Name: Machine Learning for Trading A disciplined workflow from research to live execution, with nine case studies and Al agents (Stefan Jansen)（Z-Library）
Author: Unknown Author

Page 1

(This page has no text content)

Page 2

Machine Learning for Trading Third Edition A disciplined workflow from research to live execution, with nine case studies and AI agents Stefan Jansen

Page 3

Machine Learning for Trading Third Edition Copyright © 2026 Packt Publishing All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, without the prior written permission of the publisher, except in the case of brief quotations embedded in critical articles or reviews. Every effort has been made in the preparation of this book to ensure the accuracy of the information presented. However, the information contained in this book is sold without warranty, either express or implied. Neither the author, nor Packt Publishing or its dealers and distributors, will be held liable for any damages caused or alleged to have been caused directly or indirectly by this book. Packt Publishing has endeavored to provide trademark information about all of the companies and products mentioned in this book by the appropriate use of capitals. However, Packt Publishing cannot guarantee the accuracy of this information. Portfolio Director: Sunith Shetty Relationship Lead: Tushar Gupta and Apeksha Shetty Project Manager: Shashank Desai Content Engineer: David Sugarman Technical Editor: Seemanjay Ameriya Copy Editor: David Sugarman Indexer: Tejal Soni Production Designer: Ganesh Bhadwalkar Growth Lead: Anjitha Murali First published: December 2018 Second edition: July 2020 Third edition: July 2026 Production reference: 1160726 Published by Packt Publishing Ltd. Grosvenor House 11 St Paul’s Square Birmingham B3 1RB, UK. ISBN 978-1-80324-697-0 www.packt.com

Page 4

I would like to dedicate this book to Mariana and Bastian, who are my source of happiness and who put up with me and the many hours I missed while I wrote this book. – Stefan Jansen

Page 5

Foreword I’ve read a lot of quant finance books, and most of them fall into the same trap: pile on the algorithms, sprinkle in some market data, and call it a day. Stefan Jansen’s third edition of Machine Learning for Trading doesn’t do that. It’s grounded in a way that’s almost refreshing—more of a working practitioner’s notebook than a textbook. The whole thing assumes markets are messy, non-stationary, and out to ruin your backtest, and it builds from there. A few things stuck with me. The book cares about workflow more than algorithms. Jansen treats ML in trading as a lifecycle, not a model hunt. There’s a clear line drawn between building your data infrastructure and the day-to-day grind of researching strategies. That separation sounds obvious until you realize how many people skip it. The data chapters don’t pull punches. The section on data quality is the part I’d hand to anyone starting out. Point-in-time errors, survivorship bias, the headache of corporate actions—Jansen walks through exactly how each one fabricates “alpha” that was never really there. Microstructure is explained like it matters, because it does. The breakdown of the limit order book and how prices actually form connects the tidy theory to the ugly reality of execution. Spreads, queue position, the friction that quietly eats your returns—it’s all here, and it changes how you think about whether a strategy is even tradable. Alternative data gets sober treatment. Instead of breathless hype, you get a checklist: Is there real sig- nal? Is the data clean? Are there legal landmines? What’s it going to cost to engineer? Useful questions to ask before you sink three months into a dataset. The synthetic data chapter is the one I keep thinking about. GANs, diffusion models, LLMs—all aimed at the quant’s oldest problem: we only get one history to test against. Jansen is genuinely curious about generating more, but he never lets you forget that a generator’s biases become your biases. Three ideas actually shifted how I think: Process beats the “perfect” model. Markets drift, break, and shift regimes. The edge isn’t some pristine model you discover once—it’s a research process disciplined enough to notice when your signal is decaying and flexible enough to change course. Lookahead bias quietly kills everything. Future information sneaks into the past in ways you won’t see: restated macro numbers, ignored SEC filing lags, a universe filtered down to the companies that

