Original Article

Dynamic strategic asset allocation:

Risk and return across the business
cycle
Received (in revised form): 11th March 2010

Pim van Vliet

(PhD, Vice President) is a Senior Portfolio Manager with the Quantitative Equities team of Robeco. His primary focus is Robeco’s
low-volatility strategies, including conservative equities. He joined Robeco in 2005 as a Senior Quantitative Researcher in the
Quantitative Strategies department with responsibility for the allocation research. He has published among others in the Journal of
Banking and Finance, Management Science and the Journal of Portfolio Management. He is a regular speaker at international
conferences and is a guest lecturer at several universities. Van Vliet holds a PhD and an MSc (cum laude) in Financial and Business
Economics from Erasmus University, Rotterdam.

David Blitz
(PhD, Vice President) is the co-head of Quantitative Strategies with Robeco. He is responsible for all equity-related quantitative
research. The key tools that have been developed in this area are proprietary stock selection models and portfolio optimization
algorithms. He started his career in 1995 at Robeco, and has published papers in a.o. the Journal of Portfolio Management. Blitz
holds a PhD in Business Economics and an MSc in Econometrics (cum laude) from Erasmus University, Rotterdam, and is
registered with the Dutch Securities Institute.

Correspondence: Pim van Vliet, Coolsingel 120, NL 3011 AG, Rotterdam, The Netherlands
E-mail: [email protected]

ABSTRACT We propose a practical investment framework for dynamic asset allocation

across different phases in the business cycle, which we illustrate using a sample of US data
from 1948 to 2007. We identify four phases in the business cycle and find that these capture
pronounced time variation in the risk and return properties of asset classes. Time variation is
also observed in the risk of a traditional, static strategic asset mix. In order to stabilize risk
across the business cycle, we propose a dynamic strategic asset allocation approach,
which has the potential to enhance expected return as well. The proposed investment
framework is found to be robust to variations in the variable composition of the business
cycle indicator and can easily be extended with different economic variables and/or
additional assets.
Journal of Asset Management (2011) 12, 360–375. doi:10.1057/jam.2011.12;
published online 28 April 2011

Keywords: asset allocation; economic regimes; business cycle; portfolio choice;

time-varying risk; time-varying return

INTRODUCTION determine an appropriate long-term strategic

The asset allocation decision is known to be asset allocation (SAA) policy, for example,
very important, and determines about 80–90 by engaging in an asset liability management
per cent of return variance (see, for example, study. In practice, strategic asset allocation
Brinson et al (1991); Ibbotson and Kaplan often turns into static asset allocation, with a
(2000); and Statman (2001) for a discussion of fixed allocation to different asset classes.
these findings). Thus, investors carefully Although it is known that the risk and return

& 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375
www.palgrave-journals.com/jam/
 Risk and return across the business cycle

properties of asset classes may vary over the business cycle. Moreover, in contrast to more
business cycle1, it is common practice to complex regime-switching models, which
assume constant risk/return parameters for derive regimes from the return characteristics
SAA purposes2, resulting in one portfolio of the assets themselves, our approach uses
with constant weights. However, although economic data to model the business cycle
the resulting SAA portfolio is static in directly. By focusing on economic
terms of composition, its risk and return fundamentals, the DSAA framework is
characteristics may actually exhibit significant designed to be closer to investment practice.
time variation over the cycle. The primary objective is not statistical
A popular way to exploit time variation in optimality, but to bridge the gap between
returns is to apply a tactical asset allocation research analyst and investor by concentrating
(TAA) overlay on the portfolio, as described on intuitive economic relations and
by, for example, Dahlquist and Harvey transparency. Instead of providing estimated
(2001). However, as the objective of a TAA probabilities of being in a particular regime at
program is usually to maximize returns any point in time, we can explicitly
within a certain stand-alone risk budget, for determine the prevailing phase of the
example, a tracking error or value-at-risk business cycle, which enables us to define and
limit, it does not offer a solution to compare several dynamic asset allocation
time-varying overall portfolio risk. A TAA strategies. We illustrate the DSAA framework
strategy may in fact turn out to exacerbate the with a sample of US market data over the
tendency of the SAA portfolio to become 60-year period from 1948 to 2007.
more risky during ‘bad’ economic times, Specifically, we combine four economic
which is particularly undesirable for a indicators (the credit spread, earnings yield,
risk-averse investor. As Cochrane (1999) ISM and the unemployment rate) to identify
points out, a risk-averse investor may prefer a four phases of the business cycle (expansion,
portfolio with a lower Sharpe ratio in a world slowdown, recession and recovery).6 As our
with time-varying risk and return, in case it business cycle phases do not depend on
offers protection during times of financial statistical properties of asset classes, we can
distress. Another approach is to use a easily consider a broad opportunity set
regime-switching framework, as in Ang and instead of focusing on a limited set of assets.
Bekaert (2002, 2004).3 This approach is In addition to equities, bonds and cash, we
based on statistical properties of the include small caps, value versus growth,
underlying assets and is typically used to credits and commodities in our analysis. Our
identify two regimes.4 Ang and Bekaert framework is intended to help long-term
(2004) argue that the economic mechanism investors design a transparent and practically
behind their regimes is likely the world feasible dynamic strategic asset allocation
business cycle. The main drawback of the strategy over the business cycle.
regime-switching approach is its complexity, Empirically, we find that the risk of a static
as a result of which investment professionals SAA portfolio tends to increase during bad
may be reluctant to rely on these models for times, which is undesirable for a risk-averse
real-life decision making.5 investor. In addition to risk, the average
In this article, we propose a simple and return of many assets is also found to be
transparent framework for dynamic strategic highly dependent on the prevailing economic
asset allocation (DSAA) based on the business phase. For example, most assets exhibit
cycle. In contrast to a traditional TAA above-average returns during recessions and
strategy, this DSAA framework aims at recoveries and below-average returns during
enhancing portfolio return while at the expansions and slowdowns. We show that
same time stabilizing portfolio risk across the investors can improve the performance of

