Opportunity Loss 14

1. Core Concept Overview & Intuition

Opportunity loss is the economic measure of what a decision maker forgoes when a chosen action turns out, in hindsight, not to have been the best action available for the state of nature that actually occurred. It converts a payoff table into a regret table, expressing every outcome as a non-negative shortfall from the best attainable payoff in that state. The central intuition is counterfactual: for each state of nature, identify the maximum payoff that could have been earned, then record how far each alternative falls short of that maximum. The best alternative in a given state incurs zero opportunity loss; every other alternative incurs a positive loss equal to the gap.

This reframing matters because it places all alternatives on a common scale of "cost of being wrong." A decision maker who is uncertain about which state of nature will prevail can compare alternatives by the losses they would suffer across states, rather than by raw payoffs whose magnitudes may be difficult to compare. Opportunity loss therefore supplies the natural bridge between the payoff table of a decision problem and the criteria used to choose among alternatives under uncertainty and under risk.

The concept rests on the decision-making environment established earlier: a decision problem consists of alternative courses of action, states of nature (possible events or outcomes of a random process), the probabilities attached to those states, and a decision criterion or rule for selecting the best alternative. Opportunity loss is defined relative to the first two elements — actions and states — and is computed state by state. It does not by itself prescribe a choice; it is an input to the criteria that do.

Two features of the construction deserve emphasis. First, opportunity loss is always measured against the best payoff within the same state of nature, never against a payoff in a different state. Second, the loss for the optimal action in any state is exactly zero, so each column of an opportunity-loss table contains at least one zero. These two properties make the table easy to audit: any column lacking a zero, or any negative entry, signals an arithmetic error.

2. Mathematical Formalization & Core Definitions

> Definition (Payoff table). A payoff table is a rectangular array whose rows index the alternative courses of action $a_1, a_2, \dots, a_m$ and whose columns index the states of nature $s_1, s_2, \dots, s_n$. The entry $v_{ij}$ is the payoff (profit, revenue, or net monetary consequence) received when action $a_i$ is chosen and state $s_j$ occurs. Payoffs may be negative, representing losses.

> Definition (Opportunity loss / regret). For a fixed state of nature $s_j$, let

$$
M_j = \max_{1 \le i \le m} v_{ij}
$$

> be the maximum payoff attainable in that state. The opportunity loss (also called regret) of action $a_i$ under state $s_j$ is

$$
r_{ij} = M_j - v_{ij}.
$$

The matrix $R = (r_{ij})$ is the opportunity-loss table, or regret table, associated with the payoff table $V = (v_{ij})$.

> Proposition (Basic properties of opportunity loss). For every action $a_i$ and state $s_j$:
> (i) $r_{ij} \ge 0$;
> (ii) $r_{ij} = 0$ if and only if $v_{ij} = M_j$, that is, if and only if $a_i$ is an optimal action in state $s_j$;
> (iii) for each state $s_j$ there exists at least one action $a_i$ with $r_{ij} = 0$;
> (iv) $r_{ij} = 0$ for all $j$ if and only if action $a_i$ is optimal in every state of nature.

These properties follow directly from the definition: $M_j$ is the largest entry in column $j$, so subtracting any entry of that column from $M_j$ yields a non-negative number, and the difference is zero precisely when the entry equals the column maximum.

> Definition (Prior probabilities). When the states of nature are assigned probabilities $p_1, p_2, \dots, p_n$ with $p_j \ge 0$ and

$$
\sum_{j=1}^{n} p_j = 1,
$$

> the decision problem is one of decision making under risk. When no such probabilities are available, the problem is one of decision making under uncertainty.

The opportunity-loss table is used in two distinct ways, corresponding to these two environments. Under uncertainty, the losses in each row are summarised by a criterion such as the minimax regret rule, which selects the action whose worst (largest) opportunity loss is smallest. Under risk, the losses in each row are weighted by the prior probabilities to form an expected opportunity loss, which is then minimised. The present unit establishes the construction of the table and its interpretation; the probabilistic and perfect-information elaborations belong to later units and are not developed here.

> Definition (Minimax regret criterion). For each action $a_i$ define the maximum opportunity loss

$$
R_i = \max_{1 \le j \le n} r_{ij}.
$$

> The minimax regret criterion selects the action $a_{i^{*}}$ for which

$$
R_{i^{*}} = \min_{1 \le i \le m} R_i.
$$

The minimax regret criterion is a conservative rule: it does not attempt to maximise expected payoff, but instead bounds the worst possible shortfall from the best attainable outcome. It is the natural decision rule to pair with the opportunity-loss table under uncertainty, because the table is already expressed in units of shortfall.