Crisis and Consumption Smoothing

Crisis and Consumption Smoothing


May, 2009



Abstract

The dramatic impact of the current crisis on performance of businesses across sectors and
economies have been headlining the business press for the past many months. Interestingly, the
impact of the crisis across categories of goods/services and across economies reveals several
interesting patterns. Extant reconciliations of these patterns in the popular press rely on ad-hoc
reasoning. Using historical data on currency crisis episodes across the world, this paper shows
that the observed patterns in the impact of a crisis across classes of goods and economies are a
consequence of active smoothing of consumption expenditures by consumers.

The results reveal that consumer behavior in a crisis is characterized by consumption smoothing
at various levels – inter-temporal, inter category and intra-category. In sum, these behavioral
adjustments result in significant reallocation of consumption expenditures. More importantly, the
smoothing decisions due to a crisis are distinct and independent of the impact of changes in
income and prices that accompany a crisis. Interestingly, there is marked variation in the patterns
of consumption smoothing across different types of economies. Taken together, these results have
important and interesting implications for managers, policy makers and academics.




JEL Classification: F31; M31; M38; F34

Key Words: Currency Crisis, Consumption Smoothing, Consumer Expenditure, Category Shares.

---

2
1 Introduction
What started as a problem in the mortgage industry in the United States has morphed into a crisis that
has engulfed the globe. An examination of recent newspaper headlines shows that the impact of the crisis
on businesses exhibits several interesting patterns. For instance, the roll call of rms reporting severe
contractions in their business is as long as it is illustrious (American Express, Cargill, De Beers, General
Electric, Maersk, Nokia, Procter & Gamble, Singapore Airlines, Toyota, etc.). Yet, there are several others
reporting mild or no impact of the crisis on their business (Bharati, China Mobile, McDonalds, Syngenta,
Walmart, etc.). Similarly, the impact of the crisis on a rm's business seems to vary markedly across the
world. For instance, Nokia reported sales contractions that ranged from -17% in N. America, -50% in Latin
America, -30% in EMEA (Europe, Middle East and Africa)to -6% in Asia-Paci c ex-China (J. P. Morgan
Equity Research, 17 April, 2009). In contrast, McDonald's reported 6% growth in USA, 11% growth in
APMEA (Asia-Paci c, Middle East and Africa) and 1% growth in Europe (J. P. Morgan Equity Research,
22 April, 2009).
There are several rationales advanced in the popular press for the observed patterns on the impact of
the crisis on a rm. The variation in impact of a crisis across different categories is commonly attributed
to the role of the category – necessity versus discretionary. Items such as food are considered necessities
and hence less severely impacted relative to discretionary items such as durables (automobiles) or luxuries.
Yet, there is no formal classi cation of what is a necessity versus discretionary and how this applies in the
context of a crisis. The reconciliation of variation in impact across geography is also ad-hoc and sometimes
contradictory. Here it is argued that the patterns could be a consequence of growth and development, (i.e.,
lower impact of a crisis on emerging markets such as China, India - the decoupling hypothesis), dependence
of the economy on exports relative to domestic demand (i.e., greater impact on China, Singapore, due to
their high dependence on exports relative to domestic consumption), extent of debt (i.e., more indebted
implying greater impact in Eastern European countries) and so on. In short, there is an abundance of ad-
hoc explanations for the observed patterns of impact of the crisis but a serious lack of careful and systematic

