proceed, we consider the time series properties. In all cases, the price indices are nonstationary
of first order while the sales indicator series are stationary. Based on these results, we examine
cointegration and estimate an error correction model (ECM) specification to test the impact of
the number of sales. Alternatively, we could use the Johansen (1995) procedure. However, as
we are primarily interested in the impact of the number of sales, a stationary series, the
associated parameters can be consistently estimated and tested in the ECM model, stated as
follows:
dpR = α+γPRτI + αPWP + ∑ γm+ dp- ɪ + αdpWtl + ∑ θSM+ψSF ■ V+∑ βD1t + ε
t01t-1 1t-1 m+2 t-m-1 m+2t-mkt-kkt-kjtt
This specification is estimated for each store type aggregate series. Thus, average prices
and average number of sales for each store type are used. The wholesale price index is the same
across store type. The dynamic specification (number of lags) is determined by extending the lag
until the residual series is interpretable as white noise. We start with one lag and increase the
number of lags symmetrically for all variables until autocorrelation is rejected at the 95 %
significance level. Other specification tests, such as ARCH, heteroscedasticity, normality and
functional form tests led to rejection of the null hypothesis suggesting that estimated residuals
are consistent with white noise processes (see Table 3). Finally, we examined the existence of a
short-run dynamic impact of sales by employing a Wald test. R-squares of the models range
from 0.64 to 0.73. The number of sales has a significant negative impact on the per capita
expenditures in each store type. However, this contemporaneous effect of the sale on
expenditure is found to be partly or completely offset by lagged positive effects, suggesting that
following a sale, expenditures increase. In the case of meats in BSM and DC the expenditure
reduction by sales was not entirely offset; however the effect of sales was significantly reduced.
Sales in meats have the biggest impact which ranges from 3.6 for BSM to 1.69 German Mark
for DC. Most of this reduction in expenditures is set off by price increases in the weeks
following the occurrence of sales. Thus, the long-term impact ranges from 11 to 60 pennies. A
significant impact is resulted only for BSM and DC. Sales for fruits and vegetables have a far
lower impact which is not significant for all store types. Recall that the absence of correlation in
the sales series across products suggested that cross-product price strategies are not followed
(e.g. loss leader strategies where a sale product is discounted while other product prices are
increased). However, results do indicate that intertemporal, product specific strategies are used
such that price discounts are followed by price increases. <Insert Table 4 about here>
5. Conclusions
Various models have been proposed in the literature to explain the use of promotional
measures such as simple sales prices. However, most approaches do not fit to the specific
conditions in the market for fresh foods. The most promising concepts are the Varian model and
the loss leader hypotheses. For these theories, we test the essential hypotheses for a unique data
set for the German food retail market in the period from 1995 to 2000. The data consist of
weekly retail prices for ten fresh food items (meats, fruits, and vegetables) in 131 grocery stores.
We find that pricing strategies indicate a low level of coordination between stores. Sales are an
important promotional measure that indicates the competitive market structure in this market as
predicted by the Varian model. However, the static impact of sales is often offset by dynamic
price increases. Thus, customers who are attracted by store sales prices and choose the store also
in the periods following the sales suffer from increased prices.
parameters, a panel estimation would be preferable. Because of time series properties and the extended
time component (296 weeks), the panel estimation and testing is not a standard routine. Thus, we start with
an unrestricted dynamic single equation approach for each store type that considers the non-stationarity of
the data.