Environmental Regulation and Innovation Offsets in the Bluegrass Seed Industry

Hicks, in The Theory of Wages, introduced the concept of induced innovation—change in relative factor prices can spur innovation to more efficiently use the factor that has become relatively more expensive. The concept of innovation has since produced an extensive literature about a wide range of sources of innovation and technological change (Ruttan).

Porter and van der Linde identified environmental regulation as a possible causative agent for innovation in arguing that “properly designed environmental standards can trigger innovation that may partially or more than fully offset the costs of complying with them” (Porter and van der Linde, p. 98). This hypothesis of “innovation offsets” in response to environmental regulation has been contested as not having strong theoretical foundations and contested on the assumption that inefficiencies in production are so abundant that all environmental regulation will result in cost-offsetting innovation (Palmer, Oates, and Portney). More recently, studies have attempted to provide formal theoretical foundations to the Porter hypothesis (see, for example, Ambec and Barla, or Mohr), test the validity of the hypothesis, and identify situations in which the hypothesis is likely to hold (Wagner).

While we do not argue here for or against the Porter hypothesis, we investigate the effects of an environmental restriction on the bluegrass seed industry in the state of Washington, and argue that this example has similarities to the innovation offsets described by Porter and van der Linde. In fact, we show that the magnitude of offsetting effects of observed innovation appear to have more than completely compensated for the increased costs imposed by an environmental regulation.

In 1996, Washington implemented a three-year phase-in of a permanent ban on burning bluegrass seed stubble. The ban was mandated by the state legislature and implemented by the Washington Department of Ecology after linking bluegrass field burning to increased patient loads at Spokane area hospitals and greater suffering by people with respiratory problems (Washington Department of Ecology). Prior to 1996, bluegrass seed stubble was burned after harvest in order to remove thatch, control weeds and pests, and increase yields by shocking the crown of this perennial plant. Industry representatives vehemently opposed the ban, arguing that the higher costs faced by farmers who would be forced to remove thatch mechanically would drive the bluegrass seed industry from Washington.

More than 90% of the bluegrass seed grown in the United States is grown in Washington, Oregon, and Idaho. Oregon had already imposed a ban on the burning of most bluegrass seed stubble before 1996, but Idaho had no ban until a court-imposed curtailment in 2007. So, because of the relatively close alternative locations where bluegrass seed could be grown at a lower marginal cost, a substantial reduction in Washington bluegrass seed production was anticipated.¹

However, the negative shock to the Washington bluegrass seed industry never materialized. Figure 1 shows that Washington production during the years 1997–2005 was the highest of any eight-year period in recorded history of the industry, in fact nearly two-thirds higher. In each of six consecutive years, 2000–2005, production was higher than in any year prior to the ban. Bluegrass seed production also increased in Oregon and Idaho.²

This paper seeks to explain this seeming economic paradox: marginal cost had to increase due to the ban or alternative production practices would have already been selected and in use before 1996; yet not only did aggregate state-level production fail to decrease, it actually increased. Believing that a better understanding of the economic processes at play in the Washington bluegrass seed industry will resolve this apparent paradox, we put forward and systematically test three hypotheses consistent with economic theory. Any of the hypotheses could explain, individually or collectively, this economic paradox. The first two hypotheses suggest that the increase in bluegrass seed supply is simply coincidental with the imposition of the burning ban, and if other mechanisms are controlled for, the seemingly positive effect of the ban will disappear. Our third hypothesis is that the ban itself, and anticipation of it, may have actually induced the increase in bluegrass seed supply. Listed in the order they are tested, our hypotheses are the following:

Note that the formulation of the third hypothesis stipulates that new techniques were introduced contemporaneously with the imposition of the ban. This seemingly instantaneous adjustment is not consistent with most formulations of the Porter hypothesis, where higher costs of production provide incentive to innovate. However, Oregon preceded Washington in restricting most burning and the discussion of the potential burning ban in Washington was public. Consequently, we hypothesize that farmers and researchers anticipated the burning ban before it was imposed, and were able to develop new techniques in bluegrass seed production, which were implemented when the actual ban came into effect.

Method of Analysis

The first and third hypotheses are tested using model structures based on the assumption that farmers are price-taking, profit-maximizing (i.e., perfectly competitive) firms and that the state and multicounty aggregates act as though they are also price-taking, profit-maximizing firms.³ The second hypothesis is tested by relaxing the assumption of risk-neutrality embodied in the profit-maximizing behavioral objective.

