This model can be used for studying several major issues that are relevant to market transformation program design, implementation and evaluation. One issue is whether or not the demand for a higher quality or energy efficient product is price elastic. An understanding of this issue could have a direct bearing on decisions to offer price promotions or financial subsidies to buyers of energy efficient products and services. Another issue that can be studied the extent to which public programs have an impact on increasing the price elasticity of demand of energy efficient products by creating a heightened awareness and appreciation of their benefits. A limitation of this model is that the impact of the relative prices on market shares cannot be directly estimated. To do so a model that quantifies the rate at which market share is captured due to price changes is specified.

3.2.2 Share Capture Model Form

The share capture model takes as the dependent variable the market share of product x, which is defined as the annual quantity shipped of product x divided by the total quantity shipped of product x plus product y. The generic model specification is:

where, like the market demand model, this model is estimated for the two time periods, each containing multiple years, and the terms are defined as above, with the ẞ's representing the share capture model parameters.

3.3 Research Design

For the Green Lights Partnership evaluation, fluorescent lighting ballasts data extending from 1959 through 1999 have been collected from the Bureau of the Census publication Current Industrial Reports (COMMERCE 1959-99). Based on national surveys of manufacturers, this publication provides quarterly data for thousands of products sold in the United States. These data include, at the manufacturer level, the total number of units shipped and the total value of the shipments. Among the products tracked separately are low and high power factor magnetic fluorescent lighting ballasts, for which national data are available from 1959 to the present.

Fluorescent lighting was first introduced into the U. S. market around 1940. Although nowadays low power factor magnetic ballasts are primarily thought to be used in residential applications, there are no features of the equipment that preclude using high power factor ballasts instead of

low power factor ballasts in most commercial and residential applications. Statistical evidence that these two products are substitutes is contained in their historical market shares. According to the earliest recorded data, shipments of low power factor magnetic fluorescent ballasts in 1959 made up 42 percent of the national market. Over the next fifteen years, the market share of low power factor magnetic fluorescent ballasts fell by almost a half.

Although close substitutes, high and low power factor magnetic fluorescent ballasts have different qualities. Among other things, high power factor fluorescent ballasts produce less lighting flicker and last longer than low power factor ballasts. They are therefore considered of a higher quality than their low power factor counterpart. Regarding energy efficiency it should be noted that while high power factor ballasts require that less energy be generated for their operation than low power factor ballasts, the energy savings experienced by the customer is likely to negligible. Nevertheless, while in 1974 the high power factor ballasts were, on average, three-and-a-half times more expensive per unit than low power factor ballasts, their market share swelled to 77 percent and the market share of the low power factor ballasts fell to 23 percent.

Since 1986, total manufacturer shipments and total value of shipments data also have been published for electronic fluorescent lighting ballasts. These new, improved products are substitutes for magnetic fluorescent ballasts in most lighting applications, specifically, substitutes for the high power factor magnetic fluorescent ballasts rather then the low power factor ballasts. As with the former pair of substitutes, the electronic fluorescent ballasts produce less lighting flicker than their counterparts. They are also thought to last longer; however, they have not been in the market long enough for conclusive evidence of this benefit to be collected. For these reasons, as well as for the fact that they require less energy to operate and yield appreciable energy savings to customers, electronic fluorescent ballasts are of higher quality than high power factor magnetic fluorescent ballasts.

Given the two analogous pairs of products -- from 1959 through 1985, high quality, high power factor magnetic fluorescent ballasts versus lower quality, low power factor ballasts, and from 1986 through 1999, high quality electronic fluorescent ballasts versus lower quality, high power factor magnetic ballasts -- the research design for this study rests on the premise that price response behavior related to one pair of products is, in the absence of public program interventions or public programs, similar if not identical to the price response behavior for the other pair of products. If so, then to the extent that differences in price responsiveness appear, these differences can be attributed to the impacts of public programs.

Based on the years over which the data for the three types of fluorescent ballasts are available, the research design for this study is divided into two periods. The earlier period spans 1959 through 1985 and is referred to as the comparison period. The later period spans 1986 through 1999 and is referred to as the treatment period. These period divisions track the evolution of market transformation and DSM programs. In the late 1980's and the 1990's the overwhelming emphasis of commercial sector DSM programs, and later the Green Lights Partnership, was on promoting electronic fluorescent ballasts and the companion use of T-8 lamps.

