You are on page 1of 5

Applied Economics Letters, 2001, 8, 659± 663

The impact of promotion/advertising


expenditures on citrus sales
J O S EÂ A . P A G AÂ N { , S U K H J I T S E T H I { and G OÈ K CË E A . S O Y D E M I R {
{ Department of Economics and Finance, College of Business Administration and
{ Data and Information Systems Center, University of Texas-Pan American,
Edinburgh, TX 78539, USA

This study analyses the impact of advertising expenditures on citrus sales from the
Texas Rio Grande Valley. A bivariate vector autoregressive model is estimated using
weekly data on the dollar value of advertising expenditures and carton equivalent
shipments for the 1993± 1999 growing seasons. The estimated impulse response func-
tions show that a one-time increase in advertising expenditures leads to increases in
orange sales with a one-month lag. However, the impact of advertising on grapefruit
sales is more immediate and relatively large. Carton shipments remain high for about
three weeks after a one-time advertising shock. There is also no evidence of causality
from sales to advertising. The results suggest that Federal Marketing Order regula-
tions that facilitate funds for the promotion and advertising of citrus are eŒective in
increasing the domestic consumption of oranges and grapefruit.

I. INTRODUCTION This is the ® rst study that assesses the eŒectiveness of the
advertising/promotion of citrus under USDA rules. As
Grapefruit and orange production and consumption have such, the paper provides some basic guidelines as to how
increased substantially in the USA since the beginning of these studies could be conducted for other agricultural
the 1980s. Although Florida citrus accounts for most of the commodities that fall under USDA regulation.
US orange and grapefruit production, Texas grapefruit has Using weekly data from South Texas on citrus shipments
increased its market share from 3.40% of average yearly and advertising expenditures, a vector autoregressive
total US production in the early 1980s to 14.04% of aver- (VAR) model is estimated to analyse whether advertising
age yearly total US production by the 1997± 98 season. expenditures are eŒective in promoting citrus sales.
Moreover, since Texas citrus trees are relatively young, Because advertising may in¯ uence consumer preferences,
production levels ± according to the US Department of ® rms may advertise to promote sales. Although the direc-
Agriculture (USDA) calculations ± are expected to sub- tion of causality is usually assumed to run from advertising
stantially increase over the next few years (TVCC, 1998). to sales, there may also be reverse causality from sales to
Texas Citrus is regulated by the Federal Marketing advertising if advertising expenditures are based on a cer-
Order #906, which is administered by the Texas Valley tain percentage of sales. According to some studies the
Citrus Committee. The Committee oversees the handling causal chain between advertising and sales remains ambig-
of fresh oranges and grapefruit going to the United States, uous (Ramos, 1998). Thus, it is important to study the
Canada and Mexico (TVCC, 1999). Over most of the interdependence of these variables to better understand
1990s, marketing and promotional programmes have the link between sales and advertising as this may provide
been developed with the goal of increasing citrus sales. producers with more e cient pro® t-maximizing strategies.
These programmes are partially funded by the USDA The empirical ® ndings from unit root and cointegration
and, as such, government regulations require that their tests, VAR estimation, and dynamic simulations indicate
eŒectiveness in impacting sales be analysed (DISC, 1999). that there is a lagged impact of advertising on orange sales,
Applied Economics Letters ISSN 1350± 4851 print/ISSN 1466± 4291 online # 2001 Taylor & Francis Ltd 659
http://www.tandf.co.uk/journals
DOI: 10.1080/13504850010029499
660 J. A. PagaÂn et al.
but a rather immediate impact of advertising on grapefruit advertising/promotional campaigns has important public
sales. Moreover, the magnitude of the response of grape- policy and resource allocation implications if these cam-
fruit sales to advertising is greater than that of orange sales. paigns prove to be successful (DISC, 1999).
The response of grapefruit sales remains statistically signif-
icant for a longer time period than that of orange sales.
Tests involving the existence of causality from sales to III. METHODOLOGY AND DATA
advertising do not produce results in favour of such a