Because of the previously documented sensitivity of policy-relevant conclusions to choice of functional form (e.g., Berndt and Khaled; Chalfant), we conduct our tests using supply equations based on two alternative specifications. The supply equations are specified in both linear and loglinear form:

(1)

(2)

where y_k is the quantity of the kth output, x_j is the magnitude of an exogenous nonprice variable (x₁ is the portion of acreage to which the burning ban applies, x₂ is annual rainfall in the dryland seed production area of the state), p_j is expected price of the jth output normalized (i.e., divided) by an index of prices paid for factors of production, β is a parameter to be estimated, and ɛ is the error term.⁴ The outputs include bluegrass seed (output 3) and three crops that are principal competitors with bluegrass seed production: wheat (output 4), lentils (output 5), and potatoes (output 6).⁵ Each equation is linear in parameters.

The price normalization maintains the hypothesis implied by perfectly competitive behavior that the supply equations are homogeneous of degree zero in prices. When equation (1) is estimated as a system of supply equations, cross-equation restrictions maintain symmetry of cross-partial derivatives as implied by a twice-differentiable profit function. The system is estimated by iterative seemingly unrelated regression (ITSUR) to assure that the solution converges on the covariance matrix. Because the symmetry restrictions are often rejected by an empirical test, the bluegrass supply equations in (1) and (2) are also estimated individually using ordinary least squares (OLS).

The linear equations are the first derivatives of a normalized quadratic profit function. The loglinear equations are not derived from any known closed form of the profit function, but can be treated as approximations to the first derivatives of an unknown twice-continuously differentiable profit function.⁶

Linear and loglinear bluegrass harvested acreage equations with the same regressors as in (1) and (2) are also estimated to discern whether the quantity of harvested land can be explained by relative prices, the burning ban, and rainfall. They are estimated individually using OLS. Because of lack of data, no other bluegrass input demand equations are estimated.

The burning ban was phased in between 1996 and 1998, with burning prohibited on one-third of all acreage in 1996, two-thirds of all acreage in 1997, and prohibited on all fields beginning in 1998. Therefore, x_1t takes a value of 1 for each year the full ban was in effect (1998–2005), a value of 2/3 in 1997, a value of 1/3 in 1996, and a value of 0 for all years prior to 1996.

Prior to the burning ban, most (80% of statewide production and acreage) bluegrass seed was produced on farms in the dryland region of Eastern Washington and still is mainly harvested in July and August, so the variable x_2t is the recorded rainfall at the Spokane airport from September of year t− 1 through July of year t. We argue that rainfall could have had an effect on harvested acreage in addition to yield based on farmer statements that, if low rainfall results in low yields, they do not harvest all of their planted acreage. We use rainfall at the Spokane airport because Spokane County accounted for more than 50% of statewide bluegrass seed acreage prior to the ban and remains the largest producer of bluegrass seed in the state.

We maintain the hypothesis of Saha and Shumway that the random output price received by Washington farmers for annual crops is “characterized by a Markov process, which implies that all relevant information about the next period's expected price is contained in the current period's realization” (Saha and Shumway, p. 444). This hypothesis is consistent with the findings of several studies that found that the lagged output price outperformed several alternatives, including futures prices, in empirical tests of expected output price for annual crops when using annual data (e.g., Houck et al.; Chavas, Pope, and Kao; Lim; Chavas and Holt).

However, the use of lagged output price is not an appropriate specification of expected bluegrass seed price. Bluegrass is a perennial plant. In the traditional dryland production area, it takes a year to establish a bluegrass stand, so no seed is harvested until the second year. In addition, bluegrass seed price was highly volatile prior to 1994 (see Figure 2). Consequently, expected bluegrass seed price (i.e., the numerator of p₃) is specified as a three-year moving average of Washington bluegrass seed price and lagged two years to reflect expected price at the time of planting (two years before the first harvest).

Hypothesis 1

Our first hypothesis is that bluegrass seed supply was negatively affected by the burning ban, but that effect was offset when farmers increased bluegrass seed production due to an increase in bluegrass seed price relative to competitive crop prices. To test this hypothesis using the models defined by equations (1) and (2), a negative value of the estimated parameter on the burning ban, β₁, in the bluegrass seed supply or acreage equation along with estimated parameters that conform to theory for other variables (i.e., positive value on the own-output price parameter to imply upward sloping supply or normal input, and negative value on the parameter of at least one alternative output price to imply a competitive output) would support hypothesis 1.