Page 6

happened to survive. Each one prints fake profits that vanish the moment real money is on the line. Clock time is the wrong clock. A one-minute bar at the open and a one-minute bar at lunch are not the same animal, because activity isn’t constant. Sampling by volume, dollars, or imbalance instead gives you return distributions that are far better behaved—and far better food for a model. I definitely recommend this book if you want to understand Finance in the modern AI and Agentic world. It’s a definitive guide. Antonio Gulli Sr. Engineering Director, Distinguished Engineer, Office of the CTO, Google

Page 7

Contributors About the author Stefan Jansen is the founder of Applied AI and the author of Machine Learning for Trading. For over a decade, he has built production machine learning systems across finance, insurance, and health- care—from contract intelligence that turns legacy insurance documents into structured logic; to forecasting at scale; to live trading infrastructure; and beyond, into LLMs and agents. He maintains the open-source ml4t-* libraries and the book’s companion repository, which has drawn more than 19,000 GitHub stars. He holds a master’s in economics from Harvard, an MS in computer science from Georgia Tech, and the CFA charter. Writing a book is never a solo effort. My deepest thanks go to my friends and family, who supported me throughout and put up with the long hours this book demanded. I am grateful, too, to the team at Packt who brought this third edition to life: to David Sugarman, who edited the manuscript, and to Apeksha Shetty, Shashank Desai, and Yash Basil, who guided the project through to publication—along with everyone in editorial, production, and design whose work turned a draft into a finished book. My thanks also to the technical reviewer for feedback that improved the book, and to Antonio Gulli for his generous foreword.

Page 8

About the reviewer Haitam Borqane is an AI engineer, open source advocate, and huge coffee enthusiast. He has a vast array of experience in AI application across a variety of fields, from sustainability and agriculture to finance and innovation. Professionally, he develops and deploys robust AI solutions and systems for clients to enhance their processes; on the side, he is an AI fanatic who contributes to open source models for low resource languages, while also raising awareness on the state of AI across different conferences. When not working, you can find him at your nearest cozy specialty coffee spot. I would like to thank the Packt team and author for this opportunity on this amazing book, and I would like to thank my parents and my sister for their constant support.

Page 9

(This page has no text content)

Page 10

Table of Contents Preface xxxix Chapter 1: The Process Is Your Edge 1 1.1 Why process discipline matters ............................................................................................................ 2 Four shockwaves and the fragility of assumptions • 2 A vocabulary for change • 2 Why process trumps models • 3 Process as a research-to-production system • 4 1.2 Introducing the ML4T workflow ............................................................................................................ 5 Data infrastructure • 6 Research and evidence framework • 6 Feature engineering and model development • 7 Strategy design from signals to trades • 8 Deployment and monitoring • 8 The evidence boundary • 9 1.3 Causal inference and generative AI in the workflow ............................................................................... 9 Two practical starting points and two deliverable types • 10 Causal inference as a discipline-enforcing tool • 11 Expanding capability and scaling risk with generative AI • 12 1.4 Keeping up with changing market regimes .......................................................................................... 13 Regime detection with labeling, conditioning, and monitoring • 14 Style regimes with century-long factor data • 14 What the model estimates • 14 Macro regimes and volatility environments • 15 Market validation – Volatility and drawdowns • 16 Real-time detection and risk management • 17 1.5 Independent versus institutional workflows in the real world .............................................................. 18 Solo failure modes that institutions partially avoid • 18 Decision discipline through documentation and checkpoints • 18 Constraints and asymmetric advantages • 19 Compounding returns with reusable infrastructure • 20 1.6 Summary ........................................................................................................................................... 20

