Is Capital Expenditure Contagious? An Analysis of UCC Data from Ohio and Its Neighbors

ECONOMIC CONDITIONS ARE OFTEN OBSERVED TO BE CORRELATED ACROSS SPACE AND TIME. ONE INTERPRETATION OF THIS PHENOMENON IS THAT ECONOMIC ACTIVITY IS “CONTAGIOUS.” THAT IS, GOOD OR BAD CONDITIONS IN ONE ECONOMIC AREA MAY LATER CAUSE SIMILAR CONDITIONS TO OCCUR IN NEARBY AREAS. THE PREVALENCE AND EXTENT OF THESE RELATIONSHIPS IS IMPORTANT TO UNDERSTAND FOR THOSE SEEKING TO FOSTER REGIONAL ECONOMIC DEVELOPMENT. WE FOCUS ON CAPITAL EQUIPMENT SPENDING AT THE STATE LEVEL AT A MONTHLY FREQUENCY. THIS IS POSSIBLE GIVEN OUR ACCESS TO A UNIQUE DATA SET, THE RANDALL-REILLEY CAPITAL INVESTMENT INDEX (RRCII). THIS INDEX MEASURES CAPITAL EXPENDITURE USING UNIFORM COMMERCIAL CODE (UCC) FORMS FILED EACH MONTH WITH EACH STATE’S SECRETARY OF STATE. THE DATA IS CLASSIFIED INTO THREE INDUSTRIES: AGRICULTURE, CONSTRUCTION, AND MACHINE TOOLS, AS WELL AS A COMPOSITE MEASURE. IN THIS STUDY, WE UTILIZE THE INDEX’S STATE-LEVEL DATA FOR OHIO AND ITS NEIGHBORS: MICHIGAN, INDIANA, KENTUCKY, WEST VIRGINIA, AND PENNSYLVANIA. OUR METHODOLOGY CONSISTS OF TYPICAL TIME SERIES TECHNIQUES: GRANGER CAUSALITY TESTS, VECTOR AUTOREGRESSIONS, AND THEIR ASSOCIATED IMPULSE RESPONSE FUNCTIONS. OUR INITIAL RESULTS SUGGEST THAT MICHIGAN IS THE ONLY STATE WITH A SIGNIFICANT RELATIONSHIP WITH OHIO AT THE COMPOSITE LEVEL, BUT THAT PENNSYLVANIA AND WEST VIRGINIA SHOW SOME RELATIONSHIP WITH OHIO IN CONSTRUCTION, AS DOES INDIANA WITH MACHINE TOOLS. 1 VanderHart and Zeng: Is Capital Expenditure Contagious? Published by Carroll Collected, 2015 76 INTRODUCTION Understanding the economic relationships among adjoining geographical areas is significantly important when trying to foster development in these areas. One view is that the areas may be in competition with one another, suggesting that a given area should try to distinguish itself as it competes for firms, trained workers, and other resources. An alternative view holds that the development of adjoining regions provides positive spillovers to neighboring areas, and that developmental policies that recognize this relationship and emphasize regional cooperation may be advantageous. A mixture of the two views is likely, with different effects in different developmental dimensions complicating the issue. Furthermore, the developmental effects may occur concurrently, with a lag, or with an anticipatory lead. While a significant amount of econometric research focuses on these relationships, it suffers from limitations in available data at the local level. When it is available, quality data is often limited to only a few aspects of the economy (such as employment and housing), and often can only be obtained at a low frequency (annual or perhaps quarterly, rather than monthly). These limitations make understanding the potential mixture of competitive and complementary developmental relationships difficult, as well as obscuring any correlations across time. In this paper we report on innovations in these dimensions. We use a UCC form-derived proprietary index that measures state-level expenditures on capital expenditures in three vital industries at a monthly frequency. We focus on capital expenditure in Ohio, and analyze its relationship to capital spending in its neighboring states of Michigan, Indiana, Kentucky, West Virginia, and Pennsylvania. Our results suggest that only Michigan has a significant (and complementary) relationship when the composite of capital expenditure is examined, while Pennsylvania, West Virginia, and Indiana exhibit relationships with Ohio only in specific sectors. The next section of this paper provides a brief literature review, followed by a description of the data. The methodology and results are then presented, and a brief discussion concludes the paper. LITERATURE REVIEW: COMPETITION VS. COMPLEMENTARITY The theory and practice of local economic development often is characterized as either competitive or complementary. Competitive is different jurisdictions competing to attract new facilities and their attending capital and employment, and complementary is local areas working together to attract clusters of economic activity, with suppliers, transportation, and support industries locating across several political boundaries. The competitive view is based on the simple logic that if a facility locates in one locality, it cannot also locate in another. Papers that fall into the competitive camp include Bowman (1988), Cable and Feiock (1998), and Buss (2001). Several authors criticized competitive policies designed to attract industry to a particular area as a “raceto-the-bottom” (see for example, Goetz et al., (2011)) or “smokestack chasing” (see for example, Turner (2003). Other authors note that attracting a facility to an area may benefit adjoining areas, as the economics of agglomeration cause related firms to locate not only in the initial firm’s location, but also those adjacent. Authors in this category include Audretsch and Feldman (1996), Porter (2000), and Cowell (2010). Of course both of these forces are likely to exist in any particular situation, depending on the distance among the localities, the strength of incentives, and the strength of agglomeration. Examples of authors finding this mixture include Goetz (1993), Hawkins (2010), and Delgado et al. (2012). DATA The recent empirical literature on this topic faced restrictions on the availability of useful data. A primary concern is the frequency of available data, with some authors limited to coarse annual data (see for example Rey and Montouri (1999) or Beenstock and Felsenstein (2007)). More recently others have improved on this by utilizing quarterly data (for example Owying and Wang (2009), Kueth and Pede (2011), and Brady 2 The Journal of Economics and Politics, Vol. 22 [2015], Iss. 1, Art. 6 https://collected.jcu.edu/jep/vol22/iss1/6 77 (2014)); however, even a quarterly frequency may obscure important temporal relationships when sequential events occur in the same quarter. Moreover, as Chung (2013) points out, the lower the frequency of the data used, the more likely an analysis might mistakenly attribute any movement to a common national-level shock. To their credit, a limited number of analyses obtained and used data at a monthly frequency (Park and Hewings (2012), Chung (2013)). Regardless of the frequency used, this literature also suffers from a lack of variables that accurately describe an area’s economy. While a plethora of data may exist at the national level, state-level data is often limited to employment, income, and housing. Often authors will focus on one variable of interest such as Rey and Montouri (1999), who focus on income; or Brady (2014), who examines housing prices. Some authors construct and estimate multivariate models. Kueth and Pede (2011) make use of income, unemployment, and housing prices; while Beenstock and Felsenstein utilize earnings, population, housing price, and housing stock. Other authors choose to use or construct coincident indices (Park and Hewings 2012) or use dynamic factor models (Chung 2013) to collapse several variables (typically measures of employment and building permits) into one. Note, however, that none of these models incorporate measures of capital stock or capital expenditure into their analysis (see Chung 2013 for a discussion). METHODOLOGY The treatment of spatial relationships is a relatively new econometric endeavor. Initial work in this domain can be traced to Anselin (1988) and Blanchard et al. (1992). More sophisticated and formal treatments are due to LeSage (1999), Rey and Mourtouri (1999), and Anselin (2003). In recent years some authors sought to adapt vector autoregression (VAR) techniques to incorporate spatial concerns. These are sometimes referred to as spatial vector autoregressions (SpVAR). In these models past shocks to adjoining areas are posited to affect the area of concern. Recent work includes Beenstock and Felsenstein (2007), Holly et al. (2010), Keuth and Pede (2011), and Marquez et al. (2013). All of these models rely on severe restrictions on parameter values, as without the restrictions the number of free parameters exceeds the number of observations. Other endeavors make use of dynamic factor models (Bai and Wang (2012), Chung (2013)) to reduce the number of variables considered. These authors typically use MCMC techniques to avoid under-identification. Finally, recent work by Brady (2014) relies on spatial panel data techniques, although he restricts his analysis to a single endogenous variable. GEOGRAPHIC FOCUS Two recent papers applied these techniques to Midwestern states. Park and Hewings (2012) examine Michigan, Ohio, Indiana, Wisconsin, and Illinois. Their Granger-causality tests suggest that economic fluctuations in Ohio cause and are caused by fluctuations in the other four states. They find that employment fluctuations in these states, with the exception of Illinois, coincides with national trends, but that Illinois fluctuations lag the national trend by a few months. They hypothesize that the mix of industries in these states explains this pattern, with Illinois, relatively heavy in the service sector, responding to its manufacture-heavy neighbors with a lag. Chung (2013) adds Minnesota to the list of states above. The impulse response functions derived from his complicated MCMC estimation suggest that shocks to Ohio have positive spillover effects in the other states (although the effects are muted when a more complicated multifactor approach is used); and that there are mixed results in the other direction: Shocks to Michigan, Illinois, and Minnesota actually create weak negative effects on Ohio (although this result is dependent on the specification), while Wisconsin generates a positive effect, and Indiana’s effect on Ohio varies with the specification of the model. DATA: THE RRCII Equipment Data Associates (EDA) is a division of Randall-Reilly Publishing, headquartered in Charlotte North Carolina. They purchase UCC data f