The Treasury

Global Navigation

Personal tools

You are here: Home > Publications > Guidance and Instructions > Planning > Cost Benefit Analysis > Guide to Social Cost Benefit Analysis


Guide to Social Cost Benefit Analysis

Publication Details

  • Guide to Social Cost Benefit Analysis
  • Published: 27 Jul 2015
  • Status: Current
  • Author: The Treasury

Guide to Social Cost Benefit Analysis

Published 27 Jul 2015

See also the main web page on cost benefit analysis which can be found here.

The Treasury welcomes comments and questions on the guide and on cost benefit analysis generally.  Edited versions of the comments, questions and answers (whether from the Treasury or from readers) will be posted on this page below. You can send your comments or questions to

Table of Contents

Browse section/chapter Download/Page range



Part 1: Guide to Cost Benefit Analysis

Part 2: Issues in Cost Benefit Analysis


Appendix 1: Monte Carlo Simulation

Appendix 2: Example of a ‘Rough' CBA

Appendix 3: Example CBAs


cba-guide-jul15.pdf (522 KB) pp. 78

Discussion Board

26 Mar 2015 - from Tim Denne, Director, Covec

Treatment of tax: the main two arguments appear to be that: (1) tax is already included in market prices so to remove them in one place but not the other is impractical; and (2) if tax reflects government policy, then it should be retained. Re the impracticality argument, surely the more important issue is whether to remove it from some transactions but not others is more or less distortionary, eg whether to remove taxes in some places gets the analysis closer to the outcome if taxes had been removed in all cases. This will be situation-specific. The second argument (and the example of a duty imposed to protect local industry) is an example of why tax needs to be taken off. Retaining the tax in analysis is justified if it is correcting a market distortion, eg there are negative externalities (at the margin) of imports or positive externalities (at the margin) of domestic production. But to simply assume that this is so means that there is no check on government policy. Externality-correcting taxes, for example, need to be assessed to see if they are priced correctly and adjusted accordingly. For the import duty example, the optimal decision at the margin will always be to import.

Discount rate: because CBA is an assessment of social costs and benefits, the preferable discount rate is the social rate of time preference (or an opportunity cost of consumption). The argument for using rate of return in the share market as the basis for the discount rate assumes that government-incentivised activities always displace private investment, eg because capital is limited in supply. This is not clearly so in NZ. Rather, the standard assumption is that all NPV-positive investments will be made. If there is displacement of investment associated with a specific policy (it will not occur systematically) it is better to identify and deal with that explicitly, eg through a shadow price of capital approach (adjust cost of capital to reflect opportunity cost of investment and then use SRTP). And note this is already done to some extent by using a 20% adder to the costs of (distortionary) taxes.

Economic Impact Analysis: the problem with EIAs is not really that they treat costs as benefits, but that they frequently fail to specify an adequate counterfactual. For example, if an EIA counts labour costs as a benefit, this is equivalent to a CBA counting labour costs as having a zero opportunity cost because of high unemployment (the labour costs are transferred to profit). Thus an EIA will produce the same results as a CBA if it compares a project/policy with a counterfactual in which labour is differently employed. If undertaken properly the main differences between an EIA and a CBA are the absence of externalities and consumer surpluses. However, admittedly, a properly undertaken EIA is a rare thing indeed!

Marginal vs average costs: in the long run, all costs are marginal, and average costs can be used as a proxy for long run marginal costs. It would be useful to provide some guidance on this, eg it is a frequent issue in analysis of the benefits of energy efficiency measures: in the short run only fuel costs are saved, but in the long run, so are the costs of transmission and distribution. Policy resulting in investment in long-lived energy-efficient assets should ideally be assessed against cost savings in long run marginal costs = average costs.

Period of analysis: it is suggested (p17 & p42) that impacts beyond 30 years have little impact when discounted at 8% so can be ignored. This is all subjective: $1 million still equals close to $100,000. There are better ways to deal with long time periods than ignoring costs past a certain time period, including establishing residual values for capital and/or assuming costs/benefits beyond a certain time period are constant in real terms and including a PV of future values at that point.

Benefit cost ratios: the suggestion is that cost savings should be classified as a benefit and added to the numerator. The definition of a cost saving is ambiguous and depends crucially on how you define the counter-factual against which cost-savings are measured. Often the counter-factual is uncertain, eg a frozen technology projection vs efficiency improving. NPV analysis will always rank projects in the same order regardless of how the counter-factual is defined so it becomes less important; not so BCRs.

24 Mar 2015 - from Grant Andrews, Ministry of Business, Innovation & Employment

The main weakness in using a CGE model for a BCA is that these models are usually structured in terms of national accounting concepts, and therefore may not (probably won't) include many of the effects you might be interested in.  In particular, use of national accounting concepts mean that model outcomes generally won't include impacts for which there are no market values - which are often the critical parts of the analysis.

So to take the value of life as an example: a CGE model will theoretically value the impact of lost production (GDP) from the death of an 'average' worker, but won't incorporate other values that are (legitimately) built into standard valuations.    

