This action might not be possible to undo. Are you sure you want to continue?

BooksAudiobooksComicsSheet Music### Categories

### Categories

### Categories

### Publishers

Scribd Selects Books

Hand-picked favorites from

our editors

our editors

Scribd Selects Audiobooks

Hand-picked favorites from

our editors

our editors

Scribd Selects Comics

Hand-picked favorites from

our editors

our editors

Scribd Selects Sheet Music

Hand-picked favorites from

our editors

our editors

Top Books

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Audiobooks

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Comics

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Sheet Music

What's trending, bestsellers,

award-winners & more

award-winners & more

P. 1

Risk Appetite|Views: 141|Likes: 0

Published by Ajay Devine

See more

See less

https://www.scribd.com/doc/46733728/Risk-Appetite

01/12/2011

text

original

**bond prices to say?
**

Marcelo Fernandes

Queen Mary, University of London

E-mail: m.fernandes@qmul.ac.uk

Samy Yoshima

Vision Brazil Investments

E-mail: syoshima@visionbrazil.com

Abstract: We propose a consistent estimator for the investors’ risk appetite that depends exclu-

sively on asset returns data. In particular, our estimator also has a nonparametric ﬂavor in that

it makes no parametric assumption on preferences and on the stochastic process that governs the

dynamics of asset returns. This is in stark contrast with the extant estimators in the literature that

usually require options data and assume that asset returns follow a geometric Brownian motion.

JEL classiﬁcation numbers: G12, C13, C23, E44

Keywords: asset pricing, ﬁxed-eﬀects panel regression, incomplete markets, market sentiment,

minimum-variance stochastic discount factor, risk appetite, risk tolerance.

1

1 Introduction

Investors’ shifting risk appetite may engender a pure form of contagion that does not remotely

relate to economic fundamentals. Due to the increasing interest in the role it plays in the ﬁnancial

markets, the literature oﬀers a number of indicators to gauge changes in the investors’ risk appetite.

Any risk-appetite measure must not exclusively depends on the level of uncertainty in the economy

given that its ﬂuctuation over time does not necessarily involve changes in the risk appetite.

There is an increasing interest in measuring investors’ risk appetite. The exact deﬁnition of

risk appetite though is a bit controversial (see Kumar and Persaud, 2002; Gai and Vause, 2006).

Most authors nevertheless agree that it relates to some sort of market sentiment reﬂecting the

willingness of investors to bear risk. This means that any measure of risk appetite must involve

not only investors’ attitude towards risk, but also the level of risk.

Risk appetite, by contrast, is likely to shift periodically as investors respond to episodes of

ﬁnancial distress and macroeconomic uncertainty. In adverse circumstances, investors will require

higher excess expected returns to hold each unit of risk and risk appetite will be lowit is the inverse

of the price of risk. And when the price of risk is taken together with the quantity of risk inherent

in a particular asset, the expected return required to compensate investors for holding that asset

is the risk premium. Figure 1 illustrates these concepts. It is clearly diﬃcult to disentangle risk

appetite from risk aversion in that an increase in either one of them causes asset prices to decline

and risk premia to increase.

In what follows, we formally distinguish risk appetite from risk premia and aversion. Speciﬁcally,

we propose a measure based on the variation in the ratio of risk-neutral to subjective probabilities

used by investors in evaluating the expected payoﬀ of an asset. By exploiting the linkages between

the risk-neutral and subjective probabilities that can be extracted from ﬁnancial market prices,

we follow Hayes, Panigirtzoglou and Shin (2003), Tarashev, Tsatsaronis and Karampatos (2003),

and Bollerslev, Gibson and Zhou (2004). Unlike these papers, however, we are able to extract an

indicator of market sentiment that is quite distinct from risk aversion.

This paper proposes a consistent estimator for investors’ risk appetite that relies exclusively on

asset returns data, without making any parametric assumption on preferences and on the stochastic

process that governs the dynamics of asset prices. In particular, we focus on a panel-regression

framework to derive an estimator for the demeaned realizations of the minimum-variance stochastic

discount factor (MVSDF) and then extract the market price of risk by computing the variance of

the MVSDF using a realized approach. The latter is convenient because it permits investors’ risk

2

appetite to shift over time. The precision of our risk-appetite estimator ameliorates as both the

number of time series observations and the number of assets in the economy increase. This is in

stark contrast with any option-implied risk-appetite index, whose performance does not necessarily

improve with the number of assets. Actually, the common practice is to employ option data on

only one underlying asset.

Our results give structure to the intuitive notion that investors’ risk appetite responds to eco-

nomic state variables, and should be of direct interest to market participants and monetary policy-

makers alike who study the links between the ﬁnancial markets and the overall economy. Interest-

ingly, we also ﬁnd that the estimated time-varying volatility risk premium index helps to predict

future stock market returns better than other well-established predictor variables, including the

consumption-wealth ratio (CAY) of Lettau and Ludvigson (2001).

The remainder of this paper ensues as follows. Section 2 describes the theoretical foundation

for the risk-appetite index that rests on the MVSDF approach to asset pricing. Section 3 discusses

our two-step estimation procedure. The ﬁrst step estimates the MVSDF using exclusively asset

prices data, whereas the second step consists of computing the risk-appetite index by estimating

the MVSDF-implied market price of risk. Section 4 then documents the evolution of the global

investors’ risk appetite since 1990 in light of some historical events. Section 5 oﬀers some concluding

remarks.