Page 11

Table of Contentsx Chapter 2: The Financial Data Universe 21 2.1 A Modern taxonomy of financial data .................................................................................................. 22 Market data – A hierarchy of aggregation • 22 Fundamental data – Drivers of value with release lags • 23 Alternative data – High variety, high validation burden • 23 2.2 The asset-class market data landscape ................................................................................................ 24 Equities • 25 Exchange-traded products • 26 Futures • 28 Options • 28 Digital assets • 29 Foreign exchange • 30 Fixed income • 30 Swaps and OTC derivatives (rates, credit) • 31 Commodities • 31 2.3 A due diligence framework for data sourcing ...................................................................................... 32 General data quality dimensions • 32 Finance-specific failure modes • 33 The vendor ecosystem • 37 Vendor due diligence checklist • 38 Data governance • 39 2.4 Storing data ....................................................................................................................................... 40 What we benchmark and why • 40 Benchmark context and caveats • 40 File-based storage • 41 Embedded analytics databases • 42 Server time-series databases • 42 ASOF join performance • 44 Strategic decision framework • 44 2.5 Summary ........................................................................................................................................... 45 Chapter 3: Market Microstructure 47 3.1 How microstructure impacts price formation ..................................................................................... 48 The frictions faced by liquidity providers • 48 How order types express intent in the limit order book • 49 Market design and intraday regimes • 50 3.2 The anatomy of modern market data feeds .......................................................................................... 50 The data hierarchy – From best quote to full order book • 51 Trades and quotes – The standard data package • 51 Price conventions in this chapter • 52 Sample datasets – From MBO to enriched bars • 53 Message protocols – From human-readable to fast binary • 53 Data integrity challenges • 54 3.3 From raw messages to the limit order book ......................................................................................... 54

Page 12

Table of Contents xi NASDAQ TotalView-ITCH – A case study • 55 Binary parsing at scale • 55 The LOB state machine • 56 Maintaining book integrity • 57 Single-exchange data – A local view • 58 Snapshots and analysis • 58 Empirical findings • 59 LOB stylized facts – Predictive patterns • 60 3.4 The art of sampling ............................................................................................................................ 62 Standard aggregation methods • 62 Information-driven bars • 64 Statistical properties comparison • 67 Trade-offs and selection • 68 3.5 Detecting price jumps in intraday returns ........................................................................................... 69 Continuous and jump variance • 69 The Lee–Mykland Test • 70 What the data show • 70 Why a local volatility adjustment matters • 71 Jumps as features • 71 3.6 Microstructure data quality and sessionization ................................................................................... 72 3.7 Summary ........................................................................................................................................... 73 Chapter 4: Fundamental and Alternative Data 75 4.1 The point-in-time pipeline .................................................................................................................. 76 Why restatements create leakage • 76 Bitemporal storage and as-of queries • 76 The EDGAR sourcing pipeline • 77 Handling data inconsistency • 79 4.2 Entity resolution and mapping ............................................................................................................ 80 Why resolution is critical • 80 A hierarchical resolution approach • 80 Building a master security database • 82 4.3 Fundamentals across the asset-class spectrum .................................................................................... 84 Macro and sovereign data (FX/Rates) • 85 Commodities – Physical market data • 86 Cryptocurrency and digital assets – On-chain fundamentals • 87 4.4 Understanding alternative data ........................................................................................................... 88 The alternative data landscape • 89 The evaluation framework • 89 Build versus buy for alternative data • 91 Data sources and implementation • 91 4.5 Using text data for NLP features .......................................................................................................... 92 Targeting MD&A and risk factors • 92 The engineering pipeline • 93

