Intern. J.

of Research in Marketing 25 (2008) 173182

Contents lists available at ScienceDirect

Intern. J. of Research in Marketing

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / i j r e s m a r

Does competitive entry structurally change key marketing metrics?

Marcel Kornelis a,, Marnik G. Dekimpe b, Peter S.H. Leeang c
a b c

Wageningen University and Research Center, The Netherlands Tilburg University, The Netherlands and Catholic University Leuven, Belgium University of Groningen, The Netherlands

A R T I C L E

I N F O

A B S T R A C T
To what extent does competitive entry create a structural change in key marketing metrics? New players may just be a temporal nuisance to incumbents, but could also fundamentally change the latter's performance evolution, or induce them to permanently alter their spending levels and/or pricing decisions. Similarly, the addition of a new marketing channel could permanently shift shopping preferences, or could just create a short-lived migration from existing channels. The steady-state impact of a given entry or channel addition on various marketing metrics is intrinsically an empirical issue for which we need an appropriate testing procedure. In this study, we introduce a testing sequence that allows for the endogenous determination of potential change (break) locations, thereby accounting for lead and/or lagged effects of the introduction of interest. By not restricting the number of potential breaks to one (as is commonly done in the marketing literature), we quantify the impact of the new entrant(s) while controlling for other events that may have taken place in the market. We illustrate the methodology in the context of the Dutch television advertising market, which was characterized by the entry of several late movers. We nd that the steady-state growth of private incumbents' revenues was slowed by the quasi-simultaneous entry of three new players. Contrary to industry observers' expectations, such a slowdown was not experienced in the related markets of print and radio advertising. 2008 Elsevier B.V. All rights reserved.

Article history: First received in July 14, 2004 and was under review for 8 months Area editor: Koen H. Pauwels Keywords: Multiple breaks Structural-break time-series analysis Unknown timing Late-mover entry Marketing metrics

1. Introduction Many markets experience the occasional entry of new contenders or see incumbent players add a new channel to their channel portfolio. Such entries/additions may leave some marketing metrics unaffected, create a temporal disturbance in others, and fundamentally alter the long-run evolution of still other measures. In practice, it is often hard to determine a priori which metrics will be affected, when those effects will materialize, and (provided that they are affected) what the steady-state implications will be. Deleersnyder, Geyskens, Gielens, and Dekimpe (2002), for example, documented that the addition of an Internet-based version reduced the revenues for 10 out of 85 newspapers. Nijs, Dekimpe, Steenkamp, and Hanssens (2001) concluded that new-product introductions created a signicant market-expansive effect in 30% of the studied FPCG markets. In addition, Pauwels and Srinivasan (2004) showed that store brand entry caused a structural change in some, but not all, of the investigated performance series, such as brand sales and revenues, category sales, and store trafc. As these studies illustrate,

Corresponding author. Tel.: +31 70 3358194; fax: +31 70 3358196. E-mail address: marcel.kornelis@wur.nl (M. Kornelis). 0167-8116/$ see front matter 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.ijresmar.2008.01.003

whether or not the entry of a new player structurally changes certain marketing metrics remains an empirical issue, making the use of proper testing procedures of central relevance. The development of such a testing procedure faces two fundamental challenges. First, incumbents may not only face multiple intruders, they may also experience other major events that could cause a structural break in the metric of interest. In this respect, one could think of political changes with ramications for the marketing metrics under investigation. Examples are the studies by Lamey, Deleersnyder, Dekimpe, and Steenkamp (2007), who control for the impact of the German unication in their study on the evolution of German private label share; and by El Sehity, Hoelzl, and Kirchler (2005), who study the implications of the 2002 introduction of the Euro. Another example of a major event that may cause a structural break is an unexpected product harm crisis that may impact not only the affected brand but other incumbents as well (see e.g. Van Heerde, Helsen, & Dekimpe, 2007). To avoid biased inferences on the steadystate impact of the focal competitive entry, one should account for the (potentially) confounding effects of such other events. To do so, we introduce a structural-break testing procedure that accounts for all signicant breaks in the market, in contrast to the current standard in the marketing literature of allowing for at most one break (see, among others, Deleersnyder et al., 2002; Lim, Currim, & Andrews, 2005; Nijs et al., 2001; Pauwels & Srinivasan, 2004).

174

M. Kornelis et al. / Intern. J. of Research in Marketing 25 (2008) 173182

Second, one has to allow for the fact that the change in some marketing metrics may not coincide exactly with the competitive entry, but only become evident with some delay; other metrics, in turn, may already be affected prior to the actual occurrence of the event when market participants change their behavior in anticipation (see e.g. Doyle & Saunders, 1985; Pauwels & Srinivasan, 2004, p. 368). As such, added exibility is needed to allow for such lead and lagged effects, which may hamper the a priori (i.e., exogenous) imposition of break-date locations. In this study, we develop a testing procedure that accounts for multiple structural breaks at unknown locations. We apply the proposed methodology to the demand for advertising in the Netherlands. Specically, we consider whether the quasi-joint introduction of three commercial channels in 1995 permanently altered the outlook of the industry, i.e., whether this event had implications for the steady-state growth of incumbent private and public channels and/or affected related markets (print and radio advertising). We rst position our work in earlier literature on the impact of late entrants. Next, we introduce our testing framework, which allows for multiple breaks at unknown locations. We subsequently describe our data, after which we discuss the empirical results and benchmark our ndings with various competing models. We conclude with the key implications of our study and identify several areas for future research. 2. Literature review The impact of new entries on key marketing metrics has been of interest to many researchers. Relevant empirical research has covered the areas of marketing mix effectiveness, competitive behavior, and incumbents' performance. 2.1. Marketing mix effectiveness New entrants may very well create fundamental changes in the marketing mix effectiveness of incumbents (Fok & Franses, 2004). The direction of these changes depends on various factors, such as the type of instrument or brand position. For example, Chintagunta (1999) analyzed the impact of a new entrant in the liquid laundry detergent market and documented that the new entrant tends to increase the price sensitivity, but lower the promotional sensitivity of households. Pauwels and Srinivasan (2004), in turn, demonstrated that after a store brand entry, premium brands experienced lower, while secondtier brands experienced higher, steady-state price sensitivity. 2.2. Competitive behavior The arrival of a new entrant may provoke competitive reactions from the incumbents. While it is well established in (game-) theoretical studies that the equilibrium price is expected to fall with entry (Gruca, Kumar, & Sudharshan, 1992), the empirical investigations of Fok and Franses (2004) could not conrm this hypothesis. Pauwels and Srinivasan (2004), in turn, found a decrease in the equilibrium price in only two out of four categories. Other studies have investigated the extent of competition in relation to the number of incumbents. Exploiting the cross-sectional variation in the number of players in local markets, Cleeren, Dekimpe, and Verboven (2006) found, unlike the predictions of many normative models, that the level of competition increased more in a duopoly than in a monopoly, and therefore concluded that any inferences on the competitive impact of new entrants remains an empirical issue. 2.3. Incumbents' performance implications Empirical research on the performance implications for incumbents of (late) entrants has been performed by Mahajan, Sharma, and Buzzell (1993), Nijs et al. (2001) and Van Heerde, Mela, and