2 Risk appetite and the stochastic discount factor

The uncertainty in the economy ultimately relates to the variation of consumption among the

diﬀerent states of nature, and hence it is natural to derive an investors’ risk appetite index that

depends on the stochastic discount factor (SDF) in view that the latter relates to the growth

rate of the marginal utility of consumption. As in Hansen and Jagannathan (1991), we assume a

frictionless economy in which the law of one price holds, so that

E

t

M

t+1

R

i,t+1

= 1, i ∈ {0, . . . , N} (1)

where E

t

(·) denotes the conditional expectation given the available information at time t, M

t

is

the stochastic discount factor, R

0,t

denotes the gross return on the risk-free asset, R

i,t

denotes the

gross return on the ith risky asset, and N is the number of risky assets in economy. In the CCAPM

context, the pricing equation (1) mainly illustrates the fact that consumers equate marginal rates

of substitution to prices in view that the SDF coincides with the intertemporal marginal rate of

substitution.

3

The existence of a stochastic discount factor M

t

that prices assets according to (1) only requires

the law of one price. It is not necessary to assume, for instance, that there is a complete set of

security markets. It is only to ensure the uniqueness of the SDF that one must assume away

incomplete markets. Under market incompleteness, there exists an inﬁnite number of stochastic

discount factors that correctly price all traded securities. Notwithstanding, there still exists a

unique discount factor that lies on the payoﬀ space, namely, the minimum-variance stochastic

discount factor (MVSDF). The latter coincides with the projection of any SDF onto the payoﬀ

space, and hence they share the same pricing implications (see Cochrane, 2001).

We will also implicitly assume that the MVSDF is positive given that our estimation strategy

employs a log transformation. This assumption implies that there are no arbitrage opportunities

in the payoﬀ space. It is a quite mild assumption, though somewhat stronger than the general

no arbitrage condition, which only guarantees that there exists at least one positive discount fac-

tor. Despite the fact that it requires local non-satiability, we impose no other restriction on the

preferences representation.

Section 2.1 follows Gai and Vause (2006) in deﬁning the risk appetite index as the inverse of the

market price of risk, though the estimation strategy that we derive in Section 3 diﬀers substantially

from theirs. In particular, we retrieve the market price of risk by modeling the ex-post realization

of the MVSDF integrated variance. Our estimation strategy relies on a two-step procedure. In

the ﬁrst step, we estimate the MVSDF using Araujo, Issler and Fernandes’s (2006) nonparametric

approach so as to avoid parametric assumptions on preferences representation and on the stochastic

process that governs asset prices’ dynamics. In the second step, we infer the MVSDF integrated

variance using model-free realized measures and so we must impose a semiparametric stochastic

volatility model for assets’ log-price.

Under mild assumptions on preferences, log-price processes in any frictionless arbitrage-free

market must obey a semimartingale process (Back, 1991). We thus assume, as in Bollerslev et al.

(2004), that the log-price process follows a general continuous-time stochastic volatility model, viz.

dp

i,t

= µ

i,t

(·) dt +σ

i,t

dB

i,t

dσ

2

i,t

= α

i

δ

i

−σ

2

i,t

dt +ς

i,t

(·) dW

i,t

(2)

where the instantaneous correlation between the two Brownian motions dB

i,t

and dW

i,t

is constant,

corresponding to the familiar leverage eﬀect, and the functions µ

i,t

(·) and ς

i,t

(·) satisfy the usual

no arbitrage and regularity conditions. The semiparametric speciﬁcation given by (2) is ﬂexible

enough to model most asset returns, though it is also possible to extend our framework to deal

with jump-diﬀusion processes (see Andersen, Bollerslev and Diebold, 2005).

4

2.1 Risk-appetite index

The pricing equation (1) states that, once we discount any gross return by M

t

, every security entails

zero excess return over the risk-free asset. It applies to every asset in the economy, even for the

risk-free asset. To appreciate that, one may rewrite (1) as

E

t

(M

t+1

) E

t

(R

i,t+1

) + cov

t

(M

t+1

, R

i,t+1

) = 1, (3)

where cov

t

(·) denotes the conditional covariance given the available information at time t. The ﬁrst

term relates to the risk-neutral component in that it illustrates the expected return that investors

would require to hold that asset if they were risk neutral. In turn, the covariance term corresponds

to a risk adjustment that compensates risk-averse investors to hold that risky asset.

As the risk-free rate of return is measurable on the information set available at time t, it ensues

that R

0,t+1

= 1/E

t

(M

t+1

) and hence

E

t

(R

i,t+1

−R

0,t+1

) = −R

0,t+1

cov

t

(M

t+1

, R

i,t+1

). (4)

This means that investors require a premium to hold assets whose payoﬀs are higher in periods

of high (rather than low) consumption. The usual beta-decomposition of the risk premium in (4)

then yields

E

t

(R

i,t+1

−R

0,t+1

) = −

cov

t

(M

t+1

, R

i,t+1

)

σ

2

t

(M

t+1

)

. .. .

β

i,t

× σ

2

t

(M

t+1

) R

0,t+1

. .. .

λ

t

, (5)

where σ

2

t

(·) denotes the conditional variance given the available information at time t. The beta

term corresponds to the amount of risk, whereas the market price of risk λ

t

is the expected excess

return that investors require in order to hold an additional unit of risk in equilibrium.

As in Gai and Vause (2006), we deﬁne risk appetite as the inverse of the market price of risk.

This ensures that the risk premium is decreasing with the investors’ risk appetite. This is in line,

for instance, with the view that emerging markets sovereign bonds must pay a higher risk premia

when investors are less willing to bear risk (Baek, Bandopadhyaya and Du, 2005). As is apparent

from the beta-decomposition in (5), Gai and Vause’s risk appetite measure mainly depends on the

variation in the stochastic discount factor. In view that the latter speciﬁes the marginal rate at

which the investor is willing to substitute uncertain future consumption for present consumption,

risk appetite is a function not only of the investors’ risk aversion, but also of the overall level of

uncertainty surrounding consumption prospects. The latter dependence explains why risk appetite

may vary signiﬁcantly over time, even though risk aversion is likely to remain constant.