---

3
analysis.
This research analyzes crisis episodes over the past decades and establishes a series of stylized facts
about the impact of a crisis. In particular, it shows that the impact of a crisis and its variance across different
categories and classes of economies is a consequence of active smoothing of consumption expenditures by
consumers in response to a crisis. In other words, it is the consumer's decisions on how much to spend and
the allocation of this expenditures across their consumption basket following a crisis that underpins the
observed patterns of impact of a crisis on a rm's business. The key message of the paper is that decision
makers seeking to obtain a proper assessment of the impact of the crisis on a their business need to start
by understanding the impact of a crisis on their consumer's behaviors. This is also consistent with the
recent editorial observations of Bradlow (2009) that the crisis represents a real opportunity for marketing
academia to contribute to a better understanding of the impact of a crisis.
As a rst step, it is useful to de ne what is meant by a crisis. Economic crisis may be broadly classi ed
as banking crisis and/or currency crisis. In the former, the country's nancial sector experiences a large
number of defaults, bank runs increase in frequency, and we observe massive liquidity support for banks
and bailout packages. A currency crisis, on the other hand, entails a sharp correction in the exchange rate
of a country. Currency crises are easier to measure and identify since data on exchange rates are widely
available. Banking crises, on the other hand, require much more detailed data on bank runs, bank closures,
and bailouts. Assembling data for a comprehensive set of countries over time is non-trivial, a task which
is further complicated by the fact that identifying a banking crisis involves subjective assessments. In this
paper, therefore we focus on currency crisis.1
Consumer's response to a crisis could manifest itself in several different ways.
At an aggregate level, consumers could respond to a crisis by changing how much they spend (i.e.,
altering the size of their wallet). We use aggregate indicators of consumption such as per capita
consumption expenditure and per capita retail sales to show precisely how consumers smooth their
aggregate consumption expenditures over time in the event of a crisis.
1 Reinhart and Rogoff (2009) nd that 26% of all banking crisis are also currency crisis - termed twin crisis.

---

4
Consumers can also respond to a crisis by altering the composition of their consumption expenditures
(i.e., altering the share of wallet allocated to different classes of goods and services). We show using
a four fold classi cation of consumer expenditures (spending on durable goods, non-durable goods,
semi-durable goods and services) that consumer behavior in a crisis is characterized by rich patterns
of consumption smoothing across the four different categories of consumption.
Finally, consumers could also engage in intra-category consumption smoothing by reallocating ex-
penditures across the different sub-categories within each of the four different categories. We show
using disaggregate sub-category expenditures that consumers engage in signi cant consumption
smoothing within each of the four classes of goods and services. In summary, by examining con-
sumption smoothing at various levels of analysis - inter-temporal, inter-category and intra-category
- we highlight the rich patterns of consumption smoothing that characterize consumer behavior in a
crisis.
Interestingly, there is marked variation in the patterns of consumption smoothing across different
types of economies. Taken together, these results establish a series of stylized facts about the im-
pact of a crisis which have important and interesting implications for managers, policy makers and
academics.
The rest of the paper is organized as follows. Section 2 provides a brief review of the related economics
and marketing literature. Section 3 provides a formal de nition of a crisis. Section 4 details the impact
of a crisis on aggregate consumption indicators. Section 5 uses a four fold classi cation of consumption
to illuminate the patterns of consumption smoothing across categories. Section 6 analyzes intra-category
consumption smoothing. Section 7 highlights the implications of these ndings and concludes with some
suggestions for future research.

---

5
2 Related Literature
The empirical work on currency crisis can be broadly classi ed into three categories. First, are papers that
simply establish a set of stylized facts by examining how various macroeconomic variables behaved prior
to and immediately following a crisis (Hutchison and Noy, 2005; Bordo et al 2001; Eichengreen et al,
1996. Second, some papers estimate the probability of a currency crisis (in terms of a large devaluation)
as a function of various macroeconomic indicators (Frankel and Rose, 1996; Sachs, Tornell and Velasco,
1996). A third set of papers use a non-parametric approach to evaluate the usefulness of several variables
in signaling an impending crisis (Kaminsky et al, 1998). The focus of this empirical literature has been on
broad macroeconomic aggregates such as GDP, current account, and GDP growth. Surprisingly, it is silent
on the impact of a crisis on consumption expenditure and more disaggregate variables such as retail and
category sales. The lack of research on the impact of a crisis on consumption expenditure is puzzling since
consumption expenditure is by far the biggest component of GDP and the fall in consumption expenditure
has direct welfare implications. Theoretically, it may be argued that a crisis is a temporary event, and
has no impact on consumption expenditures. This follows from the permanent income hypothesis theory
of consumption (Friedman, 1957) according to which forward looking consumers base their consumption
decisions on the expected discounted value of lifetime resources or their permanent income, and not on
current income. Only changes to permanent income triggers changes in consumption. However, this is
inconsistent with our data which shows marked shifts in consumer expenditures following a crisis.
Within economics, researchers have also documented the impact of a particular crisis on consumption
patterns within a country. Frankenberg et al. (1999) for instance nd substantial reallocation of expenditure
towards food staples such as rice during the Indonesian crisis. McKenzie (2006) uses Mexican household
income and expenditure surveys to investigate changes in consumption for the Mexico peso crisis of 1994.
Kang and Sawada (2008) show that Korean households coped with negative shocks of the 1997 Asian
crisis by reducing consumption of luxury items. However, all these are country case studies focusing on
one country and a single crisis episode. Our paper complements this literature by highlighting consistent

