top of page
  • Instagram
  • X

Job Security, Tenure, and the Price of Fear

10 hours ago
23 min read

What the tenure chart actually measures

 

Job Security, Tenure, and the Price of Fear

 

A worker can correctly recognize that a long-established job is safer than a new one and still overestimate the probability of losing it. That is the central distinction in the tenure chart: the ranking of risk is broadly understood, but the slope of the risk gradient is not fully reflected in expectations. The red expectations line remains above both measured-outcome lines across the displayed tenure categories. The result is relevant to household finance because spending decisions respond to beliefs about income security, not only to separations that eventually materialize. Yet the difference is not, by itself, a measure of irrationality or a forecast of impending weakness.

At less than six months with the current employer, the expected three-month job-loss probability is 8.2%. The measured transition into unemployment is 5.9%, while the broader measure, including laid-off workers who leave the labor force, is 7.7%. At twenty years or more, the corresponding probabilities are 3.5%, 0.9%, and 1.2%. All six numbers refer to a three-month risk comparison. They are not unemployment rates for the entire population and should not be compared directly with a monthly headline unemployment rate. The horizontal axis measures time at the current job, not age, total working experience, or time since entering the labor force.

The survey attribution matters. The chart explicitly identifies expected job loss with the New York Fed, whereas the measured outcomes come from the Current Population Survey, conducted by the Census Bureau for the Bureau of Labor Statistics. It would therefore be incorrect to describe all three lines as CPS expectations. Subjective probabilities and realized transitions are different objects produced by different measurement systems. The [New York Fed discussion of expectations and outcomes](https://libertystreeteconomics.newyorkfed.org/2026/11/can-workers-predict-job-finding-and-job-loss/) also explains that its survey asks about twelve-month job loss and converts the response to a three-month horizon. The displayed expectation is thus not simply a directly elicited quarterly forecast.

The absolute differences are instructive. For new employees, expectations exceed the unemployment-transition measure by 2.3 percentage points: 8.2 minus 5.9. Including layoffs followed by nonparticipation narrows that difference to 0.5 percentage points. For the longest-tenured group, the equivalent differences are 2.6 and 2.3 percentage points. Consequently, the broad outcome definition changes the interpretation most dramatically at short tenure. Under that definition, new employees' average expectations are relatively close to the measured frequency, while the long-tenure group retains a substantial gap. This is a more discriminating conclusion than saying that every worker exaggerates layoff risk to the same degree.

Relative comparisons emphasize a different feature. A perceived probability of 3.5% is approximately 3.9 times a measured rate of 0.9%, and approximately 2.9 times a rate of 1.2%. By contrast, 8.2% is about 1.4 times 5.9% and only about 1.1 times 7.7%. Ratios become large when their denominators are small, so the long-tenure result should always be presented together with its percentage-point gap. Calling the perception nearly four times the realized rate is arithmetically fair under the narrow definition, but it does not mean that four out of every five fears are fictitious, or that an individual respondent has been observed making the same error repeatedly.

The tenure gradient itself is much steeper in measured outcomes. Moving from the shortest to the longest category reduces expected risk by 4.7 percentage points, from 8.2% to 3.5%. The narrow measured risk falls by 5.0 points, from 5.9% to 0.9%, and the broader measure falls by 6.5 points, from 7.7% to 1.2%. In proportional terms, expectations fall by roughly 57%, whereas both measured series decline by roughly 85%. These are comparisons between groups, not the predicted experience of a particular person who waits twenty years. Distinguishing a cross-sectional gradient from a longitudinal effect is essential before turning the chart into advice about staying with an employer.

The middle categories add useful discipline. The red line does not fall smoothly at every tenure boundary; it bends upward around some intermediate groups, while measured outcomes generally trend lower. Those small reversals may reflect sampling uncertainty, different industries, age composition, or genuine heterogeneity. The figure does not display confidence bands or sample sizes, so interpreting each kink as a behavioral law would overstate what can be learned visually. Even the statement that the largest gap belongs to long-tenured workers needs qualification: the broad pattern is clear, but the exact largest point difference across every bin cannot be established by reading rounded endpoints alone. The strongest endpoint evidence concerns the persistent gap at high tenure.