& 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375 361
 van Vliet and Blitz

their strategic asset allocation by taking into or recession after the fact. Owing to this
account the risk and return properties of each hindsight, the NBER data are only suitable
asset class conditional on the phase in the for ex post explanatory analyses and not for ex
business cycle. We first examine a classic ante decision making. This is also recognized
tactical asset allocation approach, which by Gorton and Rouwenhorst (2006), who
concentrates on maximizing portfolio return use the NBER business cycle classification
during each economic phase. However, we for gaining insight into the risk and return
find that this approach is suboptimal from a properties of commodities over the cycle.7
risk perspective, as it tends to increase risk In order to address this concern, we
systematically, and during bad times in propose an alternative, forward-looking
particular. In order to address this concern, business cycle indicator. Our indicator uses
we propose a dynamic strategic asset only information that is actually available ex
allocation approach, which is successful at ante and offers the additional advantage of
stabilizing portfolio risk across the business resulting in a more balanced distribution
cycle, while at the same time offering the of observations across economic phases. In
potential to enhance portfolio return. The the appendix, we describe in detail how
key difference is that DSAA explicitly we combine four well-known economic
accounts for the interaction between tactical indicators into one overall business cycle
and strategic positions, which TAA ignores. indicator, which can take on four different
We show that our empirical results are robust states. The four phases are schematically
to changing the variables in the business cycle illustrated in Figure 1. In the ‘expansion’
model, with considerable potential for phase, the combination of four economic
further improvement. Finally, we argue that indicators is both positive and rising. In the
outsourcing a DSAA strategy to an external ‘slowdown’ phase, the level is still positive,
manager is more challenging than but conditions are worsening. In the
outsourcing a traditional TAA strategy. ‘recession’ phase, both level and direction are
Investors could try to enforce their TAA negative, whereas in the ‘recovery’ phase the
managers to behave in a DSAA-consistent level is still negative, but improving. We will
manner by imposing business cycle- refer to the expansion and slowdown phase as
dependent constraints, for example, by ‘good times’ and recession and recovery as
setting time-varying asset class bandwidths. ‘bad times’.
Alternatively, investors could decide to fully We also show that our business cycle
integrate DSAA into their own strategic asset indicator matches fairly well with the
allocation policy. ‘official’ NBER business cycle classification,
The article proceeds as follows. The next although we fully acknowledge that our
section discusses the data and methodology, method may be improved upon with a more
after which we present the empirical results.
The final section concludes.
Expansion Slowdown Recession Recovery

METHODOLOGY
Good times
Modeling the business cycle Bad times
The NBER is well known for determining
official recessionary periods. NBER data are
of little use for real-life dynamic asset Level: + Level: + Level: - Level: -
allocation purposes though, as the NBER Change: + Change: - Change: - Change: +

only classifies a period as either expansion Figure 1: Business cycle with four phases.

362 & 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375
 Risk and return across the business cycle

sophisticated approach. However, the cycle indicator. Each alternative is based on

indicator suffices for the purposes of this optimizing the asset allocation for each of the
article, namely to compare various dynamic four phases separately, where for each
strategic asset allocation approaches based on alternative we use a different set of
a business cycle framework, and to estimate restrictions. In order to make a fair assessment
the potential for risk/return improvement of the added value of the optimized dynamic
that is offered by these approaches. strategies, we compare them not only to our
Furthermore, we will show that our main base-case static SAA portfolio but also to a
results are robust to variations in the variable second SAA strategy, which is optimized
composition of the business cycle model. full sample for maximum return without
taking into account the business cycle, that is,
assuming that asset returns are IID. We
Asset allocation strategies denote this strategy by SAA-O and impose
We consider the following eight asset classes the restriction that it has the same absolute
and investment styles: US large-cap equities, risk as our base-case SAA portfolio and the
US small-cap equities, US value equities, US same constraints on asset weights as our
growth equities, US credits, US Treasuries, dynamic, business cycle-based strategies. In
commodities and cash.8,9 All returns are in the spirit of Merton (1971), our SAA-O
US dollars. The sample period is from approach represents the solution to a myopic
January 1948 to December 2007, spanning a (one period) problem, whereas our dynamic
total of 60 years. We use a monthly data asset allocation strategies, which explicitly
frequency. assume that asset returns are not IID,
As a base-case strategic asset allocation represent intertemporal hedging demands
policy, we consider a static strategic asset based on the business cycle.10 An overview is
allocation portfolio, denoted by SAA, which given in Table 1.
invests every month 25 per cent in large-cap Our first alternative that explicitly takes
equities, 25 per cent in Treasuries and the business cycle into account is a tactical
25 per cent in cash (core assets) and 5 per cent asset allocation strategy, in which we
in value equities, 5 per cent in growth optimize the portfolio in each phase of the
equities, 5 per cent in small-cap equities, cycle for maximum expected return, subject
5 per cent in credits and 5 per cent in to a 1 per cent tracking error limit. By using a
commodities (satellite assets). Next, we tracking error constraint, we ensure that the
consider several dynamic asset allocation optimized portfolios do not exhibit extreme
approaches that are based on our business deviations from the static SAA reference

Table 1: Definition asset allocation strategies

SAA-O TAA TAA-C DSAA

Panel A: Optimization approach

Full-sample versus phase-based full-sample phase-based phase-based phase-based
Panel B: Asset weight restrictions
Core assets (%) 0–100 0–100 0–100 0–100
Satellite assets 0–10 0–10 0–10 0–10
Panel C: Relative risk constraints
Tracking error limit (%) — 1 1 1
Panel D: Absolute risk constraints
Volatility limit full sample sSAA full sample — sSAA full sample sSAA full sample
Volatility limit for each phase — — — sSAA full sample

