The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
The Geography of Research and Development Activity in the U.S.* BY KRISTY BUZARD AND GERALD A. CARLINO I n the U.S., metropolitan areas contain the largest concentrations of people and jobs. Despite some drawbacks, these so-called agglomeration economies also have benefits, such as the cost savings that result from being close to suppliers and workers. Spatial concentration is even more pronounced among establishments that do basic research and development (R&D). In this article, Kristy Buzard and Jerry Carlino show that geographic concentration of R&D extends beyond locations such as Silicon Valley. In fact, many types of R&D establishments are highly concentrated geographically. Although metropolitan areas account for less than 20 percent of the total land area in the United States, they contain almost 80 percent of the nation’s population and nearly 85 percent of its jobs. Put differently, the United States has, on average, 24 jobs per square mile, but metropolitan areas average about 124 jobs per square mile. This high degree of spatial concentration of people and jobs leads to congestion costs, such as increased traffic and pollution, and higher housJerry Carlino is a senior economic advisor and economist in the Research Department of the Philadelphia Fed. This article is available free of charge at www. philadelphiafed. org/research-and-data/publications/. www.philadelphiafed.org ing costs. Congestion has become so severe in London that in February 2003, the city imposed a fee, currently £8 a day, on all vehicles entering, leaving, driving, or parking on a public road inside the Charging Zone between 7:00 a.m. and 6 p.m., Monday through Friday. New York City recently considered a similar plan. To offset these congestion costs, workers must receive higher wages, and higher wages increase firms’ costs. If congestion costs were the only thing resulting from the spatial concentration of firms, firms could easily disperse to reduce these costs. Yet they do not. This is because the negative effects of concentration make *The views expressed here are those of the authors and do not necessarily represent the views of the Federal Reserve Bank of Philadelphia or the Federal Reserve System. up only one side of the urban ledger. The positive effects of agglomeration economies — efficiency gains and cost savings that result from being close to suppliers, workers, customers, and even competitors — make up the other. Other things equal, firms will have little incentive to move if congestion costs are balanced by the benefits of agglomeration economies. While economic activity tends to be geographically concentrated, spatial concentration is even more pronounced among establishments doing basic research and development (R&D). For example, although the United States has more than 3100 counties, the 50 counties that contain the largest number of R&D labs account for almost 60 percent of all such labs, while the top 50 counties in terms of the overall number of plants across all industries account for only about one-third of all plants. More than most economic activity, R&D depends on a particular byproduct of agglomeration economies called knowledge spillovers — the continuing exchange of ideas among individuals and firms. The high geographic concentration of R&D labs creates an environment in which ideas move quickly from person to person and from lab to lab. Locations that Kristy Buzard is currently a graduate student in economics at the University of California-San Diego. Previously, she was an economic analyst in the Research Department of the Philadelphia Fed. Business Review Q3 2008 1 are dense in R&D activity encourage knowledge spillovers, thus facilitating the exchange of ideas that underlies the creation of new goods and new ways of producing existing goods. Policymakers view the success of areas such as Silicon Valley in California, the Route 128 corridor in Boston, and North Carolina’s Research Triangle as a miraculous recipe for local economic development and growth. But are these examples exceptions rather than the rule? The answer appears to be no. Equally remarkable concentrations may be found in many other types of R&D activity, such as the concentration of R&D in the pharmaceutical industry in northern New Jersey and southeastern Pennsylvania. In this article, we show that many types of R&D establishments are highly concentrated geographically. CLUSTERING OF R&D LABS Some studies have looked at the geographic clustering of economic activity in a particular industry, such as manufacturing or advertising. A study by Glenn Ellison and Edward Glaeser and one by Stuart Rosenthal and William Strange find evidence of geographic concentration of employment in many U.S. manufacturing industries. The geographic concentration of manufacturing jobs is not simply an American phenomenon, as Gilles Duranton and Henry Overman demonstrate in their analysis of manufacturing plants in the UK. A study by Mohammad Arzaghi and Vernon Henderson looks at the location pattern of firms in the advertising industry in Manhattan. They report that Manhattan accounts for 20 percent of total national employment in the ad industry, 24 percent of all advertising agency receipts, and 31 percent of media billings. They show that for an ad agency, knowledge spillovers and the benefits of networking with 2 Q3 2008 Business Review other nearby agencies are large but the benefits dissipate very quickly with distance from other ad agencies and are gone after roughly one-half of a mile. Thomas Holmes and John Stevens take a broader approach. They used employment data for all U.S. industries, not just manufacturing, and information about R&D labs. These data were not available in a machinereadable format. Since the directory lists the complete address for each establishment, we were able to assign a geographic identifier (using geocoding techniques) to 3,129 R&D labs in the U.S. in 1998.3 Policymakers view the success of areas such as Silicon Valley in California, the Route 128 corridor in Boston, and North Carolina’s Research Triangle as a miraculous recipe for local economic development and growth. But are these examples exceptions rather than the rule? The answer appears to be no. not for just a single industry, such as advertising. Among the 15 most concentrated industries, they find that six are in mining and seven are in manufacturing; only two industries fall outside mining and manufacturing (casino hotels and motion picture and video distribution). Our article differs from past studies in two ways. First, rather than looking at the geographic concentration of firms engaged in the production of goods (such as manufacturing) and services (such as advertising), we consider the spatial concentration of private R&D activity.1 Second, rather than focusing on the concentration of employment in a given industry, we look at the clustering of individual R&D labs.2 To do this, we used 1998 data from the Directory of American Research and Technology to electronically code the addresses and other A map of the spatial distribution of R&D labs reveals a striking clustering of this activity (Figure 1). In places that have little R&D activity, each dot on the map represents the location of a single R&D lab. For example, there is only one lab in Montana, represented by the dot in Flathead County. In counties with a dense clustering of labs, the dots tend to sit on top of one another, representing a concentration of labs. A prominent feature of the map is the high concentration of R&D activity in the Northeast corridor, stretching from northern Virginia to 2 The study by Paulo Guimarães, Octávio Figueiredo, and Douglas Woodward is one of only a few other studies we are aware of that look at spatial clustering at the establishment level. Specifically, they look at the geographic concentration of over 45,000 plants in 1999 for concelhos (counties) in Portugal. Duranton and Overman use plant-level data to study the locational pattern of UK manufacturing industries. 3 1 There are a number of other studies that look at innovative output across cities, such as the study by David Audretsch and Maryann Feldman. What is unique about our article is that we present information on local private R&D activity, which no one else has done. Our data on individual labs were limited to the top 1,000 U.S. public companies in terms of R&D expenditure in 1999. The 1,000 firms cover more than 95 percent of all R&D performed by public companies. Many of these firms have multiple labs. For example, the Lockheed Martin Corporation has 54 labs, and General Electric has 26. www.philadelphiafed.org FIGURE 1 Location of Total R&D Labs* * In counties with relatively little R&D activity, the dots on the map represent the location of a single R&D lab. In counties having a dense concentration of labs, the dots represent a concentration of labs. Massachusetts. There are other concentrations, such as the cluster around the Great Lakes and the concentration of labs in California’s Bay Area and in southern California. But some states that account for a relatively large share of the nation’s jobs account for a much smaller share of the nation’s R&D labs. For example, Texas ranks second among states in terms of employment, but it ranks eighth in the number of R&D labs. Similarly, Florida ranks fourth in employment, but 13th in the number of labs. However, as already noted, recent studies have shown that economic activity, especially manufacturing, also tends to be geographically clustered. We will show that R&D activity tends to be more spatially concentrated than total employment or manufacturing employment. There are 3,141 counties in the U.S., and all of them are engaged in some type of economic activity. All but 33 counties are en- www.philadelphiafed.org gaged in some form of manufacturing activity. In contrast, only 519 of these counties have at least one R&D lab, and far fewer counties have a notable concentration of labs. A simple way to quantify the concentrations of R&D relative to establishments in general or to manufacturing establishments in particular is to first compute each county’s share of total R&D labs and rank counties by descending order of this share. Moving down this ranking, we compute a cumulative total for the share of R&D labs. Next, we construct a similar ranking for establishments in general and for manufacturing establishments in particular. The top 50 counties ranked by number of R&D labs account for 58 percent of all R&D labs, while the top 50 counties ranked by number of manufacturing establishments account for only 36 percent of all manufacturing establishments and only 32 percent of all establishments. It appears that R&D labs are more highly concentrated than economic activity in general and overall manufacturing activity in particular. This is important because it means the concentration of R&D labs doesn’t simply reflect the concentration of manufacturing activity. Since R&D is more concentrated than manufacturing activity, this suggests that some factors, such as knowledge spillovers, may be a more centralizing force for R&D than they are for manufacturing activity. WHICH R&D LABS CLUSTER? Paul Krugman and David Audretsch and Maryann Feldman developed a “locational Gini coefficient” to answer the question of which manufacturing industries cluster geographically. A locational Gini coefficient shows how similar (or dissimilar) the location pattern of employment in a particular manufacturing industry is from the location pattern of overall manufacturing employment. It does this by subtracting a county’s share of national employment in manufacturing from the county’s share of national employment in a given manufacturing industry, squaring the result, and summing over locations to arrive at a single number. The squaring of the difference in shares means that larger differences contribute more than proportionately to the overall value of the index. If the squared difference takes a value of zero, employment in a particular industry is allocated across counties in exactly the same way as employment in manufacturing. That is, this would indicate that employment in a given manufacturing industry is no more or less geographically concentrated than overall manufacturing employment. At the other extreme, the locational Gini coefficient takes on values close to one when employment in a given industry is completely concentrated in one county. Business Review Q3 2008 3 Glenn Ellison and Edward Glaeser have identified a potential problem with the locational Gini coefficient. They argue that if an industry consists of a small number of large establishments, the locational Gini coefficient may take on large values, suggesting localization of the industry even if there is no agglomeration force behind the industry’s location. They refer to this as the dartboard approach to geographic concentration, using the metaphor of a few darts tossed at a dartboard randomly creating a cluster. Ellison and Glaeser have developed an alternative concentration measure — called the Ellison-Glaeser, or the EG, index — that controls for an industry’s organization. Recently, Paulo Guimarães, Octávio Figueiredo, and Douglas Woodward (GFW) have generalized the EG index to include the case where the data are in the form of establishments (labs, in our case) rather than employment shares, as in the EG index. The GFW locational Gini, or the GFW index, for R&D labs is constructed just like the locational Gini for employment except each county’s share of the nation’s labs in a given industry is used instead of the county’s employment share for the industry. As before, the GFW index for a given industry takes on a value of zero when R&D labs in the industry are not geographically more concentrated than is manufacturing employment. Following Ellison and Glaeser, Guimarães, Figueiredo, and Woodward adjust the GFW index to account for the industrial organization of the industry in question. We use the adjusted GFW index as our measure of concentration for R&D by industry.4 We find an adjusted GFW index of 0.0457 for R&D in the average industry at the county level. In studying the agglomeration patterns in the manufacturing industries, Glenn Ellison, Edward Glaeser, and William 4 Q3 2008 Business Review Kerr report an average adjusted Gini coefficient of 0.03 for manufacturing in 1997 at the metropolitan area level. (Since metropolitan areas tend to be aggregates of counties, there are more counties than metropolitan areas.) Thus, our R&D labs appear to be more spatially concentrated, on average, than is manufacturing activity.5 Our findings indicate that 256, or 68 percent, of all R&D counties have an adjusted GFW index greater than zero, suggesting that R&D labs are appreciably more concentrated than manufacturing employment. Earlier we reported that the top 50 counties ranked by number of R&D labs account for 58 percent of all R&D labs, while the top 50 counties ranked by number of manufacturing establishments account for only 36 percent of all manufacturing establishments. Thus, the concentration of labs is broadly similar when looking at the top 50 counties or the adjusted GFW index. While an adjusted GFW index for an industry could have a value greater 4 See the article by Paulo Guimarães, Octávio Figueiredo, and Douglas Woodward for details on the construction of the adjusted GFW index used in our article as well as a discussion of the EG index. Our sample consists of 376 four-digit Standard Industrial Classification industries at the county level. We chose to do our analysis based on the number of labs in a county rather than employment in these labs, since we have data on employment for only about one-half of the labs in our data set. 5 By construction, the value of both the EG index and the adjusted GFW index is directly related to the level of aggregation of the geographic area under consideration. That is, for any given industry, the EG indexes and the adjusted GFW indexes take on larger values for metropolitan areas (aggregations of counties) than the indexes do at the county level. Thus, our finding of greater average concentration of R&D labs compared with the average concentration of manufacturing employment reported in Ellison, Glaeser, and Kerr is even more striking, given that the adjusted GFW is calculated at the county level and still exceeds the average value of the EG index calculated at the MSA level. than zero, an important question is: Does this represent a significant departure from the spatial concentration of manufacturing employment? We performed a simulation procedure to determine what value of the adjusted GFW indexes constitutes a significant departure from the concentration of manufacturing employment.6 We find R&D labs in 129 of the 376 industries considered (34.3 percent) are significantly more concentrated than is manufacturing employment. Thus, of the 256 industries with an adjusted GFW index greater than zero, only about one-half — or 129 industries — represent a significant departure from the overall concentration of manufacturing employment. This shows the importance of providing statistical tests that determine whether labs in a 6 To develop measures of statistical significance for the adjusted GFW indexes, we partitioned our industries into six nonoverlapping groups based on the number of R&D labs in a given industry. The first group consists of industries that have between two and nine labs. The second group consists of industries with 10 to 30 labs, while the third group consists of industries with between 31 and 50 labs. The fourth group consists of industries with between 51 and 100 labs, while the fifth group consists of industries with 101 to 200 labs. The final group consists of industries with more than 200 labs. For each group, we performed a simulation procedure to produce a probability distribution for the adjusted GFW index. In the simulation we randomly allocated labs to counties while maintaining the counties’ share of national manufacturing employment. Therefore, if a given county has a relatively high share of the nation’s manufacturing jobs, the county is more likely to randomly be assigned more R&D labs, too. For each group the simulation produces a value for the adjusted GFW index. For each group, we performed 1,000 simulations and formed a probability distribution for the adjusted GFW indexes. From the distribution we can calculate critical values (one that’s positive and one that’s negative) that allow us to say that we are 95 percent certain that any value that exceeds the critical value indicates that labs in that grouping are significantly more concentrated than is the actual distribution of manufacturing employment. Similarly, any value that falls below the critical value indicates that labs in that grouping are significantly more dispersed than is the actual distribution of manufacturing employment. www.philadelphiafed.org given industry are significantly more concentrated (significantly more dispersed) than is the actual distribution of manufacturing employment. Our measure of concentration, the adjusted GFW index, has a maximum value of about one for R&D in five industries.7 However, there are only two R&D labs in each of these industries, so it’s not surprising to find a large value for the adjusted Gini index if the two firms are located in proximity to one another.8 Among industries with 20 or more labs, R&D tends to be most concentrated in the oil and gas field machinery industry, the computer storage devices industry, and the electronic computer industry (see the Table).9 Until now, we have looked at the concentration of R&D labs relative to the concentration of manufacturing employment. We would also like to know whether labs in a particular industry (such as pharmaceuticals) are more or less concentrated than overall TABLE Concentration of R&D Labs for Selected Industries Industry Number of Labs Adjusted GFW Indexa Concentrated Industriesb Oil & Gas Field Machinery 22 0.33 Tires and Tubes 14 0.12 Crude Petroleum & Natural Gas 14 0.10 Computer Storage Devices 34 0.08 Motor Vehicles & Car Bodies 26 0.06 Electronic Computers 57 0.06 Semiconductors 278 0.03 Prepackaged Software 359 0.03 Motor Vehicle Parts 134 0.03 36 0.03 Computer-Integrated Systems Design 105 0.02 Radio and TV Communication Equipment 185 0.02 Wood Household Furniture 11 -0.01 Gaskets, Packing, and Sealing Devices 11 -0.01 Industrial Valves 14 -0.01 Plastic Plumbing Fixtures 11 -0.01 Gray and Ductile Iron Foundries 12 -0.01 Optical Instruments and Lenses 7 They are hog production; the production of brooms and brushes; the production of fiber cans, tubes, and drums; the bottled and canned soft drinks and carbonated waters industry; and the rolling mill machinery and equipment industry. 8 There is a negative relationship between the size of the adjusted GFW index and the number of labs in an industry. However, this relationship is not strong: a correlation coefficient of -0.09 that is only marginally significant (at the 10 percent level). 