A simple frequency interpretation makes the economic distinction tangible. In an illustrative group of 1,000 otherwise comparable long-tenured workers, a 3.5% probability corresponds to 35 expected losses over three months. A measured rate of 0.9% corresponds to nine unemployment transitions, and 1.2% corresponds to twelve events under the broader definition. This is not a reconstruction of the underlying sample; it is a scale conversion. Nor are the extra 23 or 26 perceived events necessarily pure forecast error, because survey populations, event definitions, and timing may differ. Frequency translations help readers understand magnitude only when those qualifications remain attached.

Finally, the figure is a description of conditional labor-market risk, not a complete investment signal. It contains no direct evidence about current consumer spending, equity valuation, the next policy decision, or the probability of a recession. Its analytical value is narrower and more durable: measured risk depends strongly on job tenure, subjective risk appears less sensitive to that protection, and counting nonparticipation materially changes the comparison. A useful financial interpretation must preserve all three findings while separating the probability of displacement from its severity, the selection of surviving employment relationships, and the possibility that beliefs influence behavior before any worker actually loses a job.

 

Why surviving jobs become safer without proving a tenure treatment

A declining separation rate with tenure can arise even when the passage of time has no protective effect on any individual employment relationship. This is the first identification problem behind the chart. Imagine two kinds of newly formed jobs: relatively stable matches and fragile matches. If fragile matches dissolve quickly, the stock of jobs observed many years later will contain a larger share of stable matches. The average risk of the survivors falls because the composition changes. Time has sorted the population rather than necessarily changing each job. This selection mechanism can coexist with genuine improvements in security, but the cross-sectional curve cannot disentangle them.

Consider a deliberately simplified example with equal numbers of stable and fragile new jobs. Suppose their quarterly separation probabilities are 1% and 10%, respectively, and remain constant within each type. Initial average risk is 5.5%. After several quarters, more of the fragile relationships have disappeared, so the surviving group has average risk below 5.5%, even though neither type has become safer. Formally, the share of stable jobs among survivors rises with tenure because their survival probability declines more slowly. This example is not calibrated to the displayed data. Its purpose is to demonstrate why a downward hazard profile is insufficient evidence that remaining in any particular job causes the full observed reduction.

A second mechanism is learning about match quality. Hiring occurs under imperfect information: a worker cannot fully assess a manager, role, culture, or schedule before joining, and an employer cannot perfectly observe productivity at recruitment. Early employment reveals information. Poor matches may end during probation or a first project cycle, while successful relationships persist. Once both parties understand the match, the probability of a surprise incompatibility may fall. Here duration is associated with genuine resolution of uncertainty, not merely sorting across permanently different types. Yet the distinction still matters for policy: making someone wait longer in a fundamentally poor match will not necessarily reproduce the security of a mutually productive one.

A third mechanism concerns relationship-specific capital. Workers acquire knowledge of internal processes, clients, equipment, and informal coordination that may be more valuable inside the current firm than elsewhere. Employers likewise invest in recruitment and training. When ending a relationship destroys joint value, retaining the worker becomes more attractive. A productive firm may choose temporary reductions in hiring or hours rather than dismiss staff whose replacement would be costly. This is an economic incentive, not an absolute protection. A plant closure, technological change, or loss of financing can eliminate the surplus that previously supported continued employment. The same specialization that reduces routine turnover can magnify the loss when displacement finally occurs.

The link between tenure and earnings losses is developed in [Audoly, De Pace, and Fella's research on job ladders and human capital](https://www.newyorkfed.org/research/staff_reports/sr1043.html). Their structural framework studies how employer differences and accumulated skills shape the cost of displacement. The relevant insight here is conceptual: a secure match may embody valuable capital that cannot be transferred intact to the next employer. It would be inappropriate to import a model's numerical decomposition into this chart as if the populations and outcomes were identical. The research supports asking about the source of security and the portability of earnings, not declaring that tenure offers a fixed universal insurance benefit.

