Clues from networks: quantifying relational risk for credit risk evaluation of SMEs

Financial Innovation

Table 5 Discrimination performance of relational risk in relationship-specific and risk event–specific networks

Model	Metric	BF + CNAP	BF + CNLD	BF + DNAP	BF + DNLD	BF + SNAP	BF + SNLD
LR	AUC	0.723	0.729	0.730	0.734	0.725	0.731
	KS	0.417	0.429	0.427	0.442	0.417	0.425
	H	0.268	0.282	0.269	0.288	0.269	0.278
RF	AUC	0.736	0.744	0.743	0.747	0.737	0.741
	KS	0.442	0.457	0.456	0.455	0.444	0.454
	H	0.283	0.296	0.291	0.296	0.282	0.295
XGB	AUC	0.713	0.715	0.721	0.732	0.716	0.725
	KS	0.410	0.415	0.423	0.439	0.415	0.436
	H	0.250	0.239	0.244	0.276	0.244	0.251

Each metric is measured by the mean of 100 estimates obtained by repeating the outer ten-fold cross-validation procedure ten times. In Appendix 6, Tables 30 and 32 show the 95% and 90% confidence intervals of each mean, and Table 34 shows the mean of 50 estimates obtained by repeating by repeating the ten-fold cross-validation procedure five times