9 In this article, our index of concentration (the adjusted GFW index) compares the concentration of R&D labs in a given industry to the concentration of manufacturing employment in that industry. Instead of using manufacturing employment as the benchmark when constructing the adjusted GFW index, we could have used manufacturing establishments as the benchmark. In general, there’s a moderate correlation (a Spearman’s rank correlation coefficient of 0.56) between the industry ranking under the two alternative benchmarks for R&D industries with significant adjusted GFW indexes and with 20 or more labs. Following Guimarães, Figueiredo, and Woodward, we report the adjusted GFW index using manufacturing employment as the benchmark in this article to make our findings consistent with past studies, such as the one by Ellison and Glaeser. www.philadelphiafed.org Dispersed Industriesc a The adjusted GFW index for a given industry shows the sum of the squared differences of the share of employment in manufacturing from the share of labs in a given industry, adjusted to account for the industrial organization of the industry under consideration. b R&D labs in the selected industries are significantly more concentrated than manufacturing employment (5 percent level of significance). c R&D labs in the selected industries are significantly more dispersed than manufacturing employment (5 percent level of significance). Business Review Q3 2008 5 R&D labs. To get this information, we recalculated the adjusted GFW index to reflect the geographic concentration of labs in individual industries relative to the overall concentration of R&D labs (as opposed to the overall concentration of manufacturing employment). We find that 314, or 84 percent, of all R&D labs have an adjusted Gini index greater than zero; however, we find that R&D labs in only 105 of the 376 industries (28 percent) considered are significantly more concentrated than overall R&D labs.10 It’s not surprising to find less concentration of R&D by industries when the comparison is to overall R&D labs than when the comparison is to overall manufacturing employment (34.3 percent), given that R&D labs already tend to be more concentrated than manufacturing employment. Still, for the majority of industries (72 percent), labs at the industry level tend not to be more spatially concentrated than labs overall. Maps of R&D activity for individual industries (for example, software, Figure 2; pharmaceuticals, Figure 3; and chemicals, Figure 4) confirm the findings of the adjusted GFW indexes in that the location pattern of R&D activity for the majority of industries is broadly similar to the location pattern of overall R&D activity. That is, R&D activity for most industries tends to be concentrated in the Northeast corridor, around the Great Lakes, in California’s Bay Area, and in southern California. FIGURE 2 Location of Software R&D Labs FIGURE 3 Location of Pharmaceutical R&D Labs 10 We performed a simulation procedure to determine what value of the adjusted GFW indexes constitutes a significant departure from the concentration of total R&D labs. The simulation procedure is similar to the procedure used when the reference was manufacturing employment, except we now randomly allocate labs to counties while maintaining the counties’ share of national R&D labs, as opposed to the counties’ share of national manufacturing employment. 6 Q3 2008 Business Review www.philadelphiafed.org FIGURE 4 Location of Chemistry R&D Labs FIGURE 5 Location of Oil and Gas Field Machinery R&D Labs As indicated, there are a number of exceptions to the general pattern of geographic concentration just described. One exception is R&D activity in the oil and gas field machinery industry, which tends to be concentrated in Texas, especially in the Houston area, and accounts for about 60 percent of the labs doing R&D in this industry (Figure 5). Another exception is the location of R&D activity in the motor vehicle and car body industry, which tends to be concentrated in Michigan, especially in the Detroit area, and which accounts for just under 40 percent of the labs doing R&D in this industry (Figure 6). This industry comprises establishments primarily engaged in manufacturing motor vehicle parts and accessories. WHY DO R&D LABS CLUSTER? Economists have developed a number of theories to explain firms’ tendency (not just R&D labs) to cluster. Firms may attempt to minimize transport costs by locating close to a natural resource used as an input, or to their suppliers, or to their markets. Or firms may cluster to share inputs such as specialized workers. Finally, firms may cluster to take advantage of knowledge that “spills over” when firms are located near one another. Among these, the sharing of inputs and especially of knowledge spillovers is likely to be most important for R&D firms when choosing a location. Knowledge Spillovers. Economists have identified two types of knowledge spillovers thought to be important in understanding the location pattern of R&D labs: MAR spillovers and Jacobs spillovers.11 While these 11 MAR spillovers are so-called because in 1890 Alfred Marshall developed a theory of knowledge spillovers that was later extended by Kenneth Arrow and Paul Romer — hence, MAR. In 1969, Jane Jacobs developed another theory of knowledge spillovers. www.philadelphiafed.org Business Review Q3 2008 7 FIGURE 6 Location of Motor Vehicle and Car Body R&D Labs theories were originally developed to explain the concentration of industries in general, we think they are particularly important to an explanation of the clustering of R&D labs. More than most industries, R&D depends on new knowledge. Often, the latest knowledge about technological developments is valuable to firms but only for a short time. Thus, it behooves firms to set up shop as close as possible to the sources of information. The high spatial concentration of R&D activity facilitates the exchange of ideas among firms and aids in the creation of new goods and new ways of producing existing goods. MAR spillovers. According to the MAR theory of spillovers, the concentration of establishments (labs in our case) in the same industry in a common area helps knowledge travel among labs and their workers and facilitates innovation and growth.12 Employees from different establish- 8 Q3 2008 Business Review ments in the same industry exchange ideas about new products or new ways to produce goods. Often, knowledge is tacit and not easily codified and therefore requires face-to-face contact to be effectively transmitted. Having firms concentrated in a particular area is an efficient way to produce new ideas, leading to innovation and growth. People’s ability to receive ideas or knowledge is then influenced by their distance from the source of the ideas; communicating ideas is harder over longer distances. Stuart Rosenthal and William Strange consider the importance of input sharing, matching, and knowledge spillovers for manufacturing firms at the state, county, and ZIP code levels. They find that the effects of knowledge spillovers on the agglomeration of manufacturing firms tend to be quite localized, influencing agglomeration only at the ZIP code level.13 For example, many semiconductor firms have located their R&D facilities in the Silicon Valley because the area provides an environment where semiconductor firms can develop new products and new production technologies. Often, information about current developments in the semiconductor industry is shared informally. In her 1994 book, AnnaLee Saxenian describes how gathering places, such as the Wagon Wheel Bar located only a block from Intel, Raytheon, and Fairchild Semiconductor, “served as informal recruiting centers as well as listening posts; job information flowed freely along with shop talk.” Other examples include the Route 128 corridor in Massachusetts, the Research Triangle in North Carolina, and biotechnology and medical technology software firms in suburban Philadelphia. Jacobs spillovers. Jane Jacobs believed that knowledge spillovers are related to the diversity of industries (diversity of labs in our case) in an area, in contrast to MAR spillovers, which focus on firms in a common industry. Jacobs argued that an industrially diverse environment encourages innovation. Such environments include knowledge workers with varied backgrounds and interests, thereby fa- 12 Edward Glaeser, Hedi Kallal, Jose Scheinkman, and Andrei Shleifer, who coined the term MAR spillovers, pulled these various views on knowledge spillovers together in their article. 13 Several other studies have found that knowledge spillovers dissipate rapidly with distance. See, for example, the articles by Mohammad Arzaghi and J. Vernon Henderson; David Audretsch and Maryann Feldman; Wolfgang Keller; and Jed Kolko. The extent to which innovations in communication technologies are rendering face-to-face contacts obsolete is not so clear. Jess Gaspar and Edward Glaeser argue that improvements in telecommunications technology increase the demand for all interactions. So while technology may substitute for face-to face contact, this effect is offset by the greater desire for all kinds of interactions, including face-to-face contact. www.philadelphiafed.org cilitating the exchange of ideas among individuals with different perspectives. This exchange can lead to the development of new ideas, products, and processes. As John McDonald points out, both Jane Jacobs and John Jackson have noted that Detroit’s shipbuilding industry was the critical antecedent leading to the development of the auto industry in Detroit. In the 1820s, Detroit mainly exported flour. Because the industry was located north of Lake Erie along the Detroit River, small shipyards developed to build ships for the flour trade. R&D in the shipbuilding industry led to refinements and the adaptation of the internal-combustion gasoline engine to power boats on Michigan’s rivers and lakes. As it turned out, the gasoline engine, rather than the steam engine, was best suited for powering the automobile. Several of Detroit’s pioneers in the automobile industry had their roots in the boat engine industry. For example, Olds produced boat engines, and Dodge repaired them. In addition, a number of other industries in Michigan supported the development of the auto industry, such as the steel and machine tool industries. These firms engaged in R&D that led them to produce many of the components required to make cars. While other factors could be at work, the adjusted GFW indexes appear to support Jacobs’ diversity view, in that R&D labs for the vast majority of industries (almost three-quarters) tend to exhibit a common overlapping pattern of concentration. David Audretsch and Maryann Feldman used the U.S. Small Business Administration’s innovation database and focused on innovative activity for particular industries within specific MSAs. They found less industry-specific innovation in MSAs that specialized in a given industry, a finding that also supports Jacobs’ diversity thesis. www.philadelphiafed.org The Role of Natural Advantage. While it’s tempting to argue that the broadly similar geographic clustering of R&D labs in many different industries is suggestive of Jacobs externalities, this conjecture is simply based on visual inspection of a map (Figure 1). Jacobs spillovers are one possible way to account for the common overlapping pattern of concentration among R&D labs, but other forces might be at work. Rust Belt region, an area relatively low in amenities. Another natural advantage that an area may have lies in its workers and institutions, especially its universities. Universities are key players not only in creating new knowledge through the basic research produced by their faculties but also in supplying a pool of knowledge workers on which R&D depends. It is well known Universities are key players not only in creating new knowledge through the basic research produced by their faculties but also in supplying a pool of knowledge workers on which R&D depends. One such source is the natural advantages an area offers to firms that locate there. An area’s natural advantages, such as climate, soil, and mineral and ore deposits, could explain the location of some R&D labs. For example, oil deposits, an essential ingredient for testing equipment, may be largely responsible for the concentration of R&D labs in the oil and gas field machinery industry (one of the most highly concentrated industries, according to our adjusted GFW indexes) in Texas, especially in the Houston area. But the draw of ore deposits seems to be industry-specific and is therefore unlikely to account for the common overlapping pattern of concentration among R&D labs in many different industries. Of course, if R&D labs tend to be drawn to areas offering amenities such as pleasant weather, proximity to the ocean, and scenic views, this could explain the overlapping concentration in amenity-rich locations, such as the concentrations found in California. While local amenities might explain some of the concentrations of labs, the vast majority of R&D labs tend to be highly concentrated in the country’s that Silicon Valley and the Route 128 corridor became important centers for R&D as a result of their proximities to Stanford and MIT. AnnaLee Saxenian describes how Stanford’s support of local firms is an important reason for the Silicon Valley’s success. Two of Stanford’s star engineering professors, John Linvill and Fred Terman, not only drew some of the best and brightest students to Stanford, but they also trained their students (and encouraged them) to seek careers in the semiconductor industry. There is also evidence that an area’s human capital can be an important type of natural advantage. In a 2007 paper, Gerald Carlino, Satyajit Chatterjee, and Robert Hunt looked at the effect of a metropolitan area’s human capital (the share of the adult population with at least a college education) on the area’s ability to innovate (measured by patents per capita). Of the things these authors considered, by far the most powerful effect on local innovation is generated by local human capital. Specifically, a 10 percent increase in the share of the adult population with at least a Business Review Q3 2008 9 college degree is associated with an 8.6 percent increase in patents per capita. Since the share of a metropolitan area’s population with at least a college degree varied by a factor of almost six in the sample used in Carlino, Chatterjee, and Hunt’s paper, the implied gains in innovation are substantial. There is also general evidence that R&D at local universities is important for firms’ innovative activity. David Audretsch and Maryann Feldman, and Luc Anselin, Attila Varga, and Zoltan Acs found evidence of localized knowledge spillovers from university R&D to commercial innovation by private firms, even after controlling for the location of industrial R&D. However, Carlino, Chatterjee, and Hunt found that R&D at local universities has only modest effects on local innovative activity. They found that a 10 percent increase in R&D intensity of local universities is associated with less than a 1 percent increase in patent intensity. Evidence on MAR vs. Jacobs Spillovers. To more formally address the issue of the importance of industrial diversity, or, alternatively, specialization, we conducted a simple experiment. Recall that we have only one adjusted GFW index for each industry. These industry indexes can, however, be used to construct an overall adjusted GFW index for each metropolitan county. This is done by weighting each industry’s adjusted GFW index by the share of the county’s total establishment accounted for by that industry. The industry-weighted adjusted GFW indexes for a given county are then summed to arrive at an overall adjusted GFW index for each metropolitan county. The overall adjusted GFW index for a county can be correlated with a widely used index of industrial diversity.14 By construction, a county is said to be more highly specialized or less diversified as the value of the diversity index increases. Recall that as 10 Q3 2008 Business Review the value of the adjusted GFW index increases, the extent of the spatial concentration of labs in the industry also increases. A positive correlation between the overall county adjusted GFW index and the specialization index means that as the county becomes more specialized industrially, its labs are also becoming more geographically concentrated. This evidence favors MAR spillovers.15 On the other hand, if the geographic concentration of labs tends to increase as the specialization index decreases — indicating that an area is more industrially diverse (or less specialized) — this negative correlation provides evidence in favor of Jacobs spillovers. We found a positive and highly significant correlation between the overall county adjusted GFW index and the specialization measure, evidence favoring MAR spillovers. While a more definitive conclusion awaits a more complete analysis, the evidence provided in this article tends to support the importance of both Jacobs spillovers (visual inspections of maps) and MAR spillovers (statistical correlation) for R&D labs.16 CONCLUSION Most countries make sustained economic growth a principal policy objective. Although many factors contribute to economic growth, recent research has found that innovation and invention play an important role. Innovation depends on R&D, and R&D depends on, among other things, the exchange of ideas among individuals. The high spatial concentration of R&D labs creates an environment in which ideas move quickly from person to person and from lab to lab. That is, locations that are dense in R&D activity encourage knowledge spillovers, thus facilitating the exchange of ideas that underlies the creation of new goods and new ways of producing existing goods. Finally, the study by Saxenian provides a cautionary note for policymakers who view the success of areas such as Silicon Valley as a recipe for local economic development and growth. While investing in science centers to attract R&D activity is fairly common in the U.S., Saxenian’s study suggests that creating the right corporate culture to make the centers successful is more challenging. Instead of targeting industries, we suggest that policymakers consider strategies that help to establish a good business environment and which are conducive to attracting and retaining highly skilled workers. Glaeser and co-authors’ study suggests that local policymakers need to focus on life-style issues because they are important in attracting and retaining high-skill workers. One such policy is providing good public schools. Other policies might focus on reducing urban crime and providing amenities such as clean streets and public parks. BR 14 County-level specialization was measured using a Herfindahl index. A Herfindahl index measures diversification or, inversely, specialization. It is calculated by squaring and summing the share of establishments accounted for by each industry in a given county. The squaring of industry shares means that the larger industries contribute more than proportionately to the overall value of the index. Thus, as the index increases in value for a given county, this implies that the county is more highly specialized or less diversified industrially. 15 We have 847 metropolitan counties in our sample. The correlation coefficient is 0.0148 and is significant at the 1 percent level. The coefficient is small in magnitude because the average value for the diversity index is 75 times as large as the average value for the county adjusted GFW index. Despite the relatively low value of the correlation between the county adjusted GFW index and the diversity index, the relationship between these variables is economically significant, displaying an elasticity of almost one in value. 16 A more complete analysis of the role of MAR vs. Jacobs spillovers on the clustering of R&D labs should also control for an area’s natural advantages as identified in this article. www.philadelphiafed.org REFERENCES Anselin, Luc, Attila Varga, and Zoltan Acs. “Local Geographic Spillovers between University and High Technology Innovations,” Journal of Urban Economics, 42 (1997), pp. 442-48. Arzaghi, Mohammad, and J. Vernon Henderson. “Networking Off Madison Avenue,” unpublished manuscript (2005). Audretsch, David B., and Maryann P. Feldman. “R&D Spillovers and the Geography of Innovation and Production,” American Economic Review, 86 (1996), pp. 630-40. Carlino, Gerald A. “Knowledge Spillovers: Cities’ Role in the New Economy,” Federal Reserve Bank of Philadelphia Business Review (Fourth Quarter 2001), pp. 17-23. Carlino, Gerald A. “The Economic Role of Cities in the 21st Century,” Federal Reserve Bank of Philadelphia Business Review (Third Quarter 2005), pp. 9-15. Carlino, Gerald A., Satyajit Chatterjee, and Robert M. Hunt. “Urban Density and the Rate of Invention,” Journal of Urban Economics, 61 (2007), pp. 389-419. Directory of American Research and Technology, 23rd Edition. New York: R.R. Bowker, 1999. Duranton, Gilles, and Henry G. Overman. “Testing for Localization Using MicroGeographic Data,” Review of Economic Studies, 72 (2005), pp. 1077-1106. Ellison, Glenn, and Edward. L. Glaeser. “Geographic Concentration in U.S. Manufacturing Industries: A Dartboard Approach,” Journal of Political Economy, 105 (1997), pp. 889-927. www.philadelphiafed.org Ellison, Glenn, Edward L. Glaeser, and William Kerr. “What Causes Industry Agglomeration? Evidence from Coagglomeration Patterns,” Discussion Paper 2133, Harvard Institute of Economic Research (April 2007). Feldman, Maryann P., and David B. Audretsch. “Innovation in Cities: Science-Based Diversity, Specialization, and Localized Competition,” European Economic Review, 43 (1999), pp. 409-29. Gaspar, Jess, and Edward Glaeser. “Information Technology and the Future of Cities,” Journal of Urban Economics, 43 (1998), pp. 136-56. Glaeser, Edward, Hedi Kallal, Jose Scheinkman, and Andrei Shleifer. “Growth in Cities,” Journal of Political Economy, 100 (1992), pp. 1126-53. Glaeser, Edward L., Jed Kolko, and Albert Saiz. “Consumer City,” Journal of Economic Geography, 1 (2001), pp. 27-50. Guimarães, Paulo, Octávio Figueiredo, and Douglas Woodward. “Measuring the Localization of Economic Activity: A Parametric Approach,” Journal of Regional Science, 47 (2007), pp. 753-44. Holmes, Thomas J., and John J. Stevens. “Spatial Distribution of Economic Activities in North America,” in J.V. Henderson and J.