Empirical analysis on the effectiveness of the legislative framework in the maritime industry

We analyze the effectiveness of the legislative framework of the maritime industry developed by the International Maritime Organization (IMO), the International Labor Organization (ILO) and regional regulators such as the EU or the US and industry activities (inspections). With a unique combination of 310 time series using data of 44 years (1977 to 2020), we use 41 econometric models to highlight the effect of 116 main legislative milestones and their variants. We consider multiple endpoints of interest such as safety, piracy and crime, oil and chemical pollution, biosecurity, and air emissions. The results demonstrate that qualities of the fleet have improved. The time between adoption and entry into force for the events used here lies at 3.83 years on average while negative effects are equally important for the time between adoption and entry into force as compliance often is obtained prior to entry into force of a requirement. Safety related milestones lead with measured negative effects followed by pollution. Good results are found for the fishing fleet albeit only part of the legislative framework related to training and certification (STCW-F) is in force. Regional measures show stronger negative effects due to increased enforcement powers. Many of the individual milestones of interest show that the desired effect of reducing incidents or decreasing pollution can be measured including important milestones such as the IMO International Safety Management Code (ISM), the Instruments Implementation Code (III Code), the Recognized Organization Code (RO), several SOLAS specific codes dealing with stability, fire testing and construction, the IMO Member States Audit Scheme (IMSAS), MARPOL Annex I, II and VI. Although human related areas improved compared to a previous study, there is still room for improvement. Future areas of interest are ship recycling, anti-fouling, sewage, and garbage as well as the London Convention.


Introduction
The maritime industry is regulated by a complex and comprehensive international legislative framework where enforcement can be weak. There are over 50 international conventions from the International Maritime Organization (IMO) as well as several conventions related to working conditions from the International Labor Organization (ILO) conventions as well as regional legislation (European Union, USA) that impact global trade. Incident rates, loss of life rates and pollution indicators (oil and CO2 emissions) show decreasing trends over time (see Figure 1), thus indicating an overall gradual improvement of the quality of the fleet.
To further enhance enforcement and implementation of international conventions, IMO established the IMO Member State Audit Scheme which became mandatory in January 2016 after operating on a voluntary basis since 2005. Audits cover six international conventions and three protocols and are performed on a seven-year cycle. Auditors are nominated by Member States (MS) and are assessed and selected by the Secretariat based on criteria included in the Procedures for the Audit (IMO, 2015). Audit findings are addressed by Member States via a corrective action plan and implementation should be followed up after 3 to 4 years. A Member State can agree to make certain audit reports publicly available, but as a minimum those reports are shared among the Member States (IMO, 2020a) and contain the identification of root causes and corrective actions by Member States. .0002% .0003% .0004% .0005% .0006% . 0007% 1980 1985 1990 1995 2000 2005 2010 2015 2020 % TLVS incidents to total vessels .000% .001% .002% .003% .004% .005% . 006% 1980 1985 1990 1995 2000 2005 2010 2015 2020 % lost lifes to total vessels 0. More recently, the legislative focus has shifted from safety related aspects to reducing environmental impacts and a shift to sustainable maritime transport also reflected in IMO's latest theme in line with the United Nations Sustainable Development goals (UNSD) and the EU Green Deal. This has changed the regulatory landscape. According to Clarksons (Clarksons, 2022), in April 2022, uptake of green technology for fuel transmission is under way and 7.1% (27.7% by GRT) of the world fleet are considered to have "eco engines" installed and 19.7% (61.7% by GRT) have ballast water management systems installed.
