Cost savings from nuclear regulatory reform: reply.

I. Introduction

In a recent issue of this Journal, Canterbery, Reading, and Johnson (hereafter CRJ) [1] present an empirical model for nuclear power plant construction costs to simulate the "reasonable" costs which may arise from a standardized nuclear power plant design. The model is insightful in its attempt to incorporate the impact of Nuclear Regulatory Commission (NRC) regulations on nuclear power plant construction costs. The authors, however, failed to recognize that the underlying nuclear construction cost data is pooled cross-sectional in nature: that is, it has an unequal number of nuclear plants being constructed in an unequal number of years. Additionally, the authors have incorrectly modeled nuclear power construction costs on a "per-project" rather than "per-plant" basis under the aegis of avoiding problems with the inconsistent accounting treatment of the common costs of multiple units.

Both of these problems could be corrected by: (1) using a fixed effects model to incorporate differences in nuclear power plant construction costs for each year like that presented in a similar model by Joskow and Rose [2] for coal generating facilities, and (2) running a "per-plant" model and incorporating an indicator variable to account for first units in a multi-unit project as presented by Zimmerman [4]. The first modification reveals that cumulative NRC regulations are perfectly correlated with time. This modification effectively eliminates one of the most important contributions of the work; namely, the relationship between NRC regulations and nuclear power project construction costs. The second modification raises questions about the significant economies of scale found by the authors.

II. Revised Nuclear Construction Cost Model

Consider the following "per-plant" model similar in many respects to CRJ:

[Mathematical Expression Omitted].

The above is similar to CRJ because it includes variables representing size and experience. It differs in that it includes variables for first unit effects,(1) the presence of a mechanical draft cooling tower, external regional learning effects (REGION), and the fixed effects of time. Construction [TABULAR DATA FOR TABLE II OMITTED] duration was excluded from the model because the data has converted to overnight costs and adjusted for the interest costs associated with construction (e.g., Allowance for Funds Used During Construction or AFUDC), as was done by the authors. Adjusting the data for AFUDC and converting them to "overnight costs"(2) should divorce any time penalty associated with abnormally long construction durations. The model was run using data from the Utility Data Institute [3] for nuclear steam generating facilities placed into service during the period 1966-1992. Variable descriptions are included in Table I, while the descriptive statistics have been presented in Table II.

The empirical results presented in Table III are based upon an alternative nuclear construction cost model run on a "per-plant" basis rather than the "per-project" basis. CRJ have noted that the preference for the "per-project" approach was based upon an attempt to reduce the accounting inconsistencies associated with the common costs of multi-unit projects. While correct, the authors have traded one potential problem for an even greater one. Treating costs on a "per-project" basis assumes that all units in a particular nuclear power project share the same overall critical paths and construction durations. This assumption is exaggerated by the common expenditure profile used by the authors to extrapolate project costs over the various years of construction. A per-plant basis is more appropriate because it recognizes the fact that construction expenditure profiles will differ for each unit - regardless of the fact that they were constructed within the same project. An examination of the commercial operation dates of different units, within the same project, shows that different units typically experience different construction durations, and as a result follow completely different construction critical paths.

[TABULAR DATA FOR TABLE III OMITTED]

An initial run of the model presented in Table III included cumulative NRC regulations as an explanatory variable and found that this measure of federal safety regulation is perfectly collinear with time. Thus, the best indicator of regulatory impacts on nuclear construction costs has to be gleaned from the annual indicator variables (T). The results show that there are significant time-associated impacts related to the construction of nuclear power plants, holding other factors constant. The costs of plants which reached COD in 1993 were, ceteris paribus, 2.5 times more costly to build than those in the reference year 1968. These large increases in costs, while not clearly attributable to NRC-induced regulatory changes, shed some light on the potential magnitude that "other things" (like regulation) had on the costs of constructing nuclear power plants.