-F Thisse, eds., Handbook of Regional and Urban Economics, Vol. IV: Cities and Geography. Amsterdam: Elsevier, 2004. Jackson, John. “Michigan,” in R. Scott Fosler, ed., The New Role of American States. New York: Oxford University Press, 1988. Jacobs, Jane. The Economy of Cities. New York: Vintage Books, 1961. Keller, Wolfgang. “Geographic Localization of International Technology Diffusion,” American Economic Review, 92 (2002), pp. 120-42. Kolko, Jed. “Agglomeration and CoAgglomeration of Services Industries,” unpublished manuscript (April 2007). Krugman, Paul. Geography and Trade. Cambridge: MIT Press, 1991. Madden, Janice. “Creating Jobs, Keeping Jobs, and Losing Jobs: Cities and the Suburbs in the Global Economy,” unpublished manuscript (2000). Maurel, Françoise, and Béatrice Sédillot. “A Measure of the Geographic Concentration in French Manufacturing Industries,” Regional Science and Urban Economics, 29 (1999), pp. 575-604. McDonald, John F. Fundamentals of Urban Economics. Upper Saddle River, NJ: Prentice Hall, 1997. Rosenthal, Stuart, and William C. Strange. “The Determinants of Agglomeration,” Journal of Urban Economics, 50 (2001), pp. 191-229. Saxenian, AnnaLee. Regional Advantage: Culture and Competition in Silicon Valley and Route 128. Cambridge, MA: Harvard University Press, 1994. Starr, Paul. “Review of AnnaLee Saxenian,” Regional Advantage Contemporary Sociology (May 1995). Business Review Q3 2008 11 Creative Destruction and Aggregate Productivity Growth* BY SHIGERU FUJITA P roductivity growth is the engine of economic growth and is responsible for rising standards of living. But all firms do not partake equally in the nation’s productivity growth. Rather, according to economist Joseph Schumpeter’s theory, firms undergo a process of “creative destruction”: New firms that adapt to new knowledge cause the decline and eventual demise of incumbent firms. In this article, Shigeru Fujita surveys recent studies that examine the role of creative destruction in aggregate productivity growth. Productivity growth is the engine of economic growth. Firms constantly discover and implement new technologies, making it possible for them to produce new products and services or to produce existing products and services more efficiently. Productivity growth is responsible for rising living standards in the world. The figure on page 13, which plots a common measure of productivity — labor productivity — for the U.S., shows that productivity has grown Shigeru Fujita is a senior economist in the Research Department of the Philadelphia Fed. This article is available free of charge at www. philadelphiafed. org/research-and-data/publications/. 12 Q3 2008 Business Review steadily in the postwar U.S.1 economy, indicating that the economy has become wealthier over time. The smooth rise of productivity shown in the figure might suggest that all firms partake equally in the nation’s productivity growth. Joseph Schumpeter (1883-1950), one of the most influential economists of the 20th century, observed that anyone who thought so would completely miss the “essential fact about capitalism,”2 which, he argued, is the process of 1 Labor productivity is defined as the value of output less intermediate inputs (both values adjusted for inflation) produced per unit of labor input (measured as man-hours). “creative destruction.” In his famous book, Capitalism, Socialism, and Democracy, he summarized this process as one that “incessantly revolutionizes the economic structure from within, incessantly destroying the old one, incessantly creating a new one.” Of course, many other economists have deeply appreciated the importance of creative destruction in capitalism. Former Federal Reserve Chairman Alan Greenspan, for instance, argues in his latest book that creative destruction is the only way to increase productivity and therefore the only way to raise average living standards on a sustained basis. These readings suggest that the turbulent process of creation and destruction lurks beneath the smooth rise in aggregate productivity. Underlying Schumpeter’s astute observation is the fact that firms are very different from each other: They differ in terms of their managerial abilities, their location, their organization, and their know-how. These differences mean that some firms take better advantage of new knowledge and ideas than others. New and existing firms that adapt to new knowledge cause the decline and eventual demise of other firms. Schumpeter emphasized that this process of creative destruction is an “evolutionary process” whereby “every element of it takes considerable time in revealing its true features and ultimate effects,” and thus “we must judge its performance over time.” 2 *The views expressed here are those of the author and do not necessarily represent the views of the Federal Reserve Bank of Philadelphia or the Federal Reserve System. The first view is consistent with Adam Smith’s view of economic growth. See Leonard Nakamura’s article for detailed characterizations of the differences between Smith’s and Schumpeter’s views. www.philadelphiafed.org FIGURE Labor Productivity (output per hour) Log (Labor Productivity) 4.9 4.7 4.5 4.3 4.1 3.9 3.7 3.5 47 50 53 56 59 62 65 68 71 74 77 80 83 86 89 92 95 98 01 04 07 Note: Shaded areas indicate recessions. This article surveys recent studies that examine the role of creative destruction in aggregate productivity growth. These studies seek to understand the link between the productivity of individual business units and aggregate productivity, paying particular attention to the role that the birth and death of firms plays in the growth of aggregate productivity. Although Schumpeter’s idea of creative destruction has been around for more than 60 years, it is only in the last 20 years or so that economists have had access to data that make it possible to quantify — and establish beyond doubt — this “essential fact about capitalism.” LINKING INDIVIDUAL AND AGGREGATE PRODUCTIVITY To understand how growth in aggregate productivity depends on the process of creative destruction, we need a way to link individual www.philadelphiafed.org productivities to aggregate productivity. Although there are various ways to make this link, I will focus on the one proposed by Lucia Foster, John Haltiwanger, and C.J. Krizan. Their method is to take a weighted average of individual establishment productivities and it allows us to express aggregate productivity as the sum of four components, each of which has intuitive economic meaning.3 The first component represents the productivity growth of establishments that continuously exist between two dates. Obviously, if the productivity of these continuing establishments grows, aggregate productivity will grow. This first component is called 3 Note that in the literature I review in this article, an individual unit is a business establishment (or plant) that may be part of a larger firm. For this reason, I refer to an individual unit as an establishment rather than as a firm. the “within component,” reflecting the fact that this term captures the productivity gains that occur within each continuing establishment. A second component takes into account the changes in aggregate productivity that result from changes in the relative size of establishments with different productivity levels. Even if the productivity of continuing establishments were to remain constant, aggregate productivity could change because of changes in the size of the establishments with different productivity levels. For instance, if more productive establishments were to expand employment over time and less productive establishments were to shrink, aggregate productivity (which is a weighted average of individual productivities) will grow. This component is called the “between component.” The two components above measure the effects of changes in individual productivities or changes in employment shares. Because these two components are calculated by fixing either the shares or the level of individual productivities, they do not capture the effects of how the changes in the individual productivities and the changes in shares are correlated. The “cross component” measures this correlation. The positive correlation shows up as a positive contribution. Similarly, the negative correlation shows up as a negative contribution. More specifically, if the establishments with faster-growing productivity are also the ones that are increasing their shares of employment, it shows a positive contribution. Again, similarly, if the establishments with slower-growing productivity are also the ones that are decreasing their shares of employment, it shows a positive contribution. The case of the positive correlation sounds reasonable in that one may think that establishments that have higher productivity growth expand their em- Business Review Q3 2008 13 ployment shares over time, while those that have lower productivity growth shrink their shares. However, it is also possible that those that are reducing employment faster than others (for example, more aggressively restructuring) are more rapidly improving their productivity. When the two are negatively correlated, the cross component shows the negative contribution to aggregate productivity growth. The last component measures the effects of the births and deaths of establishments. If new establishments have higher-than-average productivity, their presence will contribute to growth in aggregate productivity. If exiting establishments have lowerthan-average productivity, that, too, contributes to productivity growth. The sum of these two subcomponents is called the “net entry component.” Clearly, this term is directly related to Schumpeter’s notion of creative destruction. The Process of Creative Destruction and the Accounting Framework. While the net entry component has a direct connection to the notion of creative destruction, it is important to recognize that the other components are also influenced by it. For instance, invention of a superior technology by a new entrant may encourage incumbent firms to improve their own technologies. In the accounting framework above, this effect will show up in the within component. Another possibility is that the invention of new technologies induces resource reallocation (for example, workers change jobs) across incumbent establishments. This reallocation will clearly affect the between component. Of course, the actual effects of creative destruction are likely to be more varied and subtle than any accounting framework can fully reveal.4 Nevertheless, this simple framework can shed considerable light on what 14 Q3 2008 Business Review actually happens in each establishment as new technologies emerge and old technologies die out. CREATIVE DESTRUCTION AND PRODUCTIVITY GROWTH IN MANUFACTURING Net Entry Accounts for 30 Percent of Productivity Growth over a 10-Year Period. Table 1 reports the contribution to productivity growth in the manufacturing sector.5 The first row shows the breakdown over a three five-year periods: 1977-1982, 1982-1987, and 1987-1992. Overall, it is somewhat difficult to clearly characterize the results. However, we can make two observations. First, the contribution of net entry is always around 20 percent, regardless of time period. Note that relative to the result for 10-year productivity growth, the contribution of net entry is smaller. This is consistent with the idea that the effects of creative destruction are more apparent over a longer horizon. Invention of a superior technology by a new entrant may encourage incumbent firms to improve their own technologies. 10-year period between 1977 and 1987. The first column of the row shows that aggregate labor productivity, defined as real output divided by total hours (number of workers times hours worked per worker), grew 21 percent over this period. The four columns next to the aggregate growth rate are the shares of contributions of the four terms explained above. According to the first row, the within component (77 percent) and the net entry component (29 percent) are the main contributors to productivity growth over this 10-year period. This latter finding is consistent with creative destruction. The next three rows in Table 1 present the contribution of the four components for 4 Deeper understanding of the creative destruction process requires development of the appropriate theoretical framework. Readers who are interested in such attempts can refer to a recent paper by Markus Poschke and the references therein. 5 The data in Table 1 are based on tables in the article by Lucia Foster, John Haltiwanger, and C.J. Krizan. The second observation we can make from Table 1 is that the contribution of the between component is higher when overall productivity growth is lower (and vice versa). Specifically, it is highest during 1977-1982, when productivity growth is low compared with the other two periods. This result in Table 1 is based on coarse data observations, that is, only three observations of five-year productivity growth. However, a recent study by Yoonsoo Lee, which breaks down the annual productivity growth in manufacturing from 1973 through 1997 using a similar method, also finds that the between component is higher when aggregate productivity growth is slower. To put this observation into perspective, we can note that aggregate productivity tends to move together with the business cycle, which implies that reallocation of workers from less productive establishments to more productive ones intensifies during the cyclical downturns. New and More Productive Establishments Displace Old and Less Productive Ones. Now, let’s look www.philadelphiafed.org TABLE 1 Productivity Decomposition (Manufacturing Sector) Overall Growth Rate Within Component Between Component Cross Component Net Entry Component 1977 - 1987 21.32 16.42 (77) -1.71 (8) -2.98 (-14) 6.18 (29) 1977 - 1982 2.54 3.10 (122) 2.16 (85) -3.23 (-127) 0.51 (20) 1982 - 1987 18.67 15.50 (83) 2.43 (13) -2.80 (-15) 3.55 (19) 1987 - 1992 7.17 6.74 (94) 2.37 (33) -3.51 (-49) 1.51 (21) Source: Foster, Haltiwanger, and Krizan, 2001, Tables 8.4 and 8.7. Sum of the four components equals the overall growth rate. Numbers in parentheses indicate the share of overall productivity growth explained by each component, calculated by dividing the contribution of each component by the overall growth rate (expressed as percent). The share is negative when the component contributes negatively to overall growth. more closely at the role of net entry in 10-year productivity growth. The four columns in Table 2 report productivity levels of the following three types of establishments: (i) those that existed in 1977 but disappeared 10 years later, (ii) those that did not exist in 1977 and appeared 10 years later, and (iii) those that continued to exist throughout the 10-year period. All numbers are expressed relative to the average productivity level in 1977 of the establishments that existed throughout the 10-year period. The table shows an interesting pattern. The first column (0.83) indicates that the average productivity of the establishments that failed to survive the 10-year period was 17 percent lower in 1977 than that of the establishments that successfully survived www.philadelphiafed.org the same 10-year period. One can think of these displaced establishments being replaced by new establishments, which first appeared in 1987. The average productivity level of these new establishments in 1987 is given in the second column of the table. Observe that these new establishments on average had much higher productivity than that of the displaced establishments (1.11 vs. 0.83). This pattern is clearly consistent with Schumpeter’s creative destruction insight that new and more productive firms push out old and less productive ones. Learning and Selection Play Important Roles in the Evolution of Aggregate Productivity. However, another important insight from this table is that these new establishments do not necessarily enjoy the highest productivity when they appear in the market. This is reflected in the fact that the average productivity of these entering establishments is lower than the productivity of establishments that continue to exist throughout the 10-year period (1.11 vs. 1.20). This observation is consistent with the idea that “selection” and “learning” play important roles in the evolution of establishment-level productivity. Note first that those establishments that continue to exist throughout the period (that is, survive) do so because they are able to achieve high productivity. One can view this as the “selection” process over time. Further, even though new entrants presumably have some advantages (especially over old, exiting firms) — for example, because they can take advantage of new Business Review Q3 2008 15 TABLE 2 Relative Labor Productivity for Exiting, Entering, and Continuing Establishments (Manufacturing Sector, 1977-1987) Exiting Establishments in 1977 Entering Establishments in 1987 Continuing Establishments in 1977 Continuing Establishments in 1987 0.83 1.11 1.00 1.20 Source: Foster, Haltiwanger, and Krizan, 2001, Table 8.9. Each column gives the average productivity level of each type of establishment, relative to the average productivity level of establishments in 1977 that existed throughout the 10-year period 1977-1987. Existing establishments: establishments that existed in 1977 but disappeared in 1987. Entering establishments: establishments that did not exist in 1977 but appeared in 1987. Continuing establishments: establishments that existed in both 1977 and 1987. technology or a good location — their observed productivity is not necessarily higher than the pre-existing “selected” establishments, since it takes time for these new entrants to “learn” the new technology and building organizational capability also takes time. Of course, some of these new entrants may disappear, failing to survive the competition, and only productive establishments will again be selected over time. These facts are consistent with Schumpeter’s assertion that creative destruction is an evolutionary process. As Schumpeter suspected, the facts point to the presence of rich microlevel dynamics, whereby the gradual process of learning and selection plays a key role in diffusing and propagating technological improvements. CREATIVE DESTRUCTION AND PRODUCTIVITY GROWTH IN RETAIL TRADE So far we have looked at the role of creative destruction in the manu- 16 Q3 2008 Business Review facturing sector. But the service sector employs the bulk of the U.S. workforce. For example, in 2007, 84 percent of nonfarm business employees were employed in service-providing industries. Unfortunately, data limitations prevent us from carrying out a similar analysis for the entire service sector. But Foster, Haltiwanger, and Krizan have made an important attempt to look at a key segment of the service industry, namely, the retail trade sector. While the analysis covers only one service industry, it is of particular interest given that the retail trade sector is large, employing more than 15 million workers (2007), which amounts to 11 percent of total nonfarm business employment. Moreover, it has undergone massive restructuring and reallocation since the late 1980s. In particular, it has changed its ways of doing business, mostly because of the adoption of advanced information technology (for example, improved inventory and sales tracking). Productivity Growth in Retail Trade Is Mostly Driven by Net Entry. Table 3 considers aggregate productivity growth in the retail trade sector over the 10-year period and the contributions of the four components. According to the table, the net entry component accounts for virtually all (98 percent) of the productivity growth over the 10-year period. Compared with the corresponding figure for the manufacturing sector, it is much larger, indicating the importance of net entry in the retail trade industry. This finding is consistent with the fact that job creation and destruction in this industry are explained mostly by the entry and exit of establishments. Another interesting finding in Table 3 is the large negative contribution of the cross term. As I discussed before, the cross component contributes positively if establishments with higher productivity growth also have higher employment growth or if establishments with lower productivity growth have lower employment growth. Thus, a negative contribution of this term implies that higher productivity growth at the establishment level is associated with lower employment growth and lower productivity growth is associated with higher employment growth. This appears counterintuitive if one expects more productive establishments to expand employment over time and less productive establishments to shrink employment over time. However, causality can go in the other direction as well: Downsizing of employment may have enhanced productivity growth for some establishments over the period. Establishment Births Are Driven by Expansion of Continuing Firms, While Establishment Deaths Come from Firms’ Deaths. Table 4 presents the breakdown of net entry’s contribution. The entry and exit columns of the table indicate that entry and exit account equally for net entry’s www.philadelphiafed.org large, positive contribution.6 Foster, Haltiwanger, and Krizan’s analysis does not stop there; the authors explicitly consider how much of the entry and exit of establishments reflects the entry and exit of firms as opposed to the entry and exit of establishments. Remember that the unit of observation in the analysis so far has been an “establish- 6 As in the case of the manufacturing sector, the pattern of reallocation is consistent with the idea that less productive plants are replaced by more productive plants (selection effects) and those new plants experience more rapid productivity growth than more mature incumbents (post-entry learning effects). ment,” which is defined by the physical production site (whether a manufacturing plant or a retail store). This definition leaves ownership of the establishments out of the analysis. However, bringing the notion of firms into the analysis, especially for the retail trade sector, provides a richer picture of creative destruction. Table 4 indicates that the positive contribution of entry comes mostly from entering establishments of continuing firms, whereas the large contribution of exit comes from exiting establishments of exiting firms. The authors further distinguish firms depending on whether the parent firm is a single-unit or a multi-unit firm that operates locally (one state), regionally (two to five states), or nationally (more than five states). The findings can be summarized as follows: • For continuing establishments, multiunit firms have a large productivity advantage over single-unit firms. Establishments operating locally, regionally, and nationally are, on average, 10.9 percent, 18.3 percent, and 24.1 percent more productive than single units. • Among exiting establishments, the least productive are the single units. These units are 20.9 percent less TABLE 3 Productivity Decomposition (Retail-Trade Sector) Overall Growth Rate 1987-1997 Within Component Between Component Cross Component Net Entry Component 1.83 (16) 2.74 (24) -4.46 (-39) 11.20 (98) 11.43 Source: Foster, Haltiwanger, and Krizan, 2006, Table 3. Sum of the four components equals the overall growth rate. Numbers in parentheses indicate the fraction of overall productivity growth explained by each component, calculated by dividing the contribution of each component by the overall growth rate (expressed as percent). The fraction is negative when the component contributes negatively to overall growth. TABLE 4 Productivity Decomposition: Contributions of Firm Entry and Exit (Retail-Trade Sector 1987-1997) Net Entry of Establishments Entering Establishments Continuing Firms 98 54 37 Exiting Establishments Entering Firms 17 Continuing Firms 45 Exiting Firms 3 42 Source: Foster, Haltiwanger, and Krizan, 2006, Table 3. The numbers indicate the fraction of overall productivity growth explained by each component, calculated by dividing the contribution of each component by the overall growth rate (expressed as percent). The number in the first column (98) corresponds to that in parentheses in the last column of Table 3. Entering establishments (firms): establishments (firms) that did not exist in 1987 and appeared in 1997. Existing establishments (firms): establishments (firms) that existed in 1987 but disappeared in 1997. Continuing firms: firms that existed in both 1987 and 1997. www.philadelphiafed.org Business Review Q3 2008 17 productive relative to the continuing single units, on average. The most productive among the exiting establishments are those affiliated with a national chain. These establishments are actually slightly more productive than the continuing single-unit establishments (+1.9 percent). • Among entering establishments, those associated with a national chain have a very large productivity advantage over single-unit incumbents (+24.7 percent). Clearly, these findings are consistent with views in the popular press that describe the demise of “mom and pop” stores and the increasing presence of large national chains. The creative destruction process has played a crucial role in the productivity gains in the retail trade industry. SUMMARY The availability of rich establishment-level data over the last 15 years or so has made it possible for researchers to assess Schumpeter’s assertion regarding the importance of creative destruction in aggregate productivity growth. Recent empirical studies indeed find that creative destruction plays a significant role in shaping the evolution of aggregate productivity: The evidence shows that new and relatively more productive establishments displace older and relatively less productive ones. However, new establishments are not necessarily the most productive: While new entrants have some advantages over existing establishments — for example, they can take advantage of new technology or a good location — it takes time for them to fully exploit these advantages. The facts reviewed in this article point to the importance of creative destruction but only hint at how creative destruction actually works. To fully appreciate these facts, economists have begun to build models that explicitly connect establishment-level decisions to aggregate outcomes. These models, together with the accumulating empirical evidence on establishment-level dynamics, promise to further enrich our understanding of creative destruction. BR REFERENCES Davis, Steven, John Haltiwanger, and Scott Schuh. Job Creation and Destruction. Cambridge, MA: MIT Press, 1996. Greenspan, Alan. The Age of Turbulence: Adventures in a New World. New York: Penguin Press, 2007 Poschke, Markus. “Employment Protection, Firm Selection, and Growth,” IZA Discussion Paper 3164, 2007. Foster, Lucia, John Haltiwanger, and C.J. Krizan. “Aggregate Productivity Growth: Lessons from Microeconomic Evidence,” in Charles Hulten, Edward Dean, and Michael Harper, eds., New Directions in Productivity Analysis. Chicago: University of Chicago Press, 2001. Lee, Yoonsoo. “The Importance of Reallocations in Cyclical Productivity and Returns to Scale: Evidence from Plant-level Data,” Federal Reserve Bank of Cleveland Working Paper 05-09 (2005). Schumpeter, Joseph. Capitalism, Socialism, and Democracy, Third Edition. New York: Harper and Brothers, 1942. Foster, Lucia, John Haltiwanger, and C.J. Krizan. “Market Selection, Reallocation, and Restructuring in the U.S. Retail Trade Sector in the 1990s,” Review of Economics and Statistics, 88 (2006), pp. 748-58. 18 Q3 2008 Business Review Nakamura, Leonard. “Economics and the New Economy: The Invisible Hand Meets Creative Destruction,” Federal Reserve Bank of Philadelphia Business Review (July/ August 2000). www.philadelphiafed.org APPENDIX Example of Productivity Decomposition This appendix provides a simple example to better understand the four components of aggregate productivity growth discussed in the text. Date T Date T+1 Establishments Output per worker Number of workers Output per worker Number of workers A 3 10 6 20 B 2 10 4 10 C 1 10 ---- ---- D ---- ---- 4 10 Total 2 30 5 40 In this example, I go through the decomposition of the productivity difference between the two dates T and T+1. At each point in time, there are only three establishments: at date T, they are establishments A, B, and C, and at date T+1, they are A, B, and D. That is, establishment C, which existed at date T, is replaced by a new establishment D, at T+1. The first two columns of the table summarize the information at date T. Each number in the first column gives the productivity (output per worker) of the three establishments, while the next column gives the total number of workers. The last row is economy-wide productivity, which can be calculated as a weighted average of the establishment-level productivities: Productivity of A * (Employment share of A) + Productivity of B * (Employment share of B) + Productivity of C * (Employment share of C) = 3(10/30)+ 2(10/30)+ 1(10/30) = 2. One can verify that aggregate productivity in period T+1 is 5 by doing the same calculation. Between the two dates, aggregate productivity goes up by 3 (=5-2), which amounts to a 150 percent increase in aggregate productivity. I now go over how to decompose overall improvement of productivity into four components. 1. Within component: This term measures the contribution of continuing establishments to productivity improvements. It is simply a www.philadelphiafed.org weighted average of changes in the productivity of continuing establishments, namely, A and B, in this example. To measure the effect of productivity changes that occurred “within” those establishments, employment shares are fixed at the levels of date T: (Change in productivity at A) *(A’s employment share at Date T)+ (Change in productivity at B) *(B’s employment share at Date T) = (6-3) (10/30)+(4-2)(10/30)=5/3. In this example, establishments A and B experienced productivity gains of 3 and 2, respectively. They are averaged by using their employment shares at date T. 2. Between component: This term measures how much of the overall productivity gain comes from the shift of employment from less productive establishments to more productive establishments: Even if the productivity levels of the existing units do not change over time, overall productivity can change simply because reallocating workers to more productive units improves overall productivity. It is calculated as a weighted average of the changes in employment shares, where the weights are productivity at the initial date, T, relative to overall productivity. Business Review Q3 2008 19 APPENDIX (continued) (Change in A’s share)*(Productivity difference of A from overall productivity at Date T) + (Change in B’s share)*(Productivity difference of B from overall productivity at Date T) = (20/4010/30)(3-2)+(10/40-10/30)(2-2) = 1/6. The calculation in the first set of parentheses shows that the employment share of establishment A increased from 1/3 to 1/2. Since establishment A had a productivity level of 3, which is higher than the average productivity level of 2, this term contributes positively. Similarly, the calculation in the second term captures the fact that the share of establishment B decreased, but it had the same productivity level as aggregate productivity and thus makes no contribution to aggregate productivity. 3. Cross component: This term is less intuitive, but it is simply computed by multiplying changes in shares and changes in productivity and summing them across all continuing establishments: (Change in productivity at A)*(Change in A’s share) + (Change in productivity at B)*(Change in B’s share) = (6-3)(20/40-10/30)+(4-2)(10/4010/30) = 1/3. Establishment A increased both its employment share and its productivity, and thus the first term is positive. However, part of this positive contribution is offset by the second term, which is negative because the share of establishment B decreased. 4. Net entry component: This term represents the difference between the contributions of the entry and exit components. The contribution of entry is expressed as a weighted average of the productivity of entering establishments relative to overall 20 Q3 2008 Business Review productivity at the initial date, T. In this simple example, there is only one entering establishment, namely, D. It is therefore computed as: (Productivity difference of D from overall productivity at Date T)*(D’s share at Date T+1) = (4-2)10/40=1/2 Note that at the initial period, establishment D has a higher productivity level, 4, than the overall productivity level of 2, and this positive contribution is multiplied by the share of employment at date T+1. Similarly, the contribution of exits is expressed as a weighted average of the productivity of exiting establishments. (Productivity difference of C from overall productivity at Date T)*(C’s share at Date T) = (1-2)10/30=-1/3 The calculation in the parenthesis reflects the fact that the productivity level of establishment C is lower than the average level at date T. Relative productivity is weighted by the employment share: 1/3. The net entry term is calculated as a difference between the two terms: (Contribution of entry) – (Contribution of exit) = 1/2-(-1/3) = 5/6. The exit of establishment C contributes positively to changes in overall productivity because establishment C had lower-than-average productivity, while the entry of establishment D also makes a positive contribution because it has higher-than-average productivity. This pattern is consistent with creative destruction. Summing over all four components we find that 5/3+1/6+1/3+5/6=3, which indeed gives the aggregate productivity gain observed between the two dates. www.philadelphiafed.org Ten Years After: What Are the Effects of Business Method Patents in Financial Services?* BY ROBERT M. HUNT I n recent years, the courts have determined that business methods can be patented and the United States Patent and Trademark Office has granted some 12,000 patents of this sort. Has the availability of patents for business methods increased the rate of innovation in the U.S. financial sector? The available evidence suggests that there has been no significant change in the aggregate trend of R&D investments made by financial firms. In this article, Bob Hunt discusses how recent court decisions and proposed federal legislation may change how firms enforce their patents. In addition, he outlines some of the remaining challenges that business method patents pose for financial companies. A decade has passed since American courts made clear that methods of doing business could be patented. Since then, the U.S. Patent and Trademark Office (USPTO) has granted more than 12,000 of these patents, Bob Hunt is a senior economist in the Research Department of the Philadelphia Fed. This article is available free of charge at www. philadelphiafed. org/research-and-data/publications/. www.philadelphiafed.org of which only a small share has been obtained by financial firms. A number of lawsuits have been filed, and a number of financial settlements, some involving significant sums of money, have occurred. Has the availability of patents for business methods increased the rate of innovation in the U.S. financial sector? This is a difficult question to answer, in part because our official measures are not well suited for estimating *The views expressed here are those of the author and do not necessarily represent the views of the Federal Reserve Bank of Philadelphia or the Federal Reserve System. research activity in financial services. Nevertheless, the available evidence suggests that there has been no significant change in the aggregate trend of R&D investments made by financial firms. Business method patents are probably here to stay. But recent court decisions and proposed federal legislation are likely to change how firms enforce their patents. These changes should mitigate some of the concerns raised about business method patents: that the claimed inventions are not new, are not sufficiently novel to justify the award of a patent, and are being enforced in ways that increase business risk to financial firms. Nevertheless, significant challenges remain. In particular, the boundaries of the rights being granted in at least some business method patents are not sufficiently clear. Ambiguity over these boundaries creates uncertainty for both the owners of these patents and their competitors. BACKGROUND A patent is a grant of the legal right to exclude others from making, using, or selling the patented invention for a limited period of time. If the patent is infringed, the patent owner may sue the infringer to recover lost profits. Sometimes the patent owner is able to obtain an injunction — a court order that prevents the alleged infringer from continuing to make, use, or sell the patented invention. For reasons described below, an injunction is a very powerful legal weapon in patent litigation. But not all inventions qualify for patent protection. To qualify, an Business Review Q3 2008 21 invention must satisfy a number of statutory requirements, including what the law describes as nonobviousness. This prevents the grant of a patent for an invention that would have been obvious to a practitioner in the relevant field at the time the invention was made. In other words, a patentable invention must be more than a trivial extension of what is already known (the prior art). As an example, consider one of the patents examined in the Supreme Court decision in Graham v. Deere.1 The claimed invention was a combined sprayer and cap used on bottles of household chemicals. The essential elements of the sprayer had been developed by others, but they had never been assembled in this particular way, which made possible the use of automated bottling equipment. As a result, the product was highly successful. While the Supreme Court acknowledged that long-felt need and commercial success might suggest the invention was nonobvious, in the end it decided otherwise because the differences between the product’s design and that of pre-existing products were minimal. Patentable Subject Matter. In the U.S., assuming the criteria just described are also satisfied, any process, machine, manufacture, or composition of matter, or any improvement of those things can be patented. But the courts have also identified certain categories of subject matter that cannot be patented, for example, laws of nature and abstract ideas. For at least 80 years, it was commonly believed that these limitations precluded patenting methods of doing business. This view was suddenly upended by the Federal Circuit’s State Street decision in 1998.2 That case involved a patent on a data processing system that made possible the pooling of assets in several mutual funds into a single portfolio, reducing overhead costs while maintaining the transaction information necessary for allocating gains, losses, and tax liabilities to the original funds. The district court determined that the invention in half or more of all the patents depicted in Figure 1 fall into categories of technology directly related to the provision of financial services. In addition, the vast majority of business method patents (roughly four in five) would also qualify as software patents.3 Classifying the industrial mix of the owners of business method patents can be difficult. Nevertheless, it is clear that when compared with firms in In the U.S. any process, machine, manufacture, or composition of matter, or any improvement of those things can be patented. But the courts have also identified certain categories of subject matter that cannot be patented. question was a business method and was therefore unpatentable. But the Federal Circuit concluded that, under U.S. law, there was no such thing as a subject matter exception for business methods. Business Method Patenting Grows Rapidly. The State Street decision had an almost immediate effect in terms of patenting behavior. About 1,000 patents for computer-implemented business methods were granted in each year after 1999 (Figure 1). Some examples are found in 10 Business Method Patents Granted in 2008. An inspection of random business method patents reveals that many are not directly related to the financial industry (there are many patents on postage-metering systems, for example). Nevertheless, the information and communications technology sector (for example, computers, software, and communications equipment), financial institutions are relatively minor players. Very roughly speaking, manufacturers of electronics, computers, instruments, and software account for at least a third, and likely much more, of business method patents granted in the last five years.4 In contrast, and again speaking very roughly, financial firms and providers of consumer payment services account for less than one-tenth of the total. Nevertheless, a number of financial institutions have accumulated a dozen or more these patents.5 3 See the data appendix for definitions and additional information. 4 2 1 The Supreme Court wrote a combined decision for three patent cases. The patent I describe here was at issue in Calmar, Inc. v. Cook Chemical Co. 22 Q3 2008 Business Review See also the Federal Circuit opinion in AT&T v. Excel Communications. The Federal Circuit is the sole court of appeals from federal district courts for patent cases. Federal Circuit decisions can be appealed to the Supreme Court. The leading recipients include IBM, Sony, Hewlett-Packard, Fujitsu, Hitachi, NCR, and Microsoft. 5 Among others, these include American Express, Citibank, JPMorgan Chase, Capital One, and Goldman Sachs. www.philadelphiafed.org TABLE 10 Business Method Patents Granted in 2008 Description* Company** A method and system for predicting changes in interest-rate sensitivity induced by changes in economic factors that affect the duration of assets and liabilities, including core deposits (no. 7,328,179). McGuire Performance Solutions, Inc. A method and system for calculating marginal cost curves for electricity generating plants (no. 7,333,861). NeuCo, Inc. A method of selecting sector weights and particular securities for a stock portfolio (no. 7,340,425). First Trust Portfolios A system and method of calculating prepayment and default risk, loss given default, and default correlations for the purpose of valuing a portfolio of assets (no. 7,340,431). Freddie Mac A machine and computer program that enables the pricing of auto insurance based on the risk associated with driving at particular locations and times (no. 7,343,309). International Business Machines Corp. A system and method for trading pollution emission allowances (no. 7,343,341). Chicago Climate Exchange, Inc. A computer-implemented method of computing price elasticities, choosing from one or more demand models based on goodness of fit (no. 7,343,355) i2 Technologies US, Inc. A method of assessing the capital adequacy of an automotive finance company (no. 7,346,566). Ford Motor Company A method of creating a customized payment card, based on a consumer’s instructions/images, via a website (no. 7,360,692). AT&T Delaware Intellectual Property, Inc. A method of sharing the profits generated by a payment card program, in excess of some target, with users of the card (no. 7,360,693). JPMorgan Chase Bank, N.A. * The author’s interpretation, based on the patent’s claims or description of the invention Initial assignee on the patent document ** www.philadelphiafed.org Business Review Q3 2008 23 FIGURE 1 Business Method Patents* 3,000 2,500 2,000 1,500 1,000 500 0 76 78 80 82 84 86 88 90 92 94 96 98 00 02 04 06 08e Calendar Year Patent Was Granted Source: U.S. Patent and Trademark Office and author’s calculations * These are patents in Class 705 (Data Processing: Financial, Business Practice, Management, or Cost/Price Determination) in the USPTO’s patent classification system. The 2008 total is estimated from five months of data. See the appendix for details. Between 1997 and 1999, new applications for business method patents tripled, and they have more than tripled since then. Today about 11,000 new applications for patents on business methods are filed each year, which suggests that there will be significant future growth in the number of patents granted. Over 40,000 of these applications are currently pending. ARE FINANCIAL SERVICES SPECIAL? An important question to ask is whether there are characteristics of the financial services sector that might make us think differently about how 24 Q3 2008 Business Review intellectual property influences decisions and outcomes among financial firms. For example, how do these firms protect their innovations in the absence of patents? Are there special interactions between network effects, which are important in many areas of finance, and intellectual property? What challenges does intellectual property pose for standard-setting, which is essential for coordinating the interactions of hundreds, if not thousands, of financial institutions acting on behalf of millions of clients? Protecting Innovations in the Financial Sector. In many theoretical papers, patents are considered essential for protecting the fruits of innovation. Without them, inventions might be quickly copied by imitators, leaving the inventor without a means of recovering her costs. This would reduce the incentive to do R&D in the first place and hence the rate of innovation. In practice, however, firms employ other means of protecting their innovations. Surveys of manufacturing companies in the 1980s and 1990s report that only a few industries (chemicals and pharmaceuticals) view patents as the primary means of protecting the profits generated by an invention.6 Other factors, such as lead time or proprietary knowledge maintained as a trade secret, were typically ranked as more important than patents.7 In addition, firms in most industries viewed their investments in specific manufacturing capabilities, reputation, brand names, and distribution networks as more important mechanisms than patents for protecting their innovations. Such investments are sometimes described as complementary assets. Consider the example of the semiconductor firm Intel. While the firm invests heavily in patents, much of Intel’s success is derived from its ability to design and build new factories (which produce only the latest CPU chips) 6 See the working paper by Wesley Cohen, Richard Nelson, and John Walsh. Evidence from earlier surveys is found in Edwin Mansfield’s article and the article by Richard Levin and his co-authors. 