Page 13

Table of Contentsxii Output artifact • 94 4.6 Summary ........................................................................................................................................... 95 Chapter 5: Synthetic Financial Data 97 5.1 The quant’s dilemma .......................................................................................................................... 98 The multiple testing challenge • 98 Synthetic data as simulation infrastructure • 99 5.2 Evaluating synthetic financial data ..................................................................................................... 100 Does it preserve the relevant data structure? • 100 Does it support the intended use case? • 100 Does it leak training information? • 101 Synthetic-specific failure modes • 101 Practical validation checklist • 102 5.3 Classical simulation baselines ........................................................................................................... 102 Bootstrap methods • 103 Parametric price and volatility models • 103 Limitations and how to use these baselines • 105 5.4 Generative model taxonomy .............................................................................................................. 105 5.5 GANs for financial time series ............................................................................................................ 107 Using TimeGAN for sequential data • 107 Using Tail-GAN for risk-constrained generation • 109 Using Sig-CWGAN for path signatures for temporal fidelity • 110 Using GT-GAN for irregular time series • 111 GAN training challenges • 112 5.6 Diffusion models for financial time series .......................................................................................... 112 The denoising diffusion framework • 113 Why diffusion models fit financial returns • 114 Conditional generation for regime stress testing • 114 Applying Diffusion-TS with interpretable decomposition • 115 Choosing between GANs and diffusion models • 116 5.7 LLMs for structured financial data ..................................................................................................... 117 The GReaT framework • 117 Financial applications • 117 Limitations and risks • 118 5.8 Applying the Fidelity–Utility–Privacy framework ................................................................................ 119 Reading the Diffusion-TS results • 119 Privacy–utility trade-off with DP-GAN • 120 5.9 Summary .......................................................................................................................................... 120 Chapter 6: Strategy Research Framework 123 6.1 From idea to evidence with the ML4T workflow .................................................................................. 124 The live trading loop • 124 The research loop • 125 Illustrating the research framework with case studies • 125

Page 14

Table of Contents xiii 6.2 Mapping strategies and sources of edge ............................................................................................. 126 Strategy families as feasibility filters • 127 Defining edge, alpha, and capacity • 129 Sources of edge as durability filters • 129 Using the map • 132 6.3 Defining the trading setup ................................................................................................................. 132 What the trading setup must fix • 132 Decision schedule and admissible information • 133 When a change requires a new setup version • 135 6.4 Setting objectives and evaluation metrics ........................................................................................... 136 Three metric layers, each with a different role • 136 6.5 Evaluation protocol for time series .................................................................................................... 137 Data leakage and estimation bias • 137 From k-fold cross-validation to time series evaluation • 137 Walk-forward cross-validation as the baseline • 138 Temporal buffers – Label buffer and feature buffer • 139 Holdout test set and rolling retuning • 140 Combinatorial methods and path dependence • 141 Design commitments • 141 6.6 Establishing a baseline checkpoint .................................................................................................... 142 Three preflight sanity checks • 142 The first reference run is narrow by design • 143 6.7 Search accounting and run logging .................................................................................................... 143 Trial taxonomy • 144 Minimum contents of the run log • 144 6.8 Summary .......................................................................................................................................... 145 Chapter 7: Defining the Learning Task 147 7.1 Data preprocessing and encodings ..................................................................................................... 147 Split-aware preprocessing • 148 Outliers and heavy tails • 148 Scaling and representation choices • 149 Handling missing data • 149 7.2 Label engineering ............................................................................................................................. 150 Execution conventions • 151 Fixed-horizon labels • 151 Variable-horizon labels • 153 Overlap, sample dependence, and effective sample size • 155 Label diagnostics before modeling • 156 Meta-labels for confidence-weighted sizing • 156 7.3 Univariate feature–label evaluation ................................................................................................... 157 Correctness screens – The non-negotiable first pass • 157 Evaluating continuous labels • 158 Discrete labels – Evaluating features as classifiers • 160