The remaining results are relatively consistent with those found in CRJ. Utility specific learning-by-doing effects are corroborated (LN_EXP) as are the variables representing regional industry-wide external learning effects.(3) The model also shows that first units (FIRST) tend to be more costly than trail units supporting the claim of "front-loading" common project costs onto first units as found by Zimmerman. Units with mechanical draft towers (COOLTWR) were also significantly more costly to build, closely reflecting the conclusions reached by Zimmerman.(4)

Another important conclusion of the model in Table III is that scale economies (LN_SIZE) are not strongly supported. This result may be attributed to the fact that the model is run on a per-plant rather than a per-project basis. As noted earlier, the results in Canterbery are based upon a dataset which aggregates the construction costs of multiple "units" into one "project." For instance, the costs of Palo Verde Units 1, 2, and 3 are lumped into one cost calculation as indicated on Table II of the authors' paper. It is possible that the scale-related empirical results found by the authors are due to the reduced cost of building multiple units as opposed to larger units.

III. Conclusions

This paper has shown that accounting for the pooled cross-sectional nature of nuclear power plant construction cost data yields results different than those presented in the cross-sectional analysis of Canterbery. A fixed effects model reveals perfect collinearity between the cumulative number of NRC regulations and time. Correcting the collinearity problem leaves one in the unfortunate position of being unable to statistically define a clear relationship between nuclear power plant construction costs and NRC regulations. In addition, running the model on a per-plant, as opposed to a per-project basis, calls into question the strength of the economies of scale associated with constructing large nuclear generation facilities. The lack of support for scale economies, in turn, calls into question the policy recommendations of the authors suggesting standardized designs, and accelerated construction durations for rather large nuclear generation facilities (800-2400 MW).

Table I. Variable Definitions

LNCKW = the natural logarithm of the overnight cost per KW of the ith nuclear power plant;

T = an indicator variable for the year in which the ith plant was constructed;

LNSIZE = the natural logarithm of the size of the ith plant measured in MW;

LNEXP = the natural logarithm of the cumulative number of nuclear power plants constructed by the utility constructing the ith nuclear power plant;

FIRST = a variable indicating whether the ith unit was the first constructed;

REGION = a variable indicating the nth region in which the ith plant was constructed; and

COOLTWR = a variable indicating the presence of a cooling tower for the ith plant.

David E. Dismukes Louisiana State University Baton Rouge, Louisiana 

1. Identifying first units of a multi-unit project is important because current ratemaking practices do not allow utilities to recover capital expenditures until a plant has been placed into service. Thus, multi-unit projects tend to "front-load" a significant amount of common costs onto the first unit.

2. The data was converted to overnight costs, and adjusted for each of the six Handy-Whitman regions, following the convention outlined by Canterbery, Reading, and Johnson [1, 558].

3. Census regions defined by the Department of Energy were used in the model. These include the South Atlantic (D_SATL), North Central (D_NCEN), South Central (D_SCEN), Plateau (D_PLAT), and Pacific (D_PAC). The North Atlantic region was excluded as the reference.

4. Zimmerman found that units with mechanical draft towers were between 14 and 17 percent more costly to build. This is in keeping with the 15 percent estimate presented in Table III.

References

1. Canterbery, E. Ray, Don Reading, and Ben Johnson. "Cost Savings from Nuclear Regulatory Reform: An Econometric Model." Southern Economic Journal, January 1996, 554-66.

2. Joskow, Paul L. and Rose, Nancy L. "The Effects of Technological Change, Experience, and Environmental Regulation on the Construction Cost of Coal-Burning Generating Units." Rand Journal of Economics, Spring 1985, 1-27.

3. Utility Data Institute. Electric Utility Power Plant Construction Costs. Washington: UDI/MacGraw-Hill, 1994.

4. Zimmerman, Martin B. "Learning Effects and the Commercialization of New Energy Technologies: The Case of Nuclear Power." Bell Journal of Economics, Autumn 1982, 297-310.