& 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375 363
 van Vliet and Blitz

portfolio, and we implicitly control data for each of the four economic phases,
transaction costs by forcing the optimizer to which is already a relatively short period of
focus on the most attractive bets. The asset time for strategic asset allocation purposes.
weights are required to be non-negative, to In other words, with an in-sample/
add to 1 and not exceed 10 per cent for the out-of-sample approach, the in-sample phase
five satellite assets.11 Weights for the three would already require most (or all) of our
core assets are not constrained to a maximum sample, leaving hardly any (or no) remaining
value. data for an out-of-sample test. Furthermore,
As it turns out that the base-case TAA as mentioned before, the primary objective of
approach structurally increases portfolio risk, this article is to present a framework for
we also consider an alternative tactical asset dynamic asset allocation, and the empirical
allocation approach (TAA-C, denoting data are only meant to illustrate the potential
constrained TAA), which is identical to the of such an approach. Our results do not aim
TAA approach, except for the additional to represent real-life investment strategies.
constraint that overall volatility does not
exceed overall volatility of the static SAA
portfolio. By definition, this approach RESULTS
prevents a structural increase in portfolio risk.
However, it does not offer a solution for the Risk and return across the cycle
tendency of the static SAA portfolio to We begin our empirical analysis by
become more risky during bad times. In fact, investigating the risk and return of the assets
the TAA-C approach turns out to exacerbate in our sample across the different economic
this effect. In order to stabilize portfolio risk phases. Table 2 shows the correlations
across the business cycle, we therefore between several asset classes during each
consider a final alternative, which we call a phase of the business cycle. These estimates
dynamic strategic asset allocation strategy. are based on monthly data, but we note that
With this approach, we impose the additional quarterly data yield similar outcomes. The
restriction that not only overall portfolio average correlations of equity with bonds,
volatility, but also volatility during each of the credits and commodities are 0.13, 0.29 and
four phases does not exceed the overall 0.00, respectively, over the total 60-year
volatility level of the static SAA. sample period. Interestingly, however, we
All portfolio optimizations are full sample find that equity-bond correlations are
because of data limitations. An approach negative during slowdowns and become
based on in-sample strategy development, positive during recessions and recoveries.
followed by an out-of-sample test, is This means that diversification benefits fade
practically infeasible.12 With the full-sample when they are needed most. By contrast, we
approach, we have an average of 15 years of find that during recoveries the correlation

Table 2: Key correlations across different economic phases

Equity–Bonds Equity–Credits Equity–Commodities

Panel A
Full sample 0.13 0.29 0.00
Panel B
Expansion 0.04 0.09 0.02
Slowdown 0.16 0.12 0.02
Recession 0.15 0.35 0.02
Recovery 0.38 0.45 0.12

Sample period 1948–2007.

364 & 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375
 Risk and return across the business cycle

between equities and commodities becomes 2.9 per cent in excess of cash, but this return
negative, indicating more opportunities for varies between 0.5 per cent during
diversification. slowdowns and 4.8 per cent during
Table 3 shows the annualized volatility recessions. This result is driven by the
of each asset class, as well as the static SAA fact that equity returns are highest during
portfolio, across the different phases. We find recessions and lowest during slowdowns.
that risk tends to be highest during recessions This suggests that financial markets run ahead
and recoveries (bad times). The full-sample of the business cycle by about one phase. In
volatility of the static SAA portfolio is other words, when the real economy slows
6.2 per cent, but this number varies between down, equity markets already show
5.6 per cent in good times and 6.6 per cent in disappointing returns because of the
bad times. This time-varying risk profile anticipated recession, whereas equity markets
is mainly caused by (1) the increased risk are already recovering when the real
of government and corporate bonds in bad economy is still in recession. This implies that
times and (2) the increased equity–bond financial markets do not concentrate on
correlation during bad times discussed current economic conditions, but also take
before. During recessions, the volatility into account expected future economic
of commodities decreases somewhat, and conditions. An important observation is
during recoveries the correlation of therefore that bad times for the economy are
commodities with other asset classes becomes not necessarily bad times for investors!
more negative. Equities show limited time During bad economic times, not only are
variation in risk across the four phases in the risks higher, but returns also tend to be
business cycle. higher.
Table 4 shows the excess (log-) returns of Interestingly, both the value premium
each asset class across the four phases. The (value versus growth) and the size premium
SAA portfolio yields an average return of (small versus large) are negative during

Table 3: Risk of asset classes for each economic phase

Equity (%) Value (%) Growth (%) Small (%) Credits (%) Bonds (%) Comm (%) SAA (%)

Panel A
Full sample 14.2 14.0 15.4 18.9 6.4 4.7 15.6 6.2
Panel B
Expansion 13.8 14.0 14.6 18.7 4.5 3.5 14.0 5.6
Slowdown 13.7 13.8 15.2 19.8 5.1 3.9 18.1 5.7
Recession 14.6 14.0 16.0 18.7 8.2 5.6 15.9 6.6
Recovery 13.9 13.3 15.3 18.5 6.4 5.1 15.7 6.4

Risk is defined as annualized volatility. Sample period 1948–2007.

Table 4: Annualized excess return of asset classes during each economic phase
Equity (%) Value (%) Growth (%) Small (%) Credits (%) Bonds (%) Comm (%) SAA (%)

Panel A
Full sample 5.6 6.4 4.7 6.6 0.5 0.6 1.3 2.9
Panel B
Expansion 3.7 3.2 3.9 0.9 1.0 0.4 5.7 1.8
Slowdown 0.2 1.9 1.6 2.9 3.0 0.1 0.2 0.5
Recession 10.2 11.1 9.0 12.7 1.6 1.4 3.7 4.8
Recovery 5.1 7.1 3.3 9.4 3.0 1.2 6.1 3.4

Sample period 1948–2007.