Hypothesis 2

If support is not found for hypothesis 1, a second plausible hypothesis is that bluegrass seed supply was negatively affected by the burning ban, but that effect was offset when farmers increased bluegrass seed production in response to a dramatic decrease in bluegrass seed price variability. This hypothesis was motivated by the changing pattern of historical bluegrass seed prices (see Figure 2). From the figure, it is clear that bluegrass seed price entered a period of relative stability around 1994, just prior to the phase-in of the burning ban.

While the expected impact of a change in price variability on output quantity is ambiguous (see, for example, Massell or Chavas and Larson for a complete discussion), our hypothesis embeds the expectation that lowering price variability increases supply. Although that could happen even under risk neutrality (e.g., if decisions are made sequentially and there is temporal uncertainty [see Chavas and Larson]), risk-averse behavior is a common cause of this phenomenon. Consequently, we test this hypothesis by employing Coyle's linear mean-variance model of production that allows for the possibility of risk-averse behavior in the presence of price uncertainty. This model permits a nested test of the second hypothesis. It can be implemented by a simple modification of the normalized quadratic profit function. It results in output supply functions similar to those in equation (1). The only differences from equation (1) are the inclusion of the variances and covariances of the output prices of bluegrass seed, wheat, lentils, and potatoes as independent variables in each of our supply equations. This modification adds ten independent variables per equation. To conclude that a decrease in output price variability was responsible for the increased bluegrass seed production after the burning ban was implemented, we would require a significant negative parameter on the burning ban variable, a positive coefficient on own price in the bluegrass seed equation, and a negative coefficient on bluegrass seed price variance.

Hypothesis 3

Our third hypothesis is that new techniques were introduced contemporaneously with the imposition of the burning ban that effectively reduced the marginal cost of bluegrass seed production.⁷ Hypothesis 3 is examined by both reliable but anecdotal data and a statistical model to determine geographic effects of the ban within the state. The latter is used to indirectly test the hypothesis.

This final hypothesis could provide a plausible explanation only if innovations became available contemporaneously with the implementation of the ban on burning. Such innovations would have to allow farmers to more effectively produce bluegrass seed, i.e., at lower marginal cost, while abstaining from stubble burning. Anecdotal evidence from the industry suggests that new crop rotation and watering techniques were developed in the mid-1990s in anticipation of the increased production costs of the impending ban.⁸ These new techniques allowed farmers to grow bluegrass seed on irrigated farms more efficiently than on dryland farms where most bluegrass seed had been traditionally grown. An important rotation discovery also occurred during this time that enabled bluegrass to be effectively rotated with irrigated potatoes if bluegrass was raised as an annual or biannual crop. When bluegrass is removed that quickly, stubble burning is not needed and the sod can be broken up more easily to plant potatoes. The new irrigation technology included gear drives for pivots, more efficient nozzles, and more precise water and fertilizer application capability.

We implement a procedure to indirectly test implications of the purported innovations. The goal is to determine if the anecdotal evidence can be corroborated. The anecdotal evidence from the industry implied that production of bluegrass seed would shift from the dryland farms in Eastern Washington to irrigated farms in Central Washington in order to implement the new watering and rotation techniques. We develop two sets of equations to model bluegrass seed supply and harvested acreage of bluegrass seed in dryland and irrigated areas of the state. Subject to theoretically expected signs on other parameters, the innovation hypothesis is corroborated if the ban variable in these models is estimated to have a significantly negative effect in the bluegrass seed equation in dryland areas and a significantly positive effect in irrigated areas.

While we estimate both supply and acreage equations for bluegrass seed in the irrigated and dryland regions of the state, we do not estimate the supply equations as a system with other output supplies due to the lack of county-level data for some crops. We independently estimate each regional bluegrass seed equation using OLS.

The set of regressors differs between the dryland and irrigated region equations. Common regressors include the ban variable and expected bluegrass seed price. The rainfall variable in only included in the dryland region equations. Alternative crop prices also differ between regions: wheat and dry peas are included in the dryland region, while and potatoes and lentils are included in the irrigated region.⁹

Data

The first and second hypotheses are tested using state-level data, primarily from NASS-Washington (WASS), for the period 1975–2005. Crop acreage and production are marketing-year totals. Crop prices are marketing-year average prices, calculated by WASS (USDA, 1976–2006). The wheat price variable is the sum of government payments per bushel to Washington wheat farmers and the marketing-year average cash price for wheat. Wheat payment data for 1975 to 1992 are from a special tabulation prepared by the Farm Services Agency, Kansas City, and wheat payment data for 1992–2005 are from the Farm Services Agency, Spokane (USDA 2006b). Both are annual data. Because state input prices are not sufficiently comprehensive, we use an annual U.S. index of prices paid by farmers for commodities, interest, taxes, and wage rates.¹⁰ This price index is from NASS for the period 1975–2005 (USDA, 2006a).

