Student deliverable
Tier C
Pandemic / Cultural / Methods-Applied
Financial Wellbeing and Teleworking Ability
Empirical Evidence from New Zealand's Labour Force during COVID-19
Empirical Evidence from New Zealand's Labour Force during COVID-19
The 2020 COVID-19 lockdowns split New Zealand's labour force into two trajectories that pre-shock income and education models do not fully explain. Workers in occupations whose tasks transferred to remote performance — call them teleworkable in the Dingel-Neiman sense — reported sustained subjective financial wellbeing through the lockdown period; workers in non-teleworkable occupations (retail, hospitality, manufacturing, healthcare delivery) reported sharper declines that persisted into 2021. The gap survived conditioning on income decile, education, age, and pre-shock employment status.
The conceptual framework is straightforward: **occupational teleworkability** — a job's structural capacity to be performed from home — is exogenous to any individual worker's pre-shock financial behaviour. The Dingel-Neiman teleworkability index, constructed from US O*NET task descriptors and applied to NZ ANZSCO codes via crosswalk, provides a measurable structural property of each occupation. The behavioural channel runs through dual-process accounts of financial-anxiety responses to occupational uncertainty: workers in teleworkable jobs perceive a lower probability of catastrophic income loss during stringency periods, sustaining the deliberate financial-planning behaviour that the routine-financial-decision literature has shown to dominate subjective wellbeing.
Using the Te Ara Ahunga Ora Retirement Commission survey (formerly Commission for Financial Capability) — NZ-representative COVID-19 wellbeing data, the existing analysis applies OLS and probit estimation to recover the teleworkability buffer on the four-category subjective financial wellbeing scale. Demographic controls include income decile, education, age, gender, ethnicity, employment status, and household composition.
We propose extending the existing analysis with a cross-fitted Double-ML estimator that would partial out the full demographic-and-employment covariate vector using random-forest nuisance learners, delivering a credible ATE under far weaker functional-form assumptions than the parametric controls. The paper is positioned as a documentation-stage companion to the sole-parent telecommuting natural-experiment paper (which provides cleaner identification via the lockdown timing) and to the 151-country anxiety paper (which provides the global comparison). Together the three form a pandemic-resilience research line.