New York University



Syllabus for “Machine Learning for Finance” (Fall 2017)Instructor: Igor HalperinWeek 1: IntroductionWhat is Machine Learning and how it is related to Artificial Intelligence?Differences between ML and Statistical ModelingCore paradigms of ML: Supervised, Unsupervised and Reinforcement LearningML in Finance: main applicationsDifferences between ML in Finance and ML for techWeek 2: Mathematical toolsProbability distributionsDecision theoryBayesian probabilitiesInformation theory: entropy and mutual informationThe curse of dimensionalityWeek 3: Linear regression and classification Linear regression modelsThe bias-variance decompositionRegularization Bayesian linear regression Discriminative classification models: logistic regression and Bayesian logistic regressionGenerative classification modelsRegression use case: prediction of company earningsWeek 4: Decision tree modelsBuilding Decision TreesClassification and Regression Trees (CART)Random Forest Trees Ensemble learning: Boosted TreesClassification use case: prediction of loan defaults with trees Week 5: Support Vector MachinesStatistical learning theory: learning with generalization guarantees Maximum margin separationKernel trickSupport Vector Machines (SVM) for classificationSVM for regression: Support Vector Regression (SVR)Are SVMs good for a large-scale ML?Regression use case: stock return prediction with SVR Week 6: Feed-forward neural networksPerceptron Multi-layer (feed-forward) neural networks: universal function approximationError backpropagationOptimization algorithmsDeep neural networks (deep learning)Neural network use case: Applying Deep Learning to mortgage defaultsWeek 7: Unsupervised learning and clustering methods Nearest Neighbor Methods (KNN, KD-trees)K-means clusteringHierarchical clustering methodsSpectral clusteringSelf-Organized Maps (SOM)Manifold learningClustering use case: Hierarchical clustering of stocksWeek 8: Latent variable models and dimensionality reduction Factor analysisPrincipal Component Analysis (PCA)Independent Component Analysis (ICA)Gaussian mixture modelsExpectation Maximization (EM) algorithmDimensionality reduction use case: analysis of stock returns with PCA and ICAWeek 9: Graphical models Undirected graphical models (Markov random fields)Directed graphs: Bayesian networksInference in graphsLearning in graphical modelsGraphical models use case: Bayesian network for loan applications Week 10: Unsupervised feature learning and deep learning The grand promise of deep learning: hand-engineered features no more!Restricted Boltzmann Machine (RBM) Deep Boltzmann MachinesDimensionality reduction with neural networks: the auto-encoderNeural networks use case: analysis of stocks with auto-encodersNeural networks use case: regime change detection with deep learningWeek 11: Sequence modeling with Hidden Markov Models (HMM) and Linear Dynamic Systems (LDS) Markov ModelsHidden Markov Models (HMM)Inference and Learning in HMMExtensions of HMMState-Space Models and Linear Dynamic SystemsInference and Learning in LDSExtensions of LDSParticle filtersWeek 12: Neural networks for sequence modeling Recurrent Neural Networks (RNN)Long-Short Term Memory (LSTM) Networks Training RNN and LSTMGenerative Adversarial TrainingUse case: Prediction bank closure with RNN and LSTMWeek 13: Reinforcement Learning with discrete actions Q-learningRL approach to option pricing in FinanceFitted Q-iteration with neural networksWeek 14: Reinforcement Learning with continuous actions RL with continuous actions with linear approachesPolicy gradient methods with neural networksApplications for stock trading and asset managementCourse projects and programming assignments:Students are expected to do all programming assignments and one of the course projects.Programming assignments:Prediction of company earningsPrediction of loan defaults with treesStock return prediction with Support Vector RegressionAnalysis of stock returns with PCA and ICAAutoencoder for stock returns analysisRegime change detection with variational autoencoderStock returns modeling with RNNLSTM for stock returns modelingCourse project: Autoencoders and momentum strategies in stock tradingIntraday stock price dataAutoencoder for feature extractionApplications to momentum strategies in stock tradingCourse project: Reinforcement learning for asset managementPolicy gradients with linear architecturesPolicy gradients with neural networksNo-regret learning and portfolio optimizationCourse project: Modeling mortgage defaults Mortgage data from Fannie May and Freddie MacModeling defaults with decision treesUsing SVM to predict mortgage defaultsModeling defaults with neural networksDeep learning for mortgage defaults Pre-requisitesLinear algebra Basic calculus Basic probability theoryA prior knowledge of Finance is not requiredPython programming skills (including numpy, Pandas, and iPython/Jupyter notebooks) For a refresher on linear algebra and probability theory in an amount needed for this course, see e.g. Chapters 2 and 3 in Goodfellow et. al., “Deep Learning” (2016) TextbooksNo single textbook that would cover everything for this course. Sorted by the frequency of references in the course, the list is as follows.Main references: C. Bishop, “Pattern Recognition and Machine Learning” (2006)I. Goodfellow, Y. Bengio, and A. Courville, “Deep Learning” (2016).A. Geron, “Hands-On Machine Learning with Scikit-Learn and Tensorflow” (2017)Additional references:S. Marsland, “ Machine Learning: An Algorithmic Perspective” (2009) K.P. Murphy, “Machine Learning: A Probabilistic Perspective” (2012) D. Barber, “Bayesian Reasoning and Machine Learning” (2012) N. Gershenfeld, “The Nature of Mathematical Modeling” (1999) ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download