5

Section 3 departures from Gai and Vause’s (2006) framework in that we estimate the risk-

appetite index in a completely diﬀerent manner. Gai and Vause exploit the fact that the SDF

relates to the ratio of the risk-neutral to subjective probabilities to estimate the market price of

risk using options data (see also Hayes et al., 2003; Tarashev et al., 2003; Bollerslev et al., 2004).

In contrast, we rely exclusively on asset prices data, employing the nonparametric approach put

forth by Araujo et al. (2006).

3 Estimation strategy

To conduct statistical inference, we must impose some restrictions on the stochastic nature of asset

returns and of the stochastic discount factor. For the sake of exposition, we start with the unre-

alistic assumption of conditional lognormality and homoskedasticity, even though the results still

hold in a much more general setting (see Araujo et al., 2006). It is also worth noting that condi-

tional lognormality and homoskedasticity has a long pedigree in macroeconomics (e.g., Hansen and

Singleton, 1983; Campbell, 1993; Lettau and Ludvigson, 2001), whereas conditional lognormality is

standard in ﬁnancial econometrics, especially in the context of conditional heteroskedasticity (e.g.,

Bollerslev et al., 2004; Meddahi and Renault, 2004).

Taking logs of both sides of the pricing equation (1) gives way to

ln E

t

(M

t+1

R

i,t+1

) = 0, i ∈ {1, . . . , N}. (6)

Conditional lognormality then implies that, for every i ∈ {1, . . . , N},

E

t

(m

t+1

+r

i,t+1

) +

1

2

σ

2

t

(m

t+1

+r

i,t+1

) = 0, (7)

where m

t

= ln M

t

, r

i,t

= ln R

i,t

, and the conditional variance σ

2

t

(·) given the available information

at time t is constant, say σ

2

i

, under the conditional homoskedasticity assumption. Decomposing

r

i,t+1

+m

t+1

into the projection on the information set at time t and an orthogonal error results in

r

i,t+1

+m

t+1

= E

t

(r

i,t+1

+m

t+1

) +

i,t+1

, (8)

where

i,t+1

is Gaussian with mean zero and variance σ

2

i

given the conditional lognormality and

homoskedasticity assumptions.

From (7) and (8), it follows that

r

i,t+1

= − m

t+1

−

1

2

σ

2

i

+

i,t+1

. (9)

In the context of panel-data regression, (9) corresponds to a standard unobserved ﬁxed-eﬀects model

with no explanatory variables other than time dummies, also known as the two-way ﬁxed-eﬀects

6

model (see Baltagi, 2001). Stacking the estimates of the coeﬃcients of the time dummies then

presumably provide a consistent estimate for the series of the log-SDF, whereas the ﬁxed-eﬀects

capture the individual heterogeneity stemming from the second moment of the log-returns. It turns

out however that (9) identiﬁes the coeﬃcients of the time dummies only up to a normalization

constant, though Araujo et al. (2006) show how to consistently recover the level of the MVSDF at

times t = 2, . . . , T. Under some additional conditions, they show that

´

M

t

=

¯

R

G

t

T − 1

T

¸

s=2

¯

R

s

¯

R

G

s

−1

, (10)

where

¯

R

t

and

¯

R

G

t

are respectively the cross-sectional arithmetic and geometric average of assets’

gross returns, converges in probability to the realization of the MVSDF as both the time-series and

cross-section sample sizes grow without bound.

We deem that our estimator has some interesting properties. First, the estimation procedure

is straightforward in that it relies on a standard two-way ﬁxed-eﬀects regression. In fact, the

resulting estimator is also very simple given that it depends only on averages of the asset log-

returns. Second, it postulates neither a particular preference representation nor any pricing theory.

Third, our estimator is consistent with the risk-premium for it avoids the excessive smoothness of

consumption data.

3.1 Estimating the market price of risk

The estimation procedure of the risk-appetite index consists of two steps. The ﬁrst estimates

the minimum-variance stochastic discount factors by (10), whereas the second step consists of

estimating the market price of risk by computing the MVSDF realized variance. In particular, the

risk-appetite estimator displays a semiparametric character in that it assumes that asset returns

follow a semiparametric stochastic volatility model without imposing any parametric assumption

on preferences.

To estimate the market price of risk, we must compute the conditional variance of the MVSDF.

Accordingly, we ﬁrst estimate the ex-post realization of the MVSDF integrated variance and then

condition on the past information set as in Bollerslev et al. (2004).

3.2 Discussion

Persaud (1996), Kumar and Persaud (2002), Baek et al. (2005), Bandopadhyaya and Jones (2006),

and Dupuy (2005) attempt to identify changes in risk appetite by looking at the rank correlation

between assets’ riskiness and excess returns. The idea lies on the fact that such a rank correlation

7

is presumably higher in periods of shifting risk appetite than in relatively stable periods. Although

it is simple and quite intuitive, this approach reveals several drawbacks (see discussion in Gai and

Vause, 2006). The main disadvantage refers to the fact that the rank correlation indeed identiﬁes

changes in risk appetite only under the very strong assumption of linear independence of assets

returns (Misina, 2003).

Froot and O’Connell (2003) propose to gauge investors’ conﬁdence by examining institutional

investors’ portfolio holdings and investment ﬂows. They assume that changes in portfolio holdings

are a function of expected returns, risk, wealth, and risk tolerance. Their empirical results indicate

that the risk-tolerance component explains most of the variation in the institutional investors’

portfolio holdings. Froot and O’Connell’s (2003) approach has two serious drawbacks, however.

First, it does not provide a genuine measure of risk appetite, though risk tolerance certainly relates

to the investors’ willingness to bear risk. Second, it heavily depends on the assumption that changes

in risk tolerance aﬀect holdings in an equiproportional manner, which boils down to imposing a

preferences representation with either constant absolute or relative risk aversion.