7 A trade secret is certain confidential information, such as a formula or a production technique, that a firm tries to keep from being disclosed. The firm can sue a person (or another company) for stealing or disclosing this information, but it cannot prevent others from independently discovering and using such knowledge. 8 These are described in William Silber’s article, Peter Tufano’s 1989 article, and John Caskey’s working paper. For a recent review of the literature on financial innovation, see Tufano’s 2003 book chapter. www.philadelphiafed.org more rapidly than its competitors. While those surveys focused on manufacturing firms, other researchers have documented similar lessons for innovations in financial services. For example, despite rapid imitation there appears to be persistent first-mover advantages (reflected primarily in market share) among firms developing new securities or option contracts.8 Some studies find that larger investment banks and mutual fund companies tend to innovate more frequently than smaller ones. This pattern is consistent with the idea that financial firms are able to leverage complementary assets to protect their innovations.9 In conclusion, it appears that financial firms typically protect their innovations in much the same way as do manufacturing firms. Historically, patents have not been a significant part of the story for financial firms, and yet their absence has not prevented them from investing in new products (financial instruments) or the processes required to offer them. The question is then whether the addition of financial patents to the mix can improve on the existing incentives and thus increase the rate of innovation in this sector. Network Effects and Standards. Many financial markets are subject to network effects: Users find the services provided are more valuable when there are many other users of the service. Two obvious examples include payment systems and financial exchanges. Consumers are more willing to carry a payment card when they know it will be accepted by most of the merchants they frequent. Merchants are more willing to incur costs to accept a payment card if they know there are 9 See Tufano’s 1989 article and his 1993 book chapter with Erik Sirri. www.philadelphiafed.org many potential customers who want to use them. In the case of financial exchanges, efficiency is often determined by the number of active buyers and sellers of a security. This creates a tendency to concentrate trading of an instrument on just a few (or even one) exchanges. As these examples suggest, networks are difficult to start, but once Network effects also arise from the requirements of interoperability, which is extremely important in financial services. they attain a critical mass, they often enjoy a large market share and generate considerable income. Network effects also arise from the requirements of interoperability, which is extremely important in financial services. Interoperability is accomplished via standard setting, where industry participants agree on technical features so that their systems can work together. Two examples are the specification of the layout and numbering systems of paper checks and the message formats used by automated clearinghouse (ACH) networks for direct deposit of paychecks and other transactions. Network effects have two implications for thinking about patents. First, they are an example of complementary assets that may permit financial institutions (or networks) to protect their innovations even in the absence of strong intellectual property rights. Second, networks are vulnerable to hold-up by third parties who own patents allegedly infringed by members of the network. Hold-up means that a patent owner may obtain an injunction, effectively shutting down the network. This puts the patent owner in a strong bargaining position in licensing negotiations. It is possible, then, for the patent owner to obtain income in excess of the incremental value created by the underlying invention. The source of that additional income is the value created by the size of the established network. Consider the case of Research in Motion (RIM), not only the developer of the BlackBerry but also the builder of the servers and software that make it work. RIM was sued by a patentholding company, NTP, whose primary investment was its portfolio of patents. RIM, on the other hand, had invested about $1 billion in property, equipment, and R&D. NTP won the case and was eventually granted an injunction that would shut down the RIM network in the U.S. This induced RIM to settle the litigation for about $600 million. Ironically, while NTP was very successful in court, the U.S. patent office, on re-examination, rejected many of NTP’s patent claims.10 A similar problem can arise with standard setting, since firms have limited options to make technical changes without sacrificing interoperability. Suppose a third party subsequently obtains a patent that is infringed by firms complying with the standard. The patent owner may enjoy considerable bargaining power. This is especially the case when implementing the standard requires significant up-front investments that firms will be hesitant to abandon simply to avoid infringing the patent. A key concern here is the effect of such risks on dynamic incentives. Companies may not be aware of all 10 NTP appealed at least one of those decisions. Business Review Q3 2008 25 of the patents that may arise and who owns them, at the time they are required to make their investment decisions. The risk of potential hold-up may discourage firms from investing in the first place. Such lost investment would be particularly costly, since it would otherwise enhance the network and, in turn, reinforce the positive externalities that network effects convey. Alternatively, such risk may increase the barriers that must be overcome in order for a network to reach a critical mass. In other words, some networks might never form.11 HAS THE AVAILABILITY OF BUSINESS METHOD PATENTS INCREASED FINANCIAL INNOVATION? It is always difficult to establish a cause-and-effect relationship between a policy change and subsequent economic outcomes. This is especially difficult in this case because there is no systematic data on the volume of these innovations over time. Ordinarily changes in the number of patents might be used. But in this case such changes might simply reflect the fact that obtaining business method patents became much easier after the decision in State Street. Measuring R&D. If the outputs of financial innovation are difficult to measure, another approach is to examine changes in the inputs, specifically research and development (R&D). The first items to look at, then, are the measures of R&D spending obtained from the National Science Foundation’s (NSF) regular survey of private firms. The NSF has published these data for most years since 1958. It began reporting R&D statistics for firms in finance, insurance, and real estate (FIRE) only in 1995. Its most recent estimate (2006) of R&D spending for this group of industries was only $2 billion, compared with more than $220 billion for all industries. The NSF reports that the majority (58 percent) of R&D spending in FIRE in 2003 was for computer software. The financial sector’s focus on software R&D is consistent with the mix of investment goods it purchases. In 1997, for example, companies in FIRE bought $30 billion in computers and software, making it the largest business customer of the information technology sector (accounting for 19 percent of sales). More than three-quarters of financial-sector investment, excluding structures, was devoted to the informa- 12 These statistics are from the article by Douglas Meade and his co-authors. tion, communication, and technology sector.12 Economists often examine R&D by comparing the size of these investments relative to sales or employment in the industry. According to the NSF data, financial services, including real estate, are significantly less R&D intensive than private industry as a whole (Figure 2). By these measures, the private economy as a whole enjoys a research intensity more than five times that of financial services. And while the R&D intensity of the U.S. economy has risen gradually over time, there has been no apparent change in the R&D intensity of financial services (the obvious spike in 2000 may reflect intense rewriting of computer code to address the century-date-change problem). It is quite possible that the NSF’s estimates for the financial sector do not reflect all of the R&D activity FIGURE 2 Research Intensity (R&D/Sales) Percent 4.0 All Industries 3.5 3.0 2.5 2.0 1.5 1.0 Finance, Insurance, & Real Estate 0.5 0 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 11 For a more detailed discussion, see my working paper with Samuli Simojoki and Tuomas Takalo. 26 Q3 2008 Business Review Source: National Science Foundation and author's calculations www.philadelphiafed.org that is actually occurring. The NSF’s methodology and the definition of R&D employed are derived from a long tradition of surveying R&D managers at manufacturing firms. In that sector, R&D facilities are relatively easy to identify, and members of senior management know who their R&D managers are. These factors make it relatively easy to conduct a survey of R&D patterns among manufacturing firms. For most financial institutions, the terms R&D, R&D lab, and R&D manager are largely foreign concepts.13 For example, in 2006 only six publicly traded financial firms reported any R&D in their financial statements, and the total amount they reported was only $65 million.14 No publicly held bank or insurance company reported doing any R&D in that year. The activities associated with developing new financial products, or better ways of delivering them, often fall outside the definition of R&D applied by official agencies. For example, a number of tax-court decisions conclude that research carried out by financial firms does not satisfy the IRS’s definition of R&D. The NSF excludes from its definition of R&D “other nontechnological activities… and research in the social sciences.”15 The development of a better credit scoring model or a new derivative contract would likely fall outside this definition. 13 For additional discussion of the issues in measuring R&D in finance and other service industries, see the 2005 National Research Council report and the report by Michael Gallaher, Albert Link, and Jeffrey Petrusa. Measuring Research Workers. Other data may shed additional light on both the level and the trend in R&D being performed in this sector. To do that, I compare the composition of the workforce in financial services with that of the private economy as The activities associated with developing new financial products, or better ways of delivering them, often fall outside the definition of R&D applied by official agencies. a whole. This may be a particularly informative measure for financial services, since 80 percent of R&D costs in this sector consist of wages and fringe benefits.16 The strategy is to identify those occupations that are most likely to be used for research and to count the number of these workers among financial services firms. To do that, I rely on the Occupational Employment Statistics produced by the Bureau of Labor Statistics. I defined a set of occupations I’ll call research occupations. This set includes all types of engineers and computer programmers and all scientists (including social scientists) and research managers.17 Physicians, teachers, and technicians in any of the above fields were excluded. Of course, not all work- 14 These data are from Standard & Poor’s Compustat. 15 The instructions for the survey forms used in the NSF survey of industrial R&D explicitly exclude the following listed categories: economics, expert systems, market research, actuarial and demographic research, and R&D in law. www.philadelphiafed.org 16 That statistic is derived from NSF data for 2002. The comparable share for all private firms is 53 percent. 17 See the appendix for a more complete list of occupations included in this definition. ers in these occupations and employed by financial firms are actually engaged in R&D; in fact, most are probably not. But I expect that this is also true of other industries. As long as the ratio of actual R&D workers to my broader measure remains constant over time, the broader measure should accurately capture any trend. For 2005, my occupational data identify about 3.2 million potential research workers. In that year, the NSF identified 1.1 million R&D workers in all industries (see the first column of Figure 3).18 In other words, for every three workers in these research occupations there was an R&D worker in the NSF counts. In financial services, my occupational measure identifies about 147,100 potential research workers in 2005, which is roughly five times the number of R&D workers (30,200) found by the NSF (see the second column of Figure 3). In the financial sector, about twothirds of potential research workers were either computer programmers or software engineers. The other third were actuaries, operations researchers, market researchers, or social scientists — occupations less likely to be reflected in the NSF counts, since work in these fields is not counted as R&D. In contrast, in all industries, 85 percent of potential research workers were engineers, programmers, or nonsocial scientists. The NSF count of R&D workers in the financial sector is likely to understate the actual number. As described in the previous section, 18 By R&D worker, I mean the count of (fulltime equivalent) scientists and engineers engaged in R&D as reported in the NSF survey of industrial R&D. The survey instructions indicate that the count should include “all persons engaged in scientific or engineering work at a level that requires knowledge of physical or life sciences or engineering or mathematics.” Business Review Q3 2008 27 this may result from the definition of R&D used and the greater difficulty in identifying where R&D is performed in financial organizations. A very crude estimate of the number of additional R&D workers in finance can be constructed using the relationships between my data and the NSF data for all industries. If those relationships also hold true in finance, there might have been another 20,000 R&D workers in that sector in 2005 (see the second column of Figure 3).19 About half of this amount may be attributable to the higher share of nontechnological occupations among workers who may be involved in developing new products or processes. Just as with R&D spending, we can create a measure of research intensity by calculating the share of an industry’s workforce that falls into the occupations included in my definition of potential research workers (Figure 4). There are several striking patterns. First, the potential research share of the financial workforce is about the same as for private industry as a whole. Second, after 1999, there is a rising trend for the entire economy. The pattern is more mixed in financial services, with increases in some years offset by declines in other years.20 The occupation-based measure of research intensity can be broken down to examine patterns within different segments of the financial services sector (Figure 5). Again, there does not appear to be a consistent trend for any of these five industries, but there are persistent differences across them. Using somewhat older data, we can examine even finer industry counts. In 2001, for example, insurance firms FIGURE 3 Research Workers, 2005 Thousands Thousands 3500 350 3000 Potential Research Workers - BLS* R&D Workers - NSF 2500 250 Nontechnological occupations* 3203.52 2000 300 Missing?* 200 1500 150 147.11 1000 500 1098 (34%) 50 (34%) 0 All Industries (left axis) { 9 11 30 (21%) 100 50 0 Financial Industries (right axis) Source: Author’s calculations using data from the Bureau of Labor Statistics and National Science Foundation * Potential research workers include programmers, software engineers, actuaries, mathematicians, operations researchers, statisticians, architects, cartographers, surveyors, all engineers, and all life, physical, or social scientists. Estimates of additional financial R&D workers assume that the true ratio of NSF R&D workers to potential research workers in financial services is identical to the ratio for all private industries (34 percent). About half of this amount may result from undercounting R&D workers who are actuaries, operations or market researchers, or social scientists. The remainder is categorized as potentially missing. See the appendix for additional information. FIGURE 4 Potential Research Workers* Share of workforce (Percent) 4 Financial Services 3 All Industries 2 1 0 90 19 Details on these calculations are found in the data appendix. 20 The BLS introduced a new occupational taxonomy in 1999, so we should be cautious about interpreting the decline from the level of the late 1990s. 28 Q3 2008 Business Review 93 97 98 99 00 01 02 03 04 05 06 Source: Bureau of Labor Statistics and author’s calculations * Potential research workers include programmers, software engineers, actuaries, mathematicians, operations researchers, statisticians, architects, cartographers, surveyors, all engineers, and all life, physical, or social scientists. www.philadelphiafed.org accounted for nearly half (48 percent) of potential research workers, followed by commercial banks (20 percent) and investment banks (12 percent). Do these data suggest that the financial services sector enjoys the same research intensity as other parts of the economy? Probably not. We know from NSF data that, compared with all private industries, financial firms spend significantly less on R&D per research worker.21 Adjusting for this difference, it would appear that financial services has a research intensity (roughly 1.3 percent) that is about 40 percent of that found in private industry as a whole. Still, this would be 2.5 times higher than reported in the NSF statistics. What can we conclude? First, the financial services sector is likely more research intensive than is reflected in the more traditional measures. Second, there is no clear trend in the research intensity of this industry. If financial patenting is having an effect, it is not easily discerned in any of the R&D measures presented. Finally, NSF data and my occupation-based measures show that ICT (especially software) are important technologies developed and employed in financial services. PATENT LITIGATION While there is little evidence of a change in R&D patterns in the industry, patent litigation involving financial firms has increased. In perhaps the first systematic study of suits involving financial business method patents, Josh Lerner found that they are litigated at a rate 27 times higher than patents in general.22 Defendants in these suits were typically large financial services firms or one of the financial exchanges. Plaintiffs were typically not financial companies. In several instances, they were patent-holding companies. In other words, they were not actively engaged in providing goods or services. Instead, they specialized in asserting, and sometimes litigating, patents. It also means they couldn’t be countersued for infringing someone else’s patents. Litigious plaintiffs have obtained significant damage awards and licensing revenues. These are usually paid by very large financial institutions or the technology companies that serve them. For example, in January 2006, the Lending Tree Exchange was found to infringe a patent on a method and sys- tem for making loan applications and placing them up for bid by potential lenders. The jury awarded $5.8 million in damages to the plaintiff, IMX, an award that was increased 50 percent in subsequent proceedings in the district court. In an unrelated case, the three American futures exchanges settled infringement suits, each involving the same patent, collectively paying about $50 million in licensing fees.23 Litigation Affecting Consumer Payments. Another important example of patent litigation involves the application of new technologies to an old payment instrument — the paper check. Check imaging and exchange technologies are especially important in the U.S. at this time. The Check Clearing for the 21st Century Act of 23 22 See the article by Mark Young and Gregory Corbett. See Lerner’s working paper. FIGURE 5 Potential Research Workers* by Financial Segment Share of workforce (Percent) Central Bank Credit Intermediation Insurance 12 Investment Banking Financial Exchanges 10 8 6 4 2 21 NSF data for 2003 show that for every dollar of R&D spent per full-time researcher in all industries, financial firms spent less than 40 cents. While some of this disparity may be due to the definitional issues described earlier, it’s unlikely they explain the entire difference. www.philadelphiafed.org 0 90 93 97 98 99 00 01 02 03 04 05 06 Source: Bureau of Labor Statistics and author’s calculations * Potential research workers include programmers, software engineers, actuaries, mathematicians, operations researchers, statisticians, architects, cartographers, surveyors, all engineers, and all life, physical, or social scientists. Business Review Q3 2008 29 2003 (Check 21) permits banks to process check transactions without physically presenting the original check to the issuing bank, so long as certain standards are satisfied.24 Financial institutions are making very large investments in technology in order to take advantage of the efficiencies afforded by this law. In January 2006, a company called DataTreasury sued Wells Fargo, 56 other banks, and a number of other firms that participate in the check-image clearing process. The company also sued the Clearing House Payments Co., which operates a check-image exchange network. DataTreasury owns at least six patents on processes for creating, processing, and storing digital images of paper checks. In earlier years it had sued a number of institutions and obtained licensing agreements with firms such as JPMorgan Chase, Merrill Lynch, and ATM manufacturer NCR Corporation. More recently, the ATM manufacturer Diebold struck a licensing agreement with DataTreasury in part to assuage bank customers who have grown increasingly concerned about their potential liability for patent infringement.25 SHIFTING SANDS? While financial patents are likely here to stay in the United States, they will be affected by a number of recent Supreme Court decisions and, quite possibly, new federal legislation. For the most part, this activity is prompted by more general concerns about the efficacy of our patent system, but some proposals are specifically directed at 24 Public Law 108-100, 12 U.S.C. 5001. 