It is assumed that a higher degree of enforcement of legislation will lead to the effect of interest with respect to the legislative framework, but this effect is difficult to measure because the operating environment is complex, international in nature and many factors can affect safety and environmental qualities of vessels (Knapp 2006, Knapp and Franses 2007, Bijwaard and Knapp 2009, Knapp and Heij 2020, Heij and Knapp 2012, 2018, 2019. In this study we analyze the effectiveness of the legislative framework of the maritime industry with an emphasis on IMO conventions using a unique combination of 310 time series of data of 44 years (1977 to 2020). Using econometric models, we measure the effect of legislative milestones toward reducing adverse effects that legislation was intended to be created for and account for other factors that can influence safety such as economic cycles, port state control and industry vetting inspection efforts. The time between adoption, ratification by Member States and entry into force is important to consider because it often takes time for the industry to get ready to comply with new requirements. Therefore, on most occasions the effect can already be measured prior to the formal entry into force date.
The analysis is based on a similar approach used by Knapp and Franses (2009) and expands their analysis from 45 legislative milestones to 116 milestones. We also expand on the endpoints of interest to cover all relevant areas including new areas for the environment (ship recycling, biosecurity, and air pollution) and additional endpoints for piracy besides safety, oil and chemical pollution and search and rescue (SAR) covered by Knapp and Franses (2009). A legislative milestone is either the addition of a new convention or protocol or an important amendment to the already existing one. With respect to the areas considered, 28% are related to the environment while 30% are related to safety, search, and rescue (SAR) and piracy, 6% are related to labor and working conditions of crew and the remaining are overlapping between all or parts of areas. Air pollution, biosecurity, and general ocean governance related areas such as plastics are all now areas of focus for IMO but also for the EU due to its political agenda reflected in the EU Green deal. Our analysis does not include anti-fouling and areas related to sewage and garbage due the lack of data available for this endpoint of interest. Another area not covered is the London Convention (dumping at sea) as this is felt to be outside the scope of the present analysis. An attempt is made to cover ship recycling.
Section 2 describes the data sources used as well as the creation of the legislative milestones which form the basis for the analysis. Section 3 describes the methods and model combinations used, followed by a discussion of the results in section 4 and recommendations and conclusions in section 5. Table 1 provides a list of the main data sources that were used to create the data matrix used for the analysis where three main sources are distinguished. As a first source, we incorporate information on the various legislative documents from IMO, ILO, and the EU to identify legislative milestones as well as data from IMO and ILO to create time series of the variables related to it. Second, we have data that is needed to account for the development of the world fleet over time (for each ship type), earnings to account for ship economic cycles and inspections that influence safety qualities. The third data source group is related to the creation of the end points of interest that are used to measure the effectiveness of conventions such as incidents and environmental areas such as oil pollution, chemical pollution, ship recycling, ballast water systems and air pollution related data.

Dataset and creation of milestones used in the analysis
The data is used to create monthly time series starting January 1977 to December 2020a period of 44 years or 528 months. The various types of variables and derivation of dependent variables for the econometric analysis is explained in the method section below. As the legislative requirements sometimes depend on ship type, the data distinguish between the relevant ship types when needed and we consider the following ship types: general cargo, dry bulk, container, tankers (oil tankers, chemical tankers, LPG, LNG), passenger vessels (cruise ships and other passenger ships), fishing vessels and other ship types. Tugs and pleasure yachts (large yachts with IMO numbers) are excluded.
Information about the status of IMO conventions can be obtained from the IMO home page (IMO, 2020b). Data on the ratification of conventions by Member States was obtained from the IMO Secretariat from the Global Integrated Ship Information System (GISIS). Information on ILO conventions ratifications can be accessed via the ILO home page (ILO, 2022). Information on EU legislation was obtained via CELEX (Communitatis Europeae Lex) and the Rule Check system operated by EMSA which has detailed Port State Control requirements of all legislative requirements. The adoption dates and entry into force dates of the selected milestones were mostly populated via Rule Check. In addition, this area was supplemented by general information about the start of various Port State Control Memoranda of Understanding (PMOU) based on data obtained on their websites and or information from Knapp and Franses (2009) on industry milestones such as industry vetting inspections. World Fleet data from IHS-Markit (IHSM) contains information on ship types, age, and size of vessels over time (GRT, DWT). Port state control inspections from IHSM are based on all available inspections since the inception of PSC by the Paris MoU. All MoU's are considered, and the data contain inspections, detentions, and the number of deficiencies and the 1.6 million inspections are aggregated to monthly time series data based on their inspection date. To account for ship economic cycles, monthly earnings by ship types are used from Clarksons and are corrected by inflation (see Bijwaard and Knapp, 2009).