link and, thus, there seem to be no evidence of feedback To analyse the link between advertising and sales, a vector
eŒects. autoregressive (VAR) model is employed (Sims, 1980;
Litterman, 1986; Runkle, 1987; Doan, 1988). The main
advantage of this econometric model is that there are no
speci® ed exogenous variables and no restrictions imposed
I I . C O N C E P TU A L B A C K G R O U N D
in the system. After formulating a general VAR model,
which consists in regressing each current (non-lagged) vari-
Promotion programmes such as advertising are eŒective
able in the model on all the variables in the system lagged a
when they lead to increased demand. In the case of agri-
certain number of times, one can carry out causality analy-
cultural commodities, the US Department of Agriculture
sis. For example, if one simply wants to know whether an
provides funding for the promotion and advertising of
increase in advertising leads to increasing sales then a
these goods such as oranges and grapefruit. The basic
Granger causality test would be appropriate. The test can
idea is that increasing the sales of targeted agricultural
be carried out by ® rst estimating a VAR model:
commodities leads to su ciently improved net returns for
k k
those undertaking this collective eŒort (DISC, 1999).
To accurately determine whether net returns are su - yt ˆ a0 ‡ a1 D1t ‡ ¬j yt¡j ‡ ­ j xt¡j ‡ et …1†
jˆ1 jˆ1
ciently high it is important to use appropriate statistical
methods. Many statistical papers have examined the causal where y represents citrus sales and x represents advertising
links between sales and advertising expenditures. Ashley et expenditures ± both series stationary (see Dickey and
al. (1980) and Heyse and Wei (1985) use multiple time Fuller, 1979, 1981) and D1 t is a dummy variable taking
series techniques to examine the relationship between con- on the value of one when advertising expenditures are
sumption (sales) and advertising and present evidence in zero. If ­ 1 ˆ ­ 2 ˆ ­ 3 ˆ ¢ ¢ ¢ ˆ ­ k ˆ 0, then x does not
favour of consumption leading to advertising rather than Granger cause y (Franses, 1998).
the other way around. Another advantage of a VAR system (orthogonalized) is
Zanias (1994) analyses the long-run linkage between that it allows the researcher to perform a dynamic policy
sales and advertising using Granger causality and cointe- simulation and integrate Monte Carlo methods to obtain
gration tests and ® nds that causality runs in both directions con® dence bands around the point estimates (Genberg
and, thus, that there are feedback eŒects. Supposedly, et al., 1987; Doan, 1988). More speci® cally, dynamic simu-
reverse causality from sales to advertising occurs due to lations in the form of impulse response functions allows us
the fact that in some industries advertising budgets are to analyse the likely response of y at times
set as a percentage of sales. t; t ‡ 1; t ‡ 2; . . . ; to a unitary shock in x at time t. That
Ramos (1998) analyses the relationship between sales, is, what would happen to xt over time if yt changes by one
advertising and prices for the Portuguese car market unit at time t? These response patterns are known as
using a vector error correction model. He ® nds a strong impulse response functions since they represent the beha-
one-way causality from advertising to sales complemented vior of the modeled series in response to one-time shocks.
by a two-way causality between sales and prices. If the con® dence bands ± estimated by Monte Carlo simu-
This present study diŒers from previous research in that lations ± carry the same sign as the mean response then the
it focuses on the citrus sector and uses weekly data to ana- response is statistically signi® cant leading to the rejection
lyse the postulated relationship between sales and advertis- of the null hypothesis of no impact. 1
ing. Further, the study employs a bivariate VAR model to The VAR model can also be linked to cointegration
derive the impulse response functions with con® dence analysis. The main idea behind cointegration analysis is
bands for correct statistical inference and variance decom- that there may be a mechanism causing xt and yt to
positions (Franses, 1998). Grapefruit and orange advertis- move around a long-run equilibrium (Engle and Granger,
ing is also particularly unique given that funding for citrus 1991). This equilibrium may act like an attractor such that
promotion comes from the federal government and grower if one of the series drifts away from the other then there
check-oŒfees and, as such, analysing the success/failure of will be a tendency to get back near it. At any particular

