanned risks in bond market cannot explain
the EPU-return relationship. The second column reports the results of whether the EPU-return relationship
can be explained by the dynamics of investor’s risk-aversion proxied by the squared Sharpe ratio of stock
market returns. The low R-squared (2.99%) does not support this explanation. The third column reports the
results whether the EPU-return tradeoff can be explained by macroeconomic state variables. The coefficients
of the dividend yield on the S&P 500 index, the inflation rate and the default spread are statistically
significant and the R-squared is 34.61%, implying that macroeconomic state may explain a significant part of
the EPU-return tradeoff and that the majority of this tradeoff (around two thirds) remained unexplained.
Columns 4 and 5 report the results of whether the EPU-return tradeoff can be explained by both spanned risk
and risk aversion as well as macroeconomic state. The R-squared becomes 35.21% when all proxy variables
are considered, suggesting that the majority of the EPU-return tradeoff cannot be explained by these three
explanations. In short,the results in Table 9 are again consistent with Baker, Bloom and Davis (2016) that the
EPU index can effectively measure risks broadly spanned from capital markets to areas of macroeconomy,
monetary policy, fiscal policy, national security, international trade, sovereign credit and so on.

3.7 Further Analysis

3.7.1 Categorical Knightian Uncertainty

In this section, we conduct several analyses to further understand the sources of Knightian uncertainty. As we

discussed in previous section, the EPU index is designed to capture uncertainty outside capital
market which significantly affects stock prices but cannot be spanned by conventional risk factors, risk
aversion dynamics or macroeconomic investment opportunity states. In addition to the comprehensive
uncertainty index, Baker, Bloom and Davis also use the same approach to generate uncertainty index for each
unspanned category, including uncertainties from
(1) economic policy, (2) monetary policy,
(3) fiscal policy, (4) taxes,
(5) government spending, (6) healthcare,
(7) national security, (8) government entitled programs,
(9) general regulations, (10) financial regulations,
(11) trading policy, and (12) sovereign debt and currency.
These categorical indexes individually capture Knightian uncertainty in each category and allow us to
explore the relative importance of the categorical Knightian uncertainty in investor’s total uncertainty.
Following the analysis in previous section, we first test whether the EPU-return tradeoff exists in each
category. We then investigate whether such tradeoff can be attributed to Knightian uncertainty by testing
whether market volatility, dynamic risk aversion and macroeconomic investment state can explain the
variation of each categorical EPU index. Finally, we test whether categorical Knightian uncertainty-return
tradeoff exists using a derived Knightian unertainty proxy in the tradeoff test.

The empirical results of the categorical EPU-return tradeoff tests are in Table 10. Panel A reports the results

of regressions of expected market return (approximated by next-month return) on each categorical EPU index and

conventional macroeconomic variables. The results suggests that the EPU-return tradeoff is significantly

supported by the categorical EPU index using uncertainty events in areas of economic policy, fiscal policy, taxes,

healthcare policy, nation security and government entitled program, respectively. Panel B further controls the

volatility of market portfolio returns and suggests that the insignificant positive tradeoff in Panel A in areas of

monetary policy, government spending, and regulations is driven by the negative tradeoff of market volatility-

return. The tradeoff in these areas becomes significant after the market return variance is controlled in tests. One

interesting observation is the negative but insignificant tradeoff in the financial regulation area, which implies that

the categorical EPU index in this area may be a proxy of spanned conventional risks (as we will show further

evidence in next analysis). In total, Table 10 suggests that the EPU-return tradeoff exists in 11 out of 12 categories

and significantly exists in 6 (or 9) areas.

4 Conclusions

The positive risk-return tradeoff is the most important and fundamental concept in finance while this
relationship is hard to be detected because it is challenging to find appropriate proxies of conditional risk and
expected return. Following Merton (1973) and by introducing a explicit state variable, we briefly show that
the conditional risk should contain both spanned risks and unspanned Knightian uncertainty. Borrowing the
Economic Policy Uncertainty index from Baker, Bloom and Davis (2016), which is expected to capture
unspanned risks from security, political, economic and other areas by construction, we empirically show that
incorporating unspanned Knightian uncertainty into conditional risk estimation successfully detects a risk-
return tradeoff at both short and long horizons, and this tradeoff is robust. Further analyses show that
unspanned risks themselves are able to detect a risk-return tradeoff and unspanned uncertainty estimation is
complicate and not straightforward. Moreover, our findings on the tradeoff between portfolio returns and
unspanned Knightian uncertainty also remain in corporate bond markets.
References

 Alizadeh, Sassan, Michael W. Brandt, Francis X. Diebold, 2002. Range-based estimation

of stochastic volatility models. Journal of Finance 57(3):1047-1091.