---

6
patterns in the impact of a crisis on consumption behavior of consumers for a large sample of countries over
multiple crisis episodes. Further, it analyzes aggregate consumption expenditure as well as category-level
consumption expenditures.
Researchers in marketing have studied the marketing implications of a crisis. For instance, Rubel, Naik
and Srinivasan (2007), Van Heerde, Helsen and Dekimpe (2007) and Klein and Dawar (2004) have focused
on the implications of product-harm crisis. Pennings, Wansink and Meulenberg (2002) have focused on the
implications of the mad-cow crisis. However, we have not been able to uncover any work that focuses on
the marketing implications of a economic crisis. The closest work in spirit is the literature in marketing on
the impact of business cycles. Deleersynder et al (2004) show that consumer durables are more sensitive
to business cycle uctuations than general economic indicators. Additionally, they nd that sales fall more
quickly during contractions whereas they adjust upwards more slowly during expansions. Lamey, et al
(2007) examine the impact of business cycles on private labels (a fast moving consumer good versus the
consumer durable context of their earlier work) and report largely similar results. Deleersnyder et al (2009)
show that advertising expenditures are more sensitive to business cycle uctuations than the economy as a
whole and highlight the differences in sensitivity across different media. Further, they show that cultural
and economic considerations play an important role in moderating advertising's sensitivity to the business
cycle. Our focus in this paper is on economic crisis and not business cycles. Economic crisis impacts the
periodicity of the business cycle and in that sense there is a relation to this stream of work. Additionally, our
work is very different in terms of its focus on both developed and developing countries (amongst whom,
the latter have experienced the majority of crisis episodes).
3 Currency Crisis: De nition and Measurement
A canonical currency crisis is one where investors ee a currency (sell the local currency in exchange
for safer currencies such as US dollars or Euros) because they expect it to be devalued, and much of the
pressure on the currency comes precisely because of this lack of con dence. This sort of circular logic is
the de ning feature of a currency crisis. While this is the broad and general feature of a currency crisis,

