Publikationer från Stockholms universitet
Not a member yet
31237 research outputs found
Sort by
Risk, Return, and Reputation: A Comparative Study of Sin Stocks and ESG Portfolios in the European Equity Market
No full textEuropean asset managers hold more than two trillion euros in formally labelled sustainable funds. Yet, seminal research evidence from Hong and Kacperczyk (2009) suggests that equities shunned on ethical grounds, so-called sin stocks systematically outperform in U.S markets. In this thesis we ask whether the same shunned-stock premium persists in Europe, whether ESG outperforms the broader European equity market, and lastly if said sin premium lasts in Europe’s ESG-regulated environment across different market periods. We construct two equal-weighted sin-portfolios: Traditional (tobacco, alcohol, gambling) and Extended (including coal and arms). Additionally, two ESG-portfolios comparators based on high and low levels of ESG-stringency: ESG-Select and ESG-X derived respectively from STOXX Europe ESG leaders Select 30 and STOXX600Europe ESG-X. All four portfolios are benchmarked on our broad European equity market index: STOXX600 Europe. Monthly returns covering a total span from July 1998 to April 2025 are evaluated using CAPM, Fama-French 3 Factor, Carhart-4 Factor and Fama-French 5 Factor model, supplemented by key portfolio metrics, and subperiod tests (Pre-COVID, COVID Era and ESG Momentum). Results show that both sin-portfolios consistently generate significant positive alpha in the broader equity market, while ESG-portfolios underperform across models and periods on the same benchmark. The sin premium exhibits context-dependent variation: it reaches statistical significance during the COVID Era but loses significance during the ESG Momentum window. This study extends Hong and Kacperczyk's (2009) U.S. findings to the European context, demonstrating that the shunned-stock premium persists even in markets with institutionalized ESG norms, while revealing that the premium's magnitude depends on market conditions and ethical exclusion stringency
Understanding Ambiguity: Designing, evaluating, and improving an LLM-powered virtual assistant for ambiguous task resolution in realistic scenarios
No full textIntroduction This thesis investigates how large language model (LLM)-based virtual assistants interpret ambiguous task descriptions, and how their understanding can be improved through targeted techniques. This is crucial for the development of acceptable LLM-powered virtual assistants for retrieval-augmented generation and real-world task solving. The thesis work is motivated by and written in collaboration with the company Destiny Sweden Service Center. Research Question How effectively can an LLM-powered virtual assistant interpret and solve ambiguous instructions in a realistic usage setting? How can ambiguity resolution be improved through targeted modifications to prompt and software design? Method An LLM-based virtual assistant was built on top of an existing cloud communications software platform. The assistant was then evaluated in two iterations. In the first iteration, a baseline version of the assistant was evaluated both qualitatively, to identify areas for improvement to design, and quantitatively, to measure the assistants performance. In the second iteration, the improved version of the assistant was re-evaluated quantitatively in order to determine if the improvements identified in the first iteration had any significant impact on the assistants performance. Results The results indicate that the assistant has achieved a high degree of both measurable performance and effective handling of ambiguous tasks. The results suggest that increasing LLM size had the highest impact on performance, although targeted improvements to design were also significant. Discussion These findings suggest that LLM size is the most important predictor for increasing the ambiguity resolution of virtual assistants. However, targeted prompt design and code constraint techniques seem to elicit significant enhancements to ambiguity resolution, particularly in scenarios when LLM size is limited
Between Belonging and Exclusion : A Case Study of Tied Movers and the Informal Work of Integration in Stockholm