Page 15

Table of Contentsxiv Nonparametric diagnostics – Quantile and decile analysis • 162 Preliminary feasibility checks • 162 7.4 Search accounting and multiple testing .............................................................................................. 163 Define the searched set • 163 Separate exploration from confirmation • 164 Adjustment methods • 164 Report effect sizes, not only significance • 166 7.5 From correlation to causality ............................................................................................................. 166 Why causal thinking matters for features • 167 What is a directed acyclic graph? • 167 Three structural roles • 168 From DAG vocabulary to falsification checks • 169 Mechanism plausibility checks • 170 Plausibility scorecard • 172 7.6 Summary .......................................................................................................................................... 173 Chapter 8: Financial Feature Engineering 175 8.1 Capturing and configuring the economic drivers ................................................................................ 176 8.2 Price-derived features ....................................................................................................................... 178 Trend and momentum • 180 Reversal and mean reversion to an anchor • 180 Volatility and tail-risk state • 181 Microstructure, volume, and order flow • 184 8.3 Structural and cross-instrument features ........................................................................................... 185 Carry, funding, and term structure • 185 Cross-asset structure and relative value • 187 Options-implied features • 189 8.4 Contextual and slow-moving features ................................................................................................ 191 Fundamentals and characteristics • 191 Time, calendar, and event encodings • 192 Macro and policy state • 193 8.5 Cross-cutting feature types and the limits of direct aggregation .......................................................... 193 When direct aggregation is not enough • 194 8.6 Combining features and controlling search ........................................................................................ 195 Interaction diagnostics • 195 Degrees of freedom discipline • 198 Three implementation choices that change the hypothesis • 199 8.7 Summary .......................................................................................................................................... 199 Chapter 9: Model-Based Feature Extraction 201 9.1 Diagnostics and stationarity features ................................................................................................. 202 Stationarity tests as features • 203 Structural break features • 205 Fractional differencing features • 205

Page 16

Table of Contents xv 9.2 Transforming signals to uncover hidden structure ............................................................................. 207 Kalman filter features • 207 Spectral features • 209 Wavelet features • 210 Path signature features • 211 9.3 Volatility Features ............................................................................................................................. 212 Autoregressive integrated moving average features • 213 Generalized autoregressive conditional heteroskedasticity features • 213 Asymmetric volatility – Exponential GARCH features • 214 Heterogeneous autoregressive volatility features • 215 Re-estimation cadence and online updates • 216 Rough volatility and the Hurst exponent • 216 9.4 Uncertainty features ......................................................................................................................... 217 The uncertainty-as-feature principle • 217 Stochastic volatility features • 218 ARIMA forecast uncertainty features • 219 9.5 Regime features ................................................................................................................................ 220 Observable regime rules • 220 Hidden Markov model features • 220 Markov-switching autoregressive features • 222 Distribution-based regime features • 222 Using regime features downstream • 223 9.6 Cross-sectional and panel features ..................................................................................................... 224 Cross-sectional transforms of temporal features • 224 Relative and benchmark-adjusted features • 224 Pairwise temporal features • 225 Universe-level aggregation • 225 Point-in-time handling in panels • 226 9.7 Summary .......................................................................................................................................... 226 Chapter 10: Text Feature Engineering 227 10.1 Lexical and statistical models .......................................................................................................... 228 Lexicon-based sentiment • 229 Statistical representations using bag-of-words • 229 Critical limitations • 230 10.2 Static embeddings ........................................................................................................................... 231 Learning from local context with Word2Vec • 231 Global co-occurrence statistics with GloVe • 231 Asset embeddings from portfolio data • 232 The polysemy problem • 232 10.3 Sequential models ........................................................................................................................... 233 10.4 Transformers .................................................................................................................................. 234 Query, key, value, and the attention mechanism • 235 Multi-head attention and positional information • 235

