The Cost of Looking Ahead
This chapter focuses on a price-setting firm that must decide how much of the future to calculate when it gets an opportunity to reset its price. The key question is not whether the firm has expectations, but which future contingencies it evaluates explicitly and which it compresses into a simpler belief. That distinction makes the model’s central result intelligible: looking farther ahead can improve a price decision, but the improvement is worth paying for only when the future matters enough.
Why a price decision reaches into the future
A firm that changes its price today is not making a decision about today alone. Its reset price may remain in effect for several periods because the firm does not receive another opportunity to adjust, while the model treats the price as relevant across the planning window. During that interval, demand can move, input conditions can change, and the prices of other firms can evolve. A price that is appropriate for the current state may therefore be poorly suited to the conditions under which it will continue to earn revenue.
The Federal Reserve model places this familiar pricing problem in a finite-horizon planning framework. Firms are monopolistically competitive: each sells a differentiated product and chooses a price while taking the broader economic state as given. At any date, only a fraction of firms receive an opportunity to reset. A firm that receives one must choose both the reset price and how far ahead to plan while that price may remain relevant. Firms that do not reset are treated more simply in the model, with their prices indexed to steady-state inflation. The important margin is the extra reasoning performed by a resetting firm.
To see why that reasoning has value, imagine two firms facing the same current demand. One sets its price using only current conditions and a rough assessment of what comes later. The other also calculates how persistent demand or supply disturbances will affect the future path of prices and sales. If the current state is close to normal and the disturbance is likely to fade quickly, the second calculation may barely change the recommended price. If the disturbance is large and persistent, the future path can materially change the profitability of a price chosen today. The extra calculation then has economic value.
The model represents that value through a horizon-specific pricing problem. For a horizon of k periods, the firm evaluates expected discounted profits from the present through the end of that window, taking account of the model’s relationships and shock processes during those periods. The reset price is chosen to perform well across the contingencies the firm has agreed to consider. A longer horizon does not simply mean producing a more distant forecast; it changes the set of future consequences that enter today’s price.
This is why a price-reset opportunity creates a reason to think beyond the current period. The firm is effectively purchasing a better-informed decision with analytical effort. It does not purchase certainty. Even a longer plan remains a plan under uncertainty, and it becomes obsolete if new information changes the state. But when a price will be exposed to future conditions, refusing to consider those conditions has an opportunity cost.
The model’s finite horizon also avoids a false choice between perfect foresight and no forward-looking behavior. Inside the window, the firm uses the structural relationships specified in the model. It reasons about the evolution of demand and supply disturbances and about how its price interacts with the prices and inflation it expects from other firms. Outside the window, it does not continue this detailed calculation indefinitely. Instead, it uses a continuation-value function.
That continuation value summarizes what the firm believes about the consequences of a price after the explicit planning window closes. In the model, it is not fully contingent on every future state. It reflects the firm’s past experience and its accumulated knowledge about how pricing decisions affect profits. The firm can therefore be forward looking without solving an infinite chain of contingencies. A short horizon is not backward-looking in the simple sense; it is forward looking for a limited distance and experience based beyond that distance.
The distinction matters for expectations. A firm with a horizon of k forms subjective expectations using the model’s structure only through those k periods. At a later date, the firm may plan through a shorter remaining window rather than inheriting the original plan unchanged. In that sense, the firm’s expectations are not the same as the model-consistent expectations of an agent that evaluates every future contingency. The difference is not an error in arithmetic. It is the consequence of deliberately stopping the detailed calculation.
The continuation value is also allowed to learn. After the firm makes its pricing decision, it forms an estimate of the value function associated with the realized state and the horizon it selected. It then averages that estimate with its previous continuation value. The gain applied to recent experience determines how quickly its longer-run belief changes. This gives the firm a memory of past pricing outcomes without requiring it to rebuild the entire model at every date.
Learning serves two purposes. Economically, it allows longer-run beliefs to respond eventually to persistent changes in the environment. Without learning, a structural change could leave the experience-based reference point fixed even as the world changed. Analytically, it creates a trend component in inflation that can persist beyond the initial disturbance. For the present mechanism, the main point is that the continuation value is not an empty box. It is a compact, evolving representation of what the firm has learned outside the current planning window.
There is one further modeling assumption that links an individual firm’s finite calculation to aggregate inflation. When a firm forecasts the inflation it will face during its planning window, it assumes that other firms share its horizon. A firm planning for k periods therefore evaluates future inflation under the belief that other price-setting firms also have k periods of explicit planning. This assumption is not a claim that firms observe one another’s private cognitive processes. It is a way of closing the model while preserving the idea that expectations about other firms’ prices matter for one firm’s reset price.
