Reviving Demand-Pull Perspectives: The Effect of Demand Uncertainty and Stagnancy on R&D Strategy

This paper looks at the effects of demand uncertainty and stagnancy on firms’ decisions to engage in R&D activities and the amount of financial effort devoted to it. The paper contributes to the innovation literature in three respects: first, it adds to the revived debate on demand-pull perspectives in innovation studies, by specifically looking at demand-related (lack of) incentives to invest in innovation. Also, importantly, it complements the emerging literature on barriers to innovation in a two-fold way: first, by focusing on demand-related obstacles rather than the more explored financial barriers; second, by analyzing in detail whether experiencing uncertainty or lack of demand is a sector-specific feature, namely whether firms active in high or low tech manufacturing or in knowledge intensive or low tech services are more or less dependent on demand conditions when deciding to perform R&D. We find that uncertain demand and lack of demand are perceived as two completely different barriers. While uncertainty on demand does not seem to constrain R&D efforts, the perception of lack of demand does strongly reduce not only the amount of investment in R&D but also the likelihood of firms to engage in R&D activities. We interpret this evidence in terms of the specific phase of the innovation cycle in which decisions to invest in R&D are taken. Sectoral affiliation does not seem to matter when it relates to demand conditions, supporting the conjecture that positive expectations on market demand is a structural and necessary condition to be fulfilled for all firms prior to invest in R&D.


Introduction
The closely connected influences of demand and technological opportunities on the strategic decisions of firms to innovate and the aggregate outcomes of these decisions are well established subjects of research in innovation studies, since the seminal contribution of Schmookler (1966) and followed by a fierce debate among scholars in the field (Mowery and Rosenberg, 1979). A recent contribution (Di Stefano et al., 2012) reviews this debate by examining the evolution in this research, which has in turn come down in favour of either a technology-push or demand-pull source of innovation as it has sought to disentangle their relative importance in fostering innovation.
Interestingly, no previous study has analysed the demand-pull perspective from the viewpoint of barriers to innovation. As is common within the innovation literature, analyses of the factors of innovation success are proportionally more numerous than studies of patterns of failure and the effect of the lack of incentives. As such, scholars of demand-pull perspectives seem to have overlooked lack of demand or demand uncertainty as factors hampering decisions to invest in innovation.
The emerging literature on barriers to innovation has dealt primarily with the firms' characteristics that affect their perception of barriers to innovation or, when specifically examining the actual hindrances of perceived barriers, it has paid a disproportionate amount of interest to financial barriers and limitations to the financial capacity of firms to invest in R&D (see D'Este et al., 2012, and Pellegrino and Savona, 2013, for a review of this literature). This bias toward financial obstacles might well reflect the relative "dominance" of technology-push perspectives over interest in demand-related incentives to innovate.
Rather than contrasting the two perspectives empirically, here we seek to rebalance the overall picture by attempting to disentangle the effects of lack of demand, or perceived uncertainty about demand conditions, on firms' decisions to invest in R&D and the amount of resources they devote to the activity. The paper makes a number of contributions to the innovation literature: first, it adds to the recently renewed debate on demand-pull perspectives in innovation studies, by examining demand-related (i.e., lack of) incentives to invest in innovation. Second, it complements the emerging literature on barriers to innovation in two ways: on the one hand, by focusing on demand-related obstacles rather than on the more frequently explored financial barriers; and, on the other, by analyzing in detail whether experiencing demand-related obstacles is a sector-specific feature, that is, whether firms active in high-or low-tech manufacturing or in knowledge intensive or low-tech services are more or less dependent on demand conditions when deciding to perform R&D.
We find that demand uncertainty and stagnancy are two quite distinct barriers, having substantially different effects on firms' behaviour. We interpret this evidence in terms of the specific phase in the innovation cycle in which decisions to invest in R&D are formulated.
While demand uncertainty has a weak, positive statistically significant effect on R&D plans, the perception of a lack of demand has a marked impact on not only the amount of investment in R&D but also the likelihood of firms engaging in R&D activities. Sectoral affiliation does not seem to be a factor in demand conditions, supporting the conjecture that positive expectations regarding market demand are a structural and necessary condition that has to be satisfied by all firms prior to deciding to invest in R&D. When considered from the perspective of barriers to innovation, demand-related incentives therefore seem to cut across sectoral specificities in technological opportunities.
In the section that follows we briefly review the two branches of literature mentioned above: that is, studies comparing demand-pull vs. technology-push sources of innovation and analyses of barriers to innovation. Section 3 describes the data employed in the empirical analysis; Section 4 illustrates the econometric strategy and the variables used in the estimations, while Section 5 discusses the results and provides a response to the main research question. Section 6 concludes.

