Every introductory course in economics points out that rational actors should ignore sunk costs: look forward at costs and benefits, not backward at things that can't be changed. But in the November 2011 issue of the American Economic Journal: Microeconomics, Sandeep Baliga and Jeffrey C. Ely offer an argument as to why rational actors should pay attention to sunk costs in "Mnemonomics: The Sunk Cost Fallacy as a Memory Kludge." Those who want the mathematical model and details on the follow-up laboratory experiment need to go through a library to get the article. But in the opening pages, the authors do a nice job of providing the basic intuition.
They start with a reminder of some of the classic evidence on people paying attention to sunk cost: theater tickets and production of the Concorde supersonic jet. They write: "In a classic experiment, Hal R. Arkes and Catherine Blumer (1985) sold theater season tickets at three randomly selected prices. Those who purchased at the two discounted prices attended fewer events than those who paid the full price. Hal Arkes and Peter Ayton (1999) suggest those who had “sunk” the most money into the season tickets were most motivated to use them. R. Dawkins and T. R. Carlisle (1976) call this behavior the Concorde effect. France and Britain continued to invest in the Concorde supersonic jet after it was known it was going to be unprofitable. This so-called “escalation of commitment” results in an over-investment in an activity
Here is their argument for why paying attention to sunk costs can make sense: "We provide a theory of sunk cost bias as a substitute for limited memory. We consider a model in which a project requires two stages of investment to complete. As new information arrives, a decision-maker or investor may not remember his
initial forecast of the project’s value. The sunk cost of past actions conveys information about the investor’s initial valuation of the project and is therefore an additional source information when direct memory is imperfect. This means that a rational investor with imperfect memory should incorporate sunk costs into future decisions. ... If the investor has imperfect memory of his profit forecast, a high sunk cost signals that the forecast was optimistic enough to justify incurring the high cost. For example, the willingness to incur a high sunk cost digging dry wells may signal that the oil exploration project is worth continuing. If this is the main issue the investor faces, it generates the Concorde effect as he is more likely to continue a project which was initiated at a high cost."
They point out various ways in the real world that this dynamic--that is, imperfect information about why a decision was made in the past--can cause current actors to treat sunk costs as useful information. "There are a few different ways these effects can manifest themselves in practice. Most directly, the decision maker may be an individual responsible for making the initiation and continuation decision and he may simply forget the information. An organization may also forget information or knowledge. Managerial turnover can generate organizational forgetting. In this case, we can think of the investor in the model as representing a long-lived organization headed by a sequence of short-term executives. An executive who inherits an ongoing project will not have access to all of the information available at the time of planning. Existing strategies and plans will then encode missing information and a new executive may continue to implement the plans of the old executive. In our model, data about sunk costs partially substitutes for missing information and a rational executive takes this into account."
Finally, and intriguingly, they suggest that paying attention to sunk costs may be hard-wired into people's brains as an adaptive mechanism. "Finally, we can think of sunk-cost bias as a kludge: an adaptive heuristic wherein metaphorical Nature is balancing a design tradeoff. ... In our model, the sunk cost bias is an optimal heuristic that compensates for the constraints of limited memory. This can explain the prevalence and persistence of sunk-cost bias despite its appearance, superficially, as a fallacy. To the extent that heuristics are hard-wired or built into preferences, the sunk-cost bias in observed behavior would be adapted to the “average” environment but not always a good fit in specific situations. For example, we would expect that decision-makers display a sunk-cost bias even when full memory is available, and that sometimes the bias goes in the wrong direction for the specific problem at hand."
On this final point, I'm not yet persuaded. But many economic decisions do unfold over time, and many of them may have a period of disutility or losses early in the process, later followed by compensating utility or gains. It would clearly be misguided to make a choice to pursue a path over time, then when partway into the process to forget about or undervalue the gains to come.