& 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375 365
 van Vliet and Blitz

expansions, while being positive during the premium to the maximum extent possible,
other three phases of the cycle. Also and it makes maximum use of the attractive
noteworthy is the lack of a credit risk risk/return properties of commodities.
premium (credits versus Treasuries) in our Nevertheless, it only manages to improve the
sample. Credits outperform government excess return by a statistically insignificant
bonds during recessions and recoveries, but 0.20 per cent, while also failing at stabilizing
underperform during expansions and the risk of the portfolio over the cycle.
slowdowns. Commodities deliver high The first dynamic allocation strategy that
returns during expansions and recoveries, we consider is the base-case TAA approach,
while lagging during slowdowns and which, for each phase, selects a portfolio that
recessions. In summary, we observe various is optimized for highest expected return,
pronounced cyclical patterns in the return taking into account the restrictions outlined
characteristics of the different asset classes, in the methodology section. The average
which motivate the examination of business excess return of the TAA approach is 3.71 per
cycle-based asset allocation strategies. cent per annum, compared with 2.90 per
cent for the static SAA approach. The
difference is 0.81 per cent, which, given the
Business cycle-based asset tracking error budget of 1 per cent, translates
allocation into an information ratio of 0.81. This
In this section, we compare the static SAA performance is not only economically
approach with the business cycle-based asset significant, but also highly statistically
allocation strategies defined in the significant, with an associated t-value of
methodology section. In Table 5, we show 6.29. The return differential ranges from 0.45
the optimized portfolio weights and in per cent during slowdown phases to 1.08 per
Table 6 we show the risk/return cent during recessions. However, not only
characteristics of the various approaches. portfolio return, but also portfolio risk is
The SAA-O full-sample optimized increased. In fact, the TAA portfolio exhibits
portfolio exploits the value and small-cap a systematically higher level of risk than the

Table 5: Optimized business cycle-based allocation strategies

Equity (%) Value (%) Growth (%) Small (%) Credits (%) Bonds (%) Comm (%) Rf (%)

Panel A: SAA
Static base-case 25 5 5 5 5 25 5 25
Panel B: SAA-O
Static optimized 19 10 — 10 — 25 10 26
Panel B: TAA
Expansion 39 — 3 2 — 26 10 21
Slowdown 22 10 — 9 — 31 5 23
Recession 29 10 — 8 — 34 3 15
Recovery 24 10 — 10 10 23 9 14
Panel C: TAA-C
Expansion 34 — 4 — — 19 10 33
Slowdown 14 10 — 10 — 27 4 35
Recession 25 10 — 9 — 31 1 24
Recovery 19 10 — 10 10 20 10 21
Panel D: DSAA
Expansion 39 — 5 — — 21 10 25
Slowdown 19 10 — 10 — 31 5 25
Recession 20 10 — 9 — 27 — 34
Recovery 18 10 — 10 10 19 10 23

Sample period 1948–2007.

366 & 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375
 Risk and return across the business cycle

SAA reference portfolio. Compared with the per cent during recoveries. During
static SAA portfolio, overall volatility slowdowns, recessions and recoveries, the
increases from 6.17 per cent to 6.82 per cent, TAA-C strategy overweights value stocks and
and volatility is also higher in each of the small-caps, and underweights growth stocks.
four separate economic phases. One of the During expansions, the strategy underweights
reasons for this is that the TAA portfolios value and small caps and is neutral on growth
systematically overweight equities and stocks. Credits are always underweighted,
underweight cash compared with the static except during recoveries. Commodities are
SAA portfolio. In other words, the additional overweighted during expansions and
return generated by the TAA approach is, at recoveries, neutral during slowdowns and
least partly, simply a reward for additional underweighted during recessions.
beta exposures, instead of true alpha. In fact, Although the TAA-C approach is
the outperformance of the TAA portfolio successful at controlling overall portfolio risk,
vis-à-vis the SAA portfolio exhibits a it does not succeed in achieving a stable risk.
correlation of 0.62 with the absolute return Both the static SAA strategy and the TAA-C
of the SAA portfolio. The importance strategy exhibit pronounced time-varying
of making the distinction between risk across economic phases. For the static
performances as a result of true alpha instead SAA strategy, this is simply due to the
of implicit beta in the context of tactical asset time-varying risk characteristics of asset
allocation is also stressed by Lee (2000). classes. The TAA-C strategy, however, goes
In order to address this concern, we one step further by actively reducing risk
consider the TAA-C approach, which is during low-return phases (expansions and
specifically aimed at preventing an increase in slowdowns), in order to be able to take on
overall portfolio risk. Table 6 shows that more risk during the highest return phase
overall portfolio risk of the TAA-C strategy is (recessions). In other words, instead of
indeed equal to that of the static SAA strategy. countering the tendency of the SAA strategy
Consistent with this, the structural overweight to exhibit more risk during bad economic
of equities observed for the TAA strategy is times, the TAA-C strategy turns out to
not present with the TAA-C strategy. The exacerbate this behavior. As mentioned
average excess return improvement drops to before, this is particularly undesirable for a
0.61 per cent (equivalent to an information risk-averse investor, see Cochrane (1999).
ratio of 0.61, with a t-value of 4.70), ranging In order to address this concern, we
from 0.43 per cent during expansions to 0.81 propose an approach that we call dynamic

Table 6: Risk and return characteristics of business cycle-based allocation strategies

Return Risk

SAA SAA-O TAA TAA-C DSAA SAA SAA-O TAA TAA-C DSAA
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)

Panel A
Full sample 2.90 3.10 3.71 3.51 3.38 6.17 6.17 6.82 6.17 6.17
Outperformance — 0.20 0.81 0.61 0.48 — — — — —
(t-statistic) — (1.54) (6.29) (4.70) (3.71) — — — — —
Panel B
Expansion 1.78 1.98 2.35 2.21 2.44 5.60 5.70 6.07 5.35 6.17
Slowdown 0.45 0.99 0.90 0.90 0.95 5.68 5.98 5.83 4.95 5.66
Recession 4.80 4.75 5.88 5.55 4.97 6.55 6.46 7.54 6.98 6.17
Recovery 3.37 3.76 4.42 4.18 4.12 6.36 6.16 7.06 6.32 6.17

Risk is defined as annualized volatility and return is excess return over cash. Outperformance is defined as the
return difference with the SAA reference portfolio. Sample period 1948–2007.

