The Canada-Colombia Free Trade Agreement (CCOFTA) came into force in August 2011 amidst concerns that the provisions protecting Canadian investment in Colombia could exacerbate the precarious human rights situation. The Agreement concerning Annual Reports on Human Rights and Free Trade between Canada and Colombia was negotiated to address such concerns by enshrining the first ever human rights impact assessment (HRIA) of a free trade and investment agreement (TIA) in an internationally binding instrument. This thesis builds on a growing body of international legal scholarship that has considered the duty of home states of private investors to regulate their activity in the host state so as to prevent them from causing or contributing to human rights and environmental harm. It examines state obligations found in human rights, environmental and general principles of international law to propose that while an obligation might exist for the home state to exercise unilateral regulation of its investors, in the presence of a TIA that could cause or enable private human rights or environmental harm, investor regulation through the TIA can be seen as duty for both the home and host states. In view of the absence of such regulation in the CCOFTA, this thesis will consider if the annual HRIA mechanism is an alternative for preventing human rights and environmental harm caused or enabled by the TIA. It is submitted that while HRIAs of TIAs are a novel concept for which little international practice exists, this mechanism has the capacity to provide concrete evidence of human rights or environmental harm caused or enabled by the TIA, but only if based on a methodological model that uses existing state international human rights law obligations as indicators to measure a change in the human rights situation, draws unequivocal causal links between the investment protection provisions and human rights indicators, and allows for broad public participation, especially from the most marginalized and underrepresented groups in the host state to validate its methodology and findings. While under international law all investment-exporting states might have a duty to conduct HRIA on the effects of a proposed TIA as part of the due diligence to prevent transnational harm, the enshrinement of such assessments in an internationally binding instrument triggers a duty for the home state to, on one hand use the HRIA mechanism to prevent transnational human rights or environmental harm and, on the other hand, structure its annual assessments according to the described model in order to give effect to the duty to prevent. Broad and inclusive participation of the local affected communities from the host state in the HRIA becomes an integral component of the home state duty to prevent that can be expected to reveal any negative effects on the human rights situation from the TIA provisions, as well as the type of action required from both states parties to address them.

Ekonomisk brottslighet är mer lukrativt jämfört med annan brottslighet som narkotika, häleri och människohandel. Tidiga åtgärder som försvårar att kriminella kan använda företag för brottsliga syften gör att stora kostnader för samhället kan undvikas. En genomgång av litteraturen visade också att det finns stora brister i samarbetet mellan svenska myndigheter för att upptäcka grov ekonomisk brottslighet. Idag uppdagas brotten först ofta efter att en konkurs inletts. I studier har maskininlärningsmodeller prövats för att kunna upptäcka ekonomisk brottslighet och några svenska myndigheter använder maskininlärningsmodeller för att upptäcka brott men mer avancerade metoder används idag av danska myndigheter. Bolagsverket har idag ett omfattande register för bolag i Sverige och denna studie syftar till att undersöka om maskininlärning kan användas för att identifiera misstänkta bolag, genom att använda digitalt inlämnade årsredovisningar och information ur bolagsverkets register för att kunna träna klassificeringsmodeller att identifiera misstänkta bolag. För att träna modellen så har stämningsansökningar inhämtats från Ekobrottsmyndigheten som kunnat kopplas till specifika bolag av de inlämnade årsredovisningar. Principalkomponentanalys används för att visuellt visa på skillnader mellan grupperna misstänkta och icke misstänkta bolag och analyserna visade på ett överlapp mellan grupperna och ingen tydlig klustring av grupperna. Data var obalanserat med 38 misstänkta bolag av totalt 1009 bolag och därför användes översamplingstekniken SMOTE för att skapa mer syntetiskt data och för att öka antalet i gruppen misstänkta. Två maskininlärningsmodeller Random Forest och Stödvektormaskin (SVM) jämfördes i en 10 fold korsvalidering. Där båda uppvisade en recall på runt 0.91 men där Random Forest hade en mycket högre precision och med högre accuracy. Random Forest valdes och tränades på nytt och uppvisades en recall på 0.75 när den testades på osett data bestående av 8 misstänkta av 202 bolag. Ett sänkt tröskelvärde resulterade i en högre recall men med en större antal felklassificerade bolag. Studien visar tydligt problemet med obalans i data och de utmaningar man ställs inför med mindre data. Ett större data hade möjligjort ett strängare urval på brottstyper som hade kunnat ge en mer robust modell som skulle kunna användas av bolagsverket för att lättare kunna identifiera misstänkta bolag i deras register. / Economic crimes are more lucrative compared to other crimes as drugs, selling of stolen gods, trafficing. Early preventions that make it more difficult for criminals to use companies for criminal purposes can reduce large costs for sociaty. A litterature study showed that there are large weaknesses in the collaboration between Swedish authorities to detect serious economic crimes.Today most crimes among companies that commit fraud are found after a company has declared bancruptcy. In studies, machine learning models have been tested to detect economic crimes and some swedish authorites are now using machine learning methods to detect different crimes and more advanced methods are used by the danish authorites. Bolagsverket has a large register of companies in Sweden and the aim of this study is to investigate if machinelearning can be used to detect on annual reports that have been digitaly submited and information in Bolagsverket’s register to be able to train classificationsmodels and identify companies that are suspicious. To be able to train the model lawsuits have been collected from the Swedish Economic Crime Authority that can be connected to specific companies through their digitally submited annual report. Principal component analysis is used to visually show differences between the groups suspect companies and not suspected companies and the analysis show that there is an overlap between the groups and no clear clustering between the groups. Because the dataset was unbalanced with 38 suspicious companies out of 1009 companies the oversampling tecnique SMOTE was used to create more synthethic data and more suspects in the dataset. The two machinelearnings models Random Forest and support vector machine (SVM) was compared in a 10 fold crossvalidation. Both models showed a recall on around 0.91 but Random Forest had a much higher precision with a higher accuracy. Random Forest was chosen and was trained again and showed a recall on 0.75 when it was tested on unseen data with 8 suspects out of 202 companies. Lowering the treshold resulted in a higher recall but with a larger portion of wrongly classfied companies. The study shows clearly the problem with an unbalanced dataset and the challanges with a small dataset. A larger dataset could have made it possible to make a more selective selection of certain crimes that could have resulted in a more robust model that could be used by Bolagsverket to easier identify suspicous companies in their register.

This item contains the agenda, minutes, and attachments for the Faculty Senate meeting on this date. There may be additional materials from the meeting available at the Faculty Center.