Data for rainfall at the Spokane airport were collected and compiled from a monthly data set from the National Climate Data Center (NOAA).

To test the third hypothesis, we use annual county-level acreage and production data from WASS, for the period 1985–2005 (USDA, 1975–2006). WASS has only been collecting county-level bluegrass seed production data since 1985, so we are restricted to twenty-one observations for this estimation. In addition, in order to not divulge an individual firm's production, WASS is legally restricted from revealing production data for an individual county if the number of producers is below a certain level. So, in the WASS data set, production from these counties is aggregated into an “other” category. We have county-level data for each year since 1985 for Spokane, Whitman, and Garfield counties. Each is included in the dryland region since no more than 3% of their bluegrass seed farms were irrigated in 1997 and 2002 (USDA, 1997, 2002). We also have county-level data for eight or more years for Adams, Franklin, and Walla Walla counties. Each of these is included in the irrigated region since at least 50% of their bluegrass seed farms were irrigated in 1997 and 2002 (USDA, 2002).¹¹ Besides these three counties, all other counties (typically aggregated as such in the WASS data) are included in the irrigated region. In 2002, 100% of bluegrass seed farms were irrigated in all but one of the “other” counties (USDA, 2002).¹²

As noted in Figure 1, production in the irrigated region was trivial in 1985, increased to about a quarter by 1990, and after 1997 exceeded that in the dryland region. With the exception of two counties in the irrigated region, all increases in irrigated bluegrass seed acreage between the 1992 and 2002 Census of Agriculture (USDA, 1992, 2002) were accompanied by a greater increase in total irrigated acreage. Consequently, the increase in irrigated bluegrass seed production in Washington did not come at a net reduction in irrigated acreage of other crops.

Empirical Results

The estimated coefficients for the linear system of supply equations used to test hypothesis 1 are reported in table 1. All own-price parameters are positive, implying upward sloping supply curves as implied by theory, and all are significant at the 10% level except in the wheat linear supply equation. In addition, all alternative crops with statistically significant parameters are competitive outputs. Only wheat supply is significantly related to rainfall. However, these results do not lend any support for hypothesis 1, as the burning ban variable in the bluegrass seed supply equation is estimated to be positive and statistically significant. Additionally, all cross-product supply elasticities in the bluegrass seed supply equation are so low as to suggest that alternative crop output prices have little effect on bluegrass seed supply.

The estimation results for the linear and loglinear single-equation models of bluegrass seed supply and acreage used to test hypothesis 1 are reported in table 2. Consistent with the estimated system above, the bluegrass seed own-price parameter is positive and significant in each equation. Estimated parameters on the other variables generally conform to expectations. Also like the system estimates, the coefficient on the burning ban variable is consistently positive. It is also statistically significant in three of the estimated equations. Thus, none of these estimated models supports hypothesis 1, so we conclude that an increase in bluegrass seed price relative to competitive crop prices did not explain the increased bluegrass seed production following the burning ban.

Hypothesis 2 was tested by including price variability as independent variables in the supply equations. This was done in the system described by equation (1), and results are reported in table 3. Including the variance and covariance terms does not affect the estimated sign of the burning ban parameter. Consequently, the burning ban maintains its apparent positive effect on bluegrass seed supply even after controlling for price variability. The ban parameter actually increases in magnitude and significance in comparison with the estimate in the previous supply system. While own-price parameters maintain their positive signs and increase in significance, parameters on other variables in the system do change sign (e.g., the estimated parameter on potato price in the bluegrass seed supply equation is positive in this system but not statistically significant).

The theoretical model used to test hypothesis 2 required the addition of ten variables per equation in the system (1). The quality of estimates with so many variables and so little data is suspect. In order to recoup a few degrees of freedom, we reran the estimation system excluding the covariance terms (so that only the four price variances are added to each equation), the results of which are reported in table 4. The qualitative results remained the same.