To complete the research design several issues related to the comparison and treatment periods should be noted. With respect to the last five years of the comparison period, 1981 through 1985, there are three major factors that may have affected the relative prices and quantities demanded of the different magnetic ballasts that cannot be controlled for in the scope of this study:

by the early 1980's many utilities and government agencies began promoting energy efficient magnetic fluorescent ballasts through energy audits, rebates and other public programs; however, there are no readily available data on aggregate national expenditures or levels of program efforts in this period

in 1982 the state of California adopted an energy efficiency standard for fluorescent ballasts; the standard became effective in 1983 and over the next five years four more states followed California's lead

energy efficient magnetic fluorescent ballasts, which were introduced in 1976, are not differentiated from other high power factor ballasts in the published database; according to the congressional testimony of a lighting company executive (LBNL 1995), these products were slow to be adopted, but, by 1980 had a 10 to 15 percent market share that grew by 1986 to about a 30 percent market share

On the basis of these events, the five years from 1981 through 1985 are omitted from the comparison period for this study. The 32 continuous years from 1959 through 1980 thereby comprise the estimation years for the comparison period. With respect to the treatment period, although the continuous years spanning 1986 through 1999 are included in this study, it must be noted that;

· this study does not attempt to ascertain how the 1990 national energy efficiency standard for high power factor magnetic fluorescent ballasts affected the market share of electronic ballasts, particularly since, as one study points out, it is difficult to argue that the energy efficiency standards permanently affected magnetic fluorescent ballasts prices (LBNL 1995)

since the Green Lights Partnership did not begin until 1991, no attempt is made in this study to disaggregate the changes in electronic ballasts market share between 1986 and 1990 that were due to short-term DSM program effects versus the market-transformative effects of DSM programs and other public programs; rather, all market share changes in this time span are assumed to be due to DSM rebates

national data on utility DSM program expenditures were collected by the Energy Information Administration from 1990 through 1998, only (DOE 1990-99); to use this data series as a lagged independent variable in the treatment period models, DSM program expenditure estimates for the five years of 1985 through 1989 were extrapolated based on the average annual growth rate of 18.2 percent that is found for total DSM program expenditures from 1990 through 1994

Finally, it must be noted that comprehensive national data on fluorescent ballasts stocking and inventory patterns, and annual fluorescent ballasts exports and imports, do not exist. Therefore, throughout this study it is assumed that all nationally-shipped fluorescent ballasts, and only those, are sold and purchased in the United States. Ballasts installations are assumed to occur in the calendar year in which the ballasts were shipped.

3.4 Model Specifications and Findings

3.4.1 Market Demand Model

The market demand model for the higher quality ballasts is specified in log-linear functional form based on the assumption that the price elasticity of demand is constant. To control for price endogeneity, a two-stage least squares model is specified as:

where the error term u, is assumed to be first-order autoregressive and the error terms v, and e, are not autocorrelated. For each of the two study periods,

=

log of national average annual cents per kWh, commercial sector, for year t InPRIME, log of national average prime lending rate for year 1

InKWHP,

=

=

=

=

=

=

=

=

=

log of percentage change in consumer price index from year 1-1 to year t
log of Federal Reserve Board index of industrial production, January of year!
log of total annual electric utility DSM expenditures, lagged one year, these
values are zero in the comparison period; for 1985 through 1989, in which no
published estimates are available, expenditures are projected via the average
annual growth rate in expenditures from 1990 through 1993

regression intercept of the final model

coefficient representing the price elasticity of demand

coefficient representing the percentage change in the quantity demanded of product x due to a (relative) percentage change in the national average annual retail cents per kWh in the commercial sector

coefficient representing the percentage change in the quantity demanded of product x due to a (relative) percentage change in the prime lending rate coefficient representing the percentage change in quantity demanded of

product x due to a (relative) percentage change in the change in the consumer price index

coefficients of the first-stage model

serial correlation coefficient

Like autocorrelation, the presence of heteroscedasticity in this two-stage least squares model could bias the standard errors of the model and invalidate hypothesis tests. Therefore, the model is corrected for potential heteroscedasticity through estimation of the White heteroscedasticityconsistent covariance matrix. The two-stage least squares autoregressive model is solved using a simultaneous equation estimator.

The findings of the economic analysis for the treatment and comparison periods are contained in Exhibit 2. The diagnostic statistics confirm that the models are sound. The fit of the models, as characterized by the adjusted R-squares, is very good. Based on values of the autocorrelation coefficient p of less than 1, the variables under investigation do not suffer from the presence of unit roots. This well-known problem, in which the dependent variable and at least one independent variable contain stochastic trends, causes regression results to be spurious. The Durbin-Watson statistic, an indicator of first-order serial correlation that could bias the standard errors, is within a reasonable range for each model.

The models' coefficients indicate that in the comparison period, demand for the high quality ballasts was approximately unit elastic. Unit elasticity is the kind of price responsiveness that might be expected for typical product that has relatively close substitutes, is neither a necessity nor a luxury, and is neither a very significant nor insignificant fraction of a consumer's budget. In the context of this analysis it means that, on the margin, a one percent decrease in high power factor ballasts prices led to an approximately one percent increase in the quantity demanded of this product.