 Amaya, Diego, Peter Christoffersen, Kris Jacobs, Aurelio Vasquez, 2015. Does realized skewness

predict cross-section of equity returns? Journal of Financial Economics 118:135-167.

 Ammer, John, Jianping Mei, 1996. Measuring international economic linkages. Journal of
Finance 51(5):1743-1763.

 Anderson, Evan W., Eric Ghysels, Jennifer L. Juergens, 2009. The impact of risk and
uncertainty on expected returns. Journal of Financial Economics 94: 233-263.

 Andrian, Tobias, Joshua Rossenberg, 2008. Stock returns and volatility: Pricing the short-run and
long-run components of market risk. Journal of Finance 63(6): 2997-3030.

 Baker, Scott R., Nicholas Bloom, Steven J. Davis, 2016. Measuring economic policy uncertainty.
Quarterly Journal of Economics 3: 1593-1636.

 Baker, Malcolm, Jeffrey Wurgler, 2006. Investor sentiment and the cross-section of stock returns.
Journal of Finance 61(4): 1645-1680.

 Bali, Turan, 2008. The intertemporal relation between expected returns and risk. Journal of
Financial Economics 87: 101-131.

 Bali, Turan, Stephen J, Brown, Yi Tang, 2017. Is economic uncertainty priced in the cross-
section of stock returns? Journal of Financial Economics 126: 471-489.

 Baillie, Richard T., Ramon Degennaro, 1990. Stock returns and volatility. Journal of
Financial and Quantitative Analysis 25(2): 203-214.

 Bekaert, Geert, Marie Hoerova, 2014. The VIX, the variance premium and stock volatility.
Journal of Econometrics 183: 181-192.

 Nekaert, Geert, Eric C. Engstrom, Nancy R. Xu, 2017. The time variation in risk appetite
and uncertainty. Working Paper, Columbia University.

 Baxter, Mariane, Robert G. King, 1993. Fiscal Policy in general equilibrium. American
Economic Review 83(3): 315-334.
  Belo, Frederico, Vito D. Gala, Jun Li, 2013. Government spending, political cycles, and
the cross-section of stock returns. Journal of Financial Economics 107: 305-324.

 Belo, Frederico, Jianfeng Yu, 2013. Government investment and the stock market. Journal
of Monetary Economics 60: 325-339.

 Bjornland, Hilde C., Kai Leitemo, 2009. Identifying the intderdependence between US
monetary policy and the stock market. Journal of Monetary Economics 56: 275-282.

 Blanchard, Oliver J., 1981. Output, the stock market, and interest rates. American
Economic Review 71(1): 132-143.

 Brandt, Michael, and Qiang Kang, 2004. On the relationship between conditional mean and
volatility of stock returns: A latent VAR approach. Journal of Financial Economics 72(2): 217-

257.

 Boutchkova, Maria, Hitech Doshi, Art Durnev, Alexander Molchanov, 2012. Precarious
Politics and return volatility. Review of Financial Studies 25(4): 1111-1154.
 Boyd, John H., Ross Levine, Bruce D. Smith, 2001. The impact of inflation on financial
sector performance. Journal of Monetary Economics 47: 221-248.

 Brogaard, Jonathan, Andrew Detzel, 2015. The asset-pricing implications of government

economic policy uncertainty. Management Science 61(1): 3-18.

 Campbell, John Y., 1987. Stock returns and the term structure. Journal of Financial
Economics 18(2):373-399.

 Campbell, John Y., John H. Cochrane, 1999. By force of habit: A consumption-based explanation
of aggregate stock market behavior. Journal of Political Economy 107(2): 205-251.

 Campbell, John Y., Ludger Henstchel, 1992. No news is good news: An asymmetric model of
changing volatility in stock returns. Journal of Financial Economics 31(3): 281-318.