Institutional arrangements may reinforce these mechanisms. Seniority provisions, internal promotion systems, deferred compensation, dismissal procedures, or organizational norms can affect the relative cost of laying off different workers. But those institutions differ greatly across employers and sectors. A twenty-year public-sector employee and a twenty-year worker in a cyclical private business should not automatically receive the same risk estimate. Neither should a newly hired experienced professional and a first-time labor-market entrant. The chart groups workers by one characteristic; it does not claim that every other determinant of job security has disappeared. Treating each tenure bin as a homogeneous risk class would turn a useful average into a misleading individual prediction.

Age is an especially important confounder. Long tenure is mechanically more feasible for older workers, but older workers may differ in occupation, wealth, retirement options, health, and willingness to search after a separation. Some of those differences affect the probability of displacement; others affect whether displacement is recorded as unemployment or nonparticipation. This means age composition can influence both the red expectation line and the distance between the measured lines. A rigorous empirical exercise would compare workers within age bands and occupations, or use a multivariate framework that makes the conditioning set explicit. Even then, unobserved match quality and employer selection would remain difficult to eliminate completely.

The policy counterfactual must therefore be stated carefully. Asking whether longer-tenured workers are safer is a descriptive question answered by the displayed pattern. Asking whether a worker should reject a new opportunity to preserve tenure is a decision problem involving wages, career growth, employer health, and the risk of the destination job. Asking whether a rule that increases retention would lower layoffs is a causal policy question. These questions are related but not interchangeable. A move that resets measured tenure to zero could nevertheless improve security if the new employer is financially stronger or the new skill set is more transferable. Conversely, an apparent promotion inside a deteriorating firm may preserve tenure while increasing exposure to a common shock.

For investors, the practical consequence is to distinguish a stable level relationship from a changing risk process. A large share of long-tenured workers can mechanically reduce an aggregate separation measure without indicating that each employer has become healthier. A surge in new hiring can raise the share of short-tenure workers and affect aggregate flows even during a robust expansion. Conversely, weak hiring can leave a more selected group of incumbents and make observed layoff risk look reassuring while job seekers struggle. The chart's gradient is therefore a building block for analysis, not a standalone cycle indicator. Its strongest message is that employment histories contain information, while the economic meaning of that information depends on selection, match development, and the institutions sustaining each relationship.

 

The measurement gap between unemployment and displacement

Losing a job and becoming unemployed are not equivalent events. A worker can be displaced and then retire, stop searching temporarily, enter education, manage a health problem, or take another job before the next interview. A worker can also become unemployed after a voluntary quit rather than a layoff. The choice of outcome therefore determines which risks appear in the comparison. A subjective question about losing one's current position is broader in some respects than an observed transition into unemployment, yet it may also be interpreted differently across respondents. Before diagnosing pessimism, an analyst must ask whether the expectation and realization concern the same event, population, and time interval.

The [BLS definitions of labor-force status](https://www.bls.gov/cps/definitions.htm) distinguish employment, unemployment, and being outside the labor force. Unemployment generally requires being without work, available, and actively searching, with specific treatment for temporary layoffs. Someone who is no longer working but is not searching may instead be outside the labor force. That classification is meaningful for measuring labor supply; it is not a judgment that the person suffered no economic loss. An income shock can be severe even when it does not enter the narrow unemployment measure. The broader green line in the chart addresses part of that distinction by including layoffs followed by nonparticipation, rather than adding all departures from employment indiscriminately.

At short tenure, expanding the outcome raises measured risk from 5.9% to 7.7%, an increase of 1.8 percentage points. At twenty years or more, it raises measured risk from 0.9% to 1.2%, an increase of 0.3 points. The difference between the green and blue lines should not be interpreted as the total probability of retirement, all labor-force exits, or every unrecorded layoff. It represents the additional events admitted by this particular measurement definition. Likewise, the broader series need not fully exhaust involuntary loss of a current job. It improves alignment along one margin but does not eliminate differences in timing, survey response, or the treatment of rapid reemployment.

