Table 3 Absolute prediction errors by...

Table 3

Absolute prediction errors by node centrality

Centrality quintile	Mean error	Q1 error	Median error	Q3 error	Country
Q1	2.303	0.820	1.701	3.407	Small economy
Q2	2.029	0.642	1.230	2.395	Small economy
Q3	2.152	0.804	1.605	2.707	Small economy
Q4	2.372	0.939	1.840	3.181	Small economy
Q5	2.032	0.572	1.437	2.810	Small economy
Q1	5.340	0.847	2.448	6.885	Large economy
Q2	4.097	0.874	1.927	4.808	Large economy
Q3	4.478	1.056	2.853	5.282	Large economy
Q4	4.230	0.972	2.292	5.106	Large economy
Q5	4.753	0.927	2.247	5.528	Large economy

Centrality quintile	Mean error	Q1 error	Median error	Q3 error	Country
Q1	2.303	0.820	1.701	3.407	Small economy
Q2	2.029	0.642	1.230	2.395	Small economy
Q3	2.152	0.804	1.605	2.707	Small economy
Q4	2.372	0.939	1.840	3.181	Small economy
Q5	2.032	0.572	1.437	2.810	Small economy
Q1	5.340	0.847	2.448	6.885	Large economy
Q2	4.097	0.874	1.927	4.808	Large economy
Q3	4.478	1.056	2.853	5.282	Large economy
Q4	4.230	0.972	2.292	5.106	Large economy
Q5	4.753	0.927	2.247	5.528	Large economy

Note(s): This table reports the mean and interquartile range of absolute income prediction errors across agents, grouped by centrality quintiles in the economic graph. The results indicate that GNN accuracy improves significantly with greater structural embeddedness, reinforcing the informational leverage hypothesis

Source(s): Author's simulations based on GNN-predicted income under partial observability

[ViewLarge]

Sharing Unavailable