25 See the article by Steve Bills. There are at least 63 issued U.S. patents and 123 published patent applications that contain one or more references to the phrase Check 21. 30 Q3 2008 Business Review business method patents. For example, a patent reform bill (H.R. 1908) passed by the House of Representatives in 2007 would make tax-planning methods unpatentable subject matter. In 2008, the Senate Judiciary Committee reported a bill (S. 1145) that included an amendment intended to preclude patent infringement claims against institutions processing checks in compliance with the requirements of Check 21.26 The Supreme Court Speaks. In 2006 the Supreme Court decided a case involving a patent owned by the company MercExchange that a federal district court determined was infringed by eBay’s “Buy it Now” feature on its auction website. The question was whether, in addition to damages, MercExchange was also entitled to an injunction preventing eBay’s ongoing use of this feature. On appeal, the Federal Circuit concluded that injunctions should be denied to a successful plaintiff in patent cases only under exceptional circumstances. The Supreme Court disagreed, pointing to its traditional balancing test for determining the appropriateness of a permanent injunction. On retrial, the original court concluded that an injunction was not warranted.27 In 2007 the Supreme Court decided what may become the most important patent case in at least a decade. In KSR International v. Teleflex, the court considered how to determine whether an invention consisting of a 26 A cost estimate for the bill, prepared by the Congressional Budget Office, suggests that the affected patent holders would likely sue the federal government for a taking of private property. If those suits were successful, CBO estimates that the resulting compensation payments could be as high as $1 billion. combination of pre-existing elements is obvious and therefore unpatentable. With inventions like this, courts worry about the problem of hindsight bias: A novel combination of the elements seems more obvious once it has been tried and proven to work. To prevent this, the Federal Circuit created limitations on how the prior art could be interpreted to suggest that an invention was obvious. Unless a piece of prior art actually suggested the combination of ideas, the Federal Circuit typically concluded the invention was not obvious. At the extreme, to demonstrate obviousness, all the relevant aspects of the new combination must be mentioned in a single piece of prior art. Such an approach has been criticized for being too permissive, since it presumes that a person of ordinary skill in the art has little ability or creativity. Some legal scholars and economists have argued that the standard should be related to the rate of technical progress in the field. If the standard is too low, the result is less innovation in those industries that ought to be the most innovative.28 Without specifically articulating a more appropriate standard, a unanimous Supreme Court concluded that the Federal Circuit had set the bar too low: “In many fields there may be little discussion of obvious techniques or combinations, and market demand, rather than scientific literature, may often drive design trends. Granting patent protection to advances that would occur in the ordinary course without real innovation retards progress and may, for patents combining previously known elements, deprive prior inventions of their value or utility.” Business method patents are already feeling the effects of this deci- 27 Permanent injunctions in patent cases have not disappeared. In his article, Keith Slenkovich identifies 22 district court decisions after eBay where an injunction was awarded. 28 See, for example, the article by John Barton and my 2007 law review article. www.philadelphiafed.org sion. In one case (in re Trans Texas Holdings), several issued patents for a system of inflation-adjusted deposit and loan accounts were rejected on reexamination, and the Federal Circuit upheld the decision. The rejection was based on an allegedly obvious combination of two pieces of prior art. The first was a book chapter that described how, in the 1950s, Finnish banks would adjust their loan and deposit accounts for the actual inflation that had occurred. The second was a patent granted in 1983 that described how to use a data processor (for example, a computer) to manage a set of accounts. In a separate case (Advanceme Inc v. Rapidpay), a district court invalidated a patent on a computerized method for securing a loan using future credit card receivables, arguing that the claimed invention was a predictable variation of at least five card programs already in existence. Congress Deliberates. For a number of years, there has been considerable debate over the efficacy of the patent system in facilitating innovation in high-technology industries that tend to innovate cumulatively.29 This stands in contrast to the view that in other industries, such as chemicals and pharmaceuticals, where innovations tend to be more discrete, the patent system seems to be functioning reasonably well. From this debate a consensus is emerging in favor of some limited reforms. Other proposals are more controversial. Two proposals are particularly relevant for business method patents. The first is designed to increase the quality of patents issued by increasing the information available to the patent office. That information is likely to come from interested third parties if they are afforded the opportunity to contest the issuance of a pending or recently granted patent (these are called opposition procedures). Limited forms of these procedures exist under current law, but they are used infrequently. One proposal would reduce certain disadvantages that a third party might experience in any subsequent litigation involving the patent. Under the current post-grant procedure (inter partes re-examination), a third party is precluded from using any argument in subsequent litigation that it could have raised during the reexamination proceedings.30 Under the proposal, the third party is precluded from using only the actual arguments it raised during the opposition. Another proposal stipulates that when the patent in question involves a combination invention, damages for infringement should be based on the incremental contribution of the patented feature to the value of the final product. This proposal is intended to address the problem of royalty stacking in information, communication, and technology industries where products may embody dozens or even hundreds of patented inventions. Some researchers and industry participants suspect that, in such environments, there is a tendency for courts to overestimate damages from the infringement of individual patents.31 They fear the resulting conflict over the division of profits may reduce the incentive to bring new products to market. Concerns about royalty stacking 30 See 35 U.S.C. § 315(c). The purpose of this restriction is to prevent abuse of the opposition process. 31 29 See the report by the Federal Trade Commission, the book by Stephen Merrill, Richard C. Levin, and Mark Myers, and the book by Adam Jaffe and Josh Lerner. www.philadelphiafed.org may also arise in the financial sector, especially given its reliance on ICT and the emphasis on software in its R&D. In particular, innovations in the processes used to provide financial services are typically cumulative in nature. As noted earlier, financial markets and payment systems often exhibit network effects. These effects create value for network participants that may complicate the estimation of the incremental benefit attributable to one of many patented inventions employed by a network. Patent Boundaries. Not all concerns about business method patents are likely to be resolved. One major concern about these patents, and software patents more generally, is that their abstractness makes it difficult to determine the actual boundaries of the property rights being granted. Using the jargon of patent law, these patents often suffer from ambiguous “claims.”32 This is problematic because if firms cannot determine what is protected and what is not, instances of inadvertent infringement are more likely to occur. Consider the analogy to property rights to land. If the boundary lines between properties are consistently unclear or frequently reinterpreted over time, trespassing on another’s property would be more difficult to avoid. Even worse, there may be instances in which a person makes significant improvements to his or her property only to find he or she has built partially on another’s land. The result would be more litigation, and this additional risk might deter efficient investment in the first place. The issues are described in the article by Mark Lemley and Carl Shapiro and formally modeled in Shapiro’s working paper. For some examples from actual cases, see the testimony by John Thomas. 32 In their 2008 book, James Bessen and Mike Meurer point out that appeals over the definition of claims in a business method patent occur more than six times as frequently as for (litigated) patents in general. Business Review Q3 2008 31 CONCLUSION There is, at present, very little evidence to argue that business method patents have had a significant effect on the R&D investments of financial institutions. It is possible that the availability of business method patents has encouraged more entry and R&D by start-up firms or more efficient trading of technologies. At present, however, these represent intriguing possibilities and not outcomes that have actually been measured. In short, we still cannot determine whether financial patents are creating value for the U.S. economy. Nevertheless, business method patents are becoming commonplace. Compared with many other patents, they are litigated more often. Some of this litigation has resulted in very large settlements paid by established providers of financial services. These facts, in themselves, don’t prove anything. But combined with the lack of evidence suggesting a positive effect on R&D investments, they do suggest that there is likely scope for improving on the current business method patent bargain. From the standpoint of policy, it is important to ensure that patents are granted only for new and nonobvious business methods and that those standards are rigorous. In this light, the Supreme Court’s decision in KSR and the debate over the adoption of enhanced opposition procedures appear to be positive developments. The characteristics of financial markets — in particular, network effects and the requirements of interoperability — should affect the choice of appropriate remedies for patent infringement. At least after the eBay decision, these factors may influence when a court is willing to grant an injunction or how it will determine the damages resulting from infringement. Each of these changes suggests that we may already be in the process of increasing the benefits and reducing the costs to society of financial patents. But there is likely more work to be done. BR DATA APPENDIX Counts of business method patents consist of all patents assigned to Class 705 (Data Processing: Financial, Business Practice, Management, or Cost/Price Determination) in the U.S. Patent Classification System. The United States Patent and Trademark Office (USPTO) describes Class 705 as a collection of financial and management data processing areas, including insurance, trading of financial instruments, health-care management, reservation systems, computerized postage metering, electronic shopping, auction systems, and business cryptography. For additional information, see http://www.uspto.gov/web/ menu/busmethp/class705.htm. The estimate of business method patents that are more financial in nature is based on counts of patents falling into subclasses of Class 705 based on analysis of patents performed by CHI research in 2001. These subclasses include 1, 4, 7, 10, 16, 26, 30, 33, 45, 53, and 64-80. These exclude many of the patents primarily dealing with cryptography, postage metering, and similar technologies less closely related to the provision of financial services. The definition of software patents used to calculate the software share of business methods is the one specified in the article by Bessen and Hunt. It is based on the following search of the USPTO patent full-text database: “SPEC/software OR SPEC/computer AND program ANDNOT spec/antigen OR antigenic OR chromatography ANDNOT ttl/chip OR semiconductor OR bus OR circuit OR circuitry AND ISD/$/$/yyyy AND ccl/705/$.” The analysis of occupational data is based on the Occupational Employment Statistics compiled by the Bureau of Labor Statistics (BLS) (see http://www.bls.gov/oes/home.htm). The 32 Q3 2008 Business Review BLS has used different industrial and occupational taxonomies over the years. In particular, industries were defined using the Standard Industry Classification (SIC) system up to 2001, when the BLS switched to the North American Industry Classification (NAIC) System. The BLS used its own occupational definitions in these data until 1999, when it adopted definitions based on the Census Bureau’s Standard Occupational Classification (SOC) system. In the end, I constructed two lists of industries and three lists of research occupations that were roughly comparable over time. Note that my definition of financial services excludes real estate and holding companies. Potential research occupations include computer scientists, programmers, software engineers, actuaries, mathematicians, operations researchers, statisticians, architects, cartographers, surveyors, all engineers, and all life, physical, and social scientists. Additional details are available upon request. In the text, I suggested a potential undercounting of R&D workers in financial services of about 20,000. This was derived as follows. For all industries in 2005, the ratio of potential research workers to R&D workers identified by the NSF was 2.9:1. Dividing the 147,000 potential research workers in financial services by 2.9 yields about 50,400 jobs, about 20,200 more than found by the NSF. If, however, I exclude workers in all industries who were actuaries, operations researchers, market researchers, and social scientists, the ratio of potential research workers to NSF R&D workers falls to 2.5:1. Excluding jobs in those occupations in the financial sector leaves about 98,400 potential research workers in 2005. Dividing this number by 2.5 yields about 39,400 jobs, about 9,200 more than reported in the NSF data. www.philadelphiafed.org REFERENCES Barton, John H. “Nonobviousness,” mimeo, Stanford University (2001). Bessen, James, and Robert M. Hunt. “An Empirical Look at Software Patents,” Journal of Economics and Management Strategy, 16 (2007), pp. 157–89. Bessen, James, and Michael J. Meurer. Patent Failure: How Judges, Bureaucrats, and Lawyers Put Innovators at Risk. Princeton, NJ: Princeton University Press, 2008. Bills, Steve. “Diebold Bids to Lift Image ATMs with Patent Deal,” American Banker, 172:16 (January 24, 2007). Caskey, John P. “The Evolution of the Philadelphia Stock Exchange,” Federal Reserve Bank of Philadelphia Business Review (Second Quarter 2004). Cohen, Wesley M., Richard R. Nelson, and John P. Walsh. “Protecting Their Intellectual Assets: Appropriability Conditions and Why U.S. Manufacturing Firms Patent (or Not),” NBER Working Paper 7552 (2000). Congressional Budget Office. “Cost Estimate: S. 1145, Patent Reform Act of 2007 as reported by the Senate Committee on the Judiciary on January 24, 2008.” (February 15, 2008) at http://www.cbo.gov/ ftpdoc.cfm?index=8981 Federal Trade Commission. To Promote Innovation: The Proper Balance of Competition and Patent Law and Policy. Washington, DC: Federal Trade Commission (2003). www.philadelphiafed.org Gallaher, Michael, Albert Link, and Jeffrey Petrusa. “Measuring ServiceSector Research and Development,” NIST Planning Report No. 05-01 (March 2005). Hunt, Robert M. “Economics and the Design of Patent Systems,” Michigan Telecommunications and Technology Law Review, 13 (2007), pp. 457-70. Hunt, Robert M., Samuli Simojoki, and Tuomas Takalo. “Intellectual Property Rights and Standard Setting in Financial Services: The Case of the Single European Payments Area,” Federal Reserve Bank of Philadelphia Working Paper 07-20 (2007). Jaffe, Adam B., and Josh Lerner. Innovation and Its Discontents: How Our Broken Patent System Is Endangering Innovation and Progress, and What to Do About It. Princeton, NJ: Princeton University Press, 2004. Lemley, Mark A., and Carl Shapiro. “Patent Hold-Up and Royalty Stacking,” mimeo, University of California at Berkeley (2007). Lerner, Josh. “Trolls on State Street? The Litigation of Financial Patents, 1976-2005,” mimeo, Harvard Business School (2006). Levin, Richard C., Alvin Klevorick, Richard Nelson, and Sidney Winter. “Appropriating the Returns From Industrial Research and Development,” Brookings Papers on Economic Activity (1987), pp. 783-831. Mansfield, Edwin. “Patents and Innovation: An Empirical Study,” Management Science, 32 (1986), pp. 173-81. Meade, Douglas S., Stanislaw J. Rzeznik, and Darlene C. Robinson-Smith. “Business Investment by Industry in the U.S. Economy for 1997,” Survey of Current Business, 83:11 (2003), pp. 18-70. Merrill, Stephen A., Richard C. Levin, and Mark B. Myers, eds. A Patent System for the 21st Century. Washington: National Academies Press, 2005. National Research Council. Measuring Research and Development Expenditures in the U.S. Economy. Washington DC: National Academies Press, 2005. National Science Foundation. Survey of Industrial Research and Development. Washington: NSF (various years). National Science Foundation. Survey of Industrial Research and Development Form RD-1 Instructions (2005); available at http://www.nsf.gov/statistics/ srvyindustry/ surveys/srvyindus_rd1i_2005.pdf National Science Foundation. U.S. Business R&D Expenditures Increase in 2006. NSF Info Brief 08-313, 2008. Shapiro, Carl. “Injunctions, Hold-Up, and Patent Royalties,” mimeo, University of California at Berkeley (2006). Silber, William L. “Innovation and New Contract Design in Futures Markets,” Journal of Futures Markets, 1 (1981), pp. 123-55. Sirri, Erik, and Peter Tufano. “Competition and Change in the Mutual Fund Industry,” in Samuel L. Hayes, III, ed., Financial Services: Perspectives and Challenges. Boston: Harvard Business School Press, 1993. Business Review Q3 2008 33 REFERENCES Slenkovich, Keith. “Triple Dose of Bad News to Non-Practicing Patent Holders,” IPFrontline (August 29, 2007); from http://www.ipfrontline.com/depts/article. asp?id=15866&deptid=7. Thomas, John R. Prepared statement for U.S. House of Representatives Hearing on H.R. 1908, The Patent Reform Act of 2007. Hearings before the Subcommittee on Courts, the Internet, and Intellectual Property of the Committee on the Judiciary (April 26, 2007). Tufano, Peter. “Financial Innovation and First-Mover Advantages,” Journal of Financial Economics, 25 (1989), pp. 213-40. Tufano, Peter. “Financial Innovation,” in G.M. Constantines, M. Harris, and R. Stulz, eds. Handbook of the Economics of Finance, Vol. 1a: Corporate Finance. Amsterdam: Elsevier, 2003. 