This is a competing-risks problem. Starting from employment, an individual may move to unemployment, to nonparticipation, or directly to another job. The chance of observing one destination depends partly on the availability of the others. For example, two equally vulnerable workers may experience the same dismissal risk but different measured unemployment transitions if one has an immediate outside offer. Treating their unemployment-transition probabilities as complete measures of employment insecurity would confuse displacement risk with recovery opportunities. A useful analysis distinguishes the initial shock from the state recorded after the shock. That distinction is especially important when job-finding conditions change while employers' layoff decisions remain relatively stable.

Timing introduces another layer. Monthly observations may not reveal every short spell between interviews. A worker can lose a position, search for two weeks, and start a new one before the next survey date. A measure constructed from observed states may then capture a different event history from the one the worker had in mind when answering a forecast question. Longitudinal matching can also lose respondents through nonresponse or changes in residence. Without the underlying construction and attrition diagnostics, one cannot quantify the resulting bias from the chart alone. The correct response is to treat the observed frequency as definition-dependent, not to assume that every missing event necessarily enlarges the gap by an unknown convenient amount.

Horizon conversion deserves equal care. If a twelve-month probability is P and a constant quarterly hazard is assumed, the corresponding three-month probability is q = 1 - (1 - P)^(1/4). Inverting that relationship gives P = 1 - (1 - q)^4. Thus a quarterly probability of 3.5% corresponds to approximately 13.3% over a year under that constant-hazard assumption, rather than exactly 14%. A quarterly probability of 8.2% corresponds to approximately 29.0%, rather than exactly 32.8%. These calculations illustrate compounding; they are not independent annual forecasts. Layoff hazards can vary seasonally, with business conditions, and as tenure accumulates, so the assumption should not be mistaken for a law of employment dynamics.

The distinction between a stock and a flow is equally consequential. The headline unemployment rate describes people unemployed at a point in time relative to the labor force. The chart describes a forward transition probability conditional on initially having a job and belonging to a tenure category. A low flow into unemployment can coexist with a high unemployment stock if unemployed workers have difficulty finding work. Conversely, rapid hiring can keep the stock low even if turnover is substantial. An investor who reads a low long-tenure loss rate as proof of a universally strong labor market overlooks the separate job-finding margin and the experience of people who are not currently employed.

Calibration requires more than comparing two group means. An ideally calibrated probability forecast would assign an average probability near the realized frequency among people facing equivalent conditions and answering the same question. Here, different survey samples may have different sector, age, and income compositions. The chart also pools workers rather than demonstrating that each person's subjective probability was matched to their own subsequent outcome. Reweighting, consistent event definitions, and aligned observation dates would strengthen the comparison. Confidence intervals would then help determine whether remaining differences exceed sampling noise. Without them, the endpoint gaps are economically suggestive but cannot supply a precise statistical rejection of rational expectations.

The broader lesson is not that the comparison is useless. The persistence of a substantial long-tenure gap under both outcome definitions is informative, and the short-tenure narrowing is itself evidence that definitions matter. But a disciplined interpretation separates three hypotheses: expectations may be poorly calibrated; measured outcomes may omit events respondents reasonably include; and the two surveys may describe different populations or horizons. These hypotheses can operate simultaneously. For financial analysis, the right object is an explicit range of plausible income risks, combined with a separate account of beliefs, rather than a single residual labeled irrational fear. That approach preserves the chart's warning about perceived insecurity without claiming precision the evidence does not provide.

 

A low probability can coexist with a large economic loss

Probability is only one dimension of labor-income risk. A worker whose chance of displacement is low may still have much to lose if displacement occurs. Long tenure can mean accumulated employer-specific knowledge, a favorable wage premium, internal seniority, pension expectations, or a schedule difficult to replicate elsewhere. Losing those features can matter more than the length of the initial unemployment spell. The chart contains probabilities, not conditional earnings losses, so it cannot establish which tenure group has the largest expected financial damage. A complete household-risk assessment requires both the likelihood of an event and the consequences conditional on that event.

