I’ve recently been working with Anne Marie Knott on her website and wanted to share part of her project here.

Sustainable Innovation and Growth

My overarching research goal is defining firm strategies and market conditions wherein firms operating in their own self-interest generate sustainable innovation and economic growth. An important component is entrepreneurship. Thus my work considers both established firms as well as entrants.

This interest in innovation stems from fifteen years of defense electronics R&D at Hughes Aircraft Company. Since I was equivocal about working on missile guidance systems, I sublimated: “I’m not really building weapons,…I’m pushing the knowledge frontier” Thus I bristled at government policies and firm strategies that appeared to compromise frontier growth. As examples, it seemed ex post “second sourcing” of production contracts removed firm incentives to perform their own R&D, the post-cold war R&D cuts would cause the industry to dismantle much of its R&D capability, and the post-cold war industry consolidation would reduce the level of technological competition and thus the level of innovation. I wanted to understand the extent to which these intuitions were correct. In the process I hoped to develop a better understanding of “innovation physics”, and ultimately devise firm strategies and government policies better exploiting the physics.

I began by trying to understand basic principles of firm knowledge: where does its value come from, how is it extracted, what causes it to erode, how do firms compensate for that erosion. Once I understood these basic principles I built a computational model to capture the phenomena and generate new insights about the nature of technological competition among a stable set of firms. The model yielded new hypotheses about the impact of market structure on steady-state innovation. The third stage of my research empirically tested the new hypotheses in established industries. These industries aren’t typically in the innovation limelight, but they are important because they comprise the bulk of US R&D spending. This stage was followed by a shift in focus from a stable set of firms to examine the impact of entrants. The consideration of entrants into established industries is important because of its prevalence: new firms enter established industries at an annual rate of about 11% of incumbents (firms also exit at a rate of 10%). Furthermore the topic is interesting because there is little theory addressing this churning. This fourth stage looked both at the causes and innovative consequences of churning. An additional feature of this stage is that it introduced theory and empirics linking individual behavior to market structure and outcomes. Currently, I’m extending my work to look at the impact of regulation on churning phenomena.

The aggregate contribution of my work is a parallax view of what drives innovation in established industries. Four main and mutually reinforcing insights emerge from this view. The first insight pertains to sources of competitive advantage. A dominant view in strategy is that competitive advantage emerges from position—either choice of industry or control of resources within industry. The implicit prescription is that firms need merely to find the valuable position, then “rest on their laurels”. My work reinforces the importance of position. I find that knowledge resources are valuable, and the market shares arising from those resources are remarkably stable. However I also find that in most settings the value of resources erodes rapidly due to imitation and displacing innovation (obsolescence). Therefore significant managerial effort is required to extract and replenish the resources’ value. Accordingly, it is managerial effort to renew resources, rather than properties of the resources, which provides the isolating mechanism (1) sustaining firm position.

The second insight pertains to innovation stimuli. Important views within economics and management hold that the exploitation or diffusion of knowledge inhibits innovation or exploration for new knowledge. In economics the tradeoff stems from an assumption that diffusion of knowledge reduces the returns to R&D and therefore the incentives to conduct R&D. I find instead that under many conditions diffusion of knowledge actually stimulates innovation. This occurs because the diffusion erodes the value of leaders’ positions which in turn stimulates investment to restore those positions. Thus innovation and imitation appear to be complements at the market level.

A similar view in management holds that firms must choose between exploitation of existing capabilities and exploration for new capabilities—that returns are greater to specializing in either activity than to pursuing them jointly. Following logic similar to that at the market level, I conclude that greater ability to exploit an innovation (through scale and scope) increases firm incentives to engage in exploration. This suggests exploration and exploitation are also complements at the firm level. Thus at both the market and the firm levels, the best structures for generating new innovation appear to be ones in which existing innovations are fully exploited. This is because exploitation sets up the incentives for replacement innovation. In addition to the insight itself, my work helps to define those structures.

The above dynamics rely on persistent heterogeneity of firms as the fuel for the innovation/exploitation cycle. While I can show conditions under which initial heterogeneity is preserved even when firms have identical capability and rules of behavior, a nagging concern is where the initial heterogeneity comes from. A third insight from my work is that firms differ at a more fundamental level than previously recognized. Just as individuals differ in their ability to process and generate knowledge (IQ), so do firms. These IQ differences yield persistent differences in firm behavior and position.