No full textThis thesis contributes to the sociology of migration by examining how civic inclusion is negotiated by tied movers, immigrant spouses who relocate primarily for a partner’s career. In Sweden, where integration ideals are strong but unevenly realized, tied movers often face systemic labor market barriers and symbolic exclusions that limit their civic participation. Using a qualitative online survey of members of the Fika Ladies, an informal network of predominantly tied movers in Stockholm, the study explores how participants engage with the Swedish civil sphere. An abductive thematic analysis reveals that tied movers pursue integration through three overlapping strategies: mastery, change-making, and selective belonging. Selective belonging emerges as a flexible, emotionally calibrated response to the challenges of misrecognition and mobility. Informal networks such as the Fika Ladies provide critical support yet also reflect broader social hierarchies that affect who feels central and who remains peripheral. These findings extend the emplacement concept and complicate normative models of civic inclusion by showing how belonging is negotiated through agency and constraint
Quackademic Dive into Duckburg Drollery : a linguistic analysis of humor creation in Don Rosa’s Donald Duck comics in swedish
No full textI denna uppsats undersöks hur språklig struktur samverkar med humorteoretiska strategier i skapandet av verbal humor i Don Rosas Kalle Anka-serier på svenska. Analysen av humorstrategier i de undersökta instanserna är förankrad i Attardos ’General Theory of Verbal Humor’. Humorinstanserna förekommer i ett multimodalt format där uppfattningen av både text och visuella element påverkas av olika semiotiska resurser. Vid analysen används därför Cohns teori för integrering av text och bild. Den språkstrukturella analysen genomförs med utgångspunkt i systemisk-funktionell grammatik, med generell språkvetenskaplig terminologi vid djupare analys av specifika språkliga strukturer. Analysen visar att majoriteten av de sekvenser som humorinstanserna presenteras i innehåller semantisk och språkstrukturell inkongruens på fler nivåer än den enhet som enligt det humorteoretiska ramverket identifieras som skämtets kärna. Språkliga val tycks samverka med den narrativa strategi som bygger upp hela humorinstansen, och språkstrukturella motsättningar utnyttjas för att skapa humor i flera lager utöver den huvudsakliga föreställningskrocken. De föreställningskrockar som utgör primärskämten baseras huvudsakligen på tvetydiga semantiska roller och homofoni. De språkliga strategierna visar dock att den huvudsakliga enhet som skämtet bygger på kan förstås på djupare plan än vad som innefattas i de humorteoretiska ramverken, samt att bland annat krockar mellan syntaktiska och semantiska argumentstrukturers form och funktion i hela sekvenserna bidrar till skapandet av humor.
Unga flickor som mördar : En kvalitativ studie om offentliga reaktioner på unga flickors förövarskap på medieplattformen TikTok
No full textMellan kulor och krut : En longitudinell studie om utvecklingen av och korrelationen mellan skjutningar och sprängningar i Sverige mellan 2018–2024
No full textSyftet med föreliggande studie är att undersöka och beskriva utvecklingen av och korrelationen mellan skjutningar och sprängningar i Sverige i sin helhet samt uppdelat på regionerna Stockholm, Öst, Väst, Syd och Övriga under perioden 2018–2024. En deskriptiv och jämförande analys av utvecklingen av frekvensen av skjutningar respektive sprängningar har därför genomförts på regional nivå baserat på Polismyndighetens statistik över bekräftade skjutningar och sprängningar, där bivariata korrelationsanalyser utfördes för att statistiskt undersöka sambandet mellan antalet skjutningar och sprängningar. Resultaten visade att regionerna hade delvis olika utveckling både vad gäller skjutningar och sprängningar under 2018–2024. I region Öst och Väst ökade både skjutningar och sprängningar sett till perioden i sin helhet, medan det i region Syd skedde en minskning av både skjutningar och sprängningar. I region Stockholm var det en sammantaget oförändrad nivå av skjutningar, men ökning av sprängningar, och i region Övriga var det en minskning av skjutningar och en sammantaget oförändrad nivå av sprängningar. Samtliga korrelationsanalyser visade inga signifikanta samband års- eller halvårsvis på regional nivå mellan frekvensen av skjutningar och sprängningar. Slutsatsen är därmed att även om det i vissa regioner finns en likartad utveckling mellan skjutningar och sprängningar, är sambanden inte tillräckligt starka eller konsistenta för att tala om en faktisk samvariation i frekvens mellan skjutningar och sprängningar
The Impact of Generative AI Tools in Open-Source Software Development