A useful accounting identity is expected loss = probability of displacement multiplied by expected economic loss conditional on displacement. This is an approximation when severity and event timing are simplified, but it prevents a common conceptual error. Suppose a hypothetical short-tenure worker has a 7.7% quarterly displacement probability and a conditional loss of $10,000. Expected loss is $770. Suppose a long-tenured worker has a 1.2% probability but a conditional loss of $80,000. Expected loss is $960. These are deliberately illustrative assumptions, not estimates from the chart. They show why ranking event probabilities alone need not rank expected income losses or optimal demand for insurance.

Conditional loss has several components. There is forgone income before the next job begins, offset by any benefits or severance. There may be search expenses, relocation costs, loss of health coverage, or reduced pension accrual. Once reemployed, the worker may earn less because the previous wage reflected a particularly productive match or an internal career ladder. Some losses are temporary; others last for years. The present value depends on their duration and discounting. A framework that counts only the first missed paycheck understates persistent damage, while one that assumes every displaced worker permanently loses the full previous salary overstates it. Both extremes obscure the distribution that matters for planning.

The [Chicago Fed's review of job-loss research](https://www.chicagofed.org/publications/chicago-fed-letter/2004/october-207) highlights that displacement costs depend on reemployment speed and subsequent wages, and discusses the role of specific human capital and match quality. Those mechanisms help explain why concern can remain substantial among workers with stable employment histories. They do not prove that every high-tenure worker faces larger losses or that the probabilities in the chart are rationally chosen. The important separation is between a belief about frequency and a preference over consequences. A worker can overstate a probability while still having a sound reason to maintain substantial protection against the event.

Risk aversion adds a further distinction between expected dollars and economic welfare. Losing $20,000 when cash balances are already low can be more damaging than the benefit of gaining $20,000 in a comfortable state. In a standard concave-utility framework, marginal resources are especially valuable when income falls. A small-probability adverse state can therefore justify precautions whose cost exceeds the simple expected cash loss. This is not an argument for unlimited saving or for treating every feared scenario as equally likely. It is an argument that probability calibration and optimal precaution are separate exercises. Correcting a belief does not automatically determine a household's appropriate spending or investment response.

Duration creates an additional mismatch with the chart's horizon. The event window is three months, but a displacement during that window may affect earnings long afterward. Consider a purely hypothetical worker earning $80,000 annually who finds another job at a 10% lower wage. If the $8,000 annual difference lasts five years, undiscounted forgone earnings total $40,000, before counting any initial unemployment. The arithmetic does not claim that this wage path is typical. It illustrates how a low quarterly event probability can interact with a multi-year loss. An analysis that multiplies the quarterly probability only by one quarter's salary implicitly assumes away the persistence that may motivate the worker's anxiety.

Household structure can diversify or amplify the exposure. A second earner employed in an unrelated sector may stabilize cash flow. Two earners at the same employer, or in industries affected by the same demand shock, may have highly correlated income risks. A home located in a company-dependent region can lose value just when local employment deteriorates, reducing the option to move. Employer stock and deferred equity compensation add another concentration channel. The chart does not reveal any of these correlations. Therefore, applying a single tenure-based probability to an entire household without considering joint exposure would provide false reassurance precisely where labor and financial capital are most tightly linked.

Balance-sheet liquidity is different from net worth. A long-tenured homeowner may have substantial housing equity and retirement assets but limited immediately accessible cash. If borrowing becomes harder after a job loss, assets that looked sufficient in an ordinary budget can be difficult or expensive to mobilize. Conversely, a new employee with modest wealth but low fixed expenses and a liquid reserve may absorb a short income interruption relatively well. This distinction explains why emergency-liquidity needs cannot be inferred from a worker's tenure category. A useful stress test maps the timing of required payments against accessible resources, rather than comparing total asset values with annual income in isolation.