The final insight is that entrants are an important source and stimulus to innovation even in established industries. Thus free entry while wasteful in a static sense (share stealing) is beneficial in a dynamic sense.

These insights are developed in greater detail below with links to the corresponding papers.

Managerial Effort is Required to Extract and Replenish the Value of Position.

The first insight is that managerial effort is required to extract and replenish the value of position. This insight has three components. The first is that position matters (that resources are valuable). The second is that its value erodes under competition, and the third is that managerial efforts are required to both extract value and replenish it. The primary evidence for each of these comes from a set of studies exploiting the notion of franchises as a market for organizational routines (the franchisor sells an organizational routine to franchisees). These studies are conducted in a unique setting (quick printing) that has nice experimental properties: three governance forms: franchisees, independents and company-owned establishments, a commodity product (copies), large size (20,000 firms), and maturity (founded in 1966). Each of the studies characterized the franchisor’s routines, and demonstrated that the routine generated persistently higher owner-income than that earned by independent establishments, even after controlling for brand and franchise scale. Thus position (arising from the organizational routine) matters.

The puzzle in the first paper, “The Organizational Routine Factor Market Paradox”, was how the value of the routine was sustained given that it was fully articulated in operations manuals and therefore easy to diffuse. I found that franchisees when left to their own devices, preferred to deviate from routine (both by “enhancing” the routine and by shirking), thereby diminishing its value. Thus an important role of the franchisor was discipline to enforce the routine. This was true even though the franchisees were owner-managers and therefore had perfect incentives.

These results were cross-sectional, so a second study, “Dynamic Value of Hierarchy”, examined implications over time to answer two questions. First, once franchisees have assimilated the routine, do they really need managers? Second, shouldn’t discipline be costly to the long-term survival of franchises, since it inherently narrows the variance required for adaptation? To answer these questions, I conducted a natural experiment comparing the behavior and performance of owner-managers who leave a franchise, with those who remain. I found that former franchisees abandon elements of the routine and as a result their income decays. This confirms the results from the first study that discipline matters. I also found another form of “decay”–that relative to an improving standard of best practice. Once owners leave the franchise they are slower to adopt industry innovations. Thus there are two forms of erosion: behavioral drift and resource obsolescence. Managerial efforts combat both of these to extract and replenish the value of their position. Thus managers matter. In fact, in “Nirvana Efficiency: a Comparative Test of Residual Claims and Routines”, Bill McKelvey and I show the contribution of managers in creating and enforcing routine has four times the impact of perfect incentives in explaining firm performance variance.

These studies examined behavior directly but in a setting not typically associated with innovation. In another study, “On the Accumulation of Intangible Assets”, with David Bryce and Hart Posen (then PhD students), we look at positions associated with firm R&D assets. The issue of interest was similar to that in the first paper–what protects the value of these resources. In particular we tested the hypothesis that it was the resource accumulation process that inhibited imitation–that in order to achieve a leader’s position an imitator would need to replicate the entire accumulation process. To test that hypothesis we empirically characterized the value of the R&D stocks as well as the accumulation process. As in the first study we found that the R&D resources were valuable and that the relative positions of firms arising from those resources were durable. So again position and resources matter. We also found however that the value of R&D assets depreciates rapidly. The bulk of R&D investment each year was merely that required to compensate for depreciation. Accordingly, the accumulation process is not an isolating mechanism. There are no asset mass efficiencies—entrants could catch up to leaders in three years if they matched leader investments (2). Rather, the isolating mechanism is managerial effort to continually replenish depreciating R&D stocks.

Finally, in “Persistent Heterogeneity and Sustainable Innovation”, I constructed a computational model of the dynamic conditions under which firm positions (relative market shares) are sustained in perpetuity. I find that while positions are durable, they are not static. Persistent share differences make it appear that firms are “resting on their laurels”. However, the firms with the largest shares are the ones investing the most to maintain their shares.

Innovation (Exploration) and Diffusion (Exploitation) are Complements

The second insight is that innovation and diffusion are complements. This insight emerges from the prior insights: if resources erode and if managers must work to replenish them, then diffusion may be a stimulus to innovation. “The Dynamic Value of Hierarchy”, discussed earlier, provided preliminary evidence that innovation and diffusion are compatible. Managers were both enforcing existing franchise routine and introducing innovations to that routine. However there was no evidence of a complementarity. In other words there was no evidence that enforcing the routine aided innovation, or vice versa.