Tarashev et al. (2003) oﬀer a more interesting alternative to estimate a risk-appetite index.

They show that one may gauge investors’ risk appetite by computing the level of the stochastic

discount factor in diﬀerent states of nature. Because this corresponds to computing ratio of the

quantiles of the risk-neutral and subjective distributions of asset returns, Tarashev et al. propose

a framework that relies on the estimation of the risk-neutral distribution using options data. Their

identiﬁcation and estimation strategy exhibits three drawbacks. First, they are not able to properly

control for the error-in-variable issue that arises in the ﬁrst-step estimation of the risk-neutral

and subjective distributions. Second, the estimation of the risk-neutral distribution requires some

restrictive assumptions on the dynamic behavior of the underling asset prices. Third, estimating

the ratio at a particular quantile may misrepresent investors’ overall attitude towards risk.

Gai and Vause (2006) deﬁne a risk-appetite index as the inverse of the market price of risk,

i.e., the expected excess return that investors require, in equilibrium, to hold one additional unit of

risk. Even though their deﬁnition clariﬁes that the risk-appetite index depends on the risk-neutral

distribution only through the variance of the stochastic discount factor, Gai and Vause estimate

the market price of risk by computing the variance of the ratio of the risk-neutral and subjective

probability distributions. This solves the third drawback of Tarashev et al.’s (2003) methodology,

but not the ﬁrst two.

Bollerslev et al. (2004) develop a similar methodology that focus on the diﬀerences between

8

the volatilities of the risk-neutral and subjective distributions so as to identify a volatility risk

premium. They attempt to robustify the estimation procedure by using model-free measures of

integrated volatility. In particular, they estimate the options-implied volatility using Britten-Jones

and Neuberger’s (2000) model-free estimator and the volatility of the subjective distribution using

a realized approach (Barndorﬀ-Nielsen and Shephard, 2002; Meddahi, 2002; Andersen, Bollerslev,

Diebold and Labys, 2003; Barndorﬀ-Nielsen and Shephard, 2004). Bollerslev et al. then estimate by

GMM the volatility risk premium using conditional cross-moments of the risk-neutral and subjective

expectations of the integrated volatility. Although their GMM framework solves the error-in-

variable issue in an eﬃcient manner, it still provides an incomplete picture of investors’ risk appetite

in view of their emphasis on the risk-neutral and subjective volatilities.

4 Global investors’ risk appetite

5 Conclusion

This paper proposes a novel measure of market risk appetite that relies on robust estimates of

the market price of risk. Our estimator has a nonparametric ﬂavor in that it makes no parametric

assumption on preferences and on the stochastic process that governs the dynamics of asset returns.

This is in contrast with the existing estimators in the literature that assume that asset returns follow

a geometric Brownian motion. In addition, our estimator improves as the number of assets in the

economy increases.

9

References

Andersen, T., Bollerslev, T., Diebold, F. X., Labys, P., 2003, Modeling and forecasting realizaed

volatility, Econometrica 71, 579–625.

Andersen, T. G., Bollerslev, T., Diebold, F. X., 2005, Roughing it up: Including jump components

in the measurement, modeling and forecasting of return volatility, Northwestern University,

Duke University and University of Pennsylvannia.

Araujo, F., Issler, J. V., Fernandes, M., 2006, Estimating the stochastic discount factor without a

utility function, Princeton University, Getulio Vargas Foundation, and Queen Mary, University

of London.

Back, K., 1991, Asset pricing for general processes, Journal of Mathematical Economics 20, 371–395.

Baek, I.-M., Bandopadhyaya, A., Du, C., 2005, Determinants of market-assessed sovereign risk:

Economic fundamentals or market risk appetite?, Journal of International Money and Finance

24, 533–548.

Baltagi, B. H., 2001, Econometric Analysis of Panel Data, John Wiley, Chichester.

Bandopadhyaya, A., Jones, A. L., 2006, Measuring investor sentiment in equity markets, University

of Massachusetts, Boston.

Barndorﬀ-Nielsen, O. E., Shephard, N., 2002, Analysis of realised volatility and its use in estimating

stochastic volatility models, Journal of Royal Statistical Society, Series B 64, 253–280.

Barndorﬀ-Nielsen, O. E., Shephard, N., 2004, Power and bipower variation with stochastic volatility

and jumps, Journal of Financial Econometrics 2, 1–48.

Bollerslev, T., Gibson, M., Zhou, H., 2004, Dynamic estimation of volatility risk premia and investor

risk aversion from option-implied and realized volatilities, Finance and economics discussion

series 2004-56, Federal Reserve Board, Washigton, D.C.

Britten-Jones, M., Neuberger, A., 2000, Option Prices, implied price processes, and stochastic

volatility, Journal of Finance 55, 839–866.

Campbell, J. Y., 1993, Intertemporal asset pricing without consumption data, American Economic

Review 83, 487–512.

Cochrane, J. H., 2001, Asset Pricing, Princeton University Press, Princeton.

Dupuy, P., 2005, There is little sentiment in capital markets, HSBC Asset Management Europe.

10

Froot, K. A., O’Connell, P. G. J., 2003, The risk tolerance of international investors, Working paper

10157, NBER.

Gai, P., Vause, N., 2006, Measuring investors’ risk appetite, International Journal of Central Bank-

ing 2, 167–188.

Hansen, L. P., Jagannathan, R., 1991, Implications of security market data for models of dynamic

economies, Journal of Political Economy 99, 225–262.

Hansen, L. P., Singleton, K. J., 1983, Stochastic consumption, risk aversion and the temporal

behavior of asset returns, Journal of Political Economy 91, 249–265.

Hayes, S., Panigirtzoglou, N., Shin, H. S., 2003, Liquidity and risk appetite: Evidence from equity

index option prices, Bank of England.