Manchanda (2004), among others. Mahajan et al. (1993) studied the performance of a pioneer within the instant camera market that faced a new entrant. They concluded that about 32% of the entrant's sales came from the pioneers' potential buyers, and that the entrant expanded the market's primary demand by 37%. Nijs et al. (2001), while focusing on the primary demand effects of price promotions, observed across over 500 FPCG markets that new-product introductions may actually be a more effective way to permanently expand the category. Van Heerde et al. (2004) studied the market structure dynamics that resulted from an innovative product entry into the US frozen pizza market. The innovation in this stagnant product category was found to have increased the substitutability of the existing brands and to have expanded the target market. Apart from the impact on incumbents in the focal market, competing and complementary markets may also be affected. For example, Reiss and Spiller (1989) demonstrated that, although direct and indirect ights in small airline markets were not perfect substitutes, competition in one group affected entry and exit in both groups. Berry and Waldfogel (1999a,b) analyzed the US radio broadcasting market. In their rst study, they found a large substitution effect between existing radio stations and new entrants. In the second study, evidence was found of a substitution effect between public and private stations. Deleersnyder et al. (2002) measured the impact of new online newspapers on the circulation and advertising revenues of incumbent hard-copy editions. They a priori imposed the launch dates of 85 online newspapers in the UK and the Netherlands and did not nd strong evidence of substitution effects between the Internet and print market. Our empirical illustration is situated in this third research stream, as we investigate to what extent new entrants affect key performance metrics of both incumbents and competing markets. While our approach follows the tradition of, among others, Nijs et al. (2001), Deleersnyder et al. (2002), and Pauwels and Srinivasan (2004) in that we also use structural-break time-series econometrics to investigate the steady-state implications of new market introductions, we extend their methodology in that we explicitly allow for multiple breaks at unknown locations. 3. Methodology 3.1. Testing procedure In line with recent literature in both economics (see e.g. Ben-David & Papell, 2000; Perron, 1989) and marketing (see e.g. Nijs et al., 2001; Steenkamp, Nijs, Hanssens, & Dekimpe, 2005), we dene a structural break in terms of a parameter change in the deterministic part of the model, in this case the slope and/or intercept of the deterministic growth path. This approach is consistent with the interventionanalysis approach of Box and Tiao (1975), in that unique historic events are separated from the regular noise function (Perron, 1994). We refer to Hanssens, Parsons, and Schultz (2001, p. 293296) for a review of marketing applications of intervention analyses. Conceptually, one assumes that the coefcients of the deterministic trend function are driven by long-term economic or market fundamentals which very rarely change, while regular (frequently occurring) shocks drive the stationary component of the data-generating process (Perron, 1994, pp. 114115). These regular shocks do not change the parameters of the model process (Pesaran & Samiei, 1991), while the irregular interventions are allowed to affect the parameters of the deterministic part of the model. Our proposed approach ends with the iterative procedure outlined in Ben-David and Papell (2000) where, for consecutive values of M, the null hypothesis of M breaks is tested against the alternative hypothesis of M + 1 breaks. However, since this testing procedure requires stationarity of the series under investigation, we rst apply formal unit-root tests to assess the stationary versus evolving nature

M. Kornelis et al. / Intern. J. of Research in Marketing 25 (2008) 173182

175

Fig. 1. The testing procedure.

of the series of interest. As it is well established that a failure to account for a structural break biases traditional unit-root tests towards non-stationarity (see e.g. Dekimpe & Hanssens, 2004; Perron, 1989; Zivot & Andrews, 1992), we apply unit-root tests that explicitly allow for the existence of such structural breaks in the deterministic part of the model. A graphical representation of the testing sequence is given in Fig. 1. Our testing procedure combines two research streams. The rst (depicted in the upper half of Fig. 1) tests the unit-root hypothesis against the alternative of a stationary series with structural breaks, and dates back to the pioneering work of Dickey and Fuller (1979). The tests of Zivot and Andrews (1992) and Lumsdaine and Papell (1997) belong to this research stream. In these tests, we aim to distinguish between a trend-stationary (TS) model (after allowing, perhaps, for a number of breaks) and a difference-stationary (DS) model. In the latter case, all shocks can have a persistent effect, and the magnitude of this effect can be operationalized through Campbell and Mankiw's (1987) A(1) measure (see Dekimpe & Hanssens, 1995 for a marketing application). However, recent marketing studies have questioned how realistic it is to assume that all shocks can have a permanent effect (see e.g. Deleersnyder et al., 2002 or Fok, Horvth, Paap, & Franses, 2006, p. 448). In a TS model without breaks, in contrast, none of the events that took place during the time span under investigation has a long-run effect. In between both extremes, a TS model with one or more breaks in the deterministic trend function allows a limited number of irregular or special events to have an impact on the steady-state evolution of the series, while frequently occurring, regular shocks have only a temporary effect. Once stationarity has been established, the second research stream concentrates on the assumption of parameter constancy against the alternative of a structural break and nds its origin in the seminal work of Chow (1960). The test of Ben-David and Papell (2000), depicted in the lower part of Fig. 1, belongs to this stream of research and serves to identify any remaining irregular events that may be relevant (i.e., cause a change in the deterministic part of the model). Indeed, the preceding test sequence stopped once stationarity was established, and not necessarily when all breaks had been identied. 3.2. Testing for stationarity: endogenous-break unit-root testing Two types of structural-break tests can be considered, depending on the amount of prior knowledge on the location of the break point, viz. exogenous or endogenous-break tests. In the tradition of Perron (1989), one assumes that researchers know the exact date of change, which therefore can be imposed exogenously. As a consequence, possible lead (anticipation) or lagged (delayed) effects are (most often) completely ignored or assumed to be perfectly known. The a priori imposition of a change date has been criticized from a statistical point of view as well. Selecting the most suitable change date, out of a list of potential break locations, is presumably inuenced by pretesting the data. Therefore, the researcher's choice must be viewed, at

least to some extent, as being correlated with the data. Such a correlation affects both the nite-sample and asymptotic distributions of the relevant test statistics and may inuence the test outcome(s) (see, among others, Christiano, 1992; Zivot & Andrews, 1992). We therefore opt to let the test identify the most appropriate break point and consider the endogenous-break unit-root test of Zivot and Andrews (1992), which allows for a structural break at an unknown point in time. Specically, for a sample of size T, we repeatedly estimate, for each possible break date, the following equation:
    S1 J 1 1 g t 1 ayt1 bs dst j ytj ut ; y t t 1 f t
s1 j1