34 Q3 2008 Business Review Young, Mark, and Gregory Corbett. “Futures Patent Litigation: A new Competitive Force,” Futures Industry Magazine (January/February 2005); from http://www.futuresindustry.org/fi-magazinehome.asp?a=979 eBay Inc. et al. v. MercExchange, L. L. C. 126 S. Ct. 1837 (2006). Graham v. Deere 383 U.S. 101 (1966). IMX, Inc. v. Lendingtree, LLC 469 F. Supp. 2d 203 (D. Delaware 2007). Cases Cited Advanceme Inc v. Rapidpay, LLC, et al., Case No. 6:05 CV 424 (E.D. Texas 2007). in re Trans Texas Holdings Corp. 498 F.3d 1290 (Fed Cir 2007). AT&T v. Excel Communications 172 F.3d 1352 (Fed Cir 1999). KSR International Co. v. Teleflex Inc. 127 S. Ct. 1727(2007). Calmar, Inc. v. Cook Chemical Co, 336 F.2d 110 (8th Cir. 1964) State Street v. Signature Financial Group 149 F.3d 1368 (Fed Cir 1998). DataTreasury Corporation v. Wells Fargo & Co. No. 2:06-cv-00072 (E.D. Texas 2006). www.philadelphiafed.org Innovation and Regulation in Financial Markets* A Summary of the 2007 Philadelphia Fed Policy Forum BY LORETTA J. MESTER “I nnovation and Regulation in Financial Markets” was the topic of our seventh annual Philadelphia Fed Policy Forum, held on November 30, 2007. This event, sponsored by the Bank’s Research Department, brought together economic scholars, policymakers, and market economists to discuss and debate the consequences of financial innovation and the implications for financial market regulation. The recent events in financial markets and their effects on the real sector of the economy underscore the importance of greater understanding and further research on these topics. The planning for our 2007 Policy Forum began well before the onset of the financial market disruptions in the summer of 2007. By the time of our conference on November 30, 2007, the timeliness of the topic – innovation and regulation in financial markets – could not be denied. The continued problems in the financial markets, which began with subprime mortgages but expanded to other financial instruments, the ensuing spillovers from the financial market disruptions to the real sector of the economy, and Loretta J. Mester is a senior vice president and director of research at the Federal Reserve Bank of Philadelphia. This article is available free of charge at www.philadelphiafed.org/research-and-data/ publications/. www.philadelphiafed.org the steps taken by the Federal Reserve and the U.S. Treasury to help ensure financial stability have led to various proposals for new regulatory structures to help limit systemic risk in our evolving financial markets. Given the importance of the financial markets to our economy, it is vital that we get the reforms right. Better understanding of the pros and cons of financial innovation and financial market regulation – the topic of our 2007 Policy Forum – is an important step in doing so. Charles Plosser, president of the Federal Reserve Bank of Philadelphia, provided opening remarks and outlined the Policy Forum’s three sessions. He pointed out that whenever there is innovation, regulation often follows. By its very nature innovation is *The views expressed here are those of the author and do not necessarily represent the views of the Federal Reserve Bank of Philadelphia or the Federal Reserve System. a messy process with winners and losers. Market discipline is an important part of the process, helping to weed out flawed from beneficial innovations. But the fact that there are winners and losers sets up an environment that is ripe for regulation. Our first session addressed issues in corporate governance. In financial markets, the innovation of high-yield bonds contributed to a boom in corporate restructuring and buyouts, which in turn led to changes in corporate governance structures. The boom and bust in technology stocks highlighted some of the shortcomings of these new governance structures, leading to the passage of the Sarbanes-Oxley Act in 2002. Some have argued that Congress was too quick to act. Plosser asked whether there were lessons to be learned for our current situation. Our second session examined several innovations in financial markets and the role regulation may play in helping innovations yield more efficient economic outcomes. Regulation and innovation are interrelated – regulation, or the desire to evade regulation, can help spur innovation. Some of these innovations may be inefficient and some may fail, causing painful corrections. The current situation is a case in point. Better understanding of the interplay between innovation and regulation may help us avoid these types of situations in the future. Our third session covered the role of regulation in financial markets. Technology can spur innovation, but regulation also affects the way markets function. For example, different regulatory structures can affect the competitiveness of financial markets. Business Review Q3 2008 35 As Plosser pointed out, there are subtle trade-offs in the benefits and costs of particular types of regulation, and these have implications for the health of financial markets. Thus, assessing the costs and benefits will be an important part of redesigning our financial market regulatory structure. CORPORATE GOVERNANCE1 Roberta Romano, of the Yale University Law School, began the first session with a discussion of the Sarbanes-Oxley Act and its effect on cor- Roberta Romano, Yale University Law School porate governance. Romano pointed out that the act was passed swiftly with little opposition, but since then, some flaws in the act have become apparent and four major commission reports on the act have been published. (One of these, by the Committee on Capital Market Regulation, was discussed by committee co-chair R. Glenn Hubbard in the final session of our Policy Forum.) Two criticisms are that the costs of compliance are disproportionately high for smaller public firms and that there has been an adverse impact on U.S. capital markets’ competitiveness. Some have recommended that small firms and foreign firms be exempted. The Securities and Exchange Commis- 1 Some of the presentations reviewed here and background papers are available on our website at www.philadelphiafed.org/research-and-data/ events/. 36 Q3 2008 Business Review sion (SEC) has rejected those recommendations, but it has tried to lower the compliance burden on small firms and has made it easier for foreign firms to de-register and leave U.S. markets. The rationale for the latter is that foreign firms may be less reluctant to enter U.S. markets if they feel the costs of leaving are not too high. Romano’s research has attempted to assess the probability that the act will be revised.2 This involves assessing the political climate for such a revision, which is usually difficult even when a piece of legislation’s flaws are apparent. As part of her assessment, she has examined how Sarbanes-Oxley has been covered by the business press. Coverage of Sarbanes-Oxley has increased over time, with the national press focusing more on the issue of competitiveness and the regional press covering both issues of competitiveness and small-firm impact. As coverage has increased, so have congressional hearings into Sarbanes-Oxley and the introduction of bills that call for some revisions to the act, most of these focused on exemptions for small firms and/or community banks. However, Romano pointed out that it took over 60 years to repeal the Glass-Steagall Act, which separated commercial banking from investment banking, and she thinks it would take a major shift in the political environment before revision of Sarbanes-Oxley would move quickly. She said that despite increasing dissatisfaction with the SarbanesOxley Act, it could take some time before its flaws are addressed, given the difficulty in altering the status quo within our political system. Bengt Holmstrom, of the Massachusetts Institute of Technology, continued the discussion on corporate governance. He took two perspectives: the first as a member of a board of directors and the second as an academic who has studied the issue. Holmstrom has been on the boards of several companies, including a family firm for 20 years and Nokia, the global mobile telecommunications company, for the past nine years. In his view, the academic literature on the corporate governance scandals hasn’t understood the reason the scandals occurred because it hasn’t understood the role of the board. The corporate governance discussion has focused too much on executive compensation. In Holmstrom’s view, everyone agrees that we got executive compensation wrong, but the literature attributes this to weak boards and their failure to intervene. It concludes that the corporate governance system is fundamentally flawed and therefore in need of wholesale reform, with shareholders gaining significantly more power. Holmstrom disagrees. He pointed out that if you look at the longer record of the U.S. corporate governance system, it has performed extraordinarily well and there is nothing better elsewhere in the world. He cautioned that a wholesale change would be difficult to unwind, as Romano pointed out earlier. Thus, it is important to understand the source of the corporate governance 2 See for example, Roberta Romano, “The Sarbanes-Oxley Act and the Making of Quack Corporate Governance,” Yale Law Journal, 114 (2005), pp. 1521-1611. Bengt Holmstrom, Massachusetts Institute of Technology www.philadelphiafed.org scandals in 2000 before advocating a completely different system.3 In Holmstrom’s view, flaws in the design of executive compensation schemes, which gave executives powerful incentives to act on their own behalf rather than on the behalf of stockholders, did contribute to the accounting scandals. But weak auditing systems, which allowed the executives to act in this way, were also part of the problem. Some argue that the scandals occurred because the board members weren’t strong enough against the executives, and therefore, stockholders need to be given more rights in intervening in the running of the firm. Holmstrom says that there will be costs associated with such a system. His experience suggests that boards should not be watchdogs over CEOs. Instead, the board’s role is to evaluate whether the CEO and management know what they are doing and have the ability to get the firm out of a crisis should one arise – to evaluate the management team’s capabilities for running the firm, not to determine whether the team is pilfering the firm. The board needs a trusting relationship with the CEO; it needs open communication to know how the CEO approaches problems in order to assess whether this is the right person to be running the company. It would be difficult to have such a relationship if the board were always investigating the CEO. Holmstrom believes one needs to consider how proposed reforms would affect the communication relationship between the board and CEO. If shareholders can intervene significantly into how the firm is run this could adversely affect this communication relationship. While Holmstrom believes that shareholders should have the right to fire the board, he doesn’t believe they should be able to fire one member selectively, since that might prevent board members from effectively doing their job. Holmstrom is also skeptical of some of the reforms being proposed for executive compensation schemes. For example, the use of options in executive compensation schemes arose because of problems with the performance plans used in the 1960s and 1970s. Now the pendulum has swung back to such plans in which accounting numbers are used as triggers for how much to pay people. Holmstrom prefers payment schemes that are simpler but more transparent, since he believes such plans would yield better incentives. A compensation scheme that aims to give the executive a sufficiently high stake in the firm over time should yield better incentives. Franklin Allen, of the Wharton School, University of Pennsylvania, expanded the discussion to corporate governance outside the U.S. Allen noted that the notion of a corporation’s purpose differs across countries. When you ask executives (or his MBA students) from countries where the spoken language is English and those 3 For further discussion, see Bengt Holmstrom and Steven N. Kaplan, “Corporate Governance and Merger Activity in the United States: Making Sense of the 1980s and 1990s,” Journal of Economic Perspectives, 15 (Spring 2001), pp. 121-144 and Bengt Holmstrom and Steven N. Kaplan, “The State of U.S. Corporate Governance: What’s Right and What’s Wrong,” Journal of Applied Corporate Finance, 15 (Winter 2003), pp. 8-20. www.philadelphiafed.org Franklin Allen, The Wharton School, University of Pennsylvania from non-English speaking countries whom a company is there for, you get radically different answers. Those from English-speaking countries say the company is there for the shareholders. Those from non-English speaking countries say it is there for all stakeholders–shareholders, employees, bondholders, and customers. If you ask whom the company should look after if things go bad, you again get different answers. In Japan, 97 percent say job security is most important. In Germany and France, a strong majority also favors maintaining employment. But in the U.S. and the U.K., maintaining dividends is significantly more important. These differences also mean there will be differences in corporate governance structures across countries. The U.S. and the U.K. have specific laws stating that the managers’ duty is to the shareholders’ interests. In Germany, employees have a 50 percent representation on the firm’s supervisory board, which oversees the management board. Thus, workers’ interests are taken into account in the firm’s strategic decisions. China has recently introduced mandatory representation of workers on boards. In France, while it is not mandatory to have workers on boards, workers do have the right to attend board meetings. In Finland, companies can choose whether to have workers on their boards, and many companies have chosen to have worker representation. Allen points out that despite the existence of different systems, the corporate governance literature has focused only on shareholder value, at least until recently. One question of interest is which system is better in terms of allocating society’s resources most efficiently. We know from economics that if markets are complete, there is no asymmetric information, and there is perfect competition, then maximizing shareholder Business Review Q3 2008 37 value is efficient. However, if there are market imperfections, it isn’t clear this yields the best outcome. Some of Allen’s ongoing theoretical research with Elena Carletti and Robert Marquez indicates that when there are imperfect markets, shareholders as well as workers may be better off when workers are represented on a firm’s board.4 Worker representation changes the firm’s incentives toward taking actions that reduce the chance of bankruptcy. This leads to less competition, which takes the form of higher prices (which hurt consumers), but this in turn leads to higher expected profits and in some cases higher overall market value than when the firm acts only in shareholders’ interests. Thus, it is not always the case that workers gain at the expense of shareholders. Allen and his co-authors are also investigating when firms will choose to be stakeholderoriented versus shareholder-oriented, and what happens in product markets when firms of each type compete. The auto industry provides such an example, with firms from the U.S., a shareholder-oriented system, and Germany, a stakeholder-oriented system, competing. Questions such as these become even more important as countries such as China, with different corporate governance structures, gain global economic importance. INNOVATIONS IN FINANCIAL MARKETS Our second session delved into financial market innovations with speakers who have academic, policymaking, and market practitioner experience. John Geanakoplos, of Yale University, discussed his research on the foundations of market liquid4 See Franklin Allen, Elena Carletti, and Robert Marquez, “Stakeholder Capitalism, Corporate Governance, and Firm Value,” Wharton Financial Institutions Center Working Paper 07-39, August 4, 2007. 38 Q3 2008 Business Review ity and financial crises. This research yields several policy implications for the current period of financial market distress. Geanakoplos noted that the interest rate has played a central role in economics for more than a century, but during crises, collateral levels and margins, which he sees as synonymous with leverage, become paramount. In these situations, the interest rate may not move at all, but the economy is transformed by radical shifts in margins and collateral levels. Thus, in his view policymakers may want to pay more attention to collateral levels and less attention to interest rates. Just as supply and demand determine the interest rate in equilibrium, in Geanakoplos’s theory they also determine the equilibrium margin. There is a leverage cycle in which the economy can go from having too much leverage to too little. Consider an economy that has too much leverage, that is, where margins are very low. If John Geanakoplos, Yale University there is a spate of bad news that lowers expected values but increases expected volatility, individuals may demand more margin to cover their higher risk, and the situation becomes one in which there is too little leverage in the market. Geanakoplos pointed to several historical episodes in which there were extreme changes in margins: the 1994 derivatives crisis, 1998 emerging markets debt crisis, and the 2007 subprime crisis (and a possible housing market crash, which he speculated might follow). Geanakoplos is a partner of Ellington Capital Management, a mortgage hedge fund, so he spoke as both an academic and a practitioner as he elaborated on the subprime crisis. In his view, the problems in the subprime market derive from the margin requirement, that is, the down payment, which prevents subprime borrowers from refinancing. Prior to the current crisis, when a subprime borrower’s mortgage rate reset at a higher level, a borrower that was in good standing was able to refinance at a lower rate. Now, both the decline in housing prices and the rise in down payment requirements have prevented such refinancings. In Geanakoplos’s view, the interest rate has not played the main role. In Geanakoplos’s theory, a liquidity crisis begins when bad news about an asset lowers its price. The owners of this asset had been the most optimistic buyers and they were leveraged because they wanted to invest more in the asset than their own resources permitted. The drop in the asset price hurts them more than others in the economy. Thus, wealth is redistributed away from the asset’s natural buyers, and this causes the asset price to fall more, which then causes a further drop in wealth, and so on. The crisis reaches its climax only when lenders then tighten the margin requirements, that is, the amount of collateral they require to back a loan. This tightening of margins may force investors to sell the asset, which leads to even greater declines in the asset’s value, and there may be spillovers to other asset prices if they also need to be sold. Of course, those who manage to survive the crisis can benefit from the buying opportunity provided by the low prices of the assets.5 In Geanakoplos’s www.philadelphiafed.org economic model, even a small piece of bad news, that is, one that results in just a small increase in the chance of a bad outcome, can have a large effect on the price if it results in driving the most optimistic buyers (who were the most leveraged) out of the market and increasing borrowing margins. The leverage cycle, then, has broad implications for the economy. While crises are, thankfully, rare events, changes in margins and the resulting problems happen more frequently. In some cases, bad news isn’t large enough to drive the optimistic buyers from the market and create a financial crisis. Instead, it raises uncertainty and disagreement about the future among people so that the less optimistic want to sell their assets, and the optimists want to buy up the assets being sold. Because margins have risen, in order to do that, the optimists need to sell some of their other assets, an action that causes their prices to fall. Thus, there is some contagion. The optimists also want to hold assets they can borrow money against, so they reallocate their portfolio. There is a flight to quality, a flight away from illiquidity. Geanakoplos pointed out that an important implication of the theory is that policymakers might want to focus more attention on regulating margins. Forcing people to have tighter margins in normal times and looser margins during crises can make society better off. Randall Kroszner, member of the Board of Governors of the Federal Reserve System, discussed the role of information in the development of new 5 For further discussion, see John Geanakoplos, “Liquidity, Default, and Crashes: Endogenous Contracts in General Equilibrium,” Advances in Economics and Econometrics: Theory and Applications, Eighth World Conference, Volume II, Econometric Society Monographs (2003), pp. 170-205. www.philadelphiafed.org financial products and lessons to be learned about risk management and regulation to help foster productive financial market innovations. The economy has benefited from innovations that have allowed capital to be allocated to its most productive uses and risks to be dispersed to a wide range of market participants. But innovations also create challenges when participants don’t have the necessary information to value new instruments. Kroszner described the typical life-cycle of a new instrument. When a new Randall Kroszner, Board of Governors, Federal Reserve System product is developed, there is usually an experimentation phase when market participants try to learn about the product’s performance and risk characteristics. The product’s characteristics are adjusted in response to market demand. Information is gathered to facilitate price discovery, the process that reveals the market-clearing price of the asset. Kroszner discussed how the lack of information, inadequate due-diligence to verify information, and lax risk management have created problems in the market for some structured finance products like SIVS, structured investment vehicles. Their complexity and the lack of information about where the underlying credit, legal, and operational risks reside have made these instruments hard to value. According to Kroszner, when market participants realize they lack the nec- essary information for price discovery, the price discovery process becomes disrupted, market liquidity can become impaired, and it may take a significant amount of investment in information gathering and time before the price discovery process can be revived. Investment in information gathering may also result in more standardization of the instrument. Kroszner pointed out several benefits of standardization. It can decrease complexity and increase transparency of the instruments. More standardization lowers the information-gathering costs, and also the transactions costs for the instrument, which in turn increases market liquidity. Kroszner suggested that improvements in standardization could help address some of the current challenges in the subprime market, perhaps facilitating the workout and loan-modification processes. He said that the Federal Reserve and other regulators have been actively encouraging mortgage lenders and servicers to work with borrowers at risk of losing their homes. Kroszner noted that the supervisory agencies and the industry are addressing the need for improved risk