Application to public policy

The practice of applying Bayesian decision theory to public policy needs to take into account that the policy analyst conducting the analysis is not the same person as the policy maker. In particular, the policy analyst is unlikely to know the policy maker’s prior probability beliefs and utility function (i.e. degree of risk aversion). One approach for overcoming the information gaps could involve surveying the policy maker’s prior beliefs and risk preferences and then proceeding to derive the optimal decision.

Another approach, adopted in this paper, would involve the policy analyst presenting to the policy maker a menu of policy options that differ in their risk levels. The policy maker would compare expected loss and measures of risk across policy options, and would choose the policy with risk/return properties that best meets his or her preferences.

The policy options may be ordered from low to high risk, drawing out an efficient frontier of policy options, as illustrated in Figure 3 below. Policy options lying inside the frontier would be dominated and could be discarded.

If the policy maker’s preferences are transitive, the most desired policy option may be identified through a sequence of pair-wise tests where the policy maker assesses whether the incremental benefit offered by policy option ‘n’ relative to policy option n-1 is outweighed by the increased risk:

Policy Option 2 tested against Policy Option 1, i.e. Test: PO₂≤ PO₁

Policy Option 3 tested against Policy Option 2, i.e. Test: PO₃≤ PO₂

…

Policy Option N tested against Policy Option N-1, i.e. Test: PO_N≤ PO_N-1,

where “≤ ” is read as meaning “less preferred than”.

Figure 3: Efficient policy frontier

3.3  Qualitative risk assessment

A full evaluation of the risk properties associated with each policy would require the conduct of comprehensive empirical research. The purpose of this section is to identify qualitative indicators to distinguish risk levels in broad terms in advance of more detailed empirical analysis.

Average, variance and worst-case losses

A risk averse policy maker would be concerned about the probability of a decision error, the average loss and range of losses possible if such an error did occur. Relevant considerations would include the degree of uncertainty about the level of losses and the magnitude of ‘worst-case’ losses.

Persistence vs. reversibility

The policy decision may or may not be subject to future review. A policy where a decision error would be revealed early in the post-implementation period and which could be reversed quickly and at low cost normally would be less risky than a policy that would be irreversible or reversible only slowly and at high cost.

An example of the difference is provided by the different roles of the Commerce Commission in Mergers & Acquisition applications versus some other regulatory functions. M&A decisions are clearly irreversible while decisions whether or not to impose price regulation are reversible. In addition, a decision to impose price regulation may stifle the release of further information about the competitiveness of the market whereas a decision to not impose price regulation has the advantage of allowing further observations about competitive conditions and thus whether regulation is warranted.

Monetary policy, where decisions on the Official Cash Rate are made six-weekly, is another example where policy decisions are made under considerable uncertainty but have the benefit of being reversible.

Combinations of objectives

In the case of multiple objectives, the impact of placing high or low weight on each objective depends on which other objectives are included in the policy. For example, in the case of Crown financial policy, a policy that places high weight on both tax-smoothing and agency cost would have quite different implications than if either objective received high weight alone. The former policy is balanced in the sense of creating countervailing incentives for and against the build up of financial assets, while the other policies that place high weight on one objective would result in a more extreme balance sheet structure. Thus, in most cases, the risks associated with an objective can be evaluated only by taking into consideration the other components of policy.

Implementation risk

Proposed policy options must be capable of being implemented without undue risk. One source of implementation risk arises when the policy is based on a model that assumes decision makers have more information and better capability to interpret that information than would actually be the case. This concern with implementation risk is consistent with Coase’s dictum noted above that economists should always judge alternative arrangements as they would actually operate.

Voter misperception risk

Voter misperception risk is defined as the risk that an economically desirable decision leads to outcomes that the public perceives as a mistake. For example, in the case of Crown financial policy, a desirable hedging strategy could lead to outcomes where the market value of Crown financial assets were revised downward by (say) $5 billion as an offset to upward revaluation in the tax asset. Because the Operating Balance would include the downward revision in the financial asset but exclude the increase in the tax asset, the public may view the outcome as reflecting poor economic management by the government of the day. Misperception risks drive a wedge between the interests of policy makers (as agents) and the public (as principals), potentially causing desirable policy to be over-turned in favour of a false positive or false negative position.