1 taking each possible break date, and where yt is the variable of with t interest, a constant, and t a deterministic trend variable. The variable 1) has the value t t 1 if t N t 1 and 0 otherwise, and the dummy g(t 1) has the value 1 if t N t 1 and 0 otherwise. These two variable f(t dummy variables denote a structural break in the slope and intercept, 1. The seasonal dummy variable dst takes the respectively, at time t value one in season s and zero otherwise. The presence of seasonal dummies in Eq. (1) does not affect the critical values of the unit-root test (Ghysels & Perron, 1996). ut represents a random disturbance term, , , 1, 1, a, bs, and j are parameters, and is the difference

Table 1 Empirical applications of multiple-break unit-root tests Test outcome Scope Stationary Evolving Reference GDP of thirty-two countries Allowance for: One break 16 Two breaks 24

16 8

Ben-David and Papell (1995)a Ben-David et al. (2003)a

Unemployment in twenty-six countries Allowance for: Zero breaks 0 26 Two breaks 22 4

Arestis and Biefang-Frisancho Mariscal (1999) Arestis and Biefang-Frisancho Mariscal (1999)

Per-capita GDP of twenty-ve Chinese provinces Allowance for: One break 3 22 Smyth and Inder (2004)a Two breaks 14 11 Smyth and Inder (2004)a Fourteen US macro-economic Allowance for: Zero breaks 0 One break 6 Two breaks 7 Japanese GDP Allowance for: One break Two breaks series 14 8 7 Nelson and Plosser (1982) Zivot and Andrews (1992) Lumsdaine and Papell (1997)

0 1

1 0

Iwamoto and Kobayashi (1992) Ohara (1999)

a We report the suggested evidence from the unit-root tests with a break in both intercept and trend slope, as that is the most general case.

176

M. Kornelis et al. / Intern. J. of Research in Marketing 25 (2008) 173182

operator ( y t y t y t 1). The equation includes J lags in rst differences to ensure that ut follows a white-noise process. The procedure of Perron (1989) is used to determine the maximum number of lags (J). Theoretically, one could consider the whole sample as the set of possible break dates. It is, however, advisable to leave out the beginning and end of the sample when repeatedly estimating Eq. (1), e.g. to do so in the interval [0.15T, 0.85T] (Perron, 1994, p. 136). As with traditional unit-root tests, we have to determine whether a is signicantly smaller than one. However, we now have a list of estimates for a, rather than a single estimate, since we estimated this parameter for each possible change date location t1. We select the one with the most negative t-value out of this list. This coefcient maximizes the evidence against the unit-root hypothesis. When the value of a is signicantly smaller than one, the series is 1) gives found to be stationary, and the corresponding change date (t the estimated position of the (rst) structural break. The signicance level of the associated coefcients (i.e., of 1 and 1) can then be assessed through conventional test statistics. When a is not smaller than one, the series is found to be evolving, even after explicit allowance for a single structural break. In that case, we apply the unit-root test procedure of Lumsdaine and Papell (1997), where the inference on stationarity is based on the inclusion of two jointly introduced break dummies at an a priori unknown location. The corresponding test equation is then written as: h        i 1 1 g t 1 2 f t 2 2 g t 2 ayt 1 yt t 1 f t bs dst j yt j ut :
s1 j1 S1 J

2, we again obtain an estimate for 1and t For each combination of t a. When the selected value of a is signicantly smaller than one, the series is found to be stationary, and the corresponding change dates 1and t 2) give the estimated positions of the rst two structural (t breaks. If the series is found to still be evolving, even after explicit allowance for two breaks, one can test for stationarity after allowing for three (or more) breaks. Finite-sample critical values can be obtained through a straightforward extension of the simulation approach described in Lumsdaine and Papell (1997).1 3.3. Prior practice Using Monte Carlo simulations, Kapetanios (2005) and Ohara (1999) documented the power properties of multiple-break unit-root tests. Kapetanios studied the power properties of multiple-break unitroot tests for data-generating processes with one, two, and three breaks, respectively. He concluded that (i) the results for models with permanent level shifts, trend breaks, and both level shifts and trend breaks are similar, (ii) decreasing values of a increase the power of the test, viz. the further the alternative is from the unit-root null, in terms of coefcient values, the higher the power of the tests, and (iii) the power increases when the size of the breaks increases. Ohara studied the power properties for data-generating processes with two breaks. He concluded that (i) a decreasing value of a increases the power of
1 As the number of breaks increases (relative to the number of observations), the stationary multiple-break model becomes conceptually more similar to a DS model. As indicated before, in a stationary market with multiple breaks, only the breaks can affect the steady-state performance. In a unit-root market, on the other hand, all regular shocks may affect the steady-state performance. When more and more events are classied as irregular in that they can have steady-state effects, each of them becomes less unique in determining the markets long-run performance. In such instances, one may prefer the more parsimonious unit-root model where steady-state effects are part of the regular market dynamics, especially when the extra breaks start to loose face validity. To test whether some special events have an incremental longrun impact (over the long-run impact that all other shocks have), one may apply the supWald approach to yt rather than to yt, as described in Section 3.4 (see in this respect Bai & Perron, 1998).

the test, which is in line with the ndings of Kapetantios, (ii) the twobreak test already has high power when a is smaller than 0.8 in value2, (iii) the size distortions diminish as the sample span increases, and (iv) the location of the break points hardly affects the power of the test. Moreover, there is increasing empirical support that an a priori restriction of a single structural break may lead to erroneous conclusions. For example, Ben-David and Papell (1995) analyzed the GDP series of thirty-two countries and classied sixteen of them as unit-root processes. The same series were analyzed in Ben-David, Lumsdaine, and Papell (2003). After allowing for more than one break without imposing a priori the break-date locations, half of those sixteen series were found to be stationary. Arestis and BiefangFrisancho Mariscal (1999) found that 15% of their 26 unemployment series were re-classied after allowing for more than one break. As summarized in Table 1, similar conclusions were obtained in the analysis of the per capita GDP of multiple Chinese provinces (Smyth & Inder, 2004), various US macro-economic series (Lumsdaine & Papell, 1997; Nelson & Plosser, 1982; Zivot & Andrews, 1992), and the Japanese GDP series (Iwamoto & Kobayashi, 1992; Ohara, 1999). In spite of this evidence in the economics literature, only one study in marketing has (to the best of our knowledge) allowed for two endogenously determined break dates. Van Heerde et al. (2007) applied the test procedure of Lumsdaine and Papell (1997) to determine the stationarity of some of their series. Apart from that study, the standard in marketing is to allow for (at most) a single break, which is either exogenously imposed (as done in e.g. Bronnenberg, Mahajan, & Vanhonacker, 2000; Deleersnyder et al., 2002; Nijs et al., 2001), or endogenously determined, as in the later studies of Pauwels and Srinivasan (2004), Slotegraaf and Pauwels (2008), and Srinivasan, Pauwels, Hanssens, and Dekimpe (2004), among others. Several of these studies found their results to be sensitive to the exogenous/endogenous nature of the single-break identication procedure. For example, Pauwels and Srinivasan (2004) and Srinivasan et al. (2004) describe one and four instances, respectively, in which the unit-root conclusions were reversed. These studies clearly support the importance of a exible unit-root test specication in terms of their break-date location. The practice of looking for more break locations, in contrast, has not gained much attention within the marketing literature. 3.4. Identifying potential additional break points The aim of the preceding tests was to make an inference about the potential stationarity of the series of interest, thereby allowing for a break points. The preceding test sequence (parsimonious) number of M therefore stops once stationarity has been established, and not necessarily when all breaks have been identied. Although those remaining (not yet identied) breaks do not bias the unit-root test into the direction of evolution, one has to account for them, as ignoring them may affect the inferences on the long-run impact of the other breaks. To that extent, we propose to apply the iterative procedure of Ben-David and Papell (2000) to identify any remaining breaks.3 For different M locations of t +1, one then estimates the following test equation: h    i M yt t M 1 f t M 1 g t M 1 1
m1 s 1