The severity perspective also changes the interpretation of nonparticipation. A worker who leaves the labor force after being laid off might have chosen retirement anyway, or might have retired earlier than intended because suitable work was unavailable. Those cases can have very different welfare consequences despite sharing a measured destination. The broader 1.2% and 7.7% rates improve event coverage, but they do not reveal voluntariness, lost lifetime earnings, or subjective distress. Nor does a 3.5% expectation tell us whether respondents are thinking about unemployment, demotion, reduced hours, or a forced career change. Quantitative discipline means acknowledging these unobserved margins rather than assigning them invented values.

The resulting interpretation is balanced. Long tenure is associated with much lower measured job-loss probabilities, and its perceived protection appears weaker than its measured protection. At the same time, the cost of losing a long-established position may be sufficiently large to support cautious behavior. Neither proposition cancels the other. For investment analysis, the essential questions become who bears the loss, how long it lasts, whether resources are liquid when needed, and how the shock co-moves with market returns. Those questions turn a descriptive labor chart into a structured framework for understanding household vulnerability without pretending that a single probability is a complete measure of financial security.

 

How fear reaches consumption, credit, and asset allocation

Beliefs can affect economic activity before the feared event occurs. A worker who becomes more worried about losing a job may postpone a car purchase, reduce discretionary spending, or increase a cash reserve even if employment continues uninterrupted. If many households react together, demand can weaken without a corresponding immediate surge in layoffs. That transmission mechanism explains why subjective labor-risk surveys can contain information not captured by current employment flows. But the tenure chart is a comparison across groups, not a time series showing a recent increase in fear. It supports a mechanism to investigate, not a conclusion that precautionary retrenchment is currently accelerating.

The relevant theoretical foundation is precautionary saving under uncertain income and incomplete insurance. [Carroll's buffer-stock saving research](https://www.nber.org/papers/w5788) describes how income uncertainty can lead households to hold a target reserve relative to their resources. The useful insight for this chart is that households may respond to perceived risk, while the reserve also depends on preferences, borrowing opportunities, and existing wealth. One cannot map a 2.3-percentage-point perception gap directly into a percentage decline in consumption. Such a mapping requires a model or an empirical estimate of how spending changes when beliefs change, together with information about which households actually experience that change.

An illustrative budget clarifies the timing. Consider a household with monthly essential expenses of $4,000 and liquid savings of $12,000. It has three months of essential spending covered before accounting for benefits or other income. If its desired reserve rises to four months, it needs another $4,000. Building that reserve over eight months would require diverting $500 each month from other uses. These are hypothetical amounts, not observations from the survey. They demonstrate how a change in a desired balance-sheet target can produce a temporary saving flow. Once the target is reached, the effect on monthly spending need not continue at the same rate, even if the level of perceived risk stays elevated.

Credit access can reinforce that adjustment. A household that expects to borrow easily against future income may initially maintain a smaller reserve. If it worries that lenders will reduce limits precisely when employment is lost, it may instead save in advance. Lenders can also tighten terms in response to actual deterioration in borrower risk, creating a separate channel from sentiment. Distinguishing these channels matters empirically. A rise in saving could reflect household pessimism, weaker credit supply, higher interest income, or reduced purchase opportunities. The chart alone cannot identify which force dominates. Labeling every spending slowdown a consequence of fear would be as misleading as ignoring beliefs altogether.

Heterogeneity determines the aggregate response. A wealthy household may change its portfolio composition without materially reducing consumption. A cash-constrained household may want to save more but be unable to cut essentials. Another household may reduce durable purchases sharply because those decisions are easier to postpone. Consequently, the average subjective probability does not reveal the spending response of the marginal dollar of income. Analysts need the joint distribution of beliefs, income, liquidity, and consumption sensitivity. The long-tenure group may have lower measured event risk but greater accumulated wealth, different housing obligations, or different retirement horizons. None of those characteristics can safely be inferred from tenure alone.

