Vessel valuation

Valuation accuracy report

Our valuations, scored against reality. Every model is backtested walk-forward on real sales it never trained on, so you can see how close the values land before you rely on one.

Accuracy-weighted error

±20.4%

mean absolute % error, out-of-sample

Within ±20% of price

68%

share of held-out sales

Real sales tested

4,519

transactions the models never saw

Segments covered

bulker · tanker · container · gas

Bulkers

fit on 6,738 sales · as of 24 Sept 2026

Model	Method	MAPE	Within 20%	Tested
AGE-CURVE	Statistical	±17.6%	72%	2,907
COMPS	Statistical	±17.8%	69%	2,906
GBM	Machine learning	±17.9%	70%	2,907
KNN	Machine learning	±24.3%	61%	2,907
REPL-COST	Fundamental	±34.3%	35%	2,907

Best model lands within ±17.6%, beating a naive size-bucketed $/dwt baseline (±43.8%) by 26.2 pts.

Tankers

fit on 4,005 sales · as of 24 Sept 2026

Model	Method	MAPE	Within 20%	Tested
GBM	Machine learning	±22.7%	63%	1,418
COMPS	Statistical	±22.9%	65%	1,417
AGE-CURVE	Statistical	±30.9%	50%	1,418
REPL-COST	Fundamental	±32.6%	42%	1,418
KNN	Machine learning	±36.7%	51%	1,418

Best model lands within ±22.7%, beating a naive size-bucketed $/dwt baseline (±43.7%) by 21.0 pts.

Container ships

fit on 257 sales · as of 24 Sept 2026

Model	Method	MAPE	Within 20%	Tested
GBM	Machine learning	±49.3%	31%	150
COMPS	Statistical	±58.7%	25%	150
AGE-CURVE	Statistical	±79.1%	33%	150
KNN	Machine learning	±85.9%	22%	150
REPL-COST	Fundamental	±178.5%	5%	150

Best model lands within ±49.3%, beating a naive size-bucketed $/dwt baseline (±466.4%) by 417.1 pts.

Gas carriers

fit on 129 sales · as of 24 Sept 2026

Model	Method	MAPE	Within 20%	Tested
KNN	Machine learning	±29.2%	46%	44
COMPS	Statistical	±29.7%	54%	43
AGE-CURVE	Statistical	±32.6%	48%	44
REPL-COST	Fundamental	±37.2%	50%	44
GBM	Machine learning	±48.0%	27%	44

Best model lands within ±29.2%, beating a naive size-bucketed $/dwt baseline (±51.2%) by 22.0 pts.

How this is measured

Walk-forward. Each model is refit on older sales and scored only on more recent sales it never trained on. No sale is ever tested against itself.

MAPE. The mean absolute percentage error between the model's value and the real price achieved. Lower is better; ±20% is the industry-relevant band.

Baseline lift. We compare against a naive size-bucketed $/dwt rule. The gap is the value the models add over a back-of-envelope estimate.

Figures recompute as new transactions settle, so accuracy tracks the current market rather than a frozen snapshot. Accuracy is a track record, not a guarantee: any single valuation carries the error range shown on its report.

Value a specific vessel

Run the full multi-model report on a named ship or an arbitrary spec.

Open the valuation calculator