J M S1 t m m g t m bs dst j yt j ut ; m f j1

2 In our empirical analysis, the highest value for a was found in the revenue series for the private television incumbents. In this series a = 0.43, which is well below this value. 3 Note that when starting with the Ben-David and Papell test, we condition on the location of the break-points that resulted in a trend-stationary classication of the series of interest. Indeed, changing these locations might well affect the outcome of the earlier unit-root tests, and hence, the substantive implications on the nature of the steady-state impact of all regular shocks.

M. Kornelis et al. / Intern. J. of Research in Marketing 25 (2008) 173182 Table 2 Major events in the Dutch advertising markets Advertising market Television market Radio market Major events Introductions of RTL5 (93:10), SBS6 (95:08), TMF (95:08), Veronica (95:09), and Sport7 (96:08)a. Reorganizations of the public television incumbents (91:0791:10, and 94:09), and RTL5 (97:01). Decontrolling FM frequencies in the spring of 1992. Several new FM stations (95:08). Re-organization of FM frequencies in June 1997 (97:06). Reorganizations of regional newspapers (1993 1994).

177

Print market
a

Sport7 left the market ve months after its introduction (in 96:12).

and compares its performance with the M -break model:

h J M    i S1 m m g t m bs dst j yt j ut ; 4 yt t m f t m1 s1 j1

+ 1 structural breaks, and the number of parameters of includes M the alternative model. When the test value of Eq. (5) is divided by the number of restrictions under the null hypothesis, it corresponds to the standard F-statistic (see e.g. Judge, Grifths, Carter Hill, Lutkephl, & Lee, 1985, p. 186187). Out of the obtained list of + 1)-st break location, one chooses that t M + 1 for estimates for the (M ) is maximal. This results in the M 1,t 2,,t M + 1|t which the value of W (t maximum evidence for an extra break. If its value exceeds the associated critical value, the null hypothesis is rejected, and we include the additional break into the model (see Ben-David & Papell, 2000, Table 1, for critical values). Following Ben-David and Papell (2000), we advise the location between two consecutive breaks to be separated by at least six data points. One continues this procedure of searching for additional breaks until the supWald test fails to reject the null hypothesis of no additional structural break. Because one jointly tests for additional breaks in both the intercept and trend slope, it is possible that one of the two parameters m or m is signicant, while the other is insignicant. In line with Ben-David and Papell (2000), we then remove the insignicant dummy variables. 3.5. Quantifying the steady-state impact In Eqs. (3) and (4), i gives the immediate effect of the i-th structural break on the slope of the series. However, because of the autoregressive nature of the data-generating process, i does not yet give the full steady-state impact. Instead, the steady-state impact on the series' growth rate is given by SS,i = i / (1 j) (see, e.g. BenDavid & Papell, 2000, or Deleersnyder et al., 2002 for a marketing application). Depending on the sign of this ratio, the series' pre-event

by means of the so-called supWald (or supF) test. This test can be written as:     SSR  tM SSR t 1 M jt   : T W t M1 1 ; :::; t M SSR t M 1 

M) is the sum of squared residuals under the null In Eq. (5), SSR(t structural breaks), SSR hypothesis (that is, the model in Eq. (4) with M (t M + 1) the sum of squared residuals of the alternative model (3) that

Fig. 2. The target market: TV advertising revenues and volumes.

178

M. Kornelis et al. / Intern. J. of Research in Marketing 25 (2008) 173182

Fig. 3. The related markets: radio and print advertising revenues.

steady-state growth rate is either reduced or increased. In the absence of a signicant effect on the growth of the series, we can derive in a similar way whether the event of interest caused an immediate (i) or steady-state (SS,i = i / (1 j)) level shift. In sum, our methodology allows us to quantify the steady-state size of each identied event's impact on the series' growth path. Previous marketing research limited the maximum number of potential breaks to one (so, M could only take the value of zero or one), which was often imposed a priori, rather than determined endogenously, although the study of Van Heerde et al. (2007) being the exception in that they allowed for (at most) two breaks. 4. Empirical application 4.1. The Dutch advertising market We consider the Dutch advertising market for branded goods and services in the period of 1990 to 1998, when advertising expenditures doubled to a market size of 5,054 million Dutch guilders (current prices) in 1998 (BBC/VEA Netto Mediabestedingen in Nederland, 1999, p. 65), which corresponds to approximately 2,300 million Euro. The Netherlands is a highly developed, industrialized country, similar to other European countries such as Belgium, France or Germany, both in terms of stage of economic development and in terms of media landscape (Steenkamp et al., 2005). The latter is characterized by the presence of both publicly owned and private broadcasting channels, a proliferation of radio channels, and a gradual reduction in the number of newspapers (Leeang & Van Raaij, 1993). A key event in this setting was the introduction of three commercial television channels (SBS6, TMF, and Veronica) in August (95:08) and September 1995 (95:09). We see this as a quasi-joint introduction, since all three late movers entered the market almost simultaneously. The announcement, almost a year before, caused a real media hype in the Netherlands. Between the announcement and actual entry, more than 60 press releases appeared in ve nationwide daily newspapers, which differed widely on the presumed implications for both market incumbents and related markets.