The financial-portfolio channel is similarly conditional. Human capital is the present value of future labor income, and it is usually not directly tradable. If that income behaves like a stable payment stream, it may support greater financial risk-taking than otherwise. If it falls in the same states as equity returns, the household already has an implicit exposure to market risk through its job. [Research on occupation-level income shocks and asset returns](https://www.nber.org/papers/w7905) explicitly considers that covariance in portfolio choice. The chart supplies information about one event probability, but not the covariance needed to determine an equity allocation. A tenure label cannot replace analysis of industry exposure and household concentration.

For example, an employee who receives salary, bonuses, and deferred shares from the same cyclical company may face simultaneous declines in income and invested wealth. A small average probability of displacement does not eliminate the danger of a joint adverse state. Conversely, a worker in a relatively insulated occupation may have labor income less correlated with broad equity returns, even at short tenure. These are conceptual scenarios rather than recommendations to buy or sell any asset. The principle is to assess the entire balance sheet, including nontradable income and contingent obligations, rather than treating a brokerage account as the only source of investment risk.

Aggregate asset prices introduce yet another step. Household fears might reduce demand for risky assets, weaken expected corporate revenue, or increase demand for liquid instruments. Those effects can point in different directions for different securities and can be offset by policy responses, valuations, or changes in expected inflation. There is no mechanical formula taking the chart's perceived-measured gap to an equity index target or a bond-yield forecast. A plausible narrative is not an identified pricing relationship. To establish a market signal, one would need changes in the gap over time, evidence that those changes precede returns or cash flows, and controls for information already embedded in prices.

A potentially important feedback loop nevertheless follows logically. If excessive perceived insecurity causes spending cuts, businesses may experience weaker sales and subsequently reduce labor demand, partially validating the initial concern. But the possibility of such a loop does not show that it is quantitatively important in any particular period. It also does not justify ignoring genuine risks on the premise that optimism would prevent them. The appropriate empirical question is whether belief changes predict behavior beyond what observable income fundamentals explain. Repeated surveys linked to expenditure or credit data would be more informative for that purpose than a single cross-sectional picture.

The investment use of the chart is therefore diagnostic rather than directional. It encourages analysts to ask whether households price labor-income uncertainty more conservatively than measured flows suggest, whether that conservatism changes over time, and whether vulnerable households have room to adjust spending or portfolios. It also warns against assuming that low layoffs automatically imply strong consumer demand. The starting facts remain the same: new workers report 8.2% risk against broader measured risk of 7.7%, while long-tenured workers report 3.5% against 1.2%. The economic consequences depend on liquidity, loss severity, correlated exposure, and actual behavioral responses—not on those probability gaps in isolation.

 

An evidence-based framework for monitoring the gap

The first step in using the tenure comparison is to define the question before choosing a statistic. An analyst interested in income interruption needs a different outcome from one studying involuntary employer separation, and both differ from the headline unemployment rate. A household considering a reserve needs the probability and duration of lost income, while an investor studying consumer demand needs evidence about beliefs and spending. Writing these distinctions down prevents a convenient chart from silently changing roles as the argument develops. The same three lines can inform several questions, but none supplies every missing variable required to answer them.

A monitoring framework should retain separate expectation, narrow-realization, and broader-realization series for each tenure group. The two gaps can be written as gU = expected risk minus unemployment-transition risk, and gB = expected risk minus broader measured risk. For the shortest-tenure endpoint, gU is 2.3 percentage points and gB is 0.5. For the longest-tenure endpoint, they are 2.6 and 2.3. Keeping both gaps makes the effect of the event definition visible. Collapsing them into one preferred measure risks selecting whichever comparison best supports a prior narrative. A robust conclusion should state which findings survive both definitions and which depend heavily on including nonparticipation.

Time consistency comes next. A pooled tenure gradient should not be described as a new monthly deterioration. The underlying New York Fed comparison averages a long sample and excludes an exceptional pandemic interval; that design is useful for studying a typical relationship but not sufficient for dating a current turning point. A real-time monitoring exercise would align survey collection periods, forward realization windows, and publication lags. It would distinguish information available when a forecast was made from outcomes observed afterward. Otherwise, a backtest can inadvertently use future knowledge to judge decisions that were reasonable with the information then available. The displayed figure alone is not a ready-made real-time trading history.