That evidence came from another study, “Exploration and Exploitation as Complements”. In that study I examined product development in the auto industry. I first tested for the coexistence of exploration and exploitation by examining whether auto manufacturers were both decreasing cost and increasing quality across product developments, which indeed they were. I next tried to determine if there was complementarity. To do so, I conducted a case study of innovation within Toyota. Examination of Toyota’s output and innovation history revealed the company had a sophisticated exploitation structure: 1) large scale of each product line, 2) similar interfaces across product lines (so an innovation in one line could be applied to most others), and 3) heterogeneity in product quality across product lines (vertical differentiation). This system supports intertemporal price discrimination for each innovation. Each innovation is introduced first as an option on high-end models, then is gradually cascaded to become a standard feature on high-end models as it becomes an option on lower models. This rich exploitation structure increases returns to each innovation. This increases incentives for exploration as well as the means to fund it.

Having demonstrated complementarity at the market and firm levels, the next challenge was to formalize it in a computational model. I modeled a population of firms where each firm could increase its knowledge each period in two ways: spillovers from randomly encountered rivals with superior knowledge, and creation of new knowledge. I expected the industries to converge to a steady-state where all firms had the same knowledge and innovation stopped. Instead I found that in approximately half the conditions the industry sustained innovation. Those conditions were characterized by many firms, a high degree of heterogeneity in their knowledge, and relative ease of transferring knowledge between firms. The model’s predictions differ from those in IO regarding the conditions facilitating innovation. In particular a common view is that expected returns to R&D decrease with the number of competitors and the ease of imitation. The logic reflects a forward-looking view of managerial decision-making. Managerial behavior is driven by the “carrot” of future profits. In contrast, the model I set forth reflects a “rearward” view of decision-making, responding to the “stick” of realized threats. Since the models rely on distinct assumptions, resolution is largely an empirical matter.

To date tests of market structure and innovation had been inconclusive. In large part this was because there hadn’t been good measures for two factors affecting innovation: technological opportunity and appropriability. Accordingly, Hart Posen and I developed new measures of these constructs and used them to test competing hypotheses on market structure and innovation. Those tests have largely supported the model’s inferences. “Firm R&D Behavior and Evolving Technology” examines a longitudinal sample of firms in twenty five R&D intensive industries and finds that innovation is increasing in expropriability (productive use of rival R&D), the number of firms, and the heterogeneity across firms. “Is Failure Good?” also with Hart Posen (then a PhD student), looks across markets within a single industry and also finds that innovation increases with the number of firms and their heterogeneity. Thus the empirics tend to support the rearward-looking model of innovation.

The model and empiricism just discussed pertain to the inter-firm dynamics. My earlier studies in franchises, in computers and at Toyota suggest similar dynamics occur inside the firm. “Firm Decomposition and Innovation” examines intra-firm structure and R&D dynamics. In results similar to those at the market level, I find that increasing the number of self-similar units within a firm and the degree of heterogeneity across those units increases the rate of innovation.

Fundamental Source of Persistent Heterogeneity

The work above demonstrates that persistent heterogeneity is a fuel for continuous innovation. Moreover it demonstrates that persistent heterogeneity is possible even if firms have identical capability and follow identical behaviors. This raises two new questions: 1) Where does the initial heterogeneity come from, and 2) Why should firms follow identical behaviors–why don’t they try to gain advantage, e.g., why don’t laggards match leaders’ investments for three years so that they become the leaders?

The naïve argument I set forth at the time was that laggards couldn’t afford to match leaders’ investments. But capital markets should fix that. In more recent work, “R&D-Returns Causality: Absorptive Capacity or Organizational IQ”, I find that firms differ at a more fundamental level. They don’t have identical capability. Rather they differ in something akin to individual intelligence–their ability to create new knowledge (firm IQ), measured as the output elasticity of R&D. Firms with higher IQ spend more on R&D because it is optimal to do so. Thus not only do firms differ in capability, those capability differences yield behavioral differences. What I also found however is that firm IQ is inversely related to imitation capability (output elasticity of spillovers). Firms with high IQ gain almost no benefit from rival R&D (a result that is diametrically opposed to predictions from absorptive capacity), whereas firms with low IQ tend to be extremely productive with rival R&D. These fundamental differences in capability and their corresponding optimal behaviors provide a theoretical basis for the leader/follower typologies that have existed since the strategy field’s inception.