The firm buys another period only when it pays
The horizon is not imposed as a fixed parameter. The firm begins at a short horizon and considers whether to extend it by one period. At each step it compares two alternatives. It can keep the price already chosen for the current horizon and evaluate the additional future consequences without re-solving the pricing problem. Or it can pay the incremental cognitive cost, solve the expanded problem, and use the new reset price.
The relevant benefit is the value of re-optimizing, not the value of pretending that the firm can eliminate uncertainty. More precisely, the model compares the horizon-(k+1) objective evaluated at the new price with the same objective evaluated at the old horizon-k price. Because both alternatives are evaluated over the same expanded window, the difference isolates what the additional period changes in the firm’s pricing decision. If the extra period would lead to almost the same price, re-solving has little value. If it would lead to a materially different price with higher expected profits, the benefit is larger.
The cost is incremental. A firm that has already paid to plan through k periods does not compare the full cost of planning from zero each time. It asks whether moving from k to k+1 is worth the additional amount. The assumed cost function rises with the horizon, and its increments are weakly increasing. This captures the idea that extending a detailed analysis farther into an uncertain future can become progressively harder or less economical.
In compact form, the decision is:
Extend from k to k+1 when the value gained by re-solving the price exceeds the incremental cost of thinking through one more period; otherwise stop.
That rule makes the horizon endogenous. The firm does not have one permanent level of sophistication. Its selected horizon depends on the current state because the benefit side depends on how much the future could change the price decision. The cost of attention may be stable, but the consequence of not using attention is not.
The diagram’s “stop” is not a decision to stop forecasting altogether. It is a decision to stop expanding the detailed structural calculation. The firm still has a price, a belief about future profits, and an assessment of what lies beyond the window. It simply judges that another layer of explicit analysis would not improve the price enough to justify its cost.
In the analytical cases developed by the authors, the marginal benefit of extending the horizon can be written as a positive curvature term multiplied by the square of the change in the desired reset price from k to k+1. This form carries an intuitive message. The firm is willing to plan further when the next period would materially alter its desired price. If the reset-price increment is small, the gain from re-solving is small even if the firm knows that the future is uncertain.
The square also explains why the cleanest threshold results are initially symmetric around a reference point. A disturbance that moves the desired price far above or far below that point can make an extra period valuable. The sign of the deviation matters for the price itself, but in the simplified no-learning case the size of the mismatch determines the value of additional planning. The next question is what happens when that reference point is itself moving with longer-run inflation beliefs.
The one-stage rule has a precise but limited interpretation. If the net marginal benefit of extending the horizon decreases as k grows, then once the firm reaches the first horizon at which the next extension is not worthwhile, every later extension is also not worthwhile. Under those monotonicity conditions, the sequential stopping rule coincides with the global solution to the horizon problem. The firm does not need to compare every possible horizon because the value of each additional period declines in order.
That conclusion is not automatic in every version of the model. With several disturbances that have different persistence, one component can initially offset another in the desired-price increment and then fade more quickly. The marginal benefit can fall below cost and later rise again. In such a state, a firm that searched globally might prefer a later horizon even though the one-stage rule stopped earlier. The authors retain the sequential rule as a bounded-rational, myopic planning procedure in that more general case. The distinction is important: the model does not quietly turn a convenient stopping heuristic into a universal global optimum.
What the mechanism does—and does not—say about firms
The model gives a disciplined account of a choice that standard fixed-horizon formulations suppress. A fixed horizon can be useful for isolating the consequences of finite planning, but it does not explain why firms might look farther ahead during one episode than another. Endogenizing the horizon makes the amount of forward-looking effort part of the pricing mechanism. It also gives a reason for time-varying sensitivity: when a shock makes future contingencies more consequential, firms may pay for a deeper calculation.
At the same time, the model should not be read as a direct observation of managerial cognition. The continuation-value function, the assumption that other firms share a horizon, and the sequential one-stage rule are modeling devices. They translate bounded attention and experience into a tractable price-setting problem. They do not establish that an individual firm literally computes the stated objective, knows the exact shock process, or consciously labels its own planning horizon.
The strongest interpretation is therefore structural and conditional. The model says that if planning is costly, if firms can evaluate only a finite window in detail, and if future contingencies sometimes matter more than at other times, then the selected amount of explicit planning should vary with the state. It does not yet say how large that variation is for any particular firm. Nor does it by itself identify which real-world constraints—managerial attention, information systems, uncertainty, liquidity, or market pressure—make the calculation more or less costly.
The important result to carry forward is the stop-or-continue logic. A price-setting firm looks ahead because its current price has future consequences. It stops because the next period of detailed analysis is not worth its incremental cost. The horizon can therefore move when the economic state changes. The next step is to examine what kinds of disturbances move that trade-off, how longer-run beliefs shift the reference point, and why the simple symmetry of the decision can break.