management, including more comprehensive due-diligence for new financial products, and better stresstesting to cover contingent exposures, market-wide disruptions, and potential contagion. John Bogle, founder and former CEO of The Vanguard Group, Inc., and president of Bogle Financial Markets Research Center – himself a financial markets innovator – provided his views on when innovation goes too far. In his view, innovation and entrepreneurship are major drivers of global economic growth, but financial innovation is unique because of the sharp dichotomy between the value of innovation to the financial institution and the value of innovation to the institution’s customers. Bogle believes Business Review Q3 2008 39 John Bogle, The Vanguard Group that institutions have a large incentive to favor complex and costly over simple and cheap. He estimates that the costs of the financial sector have risen from about $100 billion in 1990 to about $530 billion in 2006. These costs include annual expenses borne by mutual fund investors, brokerage commissions, investment banking fees, fess paid to hedge fund managers, and legal, accounting, marketing, and advertising costs. Bogle asks whether the costs of the financial sector have reached a level that exceeds the value of the sector’s many benefits. In Bogle’s view, two recent innovations in the banking industry – CDOs (collateralized debt obligations backed by pools of mortgages) and SIVs (structured investment vehicles) – are complex and costly vehicles of questionable benefit. He discussed a number of innovations in the mutual fund industry over the years that brought mutual fund managers high fees but ultimately losses to investors, including aggressive growth funds in the latter half of the 1960s, government-plus funds in the 1970s, adjustable-rate mortgage funds in the 1980s, and technology funds in the 1990s. Bogle explained that some mutual fund innovations have benefited fund investors. One of these is the money market fund, which he sees as one of the greatest innovations in the industry’s history. He also outlined several 40 Q3 2008 Business Review of the innovations of his own firm, Vanguard, which he established in 1974. These include a fund organizational structure that keeps investment costs down; the first market-index mutual fund, created in May 1975, that tracks the returns of the S&P 500 stock index; and tax-managed funds, introduced in 1993. Bogle concluded by suggesting that financial innovations nearly always create value for their creators, but that too often, in his view, these innovations have subtracted value from investors. REGULATION AND COMPETITIVENESS Bogle’s discussion provided an excellent segue into the final session, which addressed the proper role of regulation in capital markets. Given financial market disruptions that have taken place since the Policy Forum, this session provides particularly useful insights into thinking about the regulatory structure that is to come. R. Glenn Hubbard, dean of the Columbia University Business School and chairman of the Council of Economic Advisers from 2001 to 2003, focused his discussion on the regulation of equity markets, drawing on the work of the Committee on Capital Markets Regulation, a nonpartisan group co-chaired by Hubbard and John Thornton, former president of Goldman Sachs. Hubbard sees our financial markets as one of the important sources of the productivity boom in the U.S. over the past decade, and thus, it is important to preserve and enhance the global competitive position of U.S. capital markets. The U.S. share of equity raised in global public markets dropped from about 30 percent in 2002 to about 19 percent in 2007 (through November). There has been an increase in U.S. companies doing initial public offerings abroad and a decrease in the number of firms that are listing on U.S. equity exchanges. Economic research indicates that, on average, foreign firms still receive a benefit from listing in the U.S., but that listing premium has declined in recent years. Hubbard argued that one of the reasons for these trends is that the U.S. securities regulatory system does not do an adequate assessment of the costs and benefits of proposed and enacted regulations. The implementation of Sarbanes-Oxley could be improved to lower its costs. Another cost facing firms doing business in the U.S. is potential litigation. He cited issues surrounding auditor and director liability and securities class-action lawsuits, which have larger and more frequent settlements in the U.S. than in other financial centers. The committee recommended that a more risk-based approach be taken toward securities regulation to ensure that regulation enhances shareholder value by improving the incentives of managers, auditors, and directors, and the rights of shareholders with respect to corporate control. This would include the SEC’s performing formal cost-benefit analyses of regulations. Regarding litigation reform, the committee recommended allowing alternative dispute resolution for class actions, which might include shareholders waiving their rights to class actions at the time of the initial public offering. Hubbard suggested that the Financial Services Author- R. Glenn Hubbard, Columbia University Business School www.philadelphiafed.org ity in the United Kingdom, which is a consolidated system of financial supervision, might provide a model worth considering in the U.S., and indeed the costs and benefits of such a system are being assessed as part of the work regarding regulatory reforms needed in the aftermath of the current financial market disruptions. Annette Nazareth, who at the time of the Policy Forum was a commissioner at the SEC, was our final speaker. Nazareth argued that a well-conceived and balanced system of securities regulation gives the U.S. a competitive edge in global financial markets. The SEC’s balanced approach to securities regulation is based on the principles of competition, transparency, investor protection, and Annette Nazareth, Former Commissioner, Securities and Exchange Commission market integrity. Nazareth believes the approach has worked well and has been instrumental in establishing confidence in the U.S. securities markets, which in turn has increased market liquidity and has attracted business to the U.S. Indeed, rather than conflicting with market forces, high-quality regulation, she feels, is a complement that works with market forces. In Nazareth’s view, high-quality regulation is based on clear goals and standards. It should be minimally intrusive in the marketplace, allowing disparate business models to compete vigorously and effectively, which fosters www.philadelphiafed.org innovation. It should be flexible enough to accommodate different business models. It should promote market efficiency. Nazareth believes securities regulation has been most effective in addressing market externalities, a type of inefficiency. She outlined four types of externalities that regulation has successfully addressed: dominant markets, principal-agent conflicts, collectiveaction issues, and information asymmetries. Markets with high market power may use it anti-competitively. The U.S. has multiple competing securities markets, and the SEC has used its authority to enhance the competition in these markets. For example, the SEC mandated fair-access rules that ensured that all market participants would have access to the market. The commission also regulated the sharing of market price data, which is necessary for trading. As Bogle described in the previous session, financial intermediaries and their customers are in a principal-agent relationship, in which there are sometimes conflicts of interest. Nazareth discussed the SEC’s regulation of sales practices, which is intended to alleviate some of these conflicts. She pointed out that the U.S. has the highest level of retail investor participation anywhere in the world and attributes this to the standards set for sales practices, which inspire confidence in our markets. The SEC has helped solve collective action problems in financial markets. Nazareth discussed an example that arose in the over-thecounter credit derivatives markets. As Nazareth explained, it was discovered that there was a very serious problem of incomplete documentation on high volumes of transactions in this market. Although the securities firms realized there was an issue, none individually had the incentive or the ability to solve the problem on its own. The SEC has worked with the firms toward clearing up this problem. The SEC has advocated transparency and has mandated standardized disclosure to alleviate asymmetric information problems. Globalization has led to convergence in disclosure as well as accounting standards across countries, and this has raised market efficiency. In Nazareth’s view, these four types of market imperfections point out the need for regulation to ensure that the evolution of the marketplace benefits investors and serves the public interest. If regulation is well-designed, it will enhance competition, not stand it is way. SUMMARY The 2007 Policy Forum generated lively discussion among the program speakers and audience on the consequences of innovation in global financial markets and the implications for financial market regulation. The recent financial market disruptions, their effect on the real sector of the economy, and the feedback from the real economy to financial markets underscore the need for better understanding of financial market innovations, performance, liquidity, and regulation. It now appears clear that some reform of the financial supervisory system in the U.S. is needed and will take place. Given the vital importance of financial markets and institutions to our economic well being, it is imperative that rigorous economic modeling and empirical research be used in developing these regulatory reforms to avoid unintended consequences and to help ensure a more efficient and productive financial system that is less susceptible to systemic risk. BR Business Review Q3 2008 41 RESEARCH RAP Abstracts of research papers produced by the economists at the Philadelphia Fed You can find more Research Rap abstracts on our website at: www.philadelphiafed.org/research-and-data/ publications/research-rap/. Or view our Working Papers at: www.philadelphiafed.org/research-and-data/ publications/. REAL-TIME DATA ANALYSIS: A LOOK AT RESEARCH TO DATE This paper describes the existing research (as of February 2008) on real-time data analysis, divided into five areas: (1) data revisions; (2) forecasting; (3) monetary policy analysis; (4) macroeconomic research; and (5) current analysis of business and financial conditions. In each area, substantial progress has been made in recent years, with researchers gaining insight into the impact of data revisions. In addition, substantial progress has been made in developing better real-time data sets around the world. Still, additional research is needed in key areas, and research to date has uncovered even more fruitful areas worth exploring. Working Paper 08-4, “Frontiers of RealTime Data Analysis,” Dean Croushore, University of Richmond, and Visiting Scholar, Federal Reserve Bank of Philadelphia INSTITUTIONAL FORM AND CENTRAL BANK PERFORMANCE: SOME EMPIRICAL EVIDENCE Over the last decade, the legal and institutional frameworks governing central banks and financial market regulatory authorities throughout the world have undergone significant changes. This has created new interest in better understanding the roles played by organizational structures, accountability, and transparency, in increasing the efficiency and effectiveness of central banks in achieving their objectives and ultimately yielding better economic 42 Q3 2008 Business Review outcomes. Although much has been written pointing out the potential role institutional form can play in central bank performance, little empirical work has been done to investigate the hypothesis that institutional form is related to performance. This paper attempts to help fill this void. Working Paper 08-5, “Central Bank Institutional Structure and Effective Central Banking: Cross-Country Empirical Evidence,” Iftekhar Hasan, Rensselaer Polytechnic Institute and Bank of Finland, and Loretta J. Mester, Federal Reserve Bank of Philadelphia and The Wharton School, University of Pennsylvania FLUCTUATIONS IN UNEMPLOYMENT AND VACANCIES In a reasonably calibrated Mortensen and Pissarides matching model, shocks to average labor productivity can account for only a small portion of the fluctuations in unemployment and vacancies (Shimer (2005a)). In this paper, the author argues that if vintage-specific shocks rather than aggregate productivity shocks are the driving force of fluctuations, the model does a better job of accounting for the data. She adds heterogeneity in jobs (matches) with respect to the time the job is created in the form of different embodied technology levels. The author also introduces specific capital that, once adapted for a match, has less value in another match. In the quantitative analysis, she shows that shocks to different vintages of entrants are able to account for fluctuations in unemployment and vacancies and that, www.philadelphiafed.org in this environment, specific capital is important to decreasing the volatility of the destruction rate of existing matches. Working Paper 08-6, “Specific Capital and Vintage Effects on the Dynamics of Unemployment and Vacancies,” Burcu Eyigungor, Federal Reserve Bank of Philadelphia OPTIMAL POLICY IN A CHANNEL SYSTEM Channel systems for conducting monetary policy are becoming increasingly popular. Despite its popularity, the consequences of implementing policy with a channel system are not well understood. The authors develop a general equilibrium framework of a channel system and study the optimal policy. A novel aspect of the channel system is that a central bank can “tighten” or “loosen” its policy without changing its policy rate. This policy instrument has so far been overlooked by a large body of the literature on the optimal design of interest-rate rules. Working Paper 08-7, “Monetary Policy in a Channel System,” Aleksander Berentsen, University of Basel, and Cyril Monnet, Federal Reserve Bank of Philadelphia EXAMINING REVISIONS TO PCE INFLATION RATES This paper examines the characteristics of the revisions to the inflation rate as measured by the personal consumption expenditures price index both including and excluding food and energy prices. These data series play a major role in the Federal Reserve’s analysis of inflation. The author examines the magnitude and patterns of revisions to both PCE inflation rates. The first question he poses is: What do data revisions look like? The author runs a variety of tests to see if the data revisions have desirable or exploitable properties. The second question he poses is related to the first: Can we forecast data revisions in real time? The answer is that it is possible to forecast revisions from the initial release to August of the following year. Generally, the initial release of inflation is too low and is likely to be revised up. Policymakers should account for this predictability in setting monetary policy. Working Paper 08-8, “Revisions to PCE Inflation Measures: Implications for Monetary Policy,” Dean Croushore, University of Richmond, and Visiting Scholar, Federal Reserve Bank of Philadelphia www.philadelphiafed.org COMBINING CPI AND PCE INFLATION MEASURES: BETTER FORECASTS? Two rationales offered for policymakers’ focus on core measures of inflation as a guide to underlying inflation are that core inflation omits food and energy prices, which are thought to be more volatile than other components, and that core inflation is thought to be a better predictor of total inflation over time horizons of import to policymakers. The authors’ investigation finds little support for either rationale. They find that food and energy prices are not the most volatile components of inflation and that depending on which inflation measure is used, core inflation is not necessarily the best predictor of total inflation. However, they do find that combining CPI and PCE inflation measures can lead to statistically significant more accurate forecasts of each inflation measure, suggesting that each measure includes independent information that can be exploited to yield better forecasts. Working Paper 08-9, “Core Measures of Inflation as Predictors of Total Inflation,” Theodore M. Crone, Swarthmore College; N. Neil K. Khettry, Murray, Devine & Company; Loretta J. Mester, Federal Reserve Bank of Philadelphia, and The Wharton School, University of Pennsylvania; and Jason A. Novak, Federal Reserve Bank of Philadelphia BUSINESS METHOD PATENTS AND THE FINANCIAL SERVICES INDUSTRY A decade after the State Street decision, more than 1,000 business method patents are granted each year. Yet only one in 10 are obtained by a financial institution. Most business method patents are also software patents. Have these patents increased innovation in financial services? To address this question the author constructs new indicators of R&D intensity based on the occupational composition of financial industries. The financial sector appears more research intensive than official statistics would suggest but less than the private economy taken as a whole. There is considerable variation across industries but little apparent trend. There does not appear to be an obvious effect from business method patents on the sector’s research intensity. Looking ahead, three factors suggest that the patent system may affect financial services as it has electronics: (1) the sector’s heavy reliance on Business Review Q3 2008 43 information technology; (2) the importance of standard setting; and (3) the strong network effects exhibited in many areas of finance. Even today litigation is not uncommon; we sketch a number of significant examples affecting financial exchanges and consumer payments. The legal environment is changing quickly. The author reviews a number of important federal court decisions that will affect how business method patents are obtained and enforced. He also reviews a number of proposals under consideration in the U.S. Congress. Working Paper 08-10, “Business Method Patents and U.S. Financial Services,” Robert M. Hunt, Federal Reserve Bank of Philadelphia IS THERE A LINK BETWEEN JOBLESS RECOVERIES AND THE GREAT MODERATION? This paper uses new data on job creation and job destruction to find evidence of a link between the jobless recoveries of the last two recessions and the recent decline in aggregate volatility known as the Great Moderation. The author finds that the last two recessions are characterized by jobless recoveries that came about through contrasting margins of employment adjustment: a relatively slow decline in job destruction in 1991-92 and persistently low job creation in 2002-03. In manufacturing, he finds that these patterns followed a secular decline in the magnitude of job flows and an abrupt decline in their volatility. A structural VAR analysis suggests that these patterns are driven by a decline in the volatilities of the underlying structural shocks in addition to a shift in the response of job flows to these shocks. The shift in structural responses is broadly consistent with the change in job flow patterns observed during the jobless recoveries. Working Paper 08-11, “Job Flows, Jobless Recoveries, and the Great Moderation,” R. Jason Faberman, Federal Reserve Bank of Philadelphia WHAT EXPLAINS THE HOME BIAS IN TRADE? A large empirical literature finds that there is too little international trade and too much intra-national trade to be rationalized by observed international trade costs such as tariffs and transport costs. The literature uses frameworks in which goods are assumed to be produced in just one stage. This paper investigates whether the multi-stage nature of production helps 44 Q3 2008 Business Review explain the home bias in trade. The author shows that multi-stage production magnifies the effects of trade costs. He then calibrates a multi-stage production model to the U.S. and Canada. He solves the model with measures of trade costs constructed from data on tariffs, transport costs, and wholesale distribution margins. The model can explain about three-eighths of the Canada border effect; this is three times more than what a calibrated one-stage model can explain. The model also explains a good deal of Canada’s vertical specialization trade. Finally, a reverse engineering exercise suggests that the unknown or unobserved component of trade costs is smaller than observed trade costs. Working Paper 08-12, “Can Multi-Stage Production Explain the Home Bias in Trade?,” Kei-Mu Yi, Federal Reserve Bank of Philadelphia WORKER TURNOVER AND FIRM GROWTH The authors use establishment data from the Job Openings and Labor Turnover Survey (JOLTS) to study the micro-level behavior of worker quits and their relation to recruitment and establishment growth. They find that quits decline with establishment growth, playing the most important role at slowly contracting firms. They also find a robust, positive relationship between an establishment’s reported hires and vacancies and the incidence of a quit. This relationship occurs despite the finding that quits decline, and hires and vacancies increase, with establishment growth. The authors characterize these dynamics within a labor-market search model with on-the-job search, a convex cost of creating new positions, and multi-worker establishments. The model distinguishes between recruiting to replace a quitting worker and recruiting for a new position and relates this distinction to firm performance. Beyond giving rise to a varying quit propensity, the model generates endogenously determined thresholds for firm contraction (through both layoffs and attrition), worker replacement, and firm expansion. The continuum of decision rules derived from these thresholds produces rich firm-level dynamics and quit behavior that are broadly consistent with the empirical evidence of the JOLTS data. Working Paper 08-13, “Quits, Worker Recruitment, and Firm Growth: Theory and Evidence,” R. Jason Faberman, Federal Reserve Bank of Philadelphia, and Éva Nagypál, Northwestern University www.philadelphiafed.org