& 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375 367
 van Vliet and Blitz

strategic asset allocation, or DSAA. This portfolio risk. In this section, we will
approach is specifically designed to stabilize examine the robustness of our finding that,
risk across the business cycle. Specifically, at the very least, a business cycle-based
the portfolios are optimized subject to the approach is able to identify time-varying risk
constraint that not only overall volatility, but characteristics and can be used to adjust the
also volatility during each phase does not portfolio composition accordingly.
exceed 6.17 per cent. Looking at the We begin by examining the risk of the asset
resulting portfolios, we observe that the main classes in our sample during official NBER
change compared with the TAA-C approach expansions and contractions. The results are
is the weight of equities, which is now shown in Table 7. Consistent with the results
chosen in such a way that risk across different for our business cycle model, we observe that
economic phases is stabilized. The average the risk of each class is significantly higher
return enhancement is equal to 0.48 per cent, during ‘bad times’ (NBER contractions)
implying an information ratio of 0.48 compared with ‘good times’ (NBER
(with a t-value of 3.71). The return expansions). This provides corroborating
improvement ranges between 0.50 per cent evidence for the existence of time-varying
and 0.75 per cent during expansions, risk that is linked to the business cycle.
slowdowns and recoveries. During The business cycle model that has been
recessions, the main improvement is not an used throughout this article appears to be
increase in expected return (which is effective at identifying time-varying risk ex
enhanced by only 0.17 per cent in this phase), ante. We may wonder, however, how
but a reduction of risk, as volatility can be sensitive this result is to our choice of
seen to drop from 6.55 per cent to 6.17 per variables that, together, comprise the business
cent. Stable risk across the business cycle is cycle model. Therefore, we examined the
desirable for a risk-averse investor with a effects of leaving out any one of the four
constant risk budget. To summarize, the variables in the business cycle model. The
DSAA approach is able to stabilize risk across results of this analysis are shown in Table 8.
the business cycle, while, at the same time, We observe that removing any one of the
improving expected return. four variables in the business cycle model
does not degrade its ability to identify time-
varying risk. If anything, this aspect of the
Robustness model appears to improve somewhat, as the
The case for dynamic strategic asset allocation dispersion in volatility across economic
hinges on two premises, namely the ability to phases tends to widen.
identify time-varying risk and time-varying If DSAA is applied with the sole objective
return opportunities. In reality, the latter to stabilize portfolio risk, one might consider
condition is likely to be most challenging. tailoring the business cycle model specifically
However, even if expected returns are toward this purpose. Empirically, volatility is
considered to be unpredictable, there remains often found to be persistent, that is, if
a case for DSAA, although a simplified volatility has recently been high (low), it is
variant that is solely aimed at stabilizing likely to remain high (low) in the following

Table 7: Risk of asset classes for each economic phase

Equity (%) Value (%) Growth (%) Small (%) Credits (%) Bonds (%) Comm (%) SAA (%)

NBER expansions 13.3 13.1 14.4 17.5 5.5 4.2 14.4 5.7
NBER contractions 18.1 17.6 19.6 24.8 9.8 6.3 20.7 8.2

Risk is defined as annualized volatility. Sample period 1948–2007.

368 & 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375
 Risk and return across the business cycle

Table 8: SAA portfolio risk across the business cycle for alternative business cycle models
Regime model sensitivity analysis

Base case (%) Excl. credit Excl. earnings Excl. Excl. Plus equity
spread (%) yield (%) ISM (%) unemployment (%) volatility (%)

Expansion 5.60 5.54 5.14 5.25 5.48 5.31

Peak 5.68 5.80 5.84 6.09 5.65 5.39
Recession 6.55 6.53 6.67 6.52 6.58 6.83
Recovery 6.36 6.56 6.78 6.62 6.36 6.48

Risk is defined as annualized volatility. Sample period 1948–2007.

period. The literature, which takes a purely combination of four well-known economic
statistical approach toward identifying variables, which can take on four different
regimes in the time-series properties of asset states. The business cycle indicator is found
class returns, also tends to find alternating to relate reasonably well to the official NBER
periods with high and low volatility. Inspired business cycle. Our first empirical result is
by these results, we examined 12-month that the risk and return properties of asset
realized equity volatility as a potential classes are highly dependent on the prevailing
additional (or alternative) factor for the economic phase. This finding motivates
business cycle model.13 Consistent with the examination of various dynamic asset
what we would expect a priori, we find that allocation strategies. As a benchmark, we take
12-month realized equity volatility is above a traditional SAA portfolio, which has static
average and increasing during NBER weights, but a time-varying risk/return
contractions. The last column in Table 8 profile over the business cycle. In particular,
shows that when the variable is added to our risk tends to go up in bad times, which is
base-case business cycle model, it becomes undesirable for a risk-averse investor. An
better at identifying time-varying risk alternative, full-sample optimized static SAA
characteristics. If 12-month realized equity portfolio is neither able to stabilize risk across
volatility is added as a fifth factor, the spread the cycle nor does it succeed in significantly
between the maximum and minimum enhancing return.
volatility across economic phases increases One way investors can exploit the business
from less than 1 per cent to over 1.5 per cent. cycle is by developing a TAA strategy, which
We conclude that, if the primary objective is designed for a maximum outperformance
of a business cycle approach is considered to in each phase of the cycle. The drawback of
be stabilizing portfolio risk over time, our this approach is that absolute portfolio risk is
base-case business cycle model does not paint increased systematically, and in particular
an overly optimistic picture, as we find that once again, in bad times. In order to stabilize
even stronger results may be obtained by absolute portfolio risk and simultaneously
considering additional variables. enhance portfolio returns, we propose a
dynamic strategic asset allocation approach.
We have shown that the proposed DSAA
SUMMARY AND IMPLICATIONS approach is robust to variations in the variable
We propose a practical investment framework composition of the business cycle model and
for dynamic asset allocation across the that the approach can easily be extended with
business cycle, which we illustrate with a different economic variables and/or
sample of data for the US market over the additional assets. The DSAA framework is
period from 1948 to 2007. We construct a intended to bridge the gap between research
business cycle indicator on the basis of a analyst and investor by concentrating on