Background literature 2.1. Demand-pull perspectives revisited
The innovation literature has traditionally been somewhat ambivalent with regard to the role of demand as an incentive to innovation, besides that of technological opportunities.
As suggested by Di Stefano et al., (2012) in a recent review, the debate between demand-pull and technology-push perspectives has evolved through different stages, from the rigid adoption of opposing stances by the supporters of demand-pull (Schmookler, 1962(Schmookler, , 1966 Myers and Marquis, 1969;von Hippel, 1978von Hippel, , 1982  For the purposes of our discussion here, it should suffice to recall the main arguments in the debate, relate them to the most recent literature on barriers to innovation (Section 2.2) and formulate the conjectures (Section 2.3) that we then test empirically in the remaining of the paper.
As Fontana and Guerzoni (2008) suggest, the intuition regarding the influence of demand on innovation was sparked by the seminal contributions of Schmookler (1962;1966) and Myers and Marquis (1969), who claimed that the introduction of new products and processes is conditioned by the presence of demand or even possibly a latent demand and, in general, by positive expectations of profitability from returns to innovation. In the absence of these conditions, firms would simply not have any incentive to innovate. Moreover, the adoption and diffusion of (especially new) products are intrinsically subject to uncertainty, which would further reduce incentives to innovate. The arguments forwarded by the proponents of technology-push sources touched upon various issues, ranging from the reverse causality of the empirical relationships estimated by Schmookler (1966) and Meyers and Marquis (1969) to the difficulties of identifying the relevant demand affecting innovation incentives.
It is our contention, and one we come back to later, that market sizeand therefore expectations regarding profitabilityand demand uncertainty are very likely to refer to different levels of demand. First, positive expectations with regard to profitability and, hence, incentives to innovate, despite being intrinsically linked to the fate of the new product being launched, are affected primarily by the macro-conditions of aggregate demand and the market dynamism of the specific and related products. Even incremental product or process innovation would be hard to implement if forecasts of sales and returns to innovation were poor.
Second, while uncertainty might be linked to aggregate macro-conditions of demand, it is predominantly affected by the characteristics of the new products/services and the lack of information on users and their capabilities to adopt/benefit from the new product (see also von Tunzelmann and Wang, 2003 on user capabilities).
Of course, macro-and micro-demand conditions are likely to reinforce each other, though in the case of incremental product or process innovation, aggregate stagnancy of demand might be more influential, whereas in the case of radically new products or services it is the uncertainty that is likely to play a major role in terms of incentives to innovate (see also Fontana and Guerzoni, 2008).

