[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-119761-en":3,"doc-seo-119761-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},119761,549758146520,"Patrick","https://ap-avatar.wpscdn.com/avatar/80002397d8c0411e94?_k=1775819394049821470",8,"Research & Report","Validation of machine learning based scenario generators","Machine learning methods are increasingly used to build internal models via scenario generation, requiring validation that aligns with Solvency 2 supervisory expectations. This work analyzes how validation for data-driven scenario generators differs from classical theory-based models, focusing on market risk. It introduces two additional validation tasks: checking dependencies between risk factors and detecting an unwanted memorizing effect. Applied to an ML-based economic scenario generator, the measures produce reasonable results for market risk modeling and support both validation and model optimization.","Validation of machine learning based scenario generators  \nSolveig Flaig∗ †, Gero Junike‡  \n18.08.2023  \narXiv :2301 . 12719v2 [ q-fin .RM] 25 Aug 2023  \nAbstract  \nMachine learning methods are becoming increasingly important in the development of internal models using scenario generation. As internal models have to be validated under Solvency 2, it is imperative to understand how effective validation of these data-driven models differs from that carried out with respect to classical theory-based models. Using the specific example of market risk, we discuss the necessity of two additional validation tasks: one to check the dependencies between the risk factors used and one to detect the unwanted memorizing effect. The first task is necessary because, in this new method, the dependencies are not derived from a financial-mathematical theory but are data driven. The need for the latter task arises when the machine learning model merely repeats empirical data rather than generating new scenarios. We apply these measures to a machine learning based economic scenario generator and show that the measures lead to reasonable results for market risk modeling and can be used for validation as well as for model optimization.  \nKeywords Nearest neighbor distance, market risk modeling, Solvency 2, machine learning JEL classification C14, C45, C63, G22  \n∗ Corresponding author. Deutsche Rückversicherung AG, Kapitalanlage / Market risk management, Hansaallee 177, 40549 Düsseldorf, Germany. E-Mail: solveig.flaig@deutscherueck.de.  \n†Carl von Ossietzky Universität, Institut für Mathematik, 26111 Oldenburg, Germany.  \n‡Carl von Ossietzky Universität, Institut für Mathematik, 26111 Oldenburg, Germany. E-Mail: [gero.junike@uol.de](gero.junike@uol.de).  \n1 Introduction  \nRecently, a new class of internal models for scenario generation in the insurance and banking industry has arisen: one that uses machine learning (ML) methods. Studies that have looked at deriving a value at risk based on financial data using neural networks include Kondratyev and Schwarz (2019), Tobjörk (2021), Geller and Hainaut (2021), Fiechtner (2019) and Flaig and Junike (2022) . Regulators in several countries, too, have authored papers that address considerations associated with the use of ML methods in internal models. For example, in 2019, the Nederlandsche Bank issued a paper discussing “how AI is currently already being used in finance, and what potential applications we might expect in the near future”, see van der Burgt (2019) . In it, the authors state that future ML applications will “use broader and better data to develop predictive risk models”, which they anticipate will “increase the precision of risk assessment”, see van der Burgt (2019, p. 28 and 29) . For their part, the German regulators, Bundesbank and BaFin, in a similar paper, BaFin (2021), assert that “the use of ML methods can help to quantify risks more accurately and enhance process quality, thereby improving financial firms’ risk management”; nonetheless, these authors also note that “the limited transparency of the model’s behavior has consequences for the [...] model validation”, see BaFin (2021, p. 7) . Regulators in other countries, including France (see Dupont et al. (2020)) and Great Britain (see Jung et al. (2019)) have also issued papers on this topic.  \nOne of the key processes for internal models in insurance companies is validation, see European Union (2009, Art. 124) . BaFin (2021, p. 11), for example, notes that “supervisors are focusing on any new or much more pronounced risks that arise from ML methods. These are revealed in the data basis, validation [...], model changes and management.” Therefore, it is important to determine what additional validation measures need to be undertaken in order to prove the validity of a given model when the scenario generator is ML-based.  \nIn response to this need, the present research, using the example of a market risk scenario generator, provides a ","cbCaieiOE9osQfBp","https://ap.wps.com/l/cbCaieiOE9osQfBp","pdf",1148746,1,17,"English","en",105,"# Introduction\n## Why validation matters for ML-based internal models\n## Two validation tasks for market risk scenario generators\n# Dependency validation via nearest neighbor distance\n## Data-driven dependence vs. theory-based dependence\n# Detecting the memorizing effect","[{\"question\":\"Why are additional validation measures needed for ML-based scenario generators under Solvency 2?\",\"answer\":\"Because dependencies in ML models are data-driven rather than derived from financial-mathematical theory, and generative models may replicate training data through a memorizing effect. These issues require targeted validation beyond classical approaches.\"},{\"question\":\"What are the two additional validation tasks proposed for market risk modeling?\",\"answer\":\"The paper proposes (1) a dependency check between the risk factors used in the generator and (2) a task to detect the unwanted memorizing effect.\"},{\"question\":\"How does the approach help validate and optimize the ML-based scenario generator?\",\"answer\":\"The introduced measures lead to reasonable results in market risk modeling and can be used for both validating model behavior and supporting hyperparameter or architecture optimization.\"}]","Validation of machine learning based scenario generators | PDF",1785726168,43,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"validation-of-machine-learning-based-scenario-generators","",{"@graph":36,"@context":85},[37,54,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/validation-of-machine-learning-based-scenario-generators/119761/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":62,"encodingFormat":61,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-03",true,{"@type":65,"interactionType":66,"userInteractionCount":4},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why are additional validation measures needed for ML-based scenario generators under Solvency 2?","Question",{"text":75,"@type":76},"Because dependencies in ML models are data-driven rather than derived from financial-mathematical theory, and generative models may replicate training data through a memorizing effect. These issues require targeted validation beyond classical approaches.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What are the two additional validation tasks proposed for market risk modeling?",{"text":80,"@type":76},"The paper proposes (1) a dependency check between the risk factors used in the generator and (2) a task to detect the unwanted memorizing effect.",{"name":82,"@type":73,"acceptedAnswer":83},"How does the approach help validate and optimize the ML-based scenario generator?",{"text":84,"@type":76},"The introduced measures lead to reasonable results in market risk modeling and can be used for both validating model behavior and supporting hyperparameter or architecture optimization.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]