Incident, piracy, and pollution data are not easy to obtain in the maritime industry for such a long period, in particular pollution data. Incident, pollution, and piracy data come from four different sources (IHSM, LLIS -Lloyds List Intelligence, IMO and the United States Coast Guard (USCG)). Before monthly aggregation, the data needs to be reclassified by degree of seriousness such as very serious, serious, and less serious (IMO, 2000) as data providers have different definitions. In addition, the first event of incident is identified so that they can be classified into the correct end points of interest which will be described in section 3.
To measure the effect of environmental endpoints of interest such as biosecurity, recycling and air emissions, proxy variables are used in addition to emissions data. As mentioned, antifouling is the only endpoint of interest not covered in this analysis. The proxy variables are as follows: • percentage of the fleet, total GT or DWT recycled (Clarksons) • percentage of the fleet fitted with ballast water systems (Clarksons) • percentage of the fleet fitted with scrubbers (Clarksons) • percentage of the fleet with hybrid engines (Clarksons) Yearly CO2  Overall, the time between adoption and entry into force for the events identified in Appendix A lies at 3.83 years on average since 1977. As mentioned previously, Appendix A highlights the enhanced focus on environmental areas in the later years influenced by the need to mitigate climate change and supported by the European Union Green Deal. Table 2 provides a high-level overview of the legislation covered in the analysis with their initial adoption year of the earliest event of interest. We also account for legislation related to the fishing fleet which is treated separately and two milestones are considered. With respect to the various conventions covering liabilities, the following are considered in this analysis along with their respective Protocols:  Sea, 1996 Furthermore, unilateral legal instruments have been identified and are also included in the analysis such as the Oil Pollution Act (OPA 90) which is the response of the United States of America to the Exxon Valdez disaster and several directives, and regulations related to the phase out of oil tankers, emission and ship recycling and the ISM code are considered. Other events of interest that influence safety qualities of vessels are the various inspections conducted by ten Port State Control regimes (PSC) and industry vetting inspections (CDI, RightShip, SIRE, Greenaward). For this analysis, inspections will be grouped via Principal Components as explained in the next section. Finally, we also include the entry into force of the United Nations Convention of the Law of the Sea (1983). Table 3 provides a list of the variable groups that are used as a starting point for the analysis. Each variable group represents a set of variables used in the models and which are either the variables of interest, correction factors or dependent variables. The 166 milestones are linked to the outcome variables of interest used for the econometric analysis. Variables of main interest are: 1) Indicators for entry into force of legal instruments and amendments (0 before, 1 from the time of entry onwards), 2) Indicators for interim periods between adoption and entry into force (0 before, 0 after, and 1 in between) and 3) the number of IMO or ILO Member States which have ratified a legal instrument or protocol (count). Correction factors are indicators for seasonality, earnings to account for ship economic cycles over time (deflated) and fleet data which is used to calculate various parameters such as the number of vessels over time including by ship type, average age, and ship sizes over time, which is also needed to adjust dependent variables. A selection of dependent variables is considered depending on the area of interest and legislative requirement.

Base model and model combinations used
Each of the 116 milestones can have either one, two or all three main variables of interest (CR, IN or AD), see Table 3. Appendix B provides the main abbreviations of the milestones, descriptions and grouping into milestone summary areas (for example. safety, pollution, search, and rescue) and link to convention source which will be used in the results sections and for visualization of the findings.