Apart from the introduction of the focal private broadcasting channels, other major events took place in this and related markets, several of which are indicated in Table 2. As mentioned before, our methodology is exible enough to account for the potentially confounding effects, if any, of these additional events. We employ monthly data on advertising revenues of branded goods and services in the Netherlands for the period from January 1990 to December 1998, as collected by the BBC (Dutch abbreviation for Bureau voor BudgettenControle). Our methodology is applied on the data from January 1990 to June 1998, and the subsequent six months are used as a hold-out sample. We consider the revenue series of both private and public television incumbents. In addition, we also analyze the revenue implications for two key related markets, i.e., print and radio advertising. Other related markets (e.g. internet advertising and satellite television advertising) played a very marginal role in the Dutch advertising market of the 1990s and were not considered for lack of reliable data over the entire time span. We express the incumbents' revenue in Dutch Guilders (DFL). This common metric facilitates a comparison of the ndings across different advertising markets. Moreover, revenue generation has been argued to be a more relevant business goal to managers than mere volume creation (Srinivasan et al., 2004, p. 618). All revenue series have been deated with the Consumer Price Index (CPI) obtained from the Dutch Central Bureau of Statistics (CBS). In addition to the revenue series, we also analyze television advertising volumes (measured in seconds) since the identication of potential structural breaks in these series may provide a more detailed picture of the underlying dynamics at work. Indeed, when used in combination with the revenue ndings, they may give some important insights (although, indirectly) on what happened with the average price level. In sum, the variables that we consider in our empirical application include advertising revenues of Public TV incumbents, Private TV incumbents, Print, and Radio, and the advertising seconds of Public TV incumbents, and Private TV incumbents. The natural logarithms of these series are depicted in Figs. 2 and 3. We take logarithms of the

Table 3A Single-break unit-root test results Series (revenues) Public TV incumbents Private TV incumbents
a

Table 3B Single-break unit-root test results Test result 4.48 6.36a Date 95:09 95:09 Series (volumes) Public TV incumbents Private TV incumbents Test result 3.58 4.27 Date 95:09 93:10

Signicant at the 5% level, for which the critical value is 5.08 (Zivot & Andrews, 1992).

The critical value at the 5% level is 5.08 (Zivot & Andrews, 1992).

M. Kornelis et al. / Intern. J. of Research in Marketing 25 (2008) 173182 Table 3C Double-break unit-root test results Test result Series (revenues) Public TV incumbents Series (volumes) Public TV incumbents Private TV incumbents 6.40 Dates 91:09 and 95:09 Table 4 Single-break unit-root test results Series (related markets) Print Radio
a

179

Test result 7.97a 8.40a

Date 94:02 92:01

Signicant at the 5% level, for which the critical value is 5.08 (Zivot & Andrews, 1992).

6.36 6.60

93:07 and 96:02 92:07 and 96:11

The critical value at the 5% level is 6.82 (Lumsdaine & Papell, 1997).

variables to reduce potential heteroscedasticity, as is common in recent time-series studies on the impact of late entries, e.g. Deleersnyder et al. (2002) or Pauwels and Srinivisan (2004). In general, the rst difference of the log-transformed series is a good measure for the growth rate of the original variable (Franses & Koop, 1998), while the size of the break parameters gives information on the percentage impact of late movers on the level and growth rate of the original variable. 4.2. Break-date identication We rst established the stationarity of the series of interest through structural-break unit-root tests. The outcomes of the singlebreak test (Eq. (1)) for the two revenue series in the television market are given in Table 3A and indicate that the private incumbents series was stationary after allowing for one break (p b 0.05), while the unitroot null hypothesis was not rejected for the public incumbents series. For the volume series of both public and private incumbents (Table 3B), the unit-root null hypothesis is also not rejected when allowing for a single break. After allowing for a second break (Eq. (2)), public revenues and the two volume series were still found to be evolving (Table 3C), but became stationary (p b 0.05) after allowing for three break dates (Table 3D). The two related markets, print and radio, are both stationary (p b 0.05) after allowing for a single break (Table 4). We subsequently applied Ben-David and Papell's supWald test (Eq. (5)) to investigate whether additional breaks are still present in any of the series. No such evidence was found. We summarize the various break locations in Table 5. 4.3. Break-date interpretation We rst concentrate on the effects of the quasi-joint introduction. For the private TV revenues, an endogenously determined break point was found at 95:09. This break date coincides with the second month of the quasi-joint introduction of the late movers (95:08 and 95:09). For the private TV volumes, the break date (95:07) almost coincides with the quasi-joint introduction. For the public TV volume series, a break was identied in 96:02, which, given its proximity, can still be attributed to the event of interest, albeit with a lag of a few months, which could be due to earlier contractual agreements. Unlike some industry observers' and public policy makers' prior expectations (see e.g. Van Dijk, 1995), the related print and radio advertising markets were not structurally
Table 3D Triple-break unit-root test results Test result Series (revenues) Public TV incumbents Series (volumes) Public TV incumbents Private TV incumbents 7.98a Dates 91:09, 94:07, and 96:07

affected by the late entrants in the television market. This nding complements earlier work by Deleersnyder et al. (2002) and Silk, Klein, and Berndt (2001), who also reported very weak interdependencies, if any, between different media. However, each of these related media experienced some other structural breaks, which all have considerable face validity. The radio market, for example, experienced a break in 92:01, which coincides with the government's decontrolling of the FM frequencies in the beginning of 1992 (STER Annual Report, 1992, p.11), while the print market's advertising revenues experienced a signicant impetus from the regional newspaper's reorganization that took place in the end of 1993 and the beginning of 1994 (NDP Annual Report, 1993, p. 10; NDP Annual Report, 1994, p. 19). Moreover, the volume series experienced some further structural breaks. Volumes of both public and private TV incumbents were structurally affected by a reorganization of the public channels near the end of 1991 (STER Annual Report, 1991, p. 8), and by the introduction of RTL5. The revenue series of the public incumbents also experienced three breaks. These coincide with the reorganizations in 1991 (STER Annual Report, 1991, p. 8) and 1994 (STER Annual Report, 1994, p. 10) and the introduction of Sport7. Interestingly, the introduction of RTL5 did not cause a break in the revenue series, suggesting that the volume effects were offset by accompanying changes in the price setting for advertising space. To summarize, the above discussion indicated (i) that the identied break dates have considerable face validity, and (ii) that multiple events may actually (but do not have to) cause a break. 4.4. Quantifying the steady-state impact of the quasi-joint entry Not all TV series were affected to the same extent by the quasijoint introduction. As changes in the growth rate will eventually overtake any level shift, we focus our discussion on the former. As detailed in Table 6, the private incumbents experienced a slowdown in the growth rate of their revenue series following the entry of the three private competitors. The monthly growth rate slowed down from a pre-introduction steady-state value of 1.24% to a non-signicant postevent value of 0.27% (1.241.51). In contrast, the public TV

Table 5 The estimated break-date locations Series Revenues Public TV incumbents Date(s) 91:09 94:07 96:07 95:09 Event(s) Re-organization of public channels Re-organisation of public channels Introduction of Sport7 Quasi-joint introduction

Private TV incumbents Volumes Public TV incumbents

Private TV incumbents 8.07a 8.06a 92:01, 93:10, and 96:02 91:06, 93:10, and 95:07 Related markets Print Radio

92:01 93:10 96:02 91:06 93:10 95:07

Re-organization of public channels Introduction of RTL5 Quasi-joint introduction Re-organization of public channels Introduction of RTL5 Quasi-joint introduction

a Signicant at the 5% level, for which the critical value is 7.89 (nite-sample critical value obtained through Monte Carlo simulation, using the procedure outlined in Lumsdaine & Papell, 1997).