---

7
it can manifest itself in various ways - a sharp depreciation of the exchange rate, a depletion of foreign
exchange reserves, an increase in interest rates to shore up the currency, etc.
For purposes of this paper, we draw on the commonly accepted de nition of currency crisis (Frankel
and Rose, 1996; Hong and Tornell, 2005) - a country is said to experience a currency crisis if there is:2
at least a 20% nominal depreciation of its currency, and,3
there is also at least a 10% increase in the rate of depreciation of the exchange rate.
Currency depreciation refers to change in the natural logarithm of the nominal bilateral dollar exchange
rate (multiplied by 100) and all changes are expressed in annual rates of change. Since many countries,
such as Argentina in the 1980s experienced changes in the exchange rate of 20 per cent or more – year
after year, we also require that the change in the exchange rate, not only exceed 20 per cent, but exceed
the previous year's change in the exchange rate by a margin of at least 10 per cent. This is a conservative
speci cation and avoids counting the same crisis event multiple times.
We use data on exchange rates from International Financial Statistics to construct a dummy variable
which takes the value one in the year of the crisis. Our data spans 99 countries over the period 1960-2003.
Using the above de nition of currency crisis, we obtain 273 episodes of currency crisis which is about
6.4% of the entire sample (see Figure 1). More importantly, 87 of these 99 countries have experienced at
least one crisis testifying to the ubiquitous nature of currency crisis.4
2 Eichengreen et. al. (1996) de ne a currency crisis to include both the large depreciations that we consider here, and
also speculative attacks that are successfully warded off by the authorities. Unfortunately, unsuccessful speculative attack are
dif cult to identify even ex-post and they use sharp falls in foreign exchange reserves and/or increases in interest rates to do
this. Moreover, the majority of currency crises have historically been in developing countries where sparse data on interest rates
and foreign exchange reserves makes it dif cult to identify successful defense against speculative attacks using these variables.
Finally, reserve movements are notoriously noisy measures of exchange market intervention for almost all countries.
3 Frankel and Rose (1996) use a 25% nominal depreciation of currency to identify a crisis. However, some of the ERM crisis
countries such as Italy, Finland and Spain exhibit nominal depreciation of greater than 20% but less than 25%. Our cutoff of 20%
allows us to include these countries in our sample of countries experiencing a crisis. We also experimented with cutoffs of 15, 25
and 30%. The results remain qualitatively unaffected.
4 Table-A in the Appendix lists the variables, summary statistics and the data sources used.

---

8
4 Inter-temporal consumption smoothing
As mentioned in the introduction, one of the ways consumers could respond to a crisis is by changing
how much they spend (i.e., changing the size of the wallet). We focus in this section on analyzing the
impact of a crisis (both the magnitude and the duration of impact) on aggregate consumer expenditure
and document patterns of inter-temporal consumption smoothing. We use two methodologies to do this:
(1) the Arellano-Bond GMM estimator and (2) the Pesaran-Smith Mean-Group and Pooled-Mean-Group
Estimator.
4.1 Arellano-Bond GMM estimation
We follow Davidson et al. (1978) and Haque and Montiel (1989) in our speci cation of aggregate con-
sumption function, where we estimate per capita consumption expenditure as a function of lagged per
capita consumption, per capita income, and the in ation rate. The permanent income hypothesis implies
that consumption should evolve over time as a martingale (Hall, 1978). While the literature rejects this con-
tention (see Campbell and Mankiw 1990), it is plausible that per capita consumption expenditure depends
on its past values.5 Therefore, we estimate a consumption function where we include lagged values of
per capita consumption as regressors. GMM estimates suggest that only one lag of the dependent variable
should be included. Therefore, we estimate the following dynamic speci cation:
Cit = Cit 1 + yit +
it+ 0crisisit + ::: +
kcrisisit k + ( i +
t + it) ; i = 1; ::; N ; t = 2; ::; T
(1)
where Cit is per capita consumption in country i at time t; Cit 1 is lagged per capita consumption in
country i ; crisisit k is the crisis dummy at lag k;
is a time invariant country-speci c effect;
i
t is a
common trend component, y is per capita income and
denotes in ation, and it is the error term.6 The
country dummies should control for unobserved and time-invariant country-speci c effects while the time
5 Theories of habit-persistence in consumption for example, can also generate such lagged dependence. See Fuhrer (2000).
6 Measurement error in per capita consumption is a serious concern with the presence of the lagged dependent variables on
the right hand side. However, if this error is driven by country-speci c speci c characteristics and vary little over time, they will
be subsumed within the country speci c effect
:
i