Due to the large number of variables and possible correlations, the number of variables is reduced using Principal Component Analysis and scores (PCs) are created which are added to the model to correct for the effect of these variables. Three types of scores are used as follows (also see equation 1): • PC1 score = the PC score for inspection variables from port state control inspections • PS1 score = the PC score for inspection variables from industry inspections (SIRE, CDI, RightShip, Greenaward) • PS3 scores = the PC score of all other milestones that are not of particular interest in the model (in total 13 scores were created depending on the model type and the milestones that are included in the models which are described in Appendix A).
The model used is an econometric time series model with the addition of the PC scores. In these type of models, serial correlation is common which can lead to an underestimate of the standard errors of the parameters of interest and consequently to an exaggeration of their significance. To correct for serial correlation, two lagged variables of the dependent variable of interest (y) are included in the model denoted y(-1) and y(-2). As method of estimation, ordinary least squares (OLS) is used and the standard errors of the parameters of interest are estimated using Newey-West HAC (heteroskedasticity and autocorrelation consistent) standard errors and covariance option in EViews. The resulting basic model is given in equation 1: y = β + β1x + β2y(-1) + β3y(-2)+ β4PC1 + β5PC2 + β6PC3 + ε (1) where y is the dependent variable of interest (also see Table 4), β is the intercept, x are the explanatory variables with in vector β1 the main coefficients of interest, y(-1) and y(-2) are the lagged variables of y with β2 and β3 their respective coefficients, and finally ε are the residuals. PC1, PC2 and PC3 are the principal component scores explained previously. PC1 and PC2 is included in all models except in the fishing fleet models and PC3 changes respectively for each model.
Interpretation is concentrated on the parameters of interest (see Table 3) which are IN (indicators for entry into force of legal instruments), AD (indicators for the timing between adoption and entry into force), CR (the number of countries which have ratified a convention). Seasonality variables are also considered. The coefficient β1 of Equation 1 is called the short-run effect of x which is the main parameter of interest. The total or cumulative effect is given by a combination of the parameter of β1 and the parameters of the lagged variables (β2 and β3) in the form of β1/(1-(β2+β3)) only to be computed like this when β2+β3<1. The total effect is a scale-free value, so its absolute value is not interesting, only its relative value to other total effects. Fishing vessels are treated with separate models specific for its legislative framework. For each model type (A to E), a selection of models is presented except for ship recycling where the proxy variables turned out not be suitable. This area is therefore not covered in the present analysis. The resulting 41 total models are split up into five main groups as follows: 11 type A models dealing with general safety and ISM, 7 type B models dealing with the fishing fleet,15 type C models with different incident types as endpoint of interest (for example fire and explosion, collision, and groundings etc.), 2 type D models for loss of life as well as piracy and crime as endpoint of interest and 7 type E models covering oil and HNS pollution, ballast water uptake and air emissions. All final regression models are provided in Appendix C as supplemental material with an emphasis on the variables of interest (CR, AD, and IN).

Results and Discussions
Due to the large number of end models, it is challenging to summarize and present the results. As mentioned previously, Appendix B provides a list of the milestone abbreviations used in the regression outputs, a short description of the milestones and their classification into milestone groups which are used to summarize and visualize the results.
To summarize the results, a similar approach as used by Knapp and Franses (2009) is provided to also allow comparisons with previous results when possible. Most important is to summarize whether the effect of interest could be measured for CR, AD, and INthat is whether the effect could be measured (as statistically significant) and whether it is a positive or a negative effect where negative effects are the desired effect for model type A to E as it reduces the endpoint of interest. The exception to this rule is models E5, E6 and E7 since they measure uptake of a proxy variable such as BWT installations of scrubbers, hence a positive effect is the effect of interest.
The results are first summarized by eight milestone groups (see Figure 2) followed by conventions and then by main milestones themselves supplemented by results from the regression outputs. Based on all models listed in Table 4 and filtering out the respective relevant legislative framework for each model (for example. different ship types have different requirements) and endpoint of interest, a total of 2,341 relevant legislative events of interests are identified for the 41 models. From these events, the effect of 66.2% could not be measured (not statistically significant) while for 16.4%, the desired effect could be measured.
