Report

The effect of skills-half life on one's experience in the job market

RK Reza Khorshidi

02 March 2026

On this page 12 items

This article presents a simplified stochastic simulation to illustrate how shrinking skill half-life can mechanically increase structural misalignment in the labour market. It is a mechanism demonstration, not a forecast.
TL;DR:
In a 2-year half-life market (today), even if you are genuinely the top candidate, the system can treat you the way it used to treat a 75th-percentile candidate in an 8-year half-life market (roughly a decade ago).

1. Model Overview

We simulate:

Candidates with evolving skill vectors
Jobs with evolving requirement vectors
Hiring decisions made using lagged signals
A structural misalignment metric called regret

The goal is to understand how shrinking skill half-life changes the experience of candidates in the job market.

2. Representing Skills and Jobs

Let:

$c_{i} (t) \in R^{d}$ = candidate $i$ at time $t$
$r_{j} (t) \in R^{d}$ = job $j$ at time $t$

Similarity is measured via cosine similarity:

sim (i, j, t) = c_{i} (t)^{⊤} r_{j} (t)

Summary: Fit equals the dot product between candidate and job vectors.

3. Encoding Skill Half-Life

Quarterly update step:

Δ = 0.25

Decay parameter:

ρ (h) = 2^{- \frac{Δ}{h}}

Skill evolution:

x (t + Δ) = ρ (h) x (t) + 1 - ρ (h)^{2} ε

After $h$ years:

ρ^{\frac{h}{Δ}} = 0.5

Summary: After $h$ years, only 50% of original skill variance remains.

4. Asymmetric Drift

Jobs drift faster than candidates:

h_{jd} = 0.6 \cdot h_{cv}

Interpretation: Hiring mandates evolve faster than people reskill.

5. Adaptation Mechanism

Market mean:

\overset{r}{ˉ} (t) = \frac{1}{∣ J ∣} j \sum r_{j} (t)

Candidate update:

c_{i} (t + Δ) \leftarrow c_{i} (t + Δ) + η (\overset{r}{ˉ} (t) - c_{i} (t))

where $η = 0.10$ .

Summary: Candidates adapt, but toward a moving target.

6. Signal Lag

Hiring decisions rely on stale information:

CV lag = 2 quarters
JD lag = 1 quarter

Signal similarity:

sim_{signal} = r_{j} (t - L_{jd})^{⊤} c_{i} (t - L_{cv})

True similarity:

sim_{true} = r_{j} (t)^{⊤} c_{i} (t)

7. Structural Regret

Oracle best:

best_true (j, t) = i max r_{j} (t)^{⊤} c_{i} (t)

Regret:

regret (j, t) = best_true (j, t) - r_{j} (t)^{⊤} c_{\hat{i}} (t)

Average regret:

R (h) = E_{j, t} [regret (j, t)]

Summary: Regret measures the gap between chosen candidate and true best.

8. Experimental Setup

Parameter	Value
Dimension	50
Candidates	400
Jobs	150
Steps	60 quarters
Burn-in	30 quarters
CV Lag	2
JD Lag	1

Half-life values tested:

h \in {2, 4, 8, 16}

9. Simulation Flow

Rendering diagram…

10. Results

Y-axis definition:

regret = best_possible_true_fit - true_fit_of_chosen

11. Interpretation

Regret declines as half-life increases.
Precision matters more in stable regimes.
Drift dominates in high-volatility regimes.

At $h = 2$ , even selecting the best candidate by signal produces substantial regret.

12. What This Means

When skill half-life shrinks:

Drift dominates
Signal lag injects structural error
Misalignment becomes a system property

Recruitment and L&D must evolve toward continuous alignment systems in a high-volatility world.