4 présentations

• 
  15h30 - 15h55

  Optimization, statistical and data science methods for investment decision-making

  • Jean-François Bérubé, prés., Caisse de dépôt et placement du Québec (CDPQ)

  CDPQ invests constructively to generate sustainable returns over the long term. As a global investment group managing funds for public pension and insurance plans, CDPQ works alongside its partners to build enterprises that drive performance and progress. CDPQ is active in the major financial markets, private equity, infrastructure, real estate and private debt. As of December 31, 2022, CDPQ’s net assets totalled CAD $402 billion.
  The Quantitative Strategies and Data Science (QSDS) team has a mandate of developing technological and algorithmic capabilities needed to enable quantitative investment strategies and to support the integration of data science insights in discretionary decision-making processes. QSDS regroups professionals with expertise in finance, economics, mathematics, physics, engineering, and computer science.
  The subjects of this session have been selected to give an overview of how CDPQ leverages optimization, statistical and data science methods to enhance its investment decision-making.
• 
  15h55 - 16h20

  Optimized portfolio construction

  • Philippe Gadbois, prés., Caisse de dépôt et placement du Québec (CDPQ)

  Portfolio construction is a standard investment problem for which optimization methods have been applied with success in the last decades. It can be summarized as the identification of optimal combination of securities that maximizes expected returns under risk, diversification, and various exposure constraints. This session will explain how optimization techniques are used by portfolio managers to periodically rebalance asset weights, thus maintaining an optimal portfolio. The presentation will also highlight the pros and cons of using portfolio optimization, both in terms of effective risk management and informed decision-making.
• 
  16h20 - 16h45

  KNN-based forecasting of treasury rate variations

  • Jonathan Guymont, prés., Caisse de dépôt et placement du Québec (CDPQ)

  The K-nearest neighbors (KNN) algorithm can be used to forecast significant variations in the US Treasury rate and define a classification of the direction and magnitude of these variations. The results demonstrate the effectiveness of the KNN algorithm for this forecasting problem and highlight the importance of carefully selecting the metric for feature selection, as it can significantly impact the performance of the model. The transparency of KNN is appreciated by stakeholders, but the trade-off between capacity and explainability must be handled appropriately.
• 
  16h45 - 17h10

  ProspeXia: Making better investment decisions using machine learning

  • Guillaume Provencher, prés., Caisse de dépôt et placement du Québec (CDPQ)

  ProspeXia is a proprietary stock ranking system developed with the objective of helping portfolio managers make better investment decisions. More precisely, ProspeXia is a machine learning ranker driven by a gradient boosted trees algorithm trained on various financial factors and optimized to predict the performance rank of a company relatively to its peers, over a 12-month horizon. This session will describe the methodology used to train the model and the special cares required when working with financial data (time-series).