Kumar, M. S., Persaud, A., 2002, Pure contagion and investors’ shifting risk appetite: Analytical

issues and empirical evidence, International Finance 5, 401–426.

Lettau, M., Ludvigson, S., 2001, Resurrecting the (C)CAPM: A cross-sectional test when risk

premia are time-varying, Journal of Political Economy 109, 1238–1287.

Meddahi, N., 2002, Theoretical comparison between integrated and realized volatility, Journal of

Applied Econometrics 17, 479–508.

Meddahi, N., Renault, E., 2004, Temporal aggregation of volatility models, Journal of Econometrics

119, 355–379.

Misina, M., 2003, What does the risk-appetite index measure?, Working paper 2003-23, Bank of

Canada.

Persaud, A., 1996, Investors’ changing appetite for risk, J.P. Morgan Global FX Research.

Tarashev, N., Tsatsaronis, K., Karampatos, D., 2003, Investors’ attitude towards risk: What can

we learn from options?, BIS Quarterly Review (June), 57–65.

Wallace, H. D., Hussain, A., 1969, The use of error components model in combining cross-section

and time-series data, Econometrica 37, 55–72.

11

investors will require higher excess expected returns to hold each unit of risk and risk appetite will be lowit is the inverse of the price of risk. By exploiting the linkages between the risk-neutral and subjective probabilities that can be extracted from ﬁnancial market prices. 2002. Figure 1 illustrates these concepts. Unlike these papers. and Bollerslev.1 Introduction Investors’ shifting risk appetite may engender a pure form of contagion that does not remotely relate to economic fundamentals. Speciﬁcally. In particular. This means that any measure of risk appetite must involve not only investors’ attitude towards risk. we follow Hayes. Gai and Vause. without making any parametric assumption on preferences and on the stochastic process that governs the dynamics of asset prices. In adverse circumstances. It is clearly diﬃcult to disentangle risk appetite from risk aversion in that an increase in either one of them causes asset prices to decline and risk premia to increase. however. is likely to shift periodically as investors respond to episodes of ﬁnancial distress and macroeconomic uncertainty. the expected return required to compensate investors for holding that asset is the risk premium. Tsatsaronis and Karampatos (2003). but also the level of risk. And when the price of risk is taken together with the quantity of risk inherent in a particular asset. we are able to extract an indicator of market sentiment that is quite distinct from risk aversion. we formally distinguish risk appetite from risk premia and aversion. the literature oﬀers a number of indicators to gauge changes in the investors’ risk appetite. we focus on a panel-regression framework to derive an estimator for the demeaned realizations of the minimum-variance stochastic discount factor (MVSDF) and then extract the market price of risk by computing the variance of the MVSDF using a realized approach. Any risk-appetite measure must not exclusively depends on the level of uncertainty in the economy given that its ﬂuctuation over time does not necessarily involve changes in the risk appetite. This paper proposes a consistent estimator for investors’ risk appetite that relies exclusively on asset returns data. we propose a measure based on the variation in the ratio of risk-neutral to subjective probabilities used by investors in evaluating the expected payoﬀ of an asset. by contrast. Panigirtzoglou and Shin (2003). Due to the increasing interest in the role it plays in the ﬁnancial markets. Gibson and Zhou (2004). In what follows. Risk appetite. There is an increasing interest in measuring investors’ risk appetite. The exact deﬁnition of risk appetite though is a bit controversial (see Kumar and Persaud. The latter is convenient because it permits investors’ risk 2 . Tarashev. 2006). Most authors nevertheless agree that it relates to some sort of market sentiment reﬂecting the willingness of investors to bear risk.

Interestingly. Our results give structure to the intuitive notion that investors’ risk appetite responds to economic state variables. Section 3 discusses our two-step estimation procedure. Section 2 describes the theoretical foundation for the risk-appetite index that rests on the MVSDF approach to asset pricing. 3 . Actually.t denotes the gross return on the ith risky asset. Section 5 oﬀers some concluding remarks. and hence it is natural to derive an investors’ risk appetite index that depends on the stochastic discount factor (SDF) in view that the latter relates to the growth rate of the marginal utility of consumption. Mt is the stochastic discount factor. . the common practice is to employ option data on only one underlying asset. . N } (1) where Et (·) denotes the conditional expectation given the available information at time t. Section 4 then documents the evolution of the global investors’ risk appetite since 1990 in light of some historical events. and N is the number of risky assets in economy. we also ﬁnd that the estimated time-varying volatility risk premium index helps to predict future stock market returns better than other well-established predictor variables. i ∈ {0. R0. and should be of direct interest to market participants and monetary policymakers alike who study the links between the ﬁnancial markets and the overall economy. including the consumption-wealth ratio (CAY) of Lettau and Ludvigson (2001).t+1 = 1. The precision of our risk-appetite estimator ameliorates as both the number of time series observations and the number of assets in the economy increase. so that Et Mt+1 Ri. 2 Risk appetite and the stochastic discount factor The uncertainty in the economy ultimately relates to the variation of consumption among the diﬀerent states of nature. The ﬁrst step estimates the MVSDF using exclusively asset prices data. we assume a frictionless economy in which the law of one price holds. The remainder of this paper ensues as follows. whereas the second step consists of computing the risk-appetite index by estimating the MVSDF-implied market price of risk. .appetite to shift over time.t denotes the gross return on the risk-free asset. As in Hansen and Jagannathan (1991). Ri. the pricing equation (1) mainly illustrates the fact that consumers equate marginal rates of substitution to prices in view that the SDF coincides with the intertemporal marginal rate of substitution. This is in stark contrast with any option-implied risk-appetite index. In the CCAPM context. whose performance does not necessarily improve with the number of assets. .