Demand as a barrier to innovation: stagnancy and uncertainty
Although the literature on barriers to innovation is relatively recent, scholars have found substantial evidence for the presence and effects of perceived hindrances on the propensity and intensity of engagement in innovation activities.
A large proportion of these studies have focused their attention on analyses of the effects of financial constraints on firms' cash flow sensitivity to afford R&D investments (for a review, see Schiantarelli, 1996;Hall, 2002;Bond et al., 1999;Hottenrott and Peters, 2012).
Indeed, empirical evidence tends to confirm that encountering financial constraints significantly lowers the likelihood of firms engaging in innovative activities (Savignac, 2008), with this pattern being more pronounced in small firms and in high-tech sectors (Canepa and Stoneman, 2007;Hall, 2008;Hottenrott and Peters, 2012).
The implicit assumption behind this preferred focus of analysis is that it is essentially access to finance, financial uncertainty and information asymmetries that reduce the financial returns of R&D investments and the ability to attract external funds, thus reducing incentives to invest in R&D.
A few recent contributions have extended the analysis to non-financial obstacles to innovation, drawing primarily on evidence from innovation surveys, which allow the effects of knowledge-related obstacles (e.g., shortage of qualified employees, lack of information on technology and markets), market-related obstacles (e.g., lack of customer interest in innovative products, markets dominated by large incumbents), and barriers attributable to the need to fulfil national and international regulations) to be examined. Moreover, these innovation surveys allow researchers to look beyond the mere decision to invest in R&D and to take into account innovation outputs, such as the introduction of a new (to the market or to the firm) good or service or a new process. All these contributions point equally to the importance of the lack of access to finance and the lack of market responses to innovation.

Main conjectures
Overall, the implicit assumption behind the "bias" toward technology-push perspectives within the innovation literature is that firms plan their innovation investments in a context that is structurally and indefinitely capable of absorbing the outcomes of innovation, much in line with a blind trust in a sort of Say's Law 1 for innovative products.
This would apply both at the general macro-economic levelthat is, a general state of dynamism of aggregate consumptionand at the micro-level of analysisthat is, for the specific product/service/sector that has been introduced onto the market.
Without seeking to test the technology-push and demand-pull hypotheses empirically, here we contest this assumption and claim that if easy access to finance and the availability of 1 Put simply, Jean Baptiste Say claimed that "supply always creates its own demand"i.e., markets are able to infinitely absorb any quantity of production. The Keynesian framework overall rejected Say's Law. Here we might stretch the argument and argue that in the case of innovative products, the uncertainty of whether the launch of new products or services is going to be adopted by consumers and diffused in the markets is even higher than that affecting standard plans of production.
funds are important conditions to implement innovation investment plans, trust and positive expectations regarding the state of demand are necessary conditions for firms to enter the innovation contest and initiate innovation investment plans.
Rather than focusing on market structure issues or "lack of customer interest", we turn our attention to firms' perception of the state of demand in terms of both the lack of demand tout court and market uncertainty. As far as the latter is concerned, we are aware that some scholars (see, for instance, Czarnitzki and Toole, 2011 and 2013) have analysed the effect of market uncertainty on R&D investment behaviour from a real option theory perspective, finding that uncertainty causes a fall in R&D investments, albeit mitigated by patent protection (Czarnitzki and Toole, 2011) and firms' size and market concentration (Czarnitzki and Toole, 2013).
Here we take a more heuristic approach to uncertainty and one that is more data driven, with the aim of testing whether firms' self-reported perception of market uncertainty 2 affects their investment behaviour. Specifically, we examine whether the decision to invest in R&D and the amount of investment in R&D are affected by perceptions of these two demand-related obstacles over time and we empirically test this within a panel econometrics framework, as detailed in the next section.
Further, an important added value of this paper is the analysis it undertakes of possible sectoral differences in the way demand affects firms' propensity to invest in R&D 3 .
Our conjecture is that service firms are substantially more sensitive to the state of demand 2 As explained in Section 3, information on market uncertainty is based on responses to a specific question formulated in terms of whether "uncertain demand for innovative goods or services" is perceived as a barrier to innovation. We believe that despite the qualitative, self-report nature of the information provided by this question (in common with all CIS-based evidence), it allows us to draw a plausible picture of firms' responses to increasing levels of (perceived) uncertainty.
when planning their innovative strategies. This is in line with much of the literature on innovation in services (for a review, see Gallouj and Savona, 2009), which claims that the importance of customers and user-producer interactions in services is substantially higher than in manufacturing sectors. Accordingly, we empirically test the conjectures above for both the whole sample of firms and for sub-samples of different macro-sectors, as explained in detail below.

