Table 1.

Predictive performances for 2019 (log) overall employment levels.

Panel A – Performance on all LLMs
Predictive method	MSE	MEDSE
Corresponding quarter – last year (2018)	0.00101	0.00040
Corresponding quarter – 3-year average (2016–2018)	0.00320	0.00214
OLS	0.08510	0.04799
Regression tree	0.05669	0.02907
Random forest	0.00082	0.00026
Panel B – Performance by population size
≤ 50,000 inhabitants
Corresponding quarter – last year (2018)	0.00120	0.00042
Corresponding quarter – 3-year average (2016–2018)	0.00325	0.00189
OLS	0.08067	0.03839
Regression tree	0.04337	0.02366
Random forest	0.00107	0.00034
Between 50,000 and 200,000 inhabitants
Corresponding quarter – last year (2018)	0.00075	0.00037
Corresponding quarter – 3-year average (2016–2018)	0.00312	0.00224
OLS	0.06415	0.05289
Regression tree	0.04382	0.02943
Random forest	0.00052	0.00021
≥ 200,000 inhabitants
Corresponding quarter – last year (2018)	0.00090	0.00042
Corresponding quarter – 3-year average (2016–2018)	0.00324	0.00274
OLS	0.20479	0.18258
Regression tree	0.19768	0.08120
Random forest	0.00056	0.00020

Notes: Estimates on the 2019 sample. MSE stands for Mean Squared Error; MEDSE for Median Squared Error.