and the functions µi. Section 2. we infer the MVSDF integrated variance using model-free realized measures and so we must impose a semiparametric stochastic volatility model for assets’ log-price.t dBi.t = µi. as in Bollerslev et al.1 follows Gai and Vause (2006) in deﬁning the risk appetite index as the inverse of the market price of risk. which only guarantees that there exists at least one positive discount factor. This assumption implies that there are no arbitrage opportunities in the payoﬀ space. We thus assume. The semiparametric speciﬁcation given by (2) is ﬂexible enough to model most asset returns. Issler and Fernandes’s (2006) nonparametric approach so as to avoid parametric assumptions on preferences representation and on the stochastic process that governs asset prices’ dynamics. It is not necessary to assume. 4 .t is constant.t 2 = αi δi − σi. namely.t (2) where the instantaneous correlation between the two Brownian motions dBi.t (·) dt + σi. Bollerslev and Diebold. Under market incompleteness. viz.t 2 dσi.t (·) satisfy the usual no arbitrage and regularity conditions. though the estimation strategy that we derive in Section 3 diﬀers substantially from theirs. Notwithstanding. (2004). the minimum-variance stochastic discount factor (MVSDF). 1991). In the second step. we impose no other restriction on the preferences representation. corresponding to the familiar leverage eﬀect. In the ﬁrst step. that there is a complete set of security markets. though somewhat stronger than the general no arbitrage condition. 2005). there exists an inﬁnite number of stochastic discount factors that correctly price all traded securities. log-price processes in any frictionless arbitrage-free market must obey a semimartingale process (Back. It is a quite mild assumption. though it is also possible to extend our framework to deal with jump-diﬀusion processes (see Andersen. 2001). Under mild assumptions on preferences. we estimate the MVSDF using Araujo.t dt + ςi. In particular. there still exists a unique discount factor that lies on the payoﬀ space.The existence of a stochastic discount factor Mt that prices assets according to (1) only requires the law of one price. dpi.t (·) and ςi. we retrieve the market price of risk by modeling the ex-post realization of the MVSDF integrated variance. We will also implicitly assume that the MVSDF is positive given that our estimation strategy employs a log transformation. for instance.t and dWi. The latter coincides with the projection of any SDF onto the payoﬀ space. It is only to ensure the uniqueness of the SDF that one must assume away incomplete markets. Despite the fact that it requires local non-satiability.t (·) dWi. that the log-price process follows a general continuous-time stochastic volatility model. Our estimation strategy relies on a two-step procedure. and hence they share the same pricing implications (see Cochrane.

2 σt (Mt+1 ) βi. Bandopadhyaya and Du. whereas the market price of risk λt is the expected excess return that investors require in order to hold an additional unit of risk in equilibrium. even for the risk-free asset. As in Gai and Vause (2006). every security entails zero excess return over the risk-free asset.1 Risk-appetite index The pricing equation (1) states that.t+1 ) = −R0. In view that the latter speciﬁes the marginal rate at which the investor is willing to substitute uncertain future consumption for present consumption. with the view that emerging markets sovereign bonds must pay a higher risk premia when investors are less willing to bear risk (Baek. (3) where covt (·) denotes the conditional covariance given the available information at time t. one may rewrite (1) as Et (Mt+1 ) Et (Ri. This is in line. The beta term corresponds to the amount of risk. 2005). Ri.t+1 ) 2 × σt (Mt+1 ) R0. but also of the overall level of uncertainty surrounding consumption prospects. Gai and Vause’s risk appetite measure mainly depends on the variation in the stochastic discount factor. To appreciate that.t+1 ).t+1 − R0.t+1 = 1/Et (Mt+1 ) and hence Et (Ri. once we discount any gross return by Mt .t λt (5) 2 where σt (·) denotes the conditional variance given the available information at time t. it ensues that R0. risk appetite is a function not only of the investors’ risk aversion. It applies to every asset in the economy. As the risk-free rate of return is measurable on the information set available at time t. Ri. (4) This means that investors require a premium to hold assets whose payoﬀs are higher in periods of high (rather than low) consumption. In turn. 5 . we deﬁne risk appetite as the inverse of the market price of risk. the covariance term corresponds to a risk adjustment that compensates risk-averse investors to hold that risky asset. This ensures that the risk premium is decreasing with the investors’ risk appetite. even though risk aversion is likely to remain constant.t+1 covt (Mt+1 . Ri. The ﬁrst term relates to the risk-neutral component in that it illustrates the expected return that investors would require to hold that asset if they were risk neutral.t+1 ) = − covt (Mt+1 . The latter dependence explains why risk appetite may vary signiﬁcantly over time. As is apparent from the beta-decomposition in (5).t+1 ) = 1.t+1 ) + covt (Mt+1 .t+1 − R0.t+1 . The usual beta-decomposition of the risk premium in (4) then yields Et (Ri.2. for instance.