Not only is the IQ measure important theoretically, it also makes a methodological contribution. Since the publication of “The Rate and Direction of Inventive Activity” in 1962, the dominant measure of R&D effectiveness has been patent counts. However patent counts suffer from problems of: universality (fewer than 50% of firms conducting R&D have any patents), uniformity (there is substantial variance in patents’ economic value) and reliability (tests of the market value of patents yield anomalies). “IQ and the R&D Market Value Puzzle” with Carl Vieregger and James Yen (both PhD students) documents these problems and demonstrates that the IQ measure resolves them. For the same reasons the measure is valuable to academics, IQ is also valuable to practitioners. It allows firms both to set the optimal R&D investment (currently set using rules of thumb) and to measure the effectiveness of that investment. “New Hope for Measuring R&D Effectiveness” is an article solicited by Research and Technology Management to encourage use of the measure among practitioners. My hope is that use of the measure will improve R&D effectiveness just as TQM improved product quality, and hospital report cards reduced mortality and morbidity.

Entrepreneurship: The Role of Small Firms and Entrants in Innovation

So far the discussion of innovation has focused on a stable set of competitors. I feel this is the more important case because large firms comprise the majority of US R&D, yet there is a pervasive view is that large firms are antithetical to innovation. More recently I have begun complementing the large firm work with work examining the innovative role of small firms and entrants. This new work suggests small firms and entrants play a unique role, but one anticipated by my prior work.

The quintessential entrepreneurial milieu is Silicon Valley. Accordingly a number of scholars have examined its advantages for innovation and economic growth. The insight I find most compelling is Saxenian’s (1994) observation that Silicon Valley firms are less vertically integrated than their counterparts in Boston’s Route 128. My work (Firm R&D Behavior and Evolving Technology and Persistent Heterogeneity and Sustainable Innovation) discussed previously provide some theoretical underpinnings to this observation. Vertical disintegration is a means of increasing the number and heterogeneity of firms. Doing so accelerates the diffusion/innovation cycle.

The importance of heterogeneity in the model and empiricism further suggest that the best environments for innovation are those combining large firms and small firms (3). The large firms create a pool of spillovers to augment the innovative capability of small firms. In return, innovation by the small firms erodes leaders’ shares and forces innovation they might not otherwise undertake, something demonstrated vividly by the advent of the internet. Entrepreneurial firms were first to introduce online channels for retailing and news delivery, but their presence forced adoption by incumbent bricks and mortar retailers and printed newspapers.

This dynamic also explains a long-standing puzzle in the innovation literature, the “firm size puzzle” that large firms spend proportionately more on R&D than small firms, yet have lower R&D productivity. Lower spending and higher productivity of small firms are the joint consequence of asymmetric spillovers, where spillover asymmetry refers to the fact that firms can only gain useful knowledge from firms who know more than they do. We characterize and empirically test this in “Spillover Asymmetry and Why It Matters” (joint with Hart Posen and Brian Wu, then PhD students). Leader firms by definition are at the technological frontier. Therefore knowledge held by other firms is likely redundant. Accordingly the leader derives no benefit from spillovers, and can only achieve innovation through its own investment. In contrast, laggard firms have less knowledge than rivals and thus have greater opportunity to free-ride on the innovation of others. Measures of productivity which ignore this free-riding will overstate the productivity of R&D.

The issue of small firms leads naturally to that of entrants. There is a substantial entry even in established industries. On average new firms enter industry at the rate of 11% of the number of incumbent firms. Typically 51% of those firms fail within four years. This raises three interesting questions: 1) Why are firms entering if the most likely outcome is failure–particularly given the fact that tests of entrepreneurs show them to be risk averse, 2) Is this churning strictly wasteful or are there economic benefits to the excess entry, and 3) If there are economic benefits, are there policy implications for conditioning the extent and type of entry.