& 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375 369
 van Vliet and Blitz

intuitive economic relations and manager has been successful at stabilizing

transparency, in contrast to existing overall portfolio risk. In order to avoid these
statistically driven techniques, such as practical complexities, investors could decide
Markov regime-switching models. For to fully integrate DSAA into their internal
investors who are skeptical toward exploiting strategic investment policy. Alternatively,
time variation in asset returns, we have investors might enforce their TAA managers
shown that DSAA can still serve as a robust to behave in a DSAA-consistent manner by
tool for stabilizing portfolio risk across the imposing business cycle-dependent constraints
cycle, with considerable potential for further (for example, bandwidths) on the exposure to
improvement. each asset class. For example, in case a high-
Interestingly, the aim for stable absolute risk economic phase is identified, a TAA
performance across the business cycle manager might be restricted from
through dynamic strategic asset allocation overweighting high-risk assets such as equities.
leads to markedly different portfolios and
performance characteristics than the aim for
stable outperformance through tactical asset ACKNOWLEDGEMENTS
allocation. Absolute return investors bring We appreciate the suggestions made by Roy
down risk during bad times, whereas relative Hoevenaars, Angelien Kemna, Laurens
return investors increase risk during bad Swinkels, Artino Janssen, Jan Sytze
times. These opposing outcomes imply that it Mosselaar, Jaap van Dam, an anonymous
is essential to clearly specify the investment referee, seminar participants at the June 2009
objectives, consistent with the finding by State Street European Quantitative Forum
Binsbergen et al (2008) that a decentralized and participants of the Robeco SAA seminar
investment approach can lead to suboptimal in November 2009. Financial support by the
portfolios. Robeco Group is gratefully acknowledged.
Institutional investors who would like to Any remaining errors are the authors’
implement a business cycle-based allocation responsibility.
approach need to choose between internal
versus external management. In case of
outsourcing, the challenge is to align the NOTES
investment objectives of the investor with 1. For example, Sa-Aadu et al (2006) examine diversification
those of the external manager. Outsourcing is benefits of alternative assets across regimes and find that
commodities and real estate offer a hedge during periods
fairly straightforward if the objective is simply with low per capita consumption growth (economic bad
to enhance return. As we have seen, however, times). Gorton and Rouwenhorst (2006) find that the
a TAA approach that concentrates on diversification benefits of commodities in a balanced
portfolio vary across the different phases in the business
maximizing returns can have undesired cycle.
consequences for the overall risk profile of 2. For example, for strategic asset allocation purposes
the portfolio. A DSAA strategy with overall Hoevenaars et al (2007) and Bekkers et al (2009) suggest
using long-term historical data for estimating volatilities
and business cycle-dependent constraints on
and correlations, while deriving expected returns from a
absolute portfolio risk addresses this concern, combination of long-term historical data, economic
but outsourcing this to an external manager is theory and current market circumstances.
likely to be more challenging in practice. 3. Ang and Bekaert (2004) design a strategy that tries to
exploit time variation in returns using the switching
Risk monitoring should become more framework of Hamilton (1989). They find that dynamic
sophisticated and an ex post performance asset allocation across two regimes improves return for
evaluation of the external manager should country allocation and for allocation across equities, bonds
and cash.
not focus solely on the realized 4. The approach can be extended to model more regimes.
outperformance and tracking error of the For example, Guidolin and Timmermann (2007)
manager, but also evaluate whether the statistically identify four regimes (crash, slow-growth, bull

370 & 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375
 Risk and return across the business cycle

and recovery) to capture the joint distribution of stock and 15. research.stlouisfed.org/fred2
bond returns, which they use to derive a regime-based 16. www.econ.yale.edu/Bshiller/data.htm
strategic asset allocation strategy. They also provide a good 17. www.nber.org/cycles.html
overview of the existing literature on this subject. 18. We would like to stress that other macro or market factors
5. An econometrician is needed to estimate the regime- can also be used. These four factors are used to illustrate
switching probabilities either with maximum likelihood how economic data can be used to construct a dynamic
or Bayesian techniques. The model must be kept simple SAA framework.
with a limited number of assets because in the setup of 19. We use median instead of mean in order to reduce the
Ang and Bekaert (2004) with 2 regimes and 6 assets, impact of outliers on the resulting z-scores.
19 parameters need to be estimated and one should check 20. If the factors are normally distributed and uncorrelated,
whether the estimates are not ill-behaved. then the economic score is also standard normally
6. A detailed description of the model can be found in the distributed. However, in practice we observe that the
appendix. In the robustness section, we show that the factors are positively correlated, especially during stressful
results do not critically depend on the inclusion of any of periods.
these four factors. 21. In order to limit the transition of one phase to another
7. Actually, Gorton and Rouwenhorst (2006) go even one because of noise (signal flip-flopping), we include an
step further by distinguishing between early and late absolute threshold of 0.10. For example, if the combined
expansions and early and late recessions, on the basis of the indicator is þ 0.05 then this is within the bandwidth of
ex post identification of each expansion and recession 0.10 and þ 0.10, which means that we do not change
midpoint. This introduces an additional element of phase and keep the same phase as in the previous month.
hindsight. Another drawback of this approach is that the The same threshold applies for 1-year changes, which
frequency of the four resulting phases is quite unbalanced. should also exceed an absolute value of 0.10.
Specifically, for our 60-year sample, early and late
recessions in particular each contain only 8 per cent of the
data points, which is equivalent to fewer than 5 years of
observations. REFERENCES
8. For large-cap equity returns, we use the S&P 500 index. Ang, A. and Bekaert, G. (2002) International asset
For value and growth returns, we use the MSCI BARRA allocation with regime shifts. Review of Financial Studies
value and growth indices, which are available from 15(4): 1137–1187.
February 1975 onward, and before this we use data from Ang, A. and Bekaert, G. (2004) How do regimes affect asset
Kenneth French (BV and BG). For small caps, we use the allocation? Financial Analyst Journal 60(2): 86–99.
Russell 2000 index, backfilled with small-cap return data Bekkers, N., Doeswijk, R. and Lam, T. (2009) Strategic Asset
from Kenneth French before January 1979. Credit returns Allocation: Determining the Optimal Portfolio with Ten
are based on the Lehman US Aggregate Corporate index, Asset Classes. Working paper.
backfilled with data from Ibbotson (LT Corporate) before Binsbergen, J.H., van, R.K. and Michael, W.B. (2008)
January 1973. US Treasuries are based on the Lehman US Optimal decentralized investment management. Journal of
Aggregate Treasury index, backfilled with Ibbotson data Finance 63(4): 1849–1895.
(IT Government) before January 1973. Commodities are Brinson, G.P., Singer, B D. and Beebower, G.L. (1991)
defined as the GSCI index, backfilled with the CRB spot Determinants of portfolio performance II: An update.
index before January 1970. Cash is defined as the return Financial Analysts Journal 47(3): 4–48.
on US 30-day T-bills. Campbel, J.Y. and Shiller, R. (1987) Stock prices,
9. As US value equities and US growth equities together earnings, and expected dividends. Journal of Finance
(approximately) comprise US large-cap equities, the latter 43(3): 661–676.
asset class might be considered redundant. Nevertheless, Chen, N-F, Roll, R. and Stephen, A.R. (1986) Economic
we prefer to include each of these asset classes because it is forces and the stock market. Journal of Business 59(3):
convenient when, later on, we distinguish between core 383–403.
and satellite asset classes. Cochrane, J H. (1999) Portfolio Advice for a Multifactor
10. This argument is also given by Ang and Bekaert (2002) in World. NBERWorking paper no. 7170.
the context of their regime-switching model. Dahlquist, M. and Harvey, R.C. (2001) Global tactical asset
11. Jagannathan and Ma (2003) show that restricting portfolio allocation. Emerging Markets Quarterly 51 (Spring): 6–14.
weights is effectively a form of covariance matrix Diris, B., Palm, F. and Schotman, P. (2008) Long-term
shrinkage. Diris et al (2008) stress the importance of Strategic Asset Allocation: An Out-of-sample Evaluation.
shrinkage when determining an asset allocation strategy. Maastricht University Working paper.
12. Goyal and Welch (2008) point out that many conditional Gorton, G and Geert Rouwenhorst, K. (2006) Facts and
variables may work well in-sample, but fail out-of-sample fantasies about commodity futures. Financial Analyst Journal
after they have been documented. 62(2): 47–68.
13. A forward-looking implied volatility indicator such as Goyal, A. and Welch, I. (2008) A comprehensive look at the
VIX might be even more appealing, but the data for such empirical performance of equity premium prediction.
variables are not available with a history of 60 years. Review of Financial Studies 21(4): 1455–1508.
14. The ISM data and unemployment data are final figures Guidolin, M. and Timmermann, A. (2007) Asset allocation
and could differ from the preliminary figures published under multi-variate regime switching. Journal of Economic
earlier. Dynamics and Control 31(11): 3503–3544.

