Risk evaluation methods at individual ship and company level.
Safety management and risk profiling to identify substandard ships are of importance to theshipping industry. Whereas current methods rely heavily on detention risk and flag stateperformance, we extend the risk assessment by considering various risk dimensions and byevaluating a wide range of risk factors. Apart from detention risk, we consider also the risk ofvarious types of accidents (total loss, very serious, and serious) and damage (hull andmachinery, cargo, pollution, loss of life, and third party liabilities). Risk factors include shipparticulars like ship type and classes of companies and owners, as well as historicalinformation on past accidents, inspections, and changes of particulars. We present methods tosummarize and visualize various risk dimensions and we pay particular attention to theidentification of potentially risky companies. The empirical results are obtained by combiningrich data sets with information on ship arrivals, inspections, and accidents for the period from2006 to 2010. The presented methods and results are of interest to various stakeholders in theindustry, such as charterers, insurance companies, maritime administrations, and theInternational Maritime Organization
A ‘+’ (‘–’) denotes a positive (negative) effect that is significant at 5% level; ‘BM’ denotes a benchmark group,and ‘n/a’ means that a variable does not apply. ‘DB’ denotes Dry Bulk, ‘TLVS’ total loss and very serious, ‘S’serious, ‘HM’ hull and machinery, ‘CAR’ cargo (all ship types except passenger vessels), ‘TPL’ third partyliability, ‘POL’ pollution (for tankers), and ‘LL’ loss of life (for passenger vessels). ‘IACS’ means InternationalAssociation of Classification Societies.
8 but it decreases detention risk. Dry bulk vessels, container ships and tankers tend to haverelatively smaller accident risk as compared to general cargo vessels. Of the four consideredcompany groups, companies with unknown residency carry higher accident risk, andcompanies in the transition and developing residency groups carry lower risk. As concernsowner effects, accident risk is the lowest for the developed group. Changes in ship particularsand past risk events result in increased risk.The model-based risk probabilities can be compared with the empirical riskfrequencies. Logit models have the property that the average of the estimated individualprobabilities is equal to the empirical risk rate, defined as the number of events divided by thetotal number of observations. After aggregation to company level, this does no longer holdtrue because the companies differ in size (number of inspections or number of vessels). Table1 provides information on the model-based risk at company level, which is obtained byaveraging the individual model probabilities per company. For the inspection and accidentdata, the average is taken respectively over all inspections and over all vessels of thecompany. The model-based mean detention risk (7.4%) is slightly smaller than the empiricaldetention risk (8.3%), whereas the model-based accident risk is slightly higher than theempirical accident risk (0.23% as compared to 0.18% for TLVS, and 1.59% as compared to1.42% for S). These differences arise because large companies have larger weight than smallcompanies before aggregation, whereas all companies have equal weight after aggregation.Appendix C provides a more detailed risk comparison of small and large companies.
3.2 Visualization of risk dimensions
The risk of a vessel or company can be evaluated in several ways, as was seen in the previoussubsection. A simple tool to visualize the various risk dimensions into a two-dimensionalgraph is the so-called heat plot. One (horizontal) dimension of this graph is for preventiverisk in terms of the detention probability (PDET), and the other (vertical) dimensionintegrates various accident risks in terms of the monetary value at risk (MVR). Heat plots canbe of help to target ships for inspection and also for ISM audits to evaluate companyperformance.The MVR provides an estimate of the expected total monetary value of five damagetypes (hull and machinery, cargo, third party liability, pollution, and loss of life), taking intoaccount both the unconditional probability of an accident (of type TLVS or S) and theconditional probability of each damage type if an accident occurs. The MVR is computed as
all depend on the vessel under consideration, and the involved probabilities are obtained fromthe logit models for accidents described in Section 3.1.Heat plots based on the AMSA arrival data set will be presented in Section 4.3. Theseplots are either for individual arrivals or for company and owner averages, where the averageis taken over all arrivals of the same company or owner. Each vessel (or company or owner)takes a risk position along the detention dimension and along the MVR dimension,corresponding to a point in the heat plot with coordinates (PDET, MVR). The color of thispoint shows the relative risk position, cold (blue) for low risk and hot (red) for high risk.These colors are defined in terms of the empirical two-dimensional cumulative distribution(CDF) of (PDET, MVR) for the set of all vessels. High risk corresponds to large CDF values,and low risk to small CDF values. The risk graduation is also expressed in terms of fournumerical values, denoted by SW, NE, W, and S. Here SW (‘south-west’) is the CDF, that is,the percentage of vessels having smaller risk along both dimensions, that is, with detentionrisk smaller than PDET and with monetary value at risk smaller than MVR. In a similar way,percentages of vessels are given in the regions NE (‘north-east’, with larger risk along bothdimensions), W (‘west’, with smaller detention risk), and S (‘south’, with smaller MVR). Forexample, if the two risk dimensions are independent, then a vessel with median values forPDET and MVR will have W = 50, S = 50, SW = 25, and NE = 25