No full textIntroduction Generative AI tools like ChatGPT and GitHub Copilot transform modern software development workflows with enhanced speed and productivity. Despite the rapid adoption of these tools in corporate settings, little is known about how they work in developing open-source software (OSS), where trust, quality, and transparency are essential. The impact of such tools on OSS ecosystems and their compatibility with the fundamental ideas of sustainability and openness are examined in this thesis. Research Question This study’s central research question is: How do generative AI tools affect OSS communities’ sustainability and productivity? Two sub-questions further explore: (1) what guidelines OSS communities have implemented for responsible AI use, and (2) how projects manage technical debt and maintainer workload introduced by AI-generated code. Method The study employs a quantitative survey methodology distributed to OSS contributors and maintainers via GitHub forums and developer networks. The survey included closed and open-ended questions, capturing perspectives on AI adoption, perceived benefits, risks, and governance. Twenty-one valid responses were collected and analyzed using descriptive statistical methods. Results The results indicate that generative AI tools enhance individual productivity, especially in code generation and refactoring. These gains are often offset by increased review efforts and concerns over code quality. Respondents reported frequent issues with inconsistent AI-generated code, rising technical debt, and difficulty trusting code with unclear attribution. According to 57% of respondents, there are currently no explicit rules governing the use of AI in their open-source communities, and most of them believe that the rules, even if present, are ineffective. Discussion These results imply that although AI tools may be more efficient, their unchecked application may jeopardize open source software’s (OSS) sustainability, transparency, and legal clarity. Without open governance, the benefits of incorporating AI may be jeopardized by higher maintenance expenses and cultural misalignment. The study recommends that OSS communities develop usage guidelines, improve attribution tools, and support maintainers in AI-aware code review. Future research should explore the longitudinal impacts of AI on OSS collaboration and how community-driven policies can mitigate emerging risks
Re-evaluating the Agile Manifesto in the Era of AI: A Systematic Mapping Study
No full textIntroduction: Agile software development with a human-centric focus has been widely used in modern software development. The emerging AI shows multiple impacts on agile. However, a comprehensive understanding of how AI impacts remains limited, specifically on agile values: individuals and interactions, working software, customer collaboration, and responding to change. As AI continues to automate tasks traditionally performed by humans, there is a growing need to reassess whether agile values are still aligned with the evolving role of AI. Research Question: The research question is: "How does AI affect the Agile values: individuals and interactions, working software, customer collaboration, and responding to change?" Method: To offer a comprehensive understanding, the systematic mapping study method is used. Rather than relying on a single search query, a distinct, designed query was constructed for each agile value to retrieve relevant literature from IEEE Xplore, Scopus, and Google Scholar. This approach enabled each agile value to be treated with equal focus and analytical depth. The collected papers were then analyzed individually for each agile value using thematic analysis, focusing on identifying common themes such as benefits and challenges. The results from each analysis were systematically synthesized to form the final findings. A total of 57 papers were selected for full analysis after screening with inclusion/exclusion criteria as well as topic relevance. Results: The results show that AI enhances all four Agile values by improving individual productivity, supporting team communication, streamlining decision making, and expediting development cycles, particularly through code and test generation. Tools such as AI-powered task allocation, meeting assistants, and real-time analytics contributed to more responsive and efficient workflows. However, the study also highlights emerging challenges, including the risk of reduced human interaction due to AI's growing capabilities, the complexity of integration, and the potential for misuse stemming from limited AI literacy. Discussion: These findings underscore the urgent need for responsible AI adoption supported by robust governance frameworks that ensure transparency, accountability, and ethical integrity in AI-augmented Agile environments. Such frameworks would help to maximize the potential benefits of AI integration while navigating challenges. Moreover, the study highlights the necessity of re-evaluating and potentially redefining the core values of the Agile Manifesto—particularly those rooted in human-centric principles, such as “individuals and interactions” and “customer collaboration.” Given that AI agents take roles traditionally done by humans, these values may require reinterpretation to remain relevant and meaningful