& 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375 371
 van Vliet and Blitz

Hamilton, J.D. (1989) A new approach to the economic Table A1: Level and 1-year change of economic
analysis of non-stationary time series and the business cycle. indicators for full sample (720 months) and recession
Econometrica 57: 357–384. periods (114 months)
Hoevenaars, R., Molenaar, R. and Schotman, P. (2007)
Full sample NBER contractions
Strategic asset allocation for long-term investors: Parameter
uncertainty and prior information. Working paper, 1-year
Maastricht University. Level SD Level change
Ibbotson, R.G. and Kaplan, P.D. (2000) Does asset allocation
policy explain 40%, 90%, or 100% of performance? Credit spread 96.1 42.5 120 þ 27.8
Financial Analysts Journal 56(1): 26–33. Earnings yield (%) 7.2 3.1 9.5 þ 1.3
Jagannathan, R. and Ma, T. (2003) Risk reduction in large ISM index 54.5 8.0 43.2 10.8
portfolios: Why imposing the wrong constraints helps. Unemployment (%) 5.6 1.5 6.0 þ 1.2
Journal of Finance 58(4): 1651–1684.
Lee, W. (2000) Theory and Methodology of Tactical Asset
Allocation. New York: John Wiley & Sons.
Merton, R.C. (1971) Optimal consumption and portfolio
be a leading economic indicator) and the
rules in a continuous-time model. Journal of Economic Theory unemployment rate (which is a widely used
34: 373–413. lagging economic indicator).
Sa-Aadu, J., James, D.S. and Tiwari, A. (2006) Portfolio
Performance and Strategic Asset Allocation Across
The credit spread is defined as the
Different Economic Conditions. Working paper. difference between Baa and Aaa corporate
Statman, M. (2001) How important is asset allocation? bond spreads from Moody’s and the earnings
Journal of Asset Management 2(2): 128–135. yield is the E/P ratio of the S&P500. A high
Stock, J.H. and Watson, M W.A. (1992) A Procedure for
Predicting Recessions with Leading Indicators: credit spread or high earnings yield indicates
Econometric Issues and Recent Experience. NBER ‘contraction’, whereas low spreads or yields
Working paper series, 4014. indicate ‘expansion’. The ISM is the
seasonally adjusted US manufacturers’ survey
production index and the unemployment rate
APPENDIX is the seasonally adjusted US unemployment
rate from the Bureau of Labor Statistics.14 An
Construction of business cycle ISM value above 0.50 or low unemployment
indicator indicates ‘expansion’ and an ISM below 0.50
The philosophy behind our business cycle or high unemployment indicates
indicator is to combine a limited number of ‘contraction’. All data are obtained from
economically relevant variables using a Datastream and before 1970 are backfilled
simple and transparent model structure, using the FRED database15 and for P/E using
aiming for a reasonably good match with the Robert Shiller’s database.16
official business cycle as classified by the Figure A1 shows the monthly data for
NBER. We acknowledge that by using a each of our four business cycle indicators,
more sophisticated approach it would defined in the data section, over the full
probably be possible to develop a superior sample of 60 years. We observe different cycle
model, but this is not our main objective. lengths for each of the four factors. For the
The number of possible variables is limited credit spread level, we observe three main
because we impose the requirement that a phases: the credit spread is between 50 and
data history of at least 60 years should be 125 during most of the 1950s and 1960s,
available. We consider two market variables rising to 75–300 during the 1970s and 1980s
that have already been linked to the business and falling back to 50–125 starting from the
cycle in the literature on conditional asset 1990s onward. The earnings yield can be
pricing, namely the credit spread (Chen et al, described in four phases. It varies between 8
1986) and the earnings yield (Campbel and and 16 per cent from 1948 to 1958, falling to
Shiller, 1987), and two macro-economic 4–8 per cent during 1959–1973, going up
variables that are also clearly linked to the again to 8–15 per cent during the 1973–1985
cycle, namely the ISM (which is designed to period and then varying between 2 and 8 per