“Entrepreneurial Risk and Market Entry” with Brian Wu (then a PhD student), tackles the first question. We build a reduced form model of the entrepreneur’s entry decision which takes into account two types of uncertainty: demand uncertainty and ability uncertainty (ala Jovanovic). We find that entrepreneurs are risk averse with respect to demand variance, but are “risk seeking” or overconfident with respect to ability variance. Thus they are willing to enter markets with high levels of demand uncertainty so long as there is high dispersion in ability. A potential market failure (if entry is good) exists in markets with high demand uncertainty but low ability dispersion.

“Is Failure Good?” with Hart Posen (then a PhD student), tackles the second question. We find that excess entrants increase the efficiency of markets through three mechanisms: selection effects (highly efficient entrants displace less efficient incumbents), excess competition that stimulates investments by incumbents (as in the work discussed above), and through spillovers–incumbent assimilation of the innovations and resource enhancements of the failed firms.

Given entry generates these social benefits, there is an issue of their cost. Since the dominant outcome is failure, these benefits come at the private cost of entrants. In essence entrepreneurs are martyrs. Their overconfidence makes them willing martyrs, but there is a question of whether we can improve welfare by reducing their private costs. One way to do that is to expedite the exit of would-be failures. In “No Exit: Failure to Exit under Uncertainty”, Dan Elfenbein and I examine the timing of entrepreneurs’ exit relative to the optimal time defined by their information stream. We find that approximately 85% of exit occurs beyond the optimum. Moreover we find that the delay is due both to high thresholds for avoiding Type I error (exiting if you ultimately would have been profitable) as well as asymmetric updating from positive information versus negative information.

The caveat for all three studies is that the setting is banks, where entry is highly regulated–only entrepreneurs with substantial industry experience and financial backing are awarded charters. Accordingly in my most recent work, “The Schumpeterian Cost of Regulation on Entry and Innovation: The Case of Bail Bonds”, Erin Scott (a PhD candidate) and I examine a setting where regulations vary more substantially across markets. Here we consider the causes and consequences of entry simultaneously. One advantage of this setting is that the feature of markets which attracts entry, (potential profits), differs from the social benefits of entry (incarceration rates, failure to appear rates, and fugitive rates). We estimate a structural model of entrepreneurs’ entry decisions and incumbents’ exit decisions to conduct counterfactual simulations of unobserved market conditions (in particular unregulated markets). We find that regulation reduces entry but increases innovation relative to unregulated markets. This is exciting because these Schumpeterian benefits of innovation are above and beyond the intended benefits of regulation in the setting (correcting problems of asymmetric information and rights abuses). Combining results across the entry studies suggests that churning is good, but it is unlikely to improve welfare unless entry is restricted to qualified candidates. We also understand market conditions that increase churning and its benefits while reducing private costs.

Other Significant Streams

Induced Discrimination

This work comprises a minor research stream on an issue I found as compelling but more narrow than the topic of innovation dynamics. The issue is hiring discrimination induced by the Civil Rights Act of 1991 (CRA91). The issue came to my attention through conversations with an entrepreneur who had been sued by an employee for stress. His response was to minimize litigation risk by avoiding employees from protected groups, since they have more grounds for suit. If he was representative of other employers, this suggested the impact of CRA91 was antithetical to its intent.

I first examined whether the phenomenon existed on a wide scale, “Induced Discrimination and Firm Size: Information versus Incentive Effects”, and confirmed that CRA91 reversed a trend of increasing employment for protected groups. Second, I designed a mechanism to reverse the problem. The mechanism assumes that employers’ goal is to avoid litigious employees. Since employers currently have no means to identify litigious employees, they do the next best thing, which is to avoid employees that appear to be high risk–those in protected groups. To solve the problem, I developed a self-selection mechanism through which prospective employees reveal their litigiousness. Finally, I tested the mechanism via lab experiment, “Reversing Induced Discrimination: Theory and Experiment”. The experiment revealed a) that there is a litigious personality trait, and b) that using the mechanism to screen employees reduces litigation 85% while increasing protected group employment 3%, without imposing net new costs. Thus it removes employer incentives to discriminate while retaining penalties for employer misconduct. The Center for Technology Transfer at the University of Pennsylvania filed a patent for the mechanism, but later abandoned it. I subsequently filed a patent for an invention that improves upon this prior art, and commercialized it via the venture, Equity Benefits, LLC. Starting the venture allows me to practice what I preach in my entrepreneurship classes and text, “Venture Design”.