Of these desired non-zero effects (see Figure 2a), 39.1% are related to safety (safety management, technical areas and search and rescue) followed by 33.5% for pollution (oil, HNS, biosecurity and air pollution). Liability coverage with 10.9% combines the total of all the conventions reflecting the liability regime (for example CLC, LLMC, FUND etc.) while human related areas account for 10.8%. To be able to make a comparison with milestones with endpoint of interest of very serious incident (VS), Figures 2b and 2c show the same graph but excluding biosecurity and air pollution as these endpoints of interest do not have a seriousness classification like the incidents have. The difference is related to safety management, pollution (HNS), piracy and crime and human related areas, which show a higher percentage of negative effects for VS compared to all incidents.  Figure 3 shows the percentage of negative effects to total relevant milestones for each of the three endpoints of interest excluding milestones related to biosecurity and scrubber uptake to facilitate interpretation. Entry into force (IN) and time between adoption and entry into force (AD) are of almost equal importance while the number of signatories (CR) is not as important. When grouping this by the responsible regulator (Figure 3b), the percentage of negative effects for AD lies at 50% for the EU and 52.5% for IMO while time in-between adoption and entry into force also shows negative effects for IMO (49.2%) but this effect is less pronounced for other regulators such as the EU, ILO, the US or UNCLOS.  Figure 4, the desired effect is the negative effect except for the area related to biosecurity. A comparison is made with results from Knapp and Franses (2009) Figure 5 provides the results by main legislative area based on the present analysis. The results per convention show that even though parts of the legislative framework for the fishing fleet is not yet in force, 35% of the relevant events show a negative effect which is more than most other conventions. This is followed by the Oil Pollution Act (1990) from the US (33.3%) and EU legislation (23.5%). For IMO-BWM (biosecurity), a positive effect is desirable. Regional implementation is often stronger due to the enforcement mechanism compared to international legislation. With respect to IMO conventions, MARPOL (Annex I and II combined) leads followed by piracy measures and MAPOL Annex VI. ILO and STCW are in the middle field. Not surprisingly, some of the older conventions such as Load Lines, Tonnage, COLREG and SOLAS show lower negative results because they have been in force for a very long time and most member states are signatories and improvement in safety qualities is less measurable compared to new areas such as the environment. Figure 6 provides more detailed results by milestones (grouped together within the same area such as for instance all milestones related to air emissions or STCW etc.) and the resulting table is split between international conventions (IMO, ILO, and UN) and regional requirements (EU and US  : Results by milestone sorted by negative effects (international and regional)

Note: for the Ballast Water Convention, a positive effect is the desirable effect while for all other areas, a negative effect is of interest
At the regional level and as mentioned before, OPA90 in the US is leading followed by measures from the EU for air emissions and the ISM Code.
Complementing these findings with specific results from the regression outputs from Appendix C (supplemental material), negative effects at the 1% significance level for either CR (number of signatories), IN (entry into force) or AD (time between adoption and entry into force) are of particular interest (for type A models): • For endpoint total loss: the 1996 Protocol of the ILO Minimum Standards Convention (1976), EC Regulation 1726/2003 (accelerated phase out of single hull), MARPOL Annex I -Regulation 12A and MARPOL Annex II, the Oil Pollution Act (OPA90) of the US and IMO's piracy focus in the Gulf of Guinea.
• For endpoint very serious incidents: MARPOL Annex II and SOLAS -Goals based standards for tankers and dry bulk (2010).
• For endpoint serious incidents: MARPOL Annex I, the Bunker Convention (2001) Interesting to note is that various milestones related to air emissions are also significant and negative confirming that environmental areas overlap with safety and that measures that improve the environmental quality of vessels also show an additional effect for safety. This is understandable as a ship owner who invests in environmental compliance will also invest in safety related areas.