 Campbell, John Y., Robert J. Shiller, 1988. The dividend-price ratio and expectations of
future dividends and discounted factors. Review of Financial Studies 1: 195-228.

 Christensen, B.J., N.R., prabhala, 1998. The relation between implied and realized
volatility. Journal of Financial Economics 50: 125-150.
 Cumming, Douglas, Sofia Johan, Dan Li,2011. Exchange trading rules and stock market liquidity.

Journal of Financial Economics 99: 651-671.

 Dumas, Bernard, Bruno Solink, 1995. The world price of exchange risk. Journal of Finance 50(2):

445-479.

 Flannery, Mark J., Christopher M. James, 1984. The effect of interest rate changes on the
common stock returns of financial institutions. Journal of Finance 39(4): 1141-1153.

 Flannery, Mark J., Aris A. Protopapadakis, 2002. Macroeconomic factors do influence

aggregate stock returns. Review of Financial Studies 15(3): 751-782.

 Forbes, Kristin J., Roberto Rigobon, 2002. No contagion, only interdependence: Measuring
stock market comovements. Journal of Finance 67(5):2223-2261.

 French, Kenneth R., G. Willian Schwert, Robert F. Stambaugh, 1987. Expected stocks
returns and volatility. Journal of Financial Economics 19: 3-29.

 Ghysels, Eric, Pedro Santa-Clara, Rossen Valkanov, 2005. There is a risk-return trade-off after all.

Journal of Financial Economics 76: 509-548.

 Givoly, Dan, Arie Ovada, 1983. Year-end tax-induced sales and stock market seasonality.
Journal of Finance 38(1): 171-184.

 Glosten, Lawrence R., Ravi Jagannathan, David E. Runkle, 1993. On the relationship
between the expected value and the volatility of the nominal excess return on stocks.
Journal of Finance 68(5): 1779-1801.

 Griffin, John M., Rene Stulz, 2001. International competition and exchange rate shocks: A cross-

country industry analysis of stock returns. Review of Financial Studies 14(1): 215-241.

 Guiso, Luigi, Paola Sapienza, Luigui Zingales, 2018. Time varying risk aversion. Journal
of Financial Economics 128: 403-421.

 Guo, Hui, Robert F. Whitelaw, 2006. Uncovering the risk-return relation in the stock
market. Journal of Finance 61(8): 1433-1464.

 Harvey, Campbell R., 2001. The specification of conditional expectations. Journal of

Empirical Finance 8(5): 573-637.
 Jorion, Philippe, 1991. The pricing of exchange rate risk in the stock market. Journal of
Financial and Quantitative Analysis 26(3): 363-376.

 Joslin, Scott, Marcel Priebsch, Kenneth J. Singleton, 2014. Risk premiums in dynamic term
structure models with unspanned macro risks. Journal of Finance 69(3): 1197-1233.

Julio, Brandon, Youngsuk Yook, 2012. Political uncertainty and corporate investment cycles.
Journal of Finance 67(1): 45-83.

 Jurado, Kyle, Sydney C. Ludvigson, Serena Ng, 2015. Measuring uncertainty. American
Economic Review 105(3): 1177-1216.

 Kelly, Bryan, Lubos Pastor, Pietro Veronesi, 2016. The price of political uncertainty: Theory and

evidence from the option market. Journal of Finance 71(5): 2417-2480.

 Keynes, John Maynard, 1937. The general theory of employment. Quarterly Journal of
Economics 51(2):209-223.

 Knight, Frank H.K., 1921. Risk, return and profit. Bibliolife DBA of Bibilio Bazaar II
LLC.

 Levine, Ross, Sara Zervos, 1998. Stock markets, banks, and economic growth. American
Economic Review 88(3): 537-558.

 LeRoy, Stephen F., Larry D. Singell Jr., 1987. Knight on risk and uncertainty. Journal of
Political Economy 95(2): 394-406.

 Lettau, Martin, Sydney C. Ludvigson, 2010. Measuring and modeling variation in the risk-
return tradeoff. Handbook of Financial Econometrics, ed. by Yacine Ait-Sahalia and Lars P.
Hansen vol. 1, pp. 617-690. Elsevier Science B.V., North Holland, Amsterdam.