Data
We draw on firm level data from the Spanish Technological Innovation Panel  It is common practice in the innovation literature to focus on private manufacturing and services companies and to exclude public utilities and primary activities owing to differences in the regulatory framework in which they operate. In the case of M&A transactions, firms were eliminated from the sample in the years following the merger or acquisition.

Econometric strategy and variables
As discussed above, the main aim of this paper is to assess empirically whether and, if so, how demand-related obstacles to innovation affect two important innovative decisions taken by firms: their propensity to engage in R&D and, conditional on that, the level of investment in R&D. As stressed by a largely consolidated stream of literature, innovation and, in particular, R&D activities are processes that present high degrees of cumulativeness and irreversibility and, as a result, are characterised by a high level of persistence (see Atkinson and Stiglitz, 1960;David, 1985;Dosi, 1988;Cefis and Orsenigo, 2001). This evidence is fully supported by our data. Indeed, if we examine the transition probabilities of engaging in R&D activities (see Table 2) it emerges that almost 86% of R&D performers in one year retained this same status during the subsequent year. This percentage rises to 91% in the case of non R&D performers that did not change their status into the next period.

< INSERT TABLE 2>
This evidence suggests that the use of an autoregressive specification for the two decisions taken by a firm in relation to its R&D activities is the most suitable. Accordingly, our empirical strategy is based on the estimation of the following two equations: public support for innovation is included.
A one-period lagged value has also been considered for two indicators of whether the firm makes use respectively of patents and informal methods (registration of design, trademarks, copyrights) to protect its innovations 6 . In this case, the rationale is that the positive impact of the mechanisms of appropriability used by a firm take time to make themselves manifest.
We also use a variable recording a firm's age to control for age related effects. The In the case of the demand-related obstacles, in line with the discussion in Section 2 and the rationale underpinning this, we single out two binary variables that identify an increase (over a yearly base) in the degree of importance (irrelevant, low, medium, high) that the firms assign to the following two barriers specified as "uncertain demand for innovative goods and services" and "lack of demand for innovation" 8 . Finally, we control for possible additional negative effects of other obstacles to innovation, including a dichotomous variable recording an annual increase in the importance of the firm's level of perception of the remaining obstacle categories (cost and knowledge related obstacles, market dominated by established firms). Table A1 in the Appendix shows the list of variables, their acronyms and a detailed description.
As for the econometric methodology, in order to estimate equations (1) and (2) where ̅ refers to the within mean of the vector of explanatory variables and embodies the elements that are correlated with , while (with j = 1,2) are the initial conditions of the dependent variables that are supposed to be correlated with the individual error term.
The new equations (1) and (2), obtained by replacing the individual error terms (with j= 1,2) in the right-hand side of equation 3, are estimated using standard random effects 8 We opted to use these constructed variables in light of the high within-variation of the obstacle variables.
However, by construction, the variables take the value 0 in the case of firms persistently assessing the two barriers as highly relevant. We therefore perform robustness checks by considering instead two dichotomous variables taking the value 1 when a firm evaluates as highly relevant the lack/uncertainty of demand and 0 otherwise. The results shown in tables A3-A4 and A5 in the Appendix are remarkably consistent with those discussed in Section 5.2.

Descriptive statistics
One of the conjectures forwarded in this paper is that a firm's sectoral affiliation is a major determinant of the nature and dimension of the effects of demand obstacles on its innovative behaviour. Following the classification proposed by Eurostat and based on an aggregation of NACE manufacturing and service sectors, we identify four macro-categories: high/medium-high tech manufacturing industries (HMHt), low/medium-low tech manufacturing industries (LMLt), knowledge-intensive services sectors (KIS) and less knowledge-intensive services sectors (LKIS).  Table 4, which report the mean values (in percentages) of the two demand-related obstacles by year and sectoral categories. As is apparent, though, these variables show considerable within variation.