Composition adjustment is another necessary control. Suppose the workforce's average observed loss probability is a weighted average across tenure bins. It can rise because risk rises within bins or because employment shifts toward higher-risk short-tenure jobs. Those are economically different developments. A fixed-weight index would hold the tenure distribution constant to isolate within-group changes, while the ordinary aggregate would retain the actual composition effect. Both should be reported when possible. The same principle applies to age, sector, education, and occupation. Reweighting does not solve every identification problem, but it makes clear whether a headline movement is about changing risk or a changing mix of workers.

Sampling uncertainty should be explicit rather than buried behind decimal precision. In a simplified independent sample, a measured probability p based on n observations has approximate standard error sqrt(p(1-p)/n). If p is 1.2% and n is 1,000, that formula gives about 0.34 percentage points. A rough interval of two standard errors on either side would therefore be materially wider than the last decimal place suggests. This is only a pedagogical illustration: actual survey weights, clustering, repeated observations, and matching alter the calculation. The chart does not provide the sample counts required to apply it. Consequently, readers should not interpret small visual differences across adjacent bins as established statistically significant effects.

Calibration analysis can then become more demanding. Rather than asking only whether the average forecast exceeds the average outcome, one can compare forecast buckets with subsequent event frequencies after harmonizing definitions. A useful forecast may discriminate between safer and riskier workers even if its level is biased. Conversely, an aggregate mean can look accurate while masking offsetting errors across groups. The tenure chart suggests this distinction directly: expectations broadly rank new jobs as riskier than established ones, yet the decline is less pronounced than in measured outcomes. Separating ranking ability from level calibration helps identify whether communication should emphasize relative risk, absolute probabilities, or both.

Scenario analysis should combine those probabilities with severity and resources without disguising assumptions as data. One scenario might keep the broader measured loss rate as a baseline, another might use the subjective rate as a conservative planning input, and a third might stress both displacement frequency and reemployment wages. Each scenario should specify what changes and why. The point is not that subjective probabilities are automatically an upper bound; they may be wrong in either direction when conditions shift. Rather, using separate scenarios reveals how sensitive a decision is to the unresolved measurement gap. If a plan fails under a small probability revision, its fragility deserves attention regardless of which estimate eventually proves closer.

For market interpretation, a clear falsification standard is valuable. A story that elevated insecurity depresses demand should predict observable changes in spending, saving, or credit behavior among exposed households after accounting for income and wealth. A story that tenure itself creates protection should survive comparisons that address selection and employer composition. A story that the gap predicts asset returns should demonstrate incremental information beyond valuations and standard macro indicators, using data available at the time. These tests do not require every narrative to become a trading model. They require a distinction between an economically plausible mechanism and evidence that the mechanism explains the episode under discussion.

The chart ultimately supports a carefully bounded conclusion. Perceived three-month job-loss risk falls from 8.2% for workers with less than six months' tenure to 3.5% for those with at least 20 years. Measured unemployment transitions fall from 5.9% to 0.9%, and including laid-off workers who leave the labor force raises the endpoints to 7.7% and 1.2%. The gap remains particularly important among long-tenured workers, while the broader definition brings the short-tenure comparison much closer. These facts point toward a less steep perceived security gradient, not a demonstration that every anxious worker is irrational or that current aggregate employment is about to weaken.

The financial lesson is to respect both sides of that distinction. Measured job stability can be stronger than households perceive, and concern about losing a valuable established match can still have a rational economic foundation. Portfolio and consumption decisions depend on loss severity, liquidity, correlated exposures, and the opportunities available after displacement. An evidence-based reading neither dismisses fear because layoffs are rare nor treats fear as proof that layoffs are imminent. It uses the observed probabilities to discipline scenarios, preserves the separate role of beliefs, and updates conclusions only when new, comparable evidence shows that the underlying risks or household responses have actually changed.

Comments


© 2035 by Someo Park Investment Management LLC.

bottom of page