There are of course workarounds to build in non-market values to CGE model results, but these essentially replicate what you have to do anyway for an orthodox BCA.  As Bryce says, we can usually safely assume away the interactive effects that a CGE model incorporates unless the investment is very large relative to the economy.

24 Mar 2015 - from Eric Crampton, Head of Research at the New Zealand Initiative

I note one issue at Paragraph 114. It is not at all obvious to me that we should not count as a cost the loss of freedom due to imprisonment. If we do ignore the costs to the imprisoned, then we're begging the question: we've assumed a relevant and substantial cost to be equal to zero and so cannot help but bias the conclusion. On this point, I cannot see how we avoid David Friedman's insistence that we have to count all of the costs and all of the benefits. Chapter 15 of Law's Order on I find particularly compelling:

"If instead of treating all benefits to everyone equally we first sort people into the deserving and the undeserving, the just and the unjust, the criminals and the victims, we are simply assuming our conclusions. Benefits to bad people don't count, so rules against bad people are automatically efficient. We cannot deduce moral conclusions from economics if we start the economics by assuming the moral conclusions. What acts are or are not crimes is one of the things our theory is supposed to tell us. Murder may be a simple case, but what about speeding? What about breaking into an empty house when you are lost and starving? What about slugging someone who has the presumption to suggest that Windows is a better operating system than the Mac OS? Economic analysis provides tools for answering these questions. If we treat it, instead, as an elaborate machinery for justifying the answers we already have, we will learn little that we do not already know."

His discussion continues with more practical considerations. See discussion from page 230 onwards. Friedman's text is freely available online:

Or, consider this puzzler. Suppose that a future government proposes legalising marijuana and wants advice as to whether the policy passes CBA. On Treasury's proposed view, the harms done to those in prison count for nothing. But if the legislation passed, the continued imprisonment of those imprisoned for marijuana would surely be unjust, and so the costs imposed on them would count. So we cannot count the costs to those in prison unless we change a law that is less likely to be changed unless we count the costs to those in prison.

This has to matter when we're thinking about the severity and proportionality of punishment.

The real messes come in, to my mind, when people derive utility from seeing criminals punished. I have a hard time seeing a case for discounting those benefits, but those too might yield absurdities, depending on the distribution of vengeance-utility in the population I suppose.

I'd hit on some of this here:

24 Mar 2015 - from Sam Warburton, Ministry of Transport

I  have a question about paragraph 114 of the new guidance.

Paragraph 114:
Defining gainers and losers is particularly difficult for projects within the justice system. How should the impact on offenders be taken into account? Obviously, it wouldn't make sense to count as a cost the loss of freedom due to imprisonment. However, the cost of wrongful imprisonment should be counted. There may be other welfare impacts on prisoners that one would want to count, such as the avoidance of unsanitary conditions. Access to nutritional food might be counted as a benefit, but not access to luxury food items. In other words, social norms and the law define which features of a prison should be classified as a benefit or a cost, or should be ignored.

I do not think this is obvious at all.

What if the second sentence read 'Obviously, it wouldn't make sense to count as a cost the loss of life from the death penalty for a minor infraction'?

The welfare of all people should be included in cost-benefit analysis, that is all costs and benefits should be included. Moral/distributional judgements should then be handled by either:

  • overlaying the costs and benefits with a social welfare function that weights benefits and costs or people differently, or
  • presenting different options to decision makers that reflect different weights and letting them choose according to their moral judgements.

Access to nutritional food and luxury items are both benefits to prisoners and options that include them should count those benefits. Now, it may be simple to show that providing luxury items carries a high opportunity cost to society (there are many people who cannot afford luxury items and would be taxed to pay for others' luxury items), but this does not mean that benefits and costs should be ignored.

Another example: Revenue from fines are treated in (well conducted) CBAs as a transfer. Fines are a cost to those that pay them, and a benefit to the government, and cancel out. If the loss of money to law breakers doesn't count, then wouldn't fines be counted as a benefit? (And what terrible incentive would that give governments?)

23 Mar 2015 - from Bryce Wilkinson, Capital Economics Limited

Results from GE models are often driven by the market closure assumptions, particularly for labour, the balance of payments and the fiscal balance.  In contrast, CBA usually assumes that the project is so small that it won't change the economy-wide real wage, real exchange rate or interest rates (eg via a change in the public sector borrowing requirement)

17 Mar 2015 - from the Treasury

Two areas in the Cost Benefit Analysis Guide that we are still concerned about are the suggestion that all prices should be tax inclusive, and the discussion around the use of CGEs (general equilibrium models). While we think our comments are right, we are not as confident as we would like to be. In the case of CGEs, we are aware that the models sometimes produce results in terms of changes in GNDI that are significantly different from the results of a CBA, and we would like to understand why that is so. What exactly gives rise to benefits which are additional to those identified by a CBA? What is the nature of those benefits? We would welcome any contributions that would help clarify these issues.

Page top