For type B models for the fishing fleet, the relevant endpoints of interest such as the Cape Town agreement (2012) albeit not yet in force and the STCW-F (for AD and IN) in force since 2012 show negative effects for various endpoints of interest (very serious, serious, less serious, all incidents, hull related failures (including wrecked, stranded, grounding, flooding, foundering and capsizing) and pollution. Pollution data for the fishing fleet is very limited and this result is to be interpreted with caution.
For type C models with respect to incident type models (all degrees of seriousness), the following are negative effects of results of interest (1, 5 or 10% significance level): • For type E models with respect to pollution (oil, HNS and air emissions), the following are negative effects of results of interest (1, 5 or 10% significance level): • For oil pollution and HNS pollution: SOLAS goal-based standards (2010) In summary, many important legislative milestones and IMO Codes (for example ISM Code, III Code, RO Code) and the voluntary and mandatory IMO Member State Audit Scheme contributed towards the reduction of incidents and pollution in general including specific incident types. This is further complemented by regional legislation such as EU Directives and Regulations where IMO legislation is transferred into EU law with stronger enforcement. Many environmental milestones also serve towards the reduction of endpoints of interest in safety areas since ship owners that thrive to become environmentally compliant also invest in other areas such as safety.
For older conventions that have been in force for a long time such as Load Lines, COLREG and the Tonnage Convention, it is more difficult to filter out negative effects compared to newer areas such as environmental areas. Based on the individual incident type models and specific ship type models, negative effects could be measured for most specific areas and codesin particular towards oil pollution. Areas of improvement are still related to human related milestones including STCW and ILO Conventions albeit an improvement could be seen compared to the 2009 analysis by Knapp and Franses (2009).

Conclusions and future research
This study analyses the effect of 116 legislative milestones of the international and regional legislative regimes using 44 years of data. Based on 41 econometric models, various endpoints of interest (safety, piracy and crime, oil and chemical pollution, biosecurity, and air emissions) are considered while fishing vessels are treated with separate models specific for its legislative framework. The main conclusions of this study are as follows: The legislative framework is complex and inter-connected where legislative measures can serve multiple endpoints of interest and it is not easy to filter out the events of interest. Principal component analysis was used to group and account for the effect of variables that are not directly relevant for a particular endpoint of interest but could still have an indirect effect.
The results demonstrate that qualities of the fleet have improved. The time between adoption and entry into force for the events identified lies at 3.83 years on average since 1977. The time between adoption and entry into force shows equal importance in terms of obtaining the achieved result as ship owners will prepare often far in advance to comply at the time of entry into force.
Negative effects for areas such as safety management, technical areas related to safety and search and rescue are leading with 39.1% followed by pollution with 33.5%. Important milestones are also areas related to the liability regimes as in total, they account for 10.9% of the negative effects.
The results per convention show good results for the fishing fleet albeit part of the legislative framework is not in force at the time of the study. Regional enforcement indicates stronger effects compared to international legislation. Environmental areas such as MARPOL related areas lead, followed by piracy while ILO and STCW are found in the middle field. Older conventions such as Load Lines, Tonnage, COLREG and SOLAS show lower negative results and more effects that cannot be measured compared to all relevant events of interest. Many of the individual milestones of interest show that negative results can be measured including important activities such as the ISM Code, the III Code, IMSAS, MARPOL Annex I and II and Annex VI. Improvement areas are found within ILO and STCW although human related milestones show improvement compared to the 2009 study.
Ship recycling could not be evaluated due to the lack of data available for this endpoint of interest and is identified as further research area of interest to be covered with MARPOL Annex IV (sewage) and V (garbage), anti-fouling and the London Convention which was found to be outside the scope of this analysis. Results for the Ballast Water Convention. Air emissions related to Sox and NOx also have limited results and need to be interpreted with caution due to limited data of the proxy variables used and should be revisited in future analysis.