Design

Venture Design is an entrepreneurship textbook I developed to tackle the problem students had applying their Wharton core material to the “unstructured” problem of designing a venture. The text structures the problem by walking through each of the strategic decisions necessary for a new venture and integrating the decisions into a cohesive venture design. For each decision, the text discusses the theoretical foundations, describes a hands-on tool, and applies the tool to a case that is carried throughout the text. The most visible output of the process is an information intensive and analytically rigorous business plan. However the most useful output is a spreadsheet “simulation” stretching from demand curves for each product feature to pro forma financial statements for the venture as a whole. This simulation encourages experimentation to produce a better design pre-launch and to facilitate more intelligent adaptation post launch. The first edition was released December 2001 and has been adopted at major universities in the U.S. and Europe: Baylor, Cornell, Georgia State, Humboldt Universitat (Berlin), NYU, Penn State, Purdue, Seton Hall, SMU. Additionally, the book was a finalist for the 2002 USASBE Award for Innovative Entrepreneurship Pedagogy. Sage released a second edition December 2007 which bundles online software to make the central analysis (conjoint) accessible to a broader audience.

The design orientation evident in the text likely stems from undergraduate training in architectural design as well as several years experience in design of missile guidance systems. While the text is the most overt manifestation of that orientation, an interest in design is evident in my research as well. For example, as stated in the introduction, the goal of my work on R&D is to design firm and market structures that better exploit the “physics” of knowledge flows. Other examples of the design orientation are the mechanism design to reverse induced discrimination, the intrafirm heterogeneity work that generates an organization design, and the entrepreneurial churning work that suggests entry regulation design.

Finally, I tackle design more directly in the paper, “Trick My Routine: Redesigning Routines for Replication”, with Anuja Gupta and David Hoopes. In that work we challenge the view that routines are emergent with one suggesting they should be designed. We then characterize the design process for a firm trying to scale a new routine for single unit operations to one suitable for replication across a chain. We find that headquarters functions whose primary responsibility is maintaining and improving existing routines have difficulty assessing the design tradeoffs associated with scaling a new routine.

Current Research and Future Directions

To date my work has primarily examined innovation at the market level. This was necessary because the central constructs, e.g., isolating mechanisms, erosion and competition, involve interactions between firms, and thus are inherently defined at the market level. My future work continues with some market level issues, but focuses principally inside the firm.

The first stream of work examines whether there is an intra-firm analog to the market dynamics I modeled in Persistent Heterogeneity and Sustainable Innovation and tested in Firm R&D Behavior and Evolving Technology. What happens when there is no competition mechanism? What are the optimal number of units and degrees of heterogeneity? What is the relationship between organization structure and the R&D portfolio?

The second stream of work investigates the underpinnings of firm RQ (research quotient). A byproduct of R&D-Returns Causality: Absorptive Capacity or Organizational IQ is RQ estimates for all publicly traded US firms conducting R&D. An interesting feature of the RQs is they exhibit greater variance within industry than across industry (Reinforcing work by Rumelt 1991 and McGahan and Porter 2002). What I don’t understand however is how high RQ firms differ from their low RQ counterparts. As a first step toward that understanding, I received an NSF grant to identify the antecedents of RQ. That study comprises in-depth qualitative research characterizing the differences in R&D systems and routines of high and low RQ firms. In addition to these rich qualitative comparisons, the study generated a set of factors discriminating high RQ firms from low RQ firms. In a follow-on NSF grant, I’m matching firm RQs to their practices in the NSF Business Innovation and Research and Development Survey (BRDIS).  This study will quantify the contributions of firm practices to R&D effectiveness across the full spectrum of firms engaged in R&D.   My hope is that once firms know their RQ, and understand the factors contributing to it, they will improve it (and accordingly increase economic growth).

Footnotes

  1. Isolating mechanisms are the firm equivalent of entry barriers. They prevent share movements rather than entry.
  2. This finding matches the observation in Dunne, Roberts and Samuelson (1988) that firms who enter new industries at the same scale as incumbents meet or exceed incumbent performance.
  3. Acs and Audretsch find this in their comparison of large and small firm innovation. Large firms are more productive with their R&D, but they are most productive in industries dominated by large firms.