Why Firms Pull Back Before Demand Is Known
Firm-level uncertainty affects output when a risk-averse owner must commit to a wage bill before knowing whether customer demand will cover it. This chapter focuses on that decision: why a wider distribution of possible sales can reduce hiring even when a firm still has access to funds, and how differences in financial wealth turn one consumer-credit shock into different employment choices. The practical understanding is that net worth is not only a source of spending power; in this model, it is also a buffer against an uninsurable operating loss.
The preceding demand mechanism established that tighter customer credit can widen the relevant distribution of sales per worker. A firm now has to decide what to do with that distribution. It can hire more workers and gain more expected revenue, but it must promise wages before the customer realization arrives. The choice is therefore not between production and no production. It is between a larger expected operation with more exposure to a bad draw and a smaller operation whose downside is easier to absorb.
The wage bill is a bet on customer demand
Suppose a firm hires one additional worker. The worker adds productive capacity, so the firm can serve more customers if demand is strong. Expected sales rise with the scale of the operation. But the worker is paid whether the expected customers arrive or not. If customer matching is weak or customer purchasing capacity is low, the extra capacity produces little revenue while the wage obligation remains. The marginal worker therefore increases both the upside and the amount at risk.
This is the central risk-and-return trade-off in the model. More employment raises the mean of the firm’s sales distribution and also raises its variance. The wage bill grows with every additional worker, while the probability of receiving enough additional customers does not rise in a perfectly predictable way. At a small scale, another worker may be readily supported by the existing customer pool. At a larger scale, the firm is asking the matching process to deliver more customers, and the possible low-sales outcomes begin to carry a larger fixed wage burden.
A purely expected-profit calculation would compare the expected revenue from a worker with the expected wage cost. The model’s entrepreneur does more. Because the owner is risk-averse and cannot insure the demand realization with a complete set of contingent contracts, the owner also considers how much consumption or wealth could be lost in a weak outcome. Two hiring plans can have similar expected sales and different downside exposure. The plan with the wider loss distribution is less attractive to a risk-averse owner, even if its average return is not lower.
Imagine two firms that both forecast sales of 100 after hiring a second worker. One has a fairly concentrated range of outcomes because its customer base is stable; the other has a wider range because a small change in matching can leave the worker with very few purchasing customers. The expected-revenue calculation treats the plans as equal. The risk-averse owner does not, because the second firm must be prepared for a larger gap between wages and realized sales. The example is conceptual rather than an additional estimate from the paper, but it captures why the dispersion of demand enters the scale choice.
The source of the risk is not simply that the firm is ignorant. The firm knows the aggregate credit condition and forms expectations about customer demand. What the firm does not know at the hiring stage is the particular customer realization assigned to its workers. The owner must make a decision under a known distribution rather than wait for certainty. That is enough to make the dispersion of outcomes economically relevant.
The wage bill also determines who bears the first loss. Workers receive the promised wage independently of the firm’s realized sales in the model. The entrepreneur therefore absorbs the gap between the wage commitment and weak revenue, subject to the firm’s financial wealth and access to the non-contingent bond. This is why an increase in demand risk can reduce labor demand before any worker has been dismissed: the owner anticipates the exposure created by the next commitment.
The diagram’s arrows do not mean that risk mechanically determines a fixed hiring cut. They identify the objects in the owner’s decision. Expected sales support more hiring; the wage bill enlarges the amount exposed to a weak realization; net worth provides a buffer; and risk aversion makes the downside part of the choice. The resulting hiring level is the point at which the expected gain and the owner’s tolerance for exposure are balanced.
A firm with a large asset buffer can carry a bad customer realization without reducing consumption or violating its financial plan. A firm with little wealth faces the same customer distribution but has less room for error. The two firms can therefore choose different employment levels even when they observe the same credit condition, face the same wage, and have the same production technology. Their difference is not necessarily a difference in expected demand. It is a difference in the amount of risk they can absorb.
Net worth turns a common shock into different hiring choices
In the model, firm financial wealth enters the labor decision directly. Individual labor demand is proportional to the entrepreneur’s financial wealth, with the proportionality factor changing with the aggregate state and the wage. This captures a form of self-insurance. Safe assets allow an owner to withstand adverse idiosyncratic demand and maintain a larger wage commitment. When wealth falls, the owner reduces hiring to limit exposure because the same demand realization now has a larger consequence for the firm’s remaining resources.
It is useful to distinguish the wealth buffer from a forecast of profitability. A firm can expect a worker to generate positive average revenue and still decide not to hire if a weak customer realization would consume too much of its financial cushion. Conversely, a wealthier firm can accept the same expected revenue because it can carry the downside. The model’s result is therefore about the interaction between expected return and the capacity to bear risk, not about a claim that poorer firms always have worse technologies or weaker customer demand.
That result is deliberately different from a standard collateral story. The baseline model does not need to assume that firms suddenly lose access to external finance, that collateral values have fallen, or that a bank has imposed a new borrowing limit. A lower net-worth firm may still be able to borrow through the model’s available financial asset. The reason for its lower labor demand is that it cannot hedge the uncertain sales outcome sufficiently. It chooses a smaller operating scale because it is less able to insure the bet, not because the model has imposed a new mechanical quantity constraint on borrowing.