1
See PagaÂn and Soydemir (2000) and Soydemir (2000) for related applications.
Impact of promotion/advertising expenditures on citrus sales 661
time, external shocks may separate the series temporarily, the series are not integrated of order one, cointegration is
but there will be an overall tendency towards some type of not an issue in the estimation process.
long-run equilibrium. If variables are found to have this Figure 1 reports the response of orange shipments to a
tendency they are said to be cointegrated and a long-run one standard deviation increase in advertising expendi-
relationship between these series is established. tures, which roughly amounts to an increase of US$1878.
In all, these three econometric tools have expanded The immediate impact of the one-time advertising shock is
rapidly into economics and business, and have been widely to increase orange shipments by 2500 cartons in the ® rst
applied in studies analysing intertemporal linkages between period and, after a lag of one period, reach a maximum of
diŒerent time series. As such, the techniques are also useful 5000 carton shipments and they gradually decline in the
to gain a better understanding of the connection between following periods. The impact of a change in advertising
advertising expenditures and citrus sales. expenditures at time t has a maximum eŒect on orange
shipments at time t1 1, and it is statistically signi® cant
since the upper and lower bounds carry the same sign as
IV. ESTIMATION RESULTS
the mean response (Soydemir, 2000).
A one-time increase in advertising expenditures of the
To conduct the empirical analysis discussed above, weekly
same magnitude has a substantially larger and more pro-
data on advertising expenditures and carton equivalent
nounced impact on grapefruit sales. Based on the mean
shipments were obtained from the Texas Valley Citrus
response function, the immediate impact of a US$1878
Committee Annual Reports for the 1993± 1999 seasons.
increase in advertising expenditures is to increase grapefruit
Weekly shipments of citrus are subdivided into grapefruit
sales by roughly 10 000 cartons. The advertising-induced
(Ruby-Sweet and Rio Star varieties) and oranges (Early,
response peaks after a one period lead at approximately
Navel and Valencia varieties). Advertising data includes
15 000 units. These responses are also statistically signi® -
only actual disbursements and does not separate orange
cant as the upper and lower bands are both positive.
and grapefruit promotion. 2
The response of grapefruit shipments, reported in Fig. 2,
Before estimating the VAR model discussed above, it is
is also more persistent than the response of orange ship-
important to check the time series properties of the data as
ments to the same increase in advertising expenditures since
estimations involving non-stationary variables may pro-
the response of the former remains statistically signi® cant
duce spurious results. The most appropriate number of
until the ® fth horizon. This ® nding is consistent with the
lags to include is determined to be eight based on the lowest
view that advertising is more eŒective in the promotion of
Akaike and Schwarz’s information criterion. The residuals
grapefruit sales than orange sales. The ® nding is also a
were then examined to check for any remaining correla-
re¯ ection that most advertising and promotion activities
tions and they were found to be white noise.
concentrate on grapefruit and not orange consumption.
Table 1 reports the results of the Augmented Dickey±
Table 2 reports the variance decompositions for adver-
Fuller (ADF) with eight lags and a constant but no trend
tising expenditures and orange shipments. Variance decom-
included in the test equation. Tests involving the restricted
positions refer to a breakdown of the change in the value of
versus the unrestricted speci® cation ± including the pres-
ence of a trend ± produced results consistent with the series
being integrated of order zero. The results indicate that
both total orange sales and grapefruit sales series were
stationary at the 1% critical level whereas advertisement
expenditure series were stationary at the 5% critical level.
These results suggest that a bivariate VAR model in levels
would be the correct model speci® cation. Moreover, since