< INSERT TABLES 3 AND 4>
Our examination of possible sectoral specificities in terms of a firm's characteristics (see Table 5 for the summary statisticsmean and standard deviationof the variables presented above) reveals that some of the differences are in line with expectations.
Specifically: 1) HMHt and KIS firms appear to be more likely to engage in R&D, to invest more in R&D and to have a higher probability of receiving subsidies for their innovation activity (in line with the previous discussion) than do the other two categories; 2) firms in the manufacturing sectors show a much higher propensity to export than those active in the services sectors; 3) while no striking sectoral differences emerge with respect to the firm's propensity to use informal methods of protection (the lowest percentage being associated, as expected, with LKIS firms), HMHt firms are much more likely to protect the results of their innovation activity by means of patents than are the firms operating in the other sectors (with only 5% of LKIS firms resorting to appropriability methods of this type). If we examine the remaining variables, on average 37% of the observations refer to firms that are part of an industrial group: this percentage ranges from 34% for firms in the LMLt category to 42% for those in the MHMt group. Finally, turning to the variable ln(Size) and ln(Age), on average, firms acting in the KIS sectors appear to be younger and smaller than their counterparts in the other sectoral categories 9 . Table 6 reports the mean values of the variables for the four different firm types identified by taking into account their "demand obstacle status". More specifically we distinguish those firms that did not experience an increase in the degree of relevance assigned to either of the two obstacles, from those that report an increase in the degree of importance of only the lack of demand obstacle; only the uncertainty demand obstacle; or both types of demand obstacle. We find that firms belonging to the first category appear to present quite distinct characteristics from those presented by firms in any of the remaining groups.

< INSERT TABLE 5>
Specifically, firms that did not report any increase in the degree of relevance assigned to either of the two obstacles present higher values for all the variables considered, with the exception of the variables of other obstacles and sales growth. In contrast, and as expected, firms presenting positive values for the demand obstacle variables appear to be less R&D oriented (both in terms of the probability of conducting the activity and the level of investment) than their counterparts, and this is particularly true in the case of firms that report an increase in the level of importance of the lack of demand obstacle. This evidence is largely robust across the four sectoral categories. Albeit solely at the descriptive level, this evidence seems to suggest that, regardless of the sector, demand conditions play an important role in 9 It is worth nothing that, since we use panel data, the revealed negative relationship between R&D and age might be due to a survivorship bias. Indeed, as the subsequent surveys can only account for firms that have survived until the date of data collection, the probability that the resulting sample may be biased towards the more successful companies is not negligible. This could be particularly true for new born and young firms which are more likely to be affected by early failure.
affecting innovative firms' decisions. We test this in an econometric framework in the next section.

Econometric results
The estimation results for the propensity to engage in R&D (probit estimations) and for the amount of expenditure dedicated to R&D (tobit estimations) for the whole sample are reported in Table 7. The table shows the estimated parameters of the main variables of interest, the demand obstacles, and the control variables.
The results for the control variables present the expected signs and significance. First, both R&D decisions (whether or not to invest and how much to invest) appear to be highly persistent over time as the parameters for the initial value and the lagged dependent variables are positive and highly significant. Second, in both estimations, the traditional firm characteristics affecting decisions related to R&D expenditure present the expected sign.
Larger firms that conduct business internationally are more likely to carry out R&D activities and to devote more resources to them. Moreover, although the literature is not unanimous on this point, our results suggest that there is a negative and significant relationship between age and R&D, so that younger firms are more likely to carry out R&D activities. Third, other variables that characterise the innovation behaviour of firms, including the use of intellectual property rights and being recipients of public subsidies, also have a positive effect on R&D investments. Finally, while firms with higher levels of sales growth are more likely to engage in R&D and to invest more in R&D, the increase in the perception of other obstacles to innovation exerts, as expected, a negative and highly significant effect on both decisions taken by the firm.
The results of the estimations (Tables 8 and 9)  show that the negative effect is clearly dominant, suggesting that rather than uncertainty with regard to the demand for a single product or for a specific portfolio of products, it is the general macro-economic condition and, therefore, expectations regarding the aggregate state of the economy that affect firms' R&D strategies. This confirms our conjecture that, especially in time of crisis, demand-pull perspectives on innovation should be revisited and made better use of for (macro) policy purposes. We will return to these considerations in the concluding section.