To statistically test hypothesis 3, we estimated separate supply and acreage equations for the dryland and irrigated regions of the state. The estimation results are presented in table 5. The burning ban variable is significant in each equation. It is consistently estimated to be negative for dryland supply and acreage and positive for irrigated supply and acreage. The signs of other parameters generally comply with expectations. The only exception is the insignificant own-price parameter in the linear irrigated supply equation. Most variables have a significant impact on dryland bluegrass seed supply and acreage, but few significantly affect supply and acreage in the irrigated region. Nevertheless, the significant negative burning ban parameters in the dryland equations, the significant positive burning ban parameters in the irrigated equations, and the general theoretical consistency of other parameter estimates support the third hypothesis.

Thus, both anecdotal and statistical evidence within the state support our third hypothesis with the following reasoning. The innovations that occurred contemporaneously with the burning ban are entirely associated with irrigated farming practices. Dryland farmers faced an increase in marginal cost of production and should have decreased supply accordingly. At the same time, the purported innovations would have decreased the marginal cost of bluegrass seed production on irrigated farms. Farmers on irrigated land should have, therefore, increased supply of bluegrass seed. Using the results in table 5, we can reliably conclude that bluegrass seed acreage and supply decreased in dryland areas of the state after the burning ban, and that this decrease was more than offset on an industry-wide level by an increase in acreage and supply of bluegrass seed in areas of the state where farms are irrigated. Thus, we conclude that the measured regional impacts of the burning ban are consistent with the contemporaneous introduction of new techniques that effectively reduced the marginal cost of bluegrass seed production.

Conclusions and Implications for Decision Making

New environmental regulations inevitably and justifiably cause anxiety and apprehension among affected parties because of the frequently predicted costly effects of regulation. In this paper, we examined an example of regulation for which the aggregate consequences were the opposite of sound economic predictions and were actually favorable to some producers who were vocal opponents of the new law.

We tested three hypotheses to explain this unexpected result and found support for only one. We conclude that an increase in production efficiency can provide at least a partial explanation. Both simple observation and more complex econometric modeling of bluegrass seed production in Washington indicate that the ban on field burning, at worst, had no aggregate effect on industry supply, and, at best, strengthened the industry. The estimation of our final models documents that the burning ban was accompanied by a contemporaneous migration in production from the traditional dryland farms in Eastern Washington to irrigated farms in the Columbia Basin. This movement in production appears to have been a result of reported innovations in bluegrass seed farming technology because the innovations were associated with irrigated farming and could not be implemented on dryland farms. Our tests including price effects and price variability failed to provide a satisfactory explanation for the observed increase in statewide bluegrass seed supply. Only our regional estimates provided a satisfactory corroborative explanation for the cause of the increase in production following implementation of the environmental regulation. They provide compelling, albeit supporting, evidence that technological innovations in the irrigated region allowed farmers to produce bluegrass seed more efficiently after the burning ban.

While the effects of technological innovation offset the negative effects of the ban on statewide production, it is important to note that the predictions of a decrease in bluegrass seed production (see, for example Wandschneider et al.) held true for areas of the state where farmers could not implement the new technology. Total statewide production increased, but individual producers did not universally benefit from the increase. It is very likely that a considerable portion of bluegrass seed farmers in the dryland region incurred significant net economic costs due to the new environmental regulation.

This study has interesting decision-making implications with respect to environmental policy and innovation. The first is that while environmental regulation seems to have spurred some innovation within the bluegrass seed industry in the state of Washington, the new technology cannot be implemented by every farm producing bluegrass seed. Heterogeneity across firms in this case presents an obstacle to developing an innovation that will benefit all firms in the industry. When environmental regulation is drafted, regulators may rely explicitly or implicitly on induced innovation to offset some of the costs of complying with the regulation. However, while innovations might offset the cost of regulation on the level of the industry as a whole, it is possible that only a small portion of producers benefits.

While new technology allowed the Washington bluegrass seed industry to adapt to the field-burning ban, the second decision-making implication of our study is that there is no guarantee that similar technological innovation would have the same effect in the face of a ban in other areas. For example, irrigated farms are much closer to bluegrass seed processing centers in Washington than are the majority of irrigated farms in Idaho. It is possible that the higher costs of transporting bluegrass seed from irrigated farms in southern Idaho to processing plants in northern Idaho will prevent expansion of bluegrass seed production on Idaho's irrigated farms if the current court-imposed curtailment of field burning continues.