For the sake of exposition.. . 2004).t+1 i.t+1 = − mt+1 − 1 2 2 σi + i. 2001). and the conditional variance σt (·) given the available information 2 at time t is constant. .t+1 ) = 0. Tarashev et al. . Meddahi and Renault.t+1 ) + 1 2 2 σt (mt+1 + ri.Section 3 departures from Gai and Vause’s (2006) framework in that we estimate the riskappetite index in a completely diﬀerent manner. 2003.t = ln Ri. . 2003. Hansen and Singleton..t+1 + mt+1 = Et (ri. (6) Conditional lognormality then implies that. (8) 2 is Gaussian with mean zero and variance σi given the conditional lognormality and homoskedasticity assumptions. 3 Estimation strategy To conduct statistical inference.g. (7) 2 where mt = ln Mt .t+1 ) = 0. ri. .. . 2006). 1983. In contrast. N }. under the conditional homoskedasticity assumption. i ∈ {1. it follows that ri. whereas conditional lognormality is standard in ﬁnancial econometrics. Lettau and Ludvigson.. Et (mt+1 + ri. Gai and Vause exploit the fact that the SDF relates to the ratio of the risk-neutral to subjective probabilities to estimate the market price of risk using options data (see also Hayes et al.g. we start with the unrealistic assumption of conditional lognormality and homoskedasticity. (9) corresponds to a standard unobserved ﬁxed-eﬀects model with no explanatory variables other than time dummies. employing the nonparametric approach put forth by Araujo et al. Bollerslev et al. especially in the context of conditional heteroskedasticity (e. we must impose some restrictions on the stochastic nature of asset returns and of the stochastic discount factor. say σi . (2006). Taking logs of both sides of the pricing equation (1) gives way to ln Et (Mt+1 Ri. 2004. 1993. N }. Decomposing ri.t+1 . . we rely exclusively on asset prices data.t+1 .t+1 + mt+1 ) + where i. also known as the two-way ﬁxed-eﬀects 6 ..t+1 + mt+1 into the projection on the information set at time t and an orthogonal error results in ri. From (7) and (8). Campbell..t . Bollerslev et al. It is also worth noting that conditional lognormality and homoskedasticity has a long pedigree in macroeconomics (e. (9) In the context of panel-data regression.. . for every i ∈ {1. even though the results still hold in a much more general setting (see Araujo et al. 2004).

. In particular. Kumar and Persaud (2002). whereas the second step consists of estimating the market price of risk by computing the MVSDF realized variance. Stacking the estimates of the coeﬃcients of the time dummies then presumably provide a consistent estimate for the series of the log-SDF. Third. (2006) show how to consistently recover the level of the MVSDF at times t = 2. The ﬁrst estimates the minimum-variance stochastic discount factors by (10). Accordingly. (10) ¯ ¯G where Rt and Rt are respectively the cross-sectional arithmetic and geometric average of assets’ gross returns. whereas the ﬁxed-eﬀects capture the individual heterogeneity stemming from the second moment of the log-returns. . . we ﬁrst estimate the ex-post realization of the MVSDF integrated variance and then condition on the past information set as in Bollerslev et al. 3. . it postulates neither a particular preference representation nor any pricing theory. 2001).1 Estimating the market price of risk The estimation procedure of the risk-appetite index consists of two steps. In fact. It turns out however that (9) identiﬁes the coeﬃcients of the time dummies only up to a normalization constant. we must compute the conditional variance of the MVSDF. Second.model (see Baltagi. though Araujo et al. T . our estimator is consistent with the risk-premium for it avoids the excessive smoothness of consumption data. We deem that our estimator has some interesting properties. First. (2004). Under some additional conditions. To estimate the market price of risk. the estimation procedure is straightforward in that it relies on a standard two-way ﬁxed-eﬀects regression. 3. Bandopadhyaya and Jones (2006). they show that Mt = ¯G Rt T −1 T s=2 ¯ Rs ¯G Rs −1 . (2005). the resulting estimator is also very simple given that it depends only on averages of the asset logreturns. converges in probability to the realization of the MVSDF as both the time-series and cross-section sample sizes grow without bound. Baek et al. and Dupuy (2005) attempt to identify changes in risk appetite by looking at the rank correlation between assets’ riskiness and excess returns.2 Discussion Persaud (1996). The idea lies on the fact that such a rank correlation 7 . the risk-appetite estimator displays a semiparametric character in that it assumes that asset returns follow a semiparametric stochastic volatility model without imposing any parametric assumption on preferences.

First. 2006).is presumably higher in periods of shifting risk appetite than in relatively stable periods. however. They show that one may gauge investors’ risk appetite by computing the level of the stochastic discount factor in diﬀerent states of nature.e. First. Gai and Vause estimate the market price of risk by computing the variance of the ratio of the risk-neutral and subjective probability distributions. The main disadvantage refers to the fact that the rank correlation indeed identiﬁes changes in risk appetite only under the very strong assumption of linear independence of assets returns (Misina.’s (2003) methodology. Second. propose a framework that relies on the estimation of the risk-neutral distribution using options data. but not the ﬁrst two. Froot and O’Connell (2003) propose to gauge investors’ conﬁdence by examining institutional investors’ portfolio holdings and investment ﬂows. Froot and O’Connell’s (2003) approach has two serious drawbacks. Gai and Vause (2006) deﬁne a risk-appetite index as the inverse of the market price of risk. they are not able to properly control for the error-in-variable issue that arises in the ﬁrst-step estimation of the risk-neutral and subjective distributions. Bollerslev et al. (2003) oﬀer a more interesting alternative to estimate a risk-appetite index. Their identiﬁcation and estimation strategy exhibits three drawbacks. (2004) develop a similar methodology that focus on the diﬀerences between 8 . Their empirical results indicate that the risk-tolerance component explains most of the variation in the institutional investors’ portfolio holdings. Tarashev et al. in equilibrium. though risk tolerance certainly relates to the investors’ willingness to bear risk. to hold one additional unit of risk. risk. this approach reveals several drawbacks (see discussion in Gai and Vause. the estimation of the risk-neutral distribution requires some restrictive assumptions on the dynamic behavior of the underling asset prices. Because this corresponds to computing ratio of the quantiles of the risk-neutral and subjective distributions of asset returns. it does not provide a genuine measure of risk appetite. and risk tolerance. They assume that changes in portfolio holdings are a function of expected returns. it heavily depends on the assumption that changes in risk tolerance aﬀect holdings in an equiproportional manner. which boils down to imposing a preferences representation with either constant absolute or relative risk aversion. Third. Tarashev et al. Second. 2003). i. estimating the ratio at a particular quantile may misrepresent investors’ overall attitude towards risk. Even though their deﬁnition clariﬁes that the risk-appetite index depends on the risk-neutral distribution only through the variance of the stochastic discount factor. This solves the third drawback of Tarashev et al. the expected excess return that investors require. wealth. Although it is simple and quite intuitive..