 Lundblad, Christian, 2007. The risk return tradeoff in the long run: 1836-2003. Journal of
Financial Economics 85(1):123-150.

 Maio, Paulo, Pedro Santa-Clara, 2012. Multi-factor models and their consistency with the ICAPM.
 Journal of Financial Economics 106: 586-613.

 Manela, Asaf, Alan Moreira, 2017. News implied volatility and disaster concerns. Journal
of Financial Economics 123: 137-162.
  McGrattan, Ellen R., Edward C. Prescott, 2003. Average debt and equity returns: Puzzling?
American Economic Review 93(2):392-397.

 McQueen, Grant, V. Vance Roley, 1993. Stock prices, news, and business conditions.
Review of Financial Studies 6(3): 683-707.

 Menzly, Lior, Tano Santos, Pierto Veronesi, 2004. Understanding predictability. Journal of
Political Economy 112 (1): 1-47.

 Merton, Robert C., 1973. An intertemporal asset pricing mode. Econometrica 41: 867-888.

 Merton, Robert C., 1980. On estimating

the expected return on the market: an exploratory
investigation. Journal of Financial Economics 8: 323361.

 Nelson, Daniel B., 1991. Conditional

heteroscedasticity in asset returns: A new approach.
Econometrica 59(2): 347-370.

 Nyberg, Henri, 2012. Risk-return tradeoff in U.S. stock returns over the business cycle.

Journal of Financial and Quantitative Analysis 47(1): 137-158.

 Pastor, Lubos, Meenakshi Sinha, Bhaskaran Swaminathan, 2008. Estimating the intertemporal risk-

return tradeoff using the implied cost of capital. Journal of Finance 63(6): 2859-2897.

 Pastor, Lubos, Pietro Veronesi, 2012. Uncertainty about government policy and stock

prices. Journal of Finance 67(4):1219-1264.

 Pastor, Lubos, Pietro Veronesi, 2013. Political uncertainty and risk premia. Journal of
Financial Economics 110:520-545.

 Pastor, Lubos, Pietro Veronesi, 2017. Political uncertainty and stock returns. Working
Paper, University of Chicago.

Patelis, Alex D., 1997. Stock return predictability and the role of monetary policy. Journal of
Finance 52(5): 1951-1972.

 Reinganum, Marc R., 1983. The anomalous stock market behavior of small firms in January:
Empirical tests for tax-loss selling effect. Journal of Financial Economics 12: 89-104.

 Rigobon, Roberto, Brian Sack, 2003. Measuring the reaction of monetary policy to the stock market.
 Roll, Richard, 1977. A critique of the asset pricing theory’s tests Part I: on past and potential
 testability of the theory. Journal of Financial Economics 4(2): 129-176.

 Roll, Richard, 1992. Industrial structure and the comparative behavior of international
stock market indexes. Journal of Finance 67(1): 3-41.

 Santa-Clara, Pedro, Rossen Valkanov, 2003. The presidential puzzle: political cycles and the stock
 market. Journal of Finance 58: 1841-1872.

 Schwert, G. William, 1989. Why does stock market volatility change over time? Journal of
Finance 44 (5):1115-1153.

 Sikes, Stephanie A., 2014. The Turn-of-the-Year Effect and Tax-Loss-Selling by

Institutional Investors. Journal of Accounting and Economics 57(1): 2242.

 Telock, Paul C., 2007. Giving content to investor sentiment: The role of media in the stock market.

Journal of Finance 62(3):1139-1168.

 Thorbecke, Willem, 1997. On stock market returns and monetary policy. Journal of Finance 52(2):

635-654.

 Turner, Christopher M., Richard Startz, Charles R. Nelson, 1989. A Markov model of heteroscedas-

ticity, risk, and learning in the stock market. Journal of Financial Economics 25(1) 3-22.

 Wachter, Jessica A., 2006. A consumption-based model of the term structure of interest rates.
Journal of Financial Economics 79: 365-399.

 Wachter, Jessica A., 2013. Can time-varying risk of rare disasters explain aggregate stock
market volatility? Journal of Finance 68(3): 987-1025.

 Welch, Ivo, Amit Goyal, 2008. A comprehensive look at the empirical performance of
equity premium prediction. Review of Financial Studies 21(4):1455-1508.