372 & 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375
 Risk and return across the business cycle

Table A2: Distribution of four economic phases and transition matrix

Total no. NBER From/To Transition Matrix

Expansion (%) Slowdown (%) Recession (%) Recovery (%)

230 1 Expansion 93 3 — 3
95 6 Slowdown 6 83 11 —
238 101 Recession 2 3 90 4
144 6 Recovery 4 — 8 88

Credit Spread (Baa-Aaa) Earnings Yield

3.50 16%

3.00
12%
2.50
2.00
8%
1.50
1.00 4%
0.50
0.00 0%
1948 1958 1968 1978 1988 1998 2008 1948 1958 1968 1978 1988 1998 2008

ISM Unemployment
80 12%

70 10%

60 8%

50 6%

40 4%

30 2%

20 0%
1948 1958 1968 1978 1988 1998 2008 1948 1958 1968 1978 1988 1998 2008

Figure A1: Time series of each conditioning variable. Monthly observations over the 60-year period from 1948 to
2007.

cent during the 1985–2007 period. The ISM shows the average level and 1-year change of
factor fluctuates much more frequently, each indicator during the full sample and
passing the neutral level of 0.50 either from during NBER recession periods.
either above or below about 30 times over We observe that during NBER
this sample period. Finally, we observe that contraction periods (1) the credit spread is
the unemployment rate passes the median high and increasing, (2) earnings yield is high
value of 5.5 per cent about 15 times. Thus, and increasing, (3) ISM is low and decreasing
the different cycles last between 2 and 20 and (4) unemployment is high and increasing.
years, depending on the variable under Thus, we find that both the level and the
consideration. 1-year change contain information about
We proceed by relating the economic economic conditions. During NBER
indicators to the NBER business cycle contraction periods, we observe that the four
indicator. NBER defines 114 out of the total indicators, both in terms of level and change,
720 months in our sample as contraction deviate by about 0.5–1.0 standard deviations
period, or 16 per cent of all observations.17 from their long-term average values.
It is important to note that NBER classifies a In order to obtain a robust and broad
contraction with hindsight, whereas our indication of the condition of the US
business cycle indicator only uses information economy, we proceed by combining the four
available before the next month. Table A1 economic factors into one overall business

& 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375 373
 van Vliet and Blitz

cycle score.18 We note that the main findings for the full-sample period from 1948 to 2007.
in this article are robust to leaving out any The economic indicator varies between
one of the four indicators. We standardize the þ 1.8 (1965) and 5.0 (1982). A positive
four economic variables by deducting their score indicates ‘good times’, whereas a
full-sample medians19 and by dividing by negative score indicates ‘bad times’. In
their full-sample standard deviations. To limit general, we find that gradual increases in
the impact of outliers and individual factors, economic conditions are followed by abrupt
we further cap the individual z-scores to a downside shocks. The figure shows that a
maximum of þ 3 and a minimum of 3. negative and/or falling indicator is associated
Finally, we combine the individual z-scores with contraction periods. This finding is in
into one overall score by adding the line with the results of Table A1.
individual scores and dividing by the square We translate the combined economic
root of 4.20 indicator into four phases, depending
On the basis of the combined indicator, on its (i) level and (ii) 1-year change. For
we define four economic phases. In the example, at the end of 2007 both the level
‘expansion’ phase, the combined indicator is and change were negative, which implies
both positive and increasing. In the that the following period is classified as a
‘slowdown’ phase, the level is still positive, recession phase.
but conditions are worsening. In the Table A2 shows the distribution of the
‘recession’ phase, both level and direction are four economic phases and the accompanying
negative, whereas in the ‘recovery’ phase the transition matrix. The expansion and
level is still negative, but improving. This recession phases occur more often than
classification is consistent with Gorton and slowdown and recovery phases. This can be
Rouwenhorst (2006), who differentiate explained by our earlier observation that
between early and late expansion and early gradual increases in economic conditions
and late recession using NBER slowdown tend to be followed by rather abrupt declines
and through data. However, the advantages in economic conditions. This asymmetry
of our approach are that (i) it is applicable in causes slowdowns in particular to be rather
practice as the required data are readily short lived. The transition matrix shows that
available ex ante and (ii) the monthly the probability of staying in one phase from
observations are more evenly distributed month to month is between 83 per cent and
across economic phases, leading to more 93 per cent, whereas the probability of
statistical power. moving to another phase is 7–17 per cent.
Figure A2 shows the historical values of This translates into an average duration of
our economic indicator, together with the each phase in the business cycle of about 9
NBER contraction periods (shaded areas), months.21

2.0
1.5
1.0
0.5
0.0
J-48
J-51
J-54
J-57
J-60
J-63
J-66
J-69
J-72
J-75
J-78
J-81
J-84
J-87
J-90
J-93
J-96
J-99
J-02
J-05
J-08

-0.5
-1.0
-1.5
-2.0
-2.5
-3.0

Figure A2: Time series of the combined economic indicator combined with NBER contraction periods. Monthly
observations over the 60-year period from 1948 to 2007.

374 & 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375
 Risk and return across the business cycle

When our business cycle model indicates a the recession of the early 1990s is also
state of recession, there is a 42 per cent correctly predicted. Stock and Watson (1992)
chance that NBER will later on classify this argue that most leading indicators have
month as a part of a contraction period. We difficulty predicting this particular
find that 90 per cent of all NBER contraction contraction period.
periods falls within the model recession To summarize, we have proposed a
phase, whereas 10 per cent falls in either framework that can be used to translate
slowdown or recovery. Thus, NBER economic variables into a business cycle
contraction periods coincide strongly with indicator, which can take on four different
our economic indicator being negative and states. With the four variables proposed in
falling. All major contraction mid-points are this article, the business cycle indicator is
predicted correctly, although the exact found to match reasonably well with the
slowdown (start) and trough (end) are official business cycle, as reported by the
sometimes classified differently. Interestingly, NBER.

& 2011 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 12, 5, 360–375 375