Diebold and Labys. they estimate the options-implied volatility using Britten-Jones and Neuberger’s (2000) model-free estimator and the volatility of the subjective distribution using a realized approach (Barndorﬀ-Nielsen and Shephard. Bollerslev. 2003. Meddahi. In particular.the volatilities of the risk-neutral and subjective distributions so as to identify a volatility risk premium. it still provides an incomplete picture of investors’ risk appetite in view of their emphasis on the risk-neutral and subjective volatilities. Barndorﬀ-Nielsen and Shephard. Although their GMM framework solves the error-invariable issue in an eﬃcient manner. This is in contrast with the existing estimators in the literature that assume that asset returns follow a geometric Brownian motion. our estimator improves as the number of assets in the economy increases. Our estimator has a nonparametric ﬂavor in that it makes no parametric assumption on preferences and on the stochastic process that governs the dynamics of asset returns. Andersen. 9 . In addition. They attempt to robustify the estimation procedure by using model-free measures of integrated volatility. 2002. 4 5 Global investors’ risk appetite Conclusion This paper proposes a novel measure of market risk appetite that relies on robust estimates of the market price of risk. 2002. then estimate by GMM the volatility risk premium using conditional cross-moments of the risk-neutral and subjective expectations of the integrated volatility. Bollerslev et al. 2004).

. 839–866. 2003... I.. H. 2000. T. F.. Washigton. 1993. American Economic Review 83. Du. 2001. Journal of International Money and Finance 24. 1–48. 1991. 371–395. E. L. M.. Bollerslev.. H. J. and stochastic volatility. Determinants of market-assessed sovereign risk: Economic fundamentals or market risk appetite?. O. N. H. Barndorﬀ-Nielsen.. Measuring investor sentiment in equity markets. F. 2006.. J.. Jones. Option Prices. Modeling and forecasting realizaed volatility. Estimating the stochastic discount factor without a utility function. Baltagi. Princeton.. Journal of Royal Statistical Society.. Araujo. 2004. 2005. A.. Bollerslev. implied price processes. T. T..References Andersen. Duke University and University of Pennsylvannia. Chichester. Northwestern University. A. N.. University of Massachusetts. Dynamic estimation of volatility risk premia and investor risk aversion from option-implied and realized volatilities. Britten-Jones. Labys.-M. P. M. Shephard.. Princeton University. 2005. Campbell. There is little sentiment in capital markets. Neuberger.. Getulio Vargas Foundation. Andersen... C. O. Journal of Financial Econometrics 2. K. G. T. Econometrica 71. Econometric Analysis of Panel Data.C. Bandopadhyaya. 2005. Bandopadhyaya.. A. Federal Reserve Board. X. 253–280.. 2004. 2001. Dupuy. P. Cochrane. Gibson. University of London. B. 2006. Fernandes. T. Journal of Mathematical Economics 20. Barndorﬀ-Nielsen. John Wiley. Analysis of realised volatility and its use in estimating stochastic volatility models.. Diebold. Baek.... Issler. Intertemporal asset pricing without consumption data. modeling and forecasting of return volatility.. Back. Journal of Finance 55. 579–625. Finance and economics discussion series 2004-56. M. Diebold. and Queen Mary. 2002... Y. F. Princeton University Press. D. Zhou.. Bollerslev. X. J. 10 . Power and bipower variation with stochastic volatility and jumps. E. V. Roughing it up: Including jump components in the measurement. A. Shephard. Series B 64. 533–548. 487–512. Asset pricing for general processes. Asset Pricing. Boston. HSBC Asset Management Europe.

The risk tolerance of international investors. D.. 2003. Lettau. P. A. 2002. 11 . K. Hussain. Persaud.. A. 225–262. Jagannathan. 249–265. NBER. 355–379. 167–188. Hansen. P. Tsatsaronis. M. P. J. N.. Ludvigson. O’Connell. R. 1969. M. Journal of Applied Econometrics 17. Renault.... Journal of Econometrics 119. BIS Quarterly Review (June).. J. Gai. 2004. Morgan Global FX Research. 1996. risk aversion and the temporal behavior of asset returns. Working paper 10157. 1991. 1983.. Singleton. A.. L. 1238–1287.. 2006. N. Panigirtzoglou. Working paper 2003-23. Pure contagion and investors’ shifting risk appetite: Analytical issues and empirical evidence.. Implications of security market data for models of dynamic economies.. Journal of Political Economy 91. Liquidity and risk appetite: Evidence from equity index option prices. Investors’ attitude towards risk: What can we learn from options?. Meddahi.. International Finance 5. Meddahi. Investors’ changing appetite for risk. 2003. K. Stochastic consumption. H. E.P. Econometrica 37. M. Kumar. N.. 401–426. K. Shin. Misina. N. P. Theoretical comparison between integrated and realized volatility. L. Vause. What does the risk-appetite index measure?. Journal of Political Economy 99. 2003. Persaud. Wallace. Hayes. Journal of Political Economy 109. Karampatos.. Measuring investors’ risk appetite. D. 2002. Resurrecting the (C)CAPM: A cross-sectional test when risk premia are time-varying. S. S... Temporal aggregation of volatility models.. Bank of England. N. Hansen. H.. Tarashev. J. S. 2001. The use of error components model in combining cross-section and time-series data. G... 55–72.. Bank of Canada. 57–65.. A. 2003. International Journal of Central Banking 2.Froot. 479–508. S...

- Read and print without ads
- Download to keep your version
- Edit, email or read offline

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

CANCEL

OK

You've been reading!

NO, THANKS

OK

scribd

/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->