Uncertainty, lack of demand and R&D strategiessectoral specificities
The estimations carried out for the four groups of sectors (Tables 8 and 9 Our main conjecture is that the size of the destination market and expectations regarding profitability (that is, the perceived lack of demand and of market dynamism) are likely to have impacts other than the mere uncertainty regarding the propensity to engage in R&D and the intensity of that engagement. While the former reflects a general trust in the state of the economy and is, hence, more of a macro-condition that firms need to verify, the latter is a micro-condition concerning the specific characteristics of the product and, hence, the actual user needs that the product is supposed to satisfy. Our claim, for which we provide empirical support, is that a lack of trust in the macro-condition of demand's dynamism represents more of a deterrent for firms to even engage in innovative activities, whereas uncertainty regarding the specific demand and user needs, while still being a deterrent, are likely to be incorporated in the firms' specific R&D plans.
We have found support for this conjecture. From our analysis it emerges that while the perception of an increasing lack of demand has a significant, strong and negative effect on both the decision to invest and the amount of investment in R&D, increasing demand uncertainty does not seem to have any significant effect or to have a weakly significant positive effect (Stein and Stone, 2013). Part of the demand uncertainty might therefore be already "incorporated" in the strategic horizon of firms' decisions when they engage in an intrinsically risky and uncertain activity such as R&D.
These findings contribute to the debate on demand-pull and technology-push approaches in innovation studies from a radically novel perspective, namely, that of barriers to innovation.
The literature on barriers is increasingly important due to its obvious policy relevance.  Note: the final sample only comprises firms for which a lag of the dependent variable is available. This implies that t=2 refers to firms that are observed for at least three periods, t=3 corresponds to firms that are observed for four periods and so on.       0.000 0.000 0.000 0.000 Notes; ***, ** and * indicate significance on a 1%, 5% and 10% level, respectively. Standard errors in brackets. Time and industry dummies are included. Columns 1-3 report marginal effects.  Other obstacles Dummy=1 if the firm reports an higher degree of importance (from period t to period t+1) for at least one of the remaining obstacles variables; 0 otherwise

Independent variables (Obstacle demand variables)
Lack of demand Dummy=1 if the firm reports an higher degree of importance (from period t to period t+1) for the obstacles variables "it was not necessary to innovate due to the Lack of demand for innovation"; 0 otherwise Uncertainty Dummy=1 if the firm reports an higher degree of importance (from period t to period t+1) for the obstacles variables "Uncertain demand for innovative goods or services"; 0 otherwise Notes; ***, ** and * indicate significance on a 1%, 5% and 10% level, respectively. Standard errors in brackets. Time and industry dummies are included. Columns 1-3 report marginal effects 0.000 0.000 0.000 0.000 Notes; ***, ** and * indicate significance on a 1%, 5% and 10% level, respectively. Standard errors in brackets. Time and industry dummies are included. Columns 1-3 report marginal effects. 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Notes; ***, ** and * indicate significance on a 1%, 5% and 10% level, respectively. Standard errors in brackets. Time and industry dummies are included. Columns 1-3-5-7 report marginal effects. p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Notes; ***, ** and * indicate significance on a 1%, 5% and 10% level, respectively. Standard errors in brackets. Time and industry dummies are included. Columns 1-3-5-7 report marginal effects.