Table 1. Unit root tests in levels with intercept included in the test
equation

ADF-test Lags

Advertising expenditures 73.22* 8


Total orange sales 74.03** 8
Grapefruit sales 74.37** 8

** Signi® cant at 1% critical level


* Signi® cant at 5% critical level Fig. 1. Response of total orange sales to advertising expenditures

2
During each growing season, advertising/promotion funds are allocated ± but not necessarily spent ± during the allocated period.
662 J. A. PagaÂn et al.
Table 3. Proportion of forecast error variance k periods ahead
produced by each innovation: advertising and grapefruit shipments

Panel A: Variance decomposition of advertising expenditures


Innovation in Innovation in
ADV explaining forecast error GRPFRT explaining forecast
Period variance of ADV error variance of ADV

1 100.00 0.00
2 99.13 0.87
3 99.23 0.76
4 99.31 0.69
5 99.34 0.66
Panel B: Variance decomposition of grapefruit shipments
Innovation in Innovation in
ADV explaining forecast error GRPFRT explaining forecast
Period variance of GRPFRT error variance of GRPFRT
Fig. 2. Response of grapefruit sales to advertising expenditures
1 4.42 95.57
2 10.14 89.85
3 11.10 88.89
Table 2. Proportion of forecast error variance k periods ahead 4 11.66 88.33
produced by each innovation: advertising and orange shipments 5 11.91 88.08
Panel A: Variance decomposition of advertising expenditures
Innovation in Innovation in
ADV explaining forecast error SHIP explaining forecast error further supported by the fact that advertising expenditures
Period variance of ADV variance of ADV are not based on a certain percentage of citrus sales in the
1 100.00 0.00 region, but due to USDA Marketing Order Regulations,
2 99.99 0.00 which are determined every season.
3 99.73 0.26 Table 3 reports the variance decompositions for adver-
4 99.31 0.68
5 98.77 1.22
tising expenditures and grapefruit shipments. In panel A,
most of the variation in advertising expenditures is
Panel B: Variance decomposition of orange shipments
accounted by past advertising expenditures. Thus, and
Innovation in Innovation in
ADV explaining forecast error SHIP explaining forecast error
similarly to panel A in Table 2, advertising expenditures
Period variance of SHIP variance of SHIP are not in¯ uenced by sales.
Panel B reports the variation in grapefruit carton ship-
1 1.49 98.50 ments accounted by its past shipments and advertising
2 5.22 94.77
3 5.04 94.95 expenditures. The results indicate that approximately
4 4.81 95.18 11% of the variation in grapefruit shipments are accounted
5 4.56 95.43 for by advertising expenditures. Thus, the variance decom-
position ® ndings are also consistent with the view that ±
relative to orange shipments ± grapefruit sales are more
a variable in a given period attributable to its own innova- responsive to advertising expenditures.
tions as well as innovations in the other variables in the
system. In panel A, most of the variation in advertising V. CONCLUDING REMARKS
expenditures is accounted for by past expenditures while
only a negligible percentage is accounted for by past orange This paper assesses the causal relationship between adver-
shipments. The results in panel B indicate that about 5% of tising expenditures and citrus sales using a bivariate VAR
the variation in orange sales can be accounted for by past model. Unit root tests indicate that the series are stationary
advertisement while 95% is attributable to past orange in levels. The results from estimating the VAR model indi-
shipments. The most pronounced eŒect occurs in the sec- cate that in general grapefruit shipments are more respon-
ond period, and the eŒect gradually subsides in the follow- sive to advertising expenditures than orange shipments.
ing periods. Both orange and grapefruit shipments impulse responses
These results point to the existence of unidirectional are statistically signi® cant. The results obtained from the
causality from advertising to sales. This was formally tested variance decompositions indicate that there is no reverse
by conducting Granger-causality tests, which show that causality in the form of sales to advertising.
advertising Granger-causes citrus sales but sales do not In all, the results are consistent with the view that pro-
Granger-cause advertising expenditures. The results are motion and advertising expenditures do have a substantial
Impact of promotion/advertising expenditures on citrus sales 663
impact on orange and grapefruit sales and, thus, that they gate ¯ uctuations in the open economy: Switzerland, 1964± 81,
are an essential part of a successful pro® t-maximizing busi- Journal of Monetary Economics, 19(1), 45± 67.
ness strategy. In particular, the results point out that grape- Hamilton, J. D. (1994) Time Series Analysis. Princeton University
fruit promotion programmes are relatively more eŒective Press, Princeton, NJ.
and, as such, the success of marketing campaigns depends Heyse, J. F. and Wei, W. W. (1985) Modelling the advertising±
on grapefruit rather than on orange promotion. sales relationship through use of multiple time series tech-
niques, Journal of Forecasting, 4, 165± 81.
Litterman, R. B. (1986) Forecasting with Bayesian vector auto-
regressions ± ® ve years of experience, Journal of Business and
R E F E R E N C ES Economic Statistics, 4, 25± 38.
PagaÂn, J. and Soydemir, G. (2000) On the linkages between equity
Ashley, R., Granger, C. W. J. and Schmalensee, R. (1980)
markets in Latin America, Applied Economics Letters, 7,
Advertising and aggregate consumption: an analysis of caus-
207± 210.
ality, Econometrica, 48, 1149± 67.
Ramos, F. R. (1998) Causality among sales, advertising and
Dickey, D. A. and Fuller, W. A. (1979) Distributions of the esti-
prices: new evidence from a multivariate cointegrated system.
mators for autoregressive time series with a unit root, Journal
Unpublished manuscript. Faculty of Economics, University
of the American Statistical Association, 74, 427± 31.
Dickey, D. A. and Fuller, W. A. (1981) Likelihood ratio statistics of Porto, Porto, Portugal.
for autoregressive time series with a unit root, Econometrica, Runkle, D. E. (1987) Vector autoregressions and reality, Journal
49, 1057± 72. of Business and Economics Statistics, October, 437± 42.
DISC (1999) Texas Valley Citrus Promotion Program Evaluation. Sims, C. A. (1980) Macroeconomics and reality, Econometrica,
Final Report. Data and Information Systems Center, 48, 1± 49.
University of Texas-Pan American. Edinburg, Texas. Soydemir, G. (2000) International transmission mechanism of
Doan, T. (1988) RATS User’s Manual. VAR Econometrics. stock market movements: evidence from emerging equity
Evanston, IL. markets, Journal of Forecasting, 19, 149± 76.
Engle R. F. and Granger, C. W. J. (1991) Long-Run Economic TVCC (1998) 1997-98 Annual Report. Texas Valley Citrus
Relationships. Oxford University Press, Oxford. Committee. Mission, Texas.
Franses, P. H. (1998) Time Series Models for Business and TVCC (1999) 1998-99 Annual Report. Texas Valley Citrus
Economic Forecasting. Cambridge University Press, Committee. Mission, Texas.
Cambridge. Zanias, G. P. (1994) The long run causality, and forecasting in the
Genberg, H., Salemi, M. K. and Swoboda, A. (1987) The relative advertising-sales relationship, Journal of Forecasting, 13,
importance of foreign and domestic disturbances for aggre- 601± 10.

You might also like