Page 17

Table of Contentsxvi Transformer architectures • 236 10.5 The modern feature extraction workflow .......................................................................................... 240 Text-to-signal pipeline contract • 240 The pre-train, adapt, fine-tune cascade • 241 Fine-tuning in practice • 242 Embedding selection and long documents • 244 Text signal families and factor construction • 245 Extract-to-schema feature engineering • 248 Model interpretability • 249 10.6 Summary ........................................................................................................................................ 250 Chapter 11: The ML Pipeline 251 11.1 From inference to prediction ........................................................................................................... 252 Two modeling cultures • 252 What inference requires • 252 Why prediction demands a different approach • 253 11.2 Regularized regression .................................................................................................................... 255 Ridge regression for distributed signal • 255 LASSO regression for sparse signal • 256 Combining both priors with elastic net • 257 Standardization • 258 Hyperparameter optimization with Optuna • 259 Loss function choice • 260 Evaluating regression models • 261 Sample weighting • 262 Training loss versus trading objective • 262 11.3 Predicting direction with logistic regression ..................................................................................... 262 Binary logistic regression • 263 Multi-class extension • 264 Regularization • 264 Probability calibration • 264 From probabilities to trading signals • 265 Handling class imbalance • 266 Evaluating classification models • 266 11.4 Interpreting models with SHAP ........................................................................................................ 267 Game-theoretic attribution with SHAP • 268 Practical application in global and local analysis • 268 Building an economic narrative • 270 Limitations and failure modes • 271 Integration with the walk-forward pipeline • 272 11.5 Quantifying predictive uncertainty .................................................................................................. 272 Split-conformal prediction • 273 Adaptive conformal inference • 275 Conformalized quantile regression • 276 Conformal prediction sets for classification • 277

Page 18

Table of Contents xvii Evaluating calibration quality • 278 11.6 Case study insights .......................................................................................................................... 280 Where the signal is • 280 Choice of regularization • 281 Whether the signal persists • 282 Horizon effects • 283 Direction versus magnitude • 284 From IC to profitability • 284 11.7 Summary ........................................................................................................................................ 284 Chapter 12: Advanced Models for Tabular Data 287 12.1 From decision trees to ensembles .................................................................................................... 288 Random forests – Reducing variance with bagging • 288 The path to boosting • 289 12.2 Gradient boosting machines ............................................................................................................ 289 High performance with boosting • 290 Why GBMs excel on tabular data • 291 Inside the engines – XGBoost, LightGBM, and CatBoost • 291 Practical comparison – Which GBM to choose? • 293 Native feature importance and its limitations • 296 Learning to rank • 296 Enforcing economic logic with monotonic constraints • 297 12.3 Deep learning alternatives for tabular data ....................................................................................... 298 Tabular foundation models • 299 Modern neural baselines – TabM and strong defaults • 299 Retrieval-augmented models – TabR • 300 When to look beyond GBMs – A decision framework • 301 12.4 Advanced hyperparameter tuning with Optuna ................................................................................ 303 The tree-structured Parzen Estimator • 303 The define-by-run API • 303 GBM-specific tuning strategy • 304 Multi-objective optimization • 304 12.5 Model explainability with SHAP ....................................................................................................... 305 Dependence and interaction effects • 305 SHAP-based drift monitoring • 306 The Rashomon Effect – When equally good models disagree • 307 From explanation to uncertainty and robustness • 308 12.6 Case study insights .......................................................................................................................... 308 Where flexibility adds ranking content • 308 What the lift rests on • 310 Choosing the configuration • 310 Whether the lift persists • 311 Regression versus classification • 312 Reading the model • 313 From IC to profitability • 314

Page 19

Table of Contentsxviii 12.7 Summary ........................................................................................................................................ 314 Chapter 13: Deep Learning for Time Series 315 13.1 Recurrent networks and their limits ................................................................................................. 316 The LSTM gating mechanism • 316 Two limitations • 317 13.2 N-BEATS and explicit decomposition ................................................................................................ 318 Blocks, stacks, and doubly residual learning • 318 Basis expansion • 319 Making N-BEATS interpretable • 319 Extensions – N-BEATSx and N-HiTS • 320 Practical caveats • 320 13.3 Attention for time series .................................................................................................................. 321 Adapting Transformers for temporal data • 321 The self-attention mechanism • 322 13.4 Linear baselines versus Transformers .............................................................................................. 323 Permutation-invariance and temporal order • 323 The LTSF-Linear benchmark • 324 Results and diagnostic evidence • 324 Counterpoints and caveats • 325 13.5 Modern Transformer variants .......................................................................................