---

9
trend should capture global trends in per capita consumption. All crisis lags that were signi cant and
resulted in improvement of model t are included - in general no lags higher than three years (k = 3) turn
out to be signi cant.
It is widely recognized that per capita GDP contracts following a crisis (Bordo et al, 2001). However,
it is unclear whether the impact of a crisis on per capita consumption expenditure exceeds the impact on
per capita GDP or whether it falls short. If we think of a crisis as a transitory shock and believe in the
permanent income hypothesis then consumption smoothing would dictate a rising share of consumption in
GDP in the year of the crisis. On the other hand, if a crisis leading to a rise in uncertainty, consumers may
react by reducing consumption by more than income and increasing savings as a fraction of income. This
behavior is best explained by a precautionary motive for saving. A necessary (but not suf cient) condition
for consumption to fall by more income in the year of the crisis is that we obtain a negative and signi cant
coef cient on the contemporaneous crisis dummy.
0 > 0 would indicate inter-temporal consumption
smoothing with consumers drawing down on savings in the crisis period. Similarly, 1; 2; 3 > 0 would
capture persistence of consumption-smoothing one, two and three years after the crisis. However, if these
0s are negative this would indicate that a crisis potentially leads to a decline in consumption over and
above that dictated by income.7 Finally, the in ation variable is a proxy for various wealth effects, and
we would expect its coef cient to be negative (Hendry, 1974).
In the presence of lagged dependent variables as regressors, the xed-effects estimator is consistent
only in panels where T is large - the transformed lagged dependent variables are correlated with the trans-
formed error term but this correlation goes to zero as T gets large. While T
30 for all countries when
we use per capita consumption, this correlation should vanish and the xed-effects estimator is likely to
be consistent. However, we choose to be conservative and employ the generalized method of moments
(GMM) procedure developed by Arellano and Bond (1991) to generate consistent estimates of the para-
meters of interest and their asymptotic variance-covariance. Estimation proceeds by rst differencing the
7 We could obtain the same result by using consumption as a share of income as the dependent variable. Note that this is
equivalent to estimating (1) while constraining the coef cient on income,
= 1. However, (1) is a more general speci cation,
and in fact we can test whether
is indeed equal to one.

---

10
data - this eliminates the country-speci c effects
from the model - and instrumenting the lagged depen-
i
dent variable by appropriately lagged levels of Cit. The instruments are based on the following moment
conditions:
E [Cit s vit] = 0 for s
2; t = 3; 4; :::T
Another advantage of the Arellano-Bond technique is that it allows us to treat income and in ation as
endogenous variables. We adopt a more conservative speci cation and assume that the in ation rate and
income are endogenous in the sense that yit and it are correlated with vit and earlier shocks but uncorre-
lated with vit+1and subsequent shocks. Thus lagged values of yit and it, lagged two periods or more, are
available as instruments.8 Finally, to control for heteroskedasticity we report results using the two-step
GMM estimator and employ a nite-sample correction to the two-step covariance matrix.
To estimate (1), we use aggregate country-level data on per capita consumption expenditures from the
World Development Indicators (World Bank) over the period 1960-2003. Data on per capita income are
from Penn World Tables. For all countries, per capital consumption and income are converted to constant
2000 US dollars on a purchasing power parity (PPP) basis.9
Finally, in ation is measured using the
consumer price index, with the data derived from the World Development Indicators.
4.1.1 Results
We divide our sample into OECD and non-OECD countries to examine if there is variance in the magni-
tude and length of impact of the crisis across developed and developing countries.10 Column 1 in Table
1 presents the Arellano-Bond GMM estimates for the OECD sample, where we regress per capita con-
sumption expenditure on lagged consumption, the contemporaneous crisis dummy, as well as three lags
8 Arellano (2003b) shows that when T is large it leads to an over tting bias caused by instrument proliferation in dynamic
panels. To mitigate this concern, we use only three lags as instruments in the moment conditions.
9 Purchasing power parity conversion factor is the number of units of a country's currency required to buy the same amount of
goods and services in the domestic market as a U.S. dollar would buy in the United States. It assumes a constant real exchange
rate, which facilitates comparisons across countries and over time.
10 Originally the OECD members consisted of only rich countries. However, over time, it added middle income and de-
veloping countries such as Mexico, Poland, Korea. Since we are interested in distinguishing the crisis impact along the de-
veloped/developing country dimension, we restrict the OECD sample to the 24 countries who joined the OECD prior to 1973.
Mexico, Korea, Poland, Hungary, Slovakia and Czech Republic, all of whom joined the OECD in the 1990s are classi ed as
non-OECD.

---