This distinction matters when interpreting an observed hiring decline. In real firms, weaker employment can reflect lower expected demand, a shortage of cash, a covenant, a higher cost of borrowing, or a more cautious response to risk. The Federal Reserve model isolates the last channel under its assumptions. It shows that a risk response can exist even when the firm’s financing problem is not represented as a denied loan. The model does not establish that risk aversion is the only explanation for any real-world hiring decision.
The relationship between net worth and hiring also creates persistence. After a weak demand realization, a firm’s assets can be lower. Lower wealth then reduces the amount of labor the owner is willing to hire in the next period, which lowers future production and the opportunity to rebuild wealth. In an expansion, the opposite logic can operate: stronger sales improve wealth, a larger buffer supports more hiring, and higher production creates more expected revenue. The source’s closed-form policies make this feedback tractable, but the economic idea is broader—financial strength changes the amount of operating risk a firm can carry.
Cross-sectional heterogeneity is essential to the aggregate result. If every firm had the same wealth, customer realization, and capacity, the economy would respond more uniformly to a credit shock. With heterogeneous firms, some owners remain willing to expand while others cut back. The aggregate employment response is the sum of those different choices. The same aggregate credit condition can therefore produce a broad distribution of firm-level adjustments rather than one representative-firm response.
This aggregation also explains why an economy can experience a meaningful employment response without every firm making a dramatic change. Firms near the margin of tolerable exposure may cut one hiring decision, postpone an expansion, or accept a smaller workforce. Firms with more wealth may continue to expand. The aggregate statistic records the sum of these discrete choices, while the cross-sectional distribution reveals that the adjustment was uneven. A representative-firm description would hide the role of the financial buffer in deciding which firms move.
The model also links customer traffic to employment. The customer-traffic evidence used for calibration is procyclical, and the model generates the expected comovement: better aggregate credit raises expected sales, makes the hiring bet less risky, and increases labor demand. Worse credit lowers expected sales and increases the risk attached to labor, so firms revise production plans downward. The aggregate movement in employment is consequently built from both the level of expected demand and the cross-sectional risk borne by individual owners.
The calibrated responses give this mechanism a scale, while remaining model results rather than forecasts. A positive one-percent innovation to aggregate credit raises model output by about 0.8 percent and employment by roughly 0.6 percent on impact. Sales per worker rises by approximately 0.18 percent. The larger output response comes primarily through labor demand, with a smaller productivity component. The model also reports a sharp decline in demand uncertainty after an improvement in credit conditions, because the sales-per-worker distribution becomes more compressed around the capacity margin.
These numbers should not be read as saying that a one-percent change in any observed consumer-credit series will generate exactly those movements. They are impulse responses from a calibrated economy with a particular customer distribution, capacity ceiling, labor supply, and incomplete-insurance structure. Their value is to show how the behavioral mechanism can connect a credit innovation to economically meaningful employment and output changes.
What the model reproduces—and what it leaves out
The model is useful partly because it is evaluated against cross-sectional as well as aggregate moments. It reproduces substantial dispersion in employment and sales and captures the qualitative fact that sales-growth dispersion is countercyclical. The reported correlation between model sales-growth dispersion and the business cycle is −0.170, close in direction and size to the data’s −0.194. This is evidence that the capacity-and-matching mechanism can generate the central cyclical pattern without requiring an exogenous volatility shock.
The fit is not exact. The paper notes that the parsimonious model overstates the level of cross-sectional dispersion in both employment and sales. Its quantitative fit for employment-dispersion cyclicality is also weaker than its fit for sales-growth dispersion. That limitation is informative. A single underlying source of demand heterogeneity can reproduce the main sales pattern, but it does not capture every force shaping firm employment distributions. The result supports the mechanism’s relevance, not the claim that it is a complete model of firm dynamics.
Several assumptions carry the result. Entrepreneurs are risk-averse; firms have incomplete access to insurance; labor is hired before the idiosyncratic customer realization; workers are paid regardless of achieved sales; and production per worker is capped. Remove the capacity ceiling and changes in credit would be more likely to act as a shift in expected demand without materially changing dispersion. Give firms complete contingent insurance and the downside distribution would have less influence on the hiring decision. Let wages adjust to realized sales, and the owner’s exposure would be different.
The effect is also state-dependent. When firms have substantial unused capacity, a rise in customer purchasing power can be translated into more sales per worker before the ceiling binds. When many firms are already close to capacity, the same change affects the distribution around the ceiling and has a stronger risk interpretation. Labor-supply elasticity changes how the aggregate adjustment is divided between employment and wages. These are not secondary calibration details; they determine which margin bears the shock.
The conclusion is therefore narrower and more useful than “uncertainty causes recessions.” In this framework, a change in consumer credit changes the distribution of sales. A risk-averse owner converts that distribution into a hiring choice. Financial wealth determines how much downside the owner can absorb, and the collection of those choices changes aggregate employment and output. The next synthesis compares the direct level response with the additional effect of endogenous uncertainty, asking how much the distributional channel adds and when its contribution is likely to be strongest.