Research

Working paper · 2026

Before the Factory Opens: Anticipation and Spatial Adjustment to Announced and Delayed Local Shocks

Solo-authored · Presented at the Kentucky Economic Association Annual Conference, October 2026

How much do people move before a large employer actually arrives? I develop a continuous-time spatial mean field game, built on coupled Hamilton–Jacobi–Bellman and Kolmogorov forward equations, to study anticipatory relocation around Intel’s announced Ohio semiconductor investment. The model is calibrated through an exact discrete inverse equilibrium to 969,829 worker residences from Census LODES, together with BLS QCEW data and publicly announced project employment and pay. The agents stand for the region’s whole resident labor force, in every industry, not only the people who might take a job at the plant. Roughly 46.7% of the eventual model-implied spatial adjustment occurs before the initially expected opening date, and later information revisions about project delays generate a partial reversal. Extending the model to two-dimensional geography reduces predicted reallocation by about 16% while leaving the timing result nearly unchanged.

• Mean field games
• Spatial equilibrium
• Anticipation
• Place-based policy
• Census LODES
• BLS QCEW

# Exact baseline: detailed balance at each cell face gives the value gap
# that holds the 2021 LODES distribution perfectly still.
dp = p[1:] - p[:-1]
V[1:] = np.cumsum(m0.c * D * dp / np.minimum(p[:-1], p[1:]))
V -= p @ V
a = drift(V, g, m0)
Q = controlled_Q(a, diffusion_Q(g, m0), g)

# Back out the baseline amenity b from the stationary HJB.
F = m0.kappa * (density(p, g) + m0.eps) ** m0.gamma
b = m0.rho * V - Q @ V + F + cost(a, m0)

# Mean field game fixed point: iterate HJB backward and KFE forward
# until the population path workers expect is the one they create.
for _ in range(m.dyn_max):
    V, A, maxh = backward(pp)   # choices, given where everyone will be
    br, maxk = forward(A)       # where everyone ends up, given those choices
    fp = float(np.max(np.abs(br - pp)))
    if fp < m.dyn_tol:
        break
    pp = (1 - m.dyn_relax) * pp + m.dyn_relax * br
    pp[0] = p0

Work in progress

Time, Tools, Technology: Identifying the Educational Returns to Technology Use

Solo-authored

This paper extends my honors thesis to identify which specific devices, activities, and settings are associated with differences in student achievement in the 2018 Programme for International Student Assessment (PISA). Estimates are combined across PISA’s multiply imputed datasets using Rubin’s Rules, and the robustness procedures developed in the thesis are carried forward.

• Economics of education
• Multiple imputation
• PISA 2018

Manuscript available soon.

Honors thesis · University of Louisville · 2026

Educational Technology and Cognition: PISA 2018 Evidence on Technology Use and Student Performance

Defended before a three-member faculty committee, Spring 2026 · Presented at the Kentucky Economic Association Annual Conference, October 2026

Using the 2018 PISA data, this thesis examines how the ways students use educational technology relate to achievement in mathematics, reading, and science. I construct a cross-country sample that accounts for differences across assessment booklets and excludes countries with insufficient observations, estimate multivariate regression models and marginal effects conditional on external controls, and test robustness with variance inflation factor diagnostics. Home technology is the strongest statistically significant technology-related predictor of achievement, socioeconomic status accounts for a substantial share of the observed relationship, and the associations differ across subjects.

• Economics of education
• Applied econometrics
• PISA 2018