94:02 92:01

Reorganization of newspapers Decontrolling of FM frequencies

180

M. Kornelis et al. / Intern. J. of Research in Marketing 25 (2008) 173182 Table 7 Comparison with Zivot and Andrews' single-break approach Quasi-joint entry UR outcome Date Single-break approach Public TV revenues Evolving Public TV volumes Evolving Private TV volumes Evolving Multiple-break approach Public TV revenues Stable Public TV volumes Stable Private TV volumes Stable Trend break Regular shock persistency 0.69 (3.34) 0.51 (7.42) 0.80 (3.62)

incumbents showed a positive trend break in their volume series, where the non-signicant pre-event growth rate (0.06%) increased to a signicant post-event rate of 1.04%. However, since the steady-state growth path in the public TV revenues were not affected by the quasijoint introduction, the potential revenue implications from an increase in the number of advertising seconds were completely offset by a negative pressure on the average price level (as reported also in, e.g., NRC Handelsblad, 1996). Overall, our ndings conrm the idea that the more similar (i.c. private) incumbents were most affected by the late entry of three additional new players, as their more similar positioning made them more vulnerable to customer switching (Berry & Waldfogel, 1999b; Rooderkerk, Van Heerde, & Bijmolt, 2006). 4.5. Benchmarking To see whether the aforementioned substantive conclusions would be affected if we had restricted the maximum number of break dates to one, as is now common practice in the marketing literature, we implemented the test of Zivot and Andrews (1992), in line with (among others) Pauwels and Srinivasan (2004), Slotegraaf and Pauwels (2008), and Srinivasan et al. (2004). Since this strategy is also the rst step in our testing framework, both approaches led to the same outcomes in those instances where also our testing strategy resulted in one structural break, viz. private television revenues and the related radio and print series. In those cases where our approach identied multiple breaks (the volume series of both public and private incumbents and the revenue series for the public channels), however, the substantive implications differed considerably. As indicated in Table 7, the three series were found to be evolving in the single-break tests, implying that all shocks would have a steady-state effect (given by Campbell and Mankiw's regular shock persistency in the nal column of Table 7), as opposed to our nding that only a limited number of known major events have an impact on the series' steady-state performance. Conceptually, the latter is a more appealing scenario, as also argued in Deleersnyder et al. (2002) and Fok et al. (2006). As for the impact of our focal event, we found a structural break in the slope of two of these three series, while this was not the case with the single-break tests. We compared both approaches more formally in terms of their within-sample t. The multiple-break approach resulted in all three cases in a better within-sample t, as reected in a lower (more negative) value of the AIC criterion, i.e., 1.41, 1.52, and 2.01, versus 1.01, 1.13, and 1.53 for, respectively, the public TV revenues, public TV volumes, and private TV volumes. We also compared the one-step ahead forecasting performance of both approaches in a hold-out sample of six observations. Across the three series, a nonparametric sign test (also used in Franses & Kleibergen, 1996) revealed no signicant difference (p = 0.17).

No trend break No trend break No trend break

No trend break Zero by denition 96:02 0.98% (4.14) Zero by denition 95:07 1.26% ( 3.16) Zero by denition

The corresponding t-values are given in parentheses.

The previous comparison focused on the number of break points (3 versus 1). As an alternative benchmark model, we also considered the situation where the number of breaks (3) is xed, but where the location of the break points is varied. To that extent, we estimated for each of our three series with three break points 18 additional models, in which we consecutively moved the location of one of the identied break points 1, 2 or 3 positions back- or forward. Hence, we estimated a total of 54 competing 3-break-models. To ensure a fair comparison, we recomputed (using the procedure advocated by Perron, 1994) the number of lagged dependent variables in each of the test equations. Indeed, if we had imposed every time the number of terms found for our focal model, we would bias the results in favor of our specication. After changing the location of these break points, we could no longer reject the unit-root null hypothesis. We therefore estimated all 54 models in rst differences, while allowing for three structural breaks.4 These were captured through pulse (to capture a level shift) and step (to allow for slope change) dummies. However, this consistently resulted in a worse t. For example, the average AIC value (across 18 specications) for the competing models was 1.00, 1.01, and 1.61 for the public TV revenues, public TV volumes, and private TV volumes, respectively (details on the AIC of all 54 individual models can be obtained from the rst author upon request). This clearly suggests that also the location of the break points matters. 5. Conclusion We introduced a testing procedure that allows for multiple breaks, which does not require an a priori imposition of the location of the various break points. The former is needed to avoid omitted-variable biases in quantifying the impact of a given event, while the latter property recognizes that lead and lagged effects may be present, in which case the arrival time of the late entrant may not coincide exactly with observed changes in the incumbents performance series. Whereas the need to endogenously determine potential break points has been recognized in prior marketing studies, no attention had been given to the determination of the appropriate number of breaks. We compared our results with a testing procedure that allows for a single break, and demonstrated that this more restrictive approach (i) affected the substantive insights, (ii) resulted in poorer withinsample t, and (iii) did not improve the forecasting performance of the models. As such, we concur fully with Allen and Fildes (2005, p. 901) that a more careful testing for structural breaks should become an automatic aspect of specication testing in time-series models, and

Table 6 The steady-state effects of the quasi-joint introduction on growth rates Series Trend before Revenues Public TV incumbents Private TV incumbents Volumes Public TV incumbents Private TV incumbents Related markets Print Radio 3.05% (10.7) 1.24% (8.87) Late-mover entry Date Trend break Trend after

No trend break 95:09 1.51% (4.88)

0.27% ( 1.01)

0.06% (0.37) 1.62% (4.34)

96:02 95:07

0.98% (4.14) 1.26% (3.16)

1.04% (6.59) 0.36% (2.23)

4 Alternatively, we could have augmented the number of potential breaks in an attempt to nevertheless obtain trend stationarity, but this would no longer have resulted in a model comparison across specications with a common number (3) of breaks.

0.73% (15.3) 1.23% (25.0)

No trend break No trend break

The corresponding t-values are given in parentheses.

M. Kornelis et al. / Intern. J. of Research in Marketing 25 (2008) 173182

181

with Franses and Kleibergen (1996, p. 288) that it is worthwhile to investigate the presence of structural breaks within the sample while selecting between a DS and a TS process. This complements Franses' (2005) call to adequately test variables' time-series properties along with the constancy of the resulting model parameters. Through our empirical illustration of this procedure, we contributed to the empirical knowledge base on the impact of new entrants on the revenues of advertising markets. We investigated the impact over time of three late movers in the Dutch television market on the performance of market incumbents in both the penetrated (i.e., television advertising) and two related (i.e., print and radio) advertising markets. The entry of three new players in the Dutch TV market had a profound impact on the subsequent evolution of the television advertising market, both in terms of revenues (for the private incumbents) and price changes (for both private and public incumbents). However, the fear of many industry observers that the related print and radio markets would also be affected was not substantiated. Of course, the Netherlands is not the only country where late entrants appeared in advertising markets. It would be interesting to identify the most important impacts in other countries as well, which would allow one to derive empirical generalizations and to identify various underlying drivers for the presence/absence of such an impact. Moreover, advertising spending markets are not the sole domain for late-mover entries. Other environments that face new entrants include retail markets (see e.g. Gielens, Van de Gucht, Steenkamp, & Dekimpe, in press), and markets for store brands (see e.g. Pauwels & Srinivasan, 2004), durables (e.g. Mahajan et al., 1993), and public services (see e.g. Newbury, 1998). Our proposed methodology is exible enough to capture the performance consequences of competitive entries in these contexts as well. Moreover, competitive entries could, as reviewed in Section 2, alter the incumbents' pricing and/or spending patterns. As time-series econometrics are often used to infer competitive reaction patterns (see e.g. Horvth, Leeang, Wierenga, & Wittink, 2005; Steenkamp et al., 2005), our testing framework is directly applicable in those settings, as in other situations where major interventions take place. In this respect, one could consider to what extent the introduction of loyalty programs (cf. Leenheer, Van Heerde, Bijmolt, & Smits, 2007) or additional channels (e.g. Verhoef, Neslin, & Vroomen, 2007) change the underlying data-generating process of some key metrics. More research is needed, however, to infer how competitive entries (or other discrete events) might affect the effectiveness of various marketing activities (a third area identied in Section 2). Pauwels and Srinivasan (2004) estimated separate pre- and postentry models to quantify the impact of a private label introduction on customers' price sensitivity. In so doing, they implicitly assumed that the real break date was known a priori, and that there was at most one break date. It would be useful to study how our testing procedure could be used in varying-parameter models. A promising avenue would be to allow for an unknown number of break points at unknown locations in Dynamic Linear Models (see e.g. Ataman, Van Heerde, & Mela, in press for a recent review on DLM models). Also in that tradition, one typically considers only a xed number of a priori determined break points (see e.g. Van Heerde et al., 2004, 2007), and the idea to allow for multiple breaks at unknown locations may also be worthwhile in those settings. Acknowledgements The paper is based on the rst author's PhD thesis written at the University of Groningen, which was kindly funded by the Netherlands Organization for Scientic Research (NWO) under project number 510-01-0402. The second author acknowledges support from the Flemish Organization for Scientic Research (FWO) under grant number G.0116.04. We also thank three anonymous reviewers, the Area Editor, and IJRM editors Hubert Gatignon and Stefan Stremersch for constructive comments during the review process.

References
Allen, P. G., & Fildes, R. (2005). Levels, differences and ECMs: Principles for improved econometric forecasting. Oxford Bulletin of Economics and Statistics, 67(s1), 881904. Arestis, P. h., & Biefang-Frisancho Mariscal, I. (1999). Unit roots and structural breaks in OECD unemployment. Economics Letters, 65(2), 149156. Ataman, B., Van Heerde, H.J., & Mela, C.F., (in press). Building brands. Marketing Science, forthcoming. Bai, J., & Perron, P. (1998). Estimating and testing linear models with multiple structural changes. Econometrica, 66(1), 4778. BBC/VEA Netto mediabestedingen in Nederland. (1999). Amsterdam: BBC (in Dutch). Ben-David, D., Lumsdaine, R. L., & Papell, D. H. (2003). Unit roots, postwar slowdowns and long-run growth: Evidence from two structural breaks. Empirical Economics, 28(2), 303319. Ben-David, D., & Papell, D. H. (1995). The great wars, the great crash, and steady state growth: Some new evidence about an old stylized fact. Journal of Monetary Economics, 36(3), 453475. Ben-David, D., & Papell, D. H. (2000). Some evidence on the continuity of the growth process among the G7 countries. Economic Inquiry, 38(2), 320330. Berry, S. T., & Waldfogel, J. (1999). Free entry and social inefciency in radio broadcasting. RAND Journal of Economics, 30(3), 397420. Berry, S. T., & Waldfogel, J. (1999). Public radio in the United States: Does it correct market failure of cannibalize commercial stations? Journal of Public Economics, 71(2), 189211. Box, G. E. P., & Tiao, G. C. (1975). Intervention analysis with applications to economic and environmental problems. Journal of the American Statistical Association, 70(349), 7079. Bronnenberg, B. J., Mahajan, V., & Vanhonacker, W. R. (2000). The emergence of market structure in new repeat-purchase categories: The interplay of market share and retailer distribution. Journal of Marketing Research, 37(1), 1631. Campbell, J. Y., & Mankiw, N. G. (1987). Are output uctuations transitory? Quarterly Journal of Economics, 102(4), 857880. Chintagunta, P. K. (1999). Measuring the effects of new brand introduction on inter-brand strategic interaction. European Journal of Operational Research, 118(2), 315331. Chow, G. C. (1960). Tests of equality between sets of coefcients in two linear regressions. Econometrica, 28(3), 591605. Christiano, L. J. (1992). Searching for a break in GNP. Journal of Business and Economic Statistics, 10(3), 237250. Cleeren, K., Dekimpe, M. G., & Verboven, F. (2006). Competition in local-service sectors. International Journal of Research in Marketing, 23(4), 357367. Dekimpe, M. G., & Hanssens, D. M. (1995). The persistence of marketing effects on sales. Marketing Science, 14(1), 121. Dekimpe, M. G., & Hanssens, D. M. (2004). Persistence modeling for assessing marketing strategy performance. In D. Lehmann & C. Moorman (Eds.), Assessing marketing strategy performance (pp. 6993). Cambridge: Marketing Science Institute. Deleersnyder, B., Geyskens, I., Gielens, K., & Dekimpe, M. G. (2002). How cannibalistic is the internet channel? A study of the newspaper industry in the United Kingdom and the Netherlands. International Journal of Research in Marketing, 19(4), 337348. Dickey, D. A., & Fuller, W. A. (1979). Distribution of the estimation for autoregressive time series with a unit root. Journal of the American Statistical Association, 74(366), 427431. Doyle, P., & Saunders, J. (1985). The lead effects of marketing decisions. Journal of Marketing Research, 22(1), 5465. El Sehity, T., Hoelzl, E., & Kirchler, E. (2005). Price developments after a nominal shock: Benford's Law and psychological pricing after the Euro introduction. International Journal of Research in Marketing, 22(4), 471480. Fok, D., & Franses, Ph. H. (2004). Analyzing the effects of a brand introduction on competitive structure using a market share attraction model. International Journal of Research in Marketing, 21(2), 159177. Fok, D., Horvth, Cs., Paap, R., & Franses, Ph. H. (2006). A hierarchical Bayes error correction model to explain dynamic effects of price changes. Journal of Marketing Research, 43(3), 443461. Franses, Ph. H. (2005). On the use of econometric models for policy simulation in marketing. Journal of Marketing Research, 42(1), 414. Franses, Ph. H., & Kleibergen, F. (1996). Unit roots in the NelsonPlosser data: Do they matter for forecasting? International Journal of Forecasting, 12(2), 283288. Franses, Ph. H., & Koop, G. (1998). On the sensitivity of unit root inference to nonlinear data transformations. Economics Letters, 59(1), 715. Ghysels, E., & Perron, P. (1996). The effect of linear lters on dynamic time series with structural change. Journal of Econometrics, 70(1), 6997. Gielens, K., Van de Gucht, L.M., Steenkamp, J.-B.E.M., & Dekimpe, M.G. (in press). Dancing with a giant: The effect of Wal-Mart's entry into the UK on the performance of European retailers. Journal of Marketing Research, forthcoming. Gruca, T. S., Kumar, K. R., & Sudharshan, D. (1992). An equilibrium analysis of defensive response to entry using a coupled response function model. Marketing Science, 11(4), 348358. Hanssens, D. M., Parsons, L. J., & Schultz, R. L. (2001). Market response models, 2nd ed. Boston: Kluwer Academic Publishers. Horvth, Cs., Leeang, P. S. H., Wierenga, J. E., & Wittink, D. R. (2005). Competitive reaction- and feedback effects based on VARX models of pooled store data. International Journal of Research in Marketing, 22(4), 415426. Iwamoto, Y., & Kobayashi, H. (1992). Testing for a unit root in Japanese GNP. Japan and the World Economy, 4(1), 1737. Judge, G. G., Grifths, W. E., Carter Hill, R., Lutkephl, H., & Lee, T. -C. (1985). The theory and practice of econometrics, 2nd ed. New York: John Wiley and Sons. Kapetanios, G. (2005). Unit-root testing against the alternative hypothesis of up to m structural breaks. Journal of Time Series Analysis, 26(1), 123133.

182

M. Kornelis et al. / Intern. J. of Research in Marketing 25 (2008) 173182 Pesaran, M. H., & Samiei, H. (1991). Persistence, seasonality and trend in UK egg production. Applied Economics, 23(3), 479484. Reiss, P. C., & Spiller, P. T. (1989). Competition and entry in small airline markets. Journal of Law and Economics, 32(2), S179S202. Rooderkerk, R. P., Van Heerde, H. J., & Bijmolt, T. H. A. (2006). The impact of context effects on optimal product lines.: Tilburg University Working Paper. Silk, A. J., Klein, L. R., & Berndt, E. R. (2001). The emerging position of the Internet as an advertising medium. Netnomics, 3(2), 129148. Slotegraaf, R. J., & Pauwels, K. (2008). The impact of brand equity and innovation on the long-term effectiveness of promotions. Journal of Marketing Research, 45(3), 293306. Smyth, R., & Inder, B. (2004). Is Chinese provincial real GDP per capita nonstationary? Evidence from multiple trend break unit root tests. China Economic Review, 15(1), 124. Srinivasan, S., Pauwels, K., Hanssens, D. M., & Dekimpe, M. G. (2004). Do promotions benet manufacturers, retailers, or both? Management Science, 50(5), 617629. Steenkamp, J. -B. E. M., Nijs, V. R., Hanssens, D. M., & Dekimpe, M. G. (2005). Competitive reactions to advertising and promotions attacks. Marketing Science, 24(1), 3554. Ster Annual Report. (1991). Hilversum: STER (in Dutch). Ster Annual Report. (1992). Hilversum: STER (in Dutch). Ster Annual Report. (1994). Hilversum: STER (in Dutch). Van Dijk, P. (1995, March 4). Lang leve de krant.Algemeen Dagblad Amsterdam: PCM uitgevers (in Dutch). Van Heerde, H. J., Helsen, K., & Dekimpe, M. G. (2007). The impact of a product-harm crisis on marketing effectiveness. Marketing Science, 26(2), 230245. Van Heerde, H. J., Mela, C. F., & Manchanda, P. (2004). The dynamic effect of innovation on market structure. Journal of Marketing Research, 41(2), 166183. Verhoef, P. C., Neslin, S. A., & Vroomen, B. (2007). Multichannel customer management: Understanding the research-shopper phenomenon. International Journal of Research in Marketing, 24(2), 129148. Zivot, E., & Andrews, D. W. K. (1992). Further evidence on the great crash, the oil-price shock, and the unit root hypothesis. Journal of Business and Economic Statistics, 10(3), 251270.

Lamey, L., Deleersnyder, B., Dekimpe, M. G., & Steenkamp, J. -B. E. M. (2007). How business cycles contribute to private-label success: Evidence from the United States and Europe. Journal of Marketing, 71(1), 115. Leeang, P. S. H., & Van Raaij, W. F. (1993). The changing consumer in the Netherlands: Recent changes in environmental variables and their consequences for future consumption and marketing. International Journal of Research in Marketing, 10(4), 345363. Leenheer, J., Van Heerde, H. J., Bijmolt, T. H. A., & Smits, A. (2007). Do loyalty programs really enhance behavioral loyalty? An empirical analysis accounting for selfselecting members. International Journal of Research in Marketing, 24(1), 3147. Lim, J., Currim, I. S., & Andrews, R. L. (2005). Consumer heterogeneity in the longer-term effects of price promotions. International Journal of Research in Marketing, 22(4), 441457. Lumsdaine, R. L., & Papell, D. H. (1997). Multiple trend breaks and the unit-root hypothesis. The Review of Economics and Statistics, 79(2), 212218. Mahajan, V., Sharma, S., & Buzzell, R. D. (1993). Assessing the impact of competitive entry on market expansion and incumbent sales. Journal of Marketing, 57(3), 3952. NDP Annual Report. (1993). Amsterdam: NDP (in Dutch). NDP Annual Report. (1994). Amsterdam: NDP (in Dutch). Nelson, C. R., & Plosser, C. I. (1982). Trends and random walks in macroeconomic time series. Journal of Monetary Economics, 10(2), 139162. Newbury, D. M. (1998). Freer electricity markets in the UK: A progress report Cambridge, UK: Department of Applied Economics. Nijs, V. R., Dekimpe, M. G., Steenkamp, J. -B. E. M., & Hanssens, D. M. (2001). The category demand effects of price promotions. Marketing Science, 20(1), 122. NRC Handelsblad (1996, September 16). Inkomsten omroepen uit de STER dalen sterk.NRC Handelsblad Amsterdam: PCM uitgevers (in Dutch). Ohara, H. I. (1999). A unit root test with multiple trend breaks: A theory and an application to US and Japanese macroeconomic time-series. The Japanese Economic Review, 50(3), 266290. Pauwels, K., & Srinivasan, S. (2004). Who benets from store brand entry? Marketing Science, 23(3), 364390. Perron, P. (1989). The great crash, the oil price shock, and the unit root hypothesis. Econometrica, 57(6), 13611401. Perron, P. (1994). Trend, unit root and structural change in macroeconomic time series. In B. Bhaskara Rao (Ed.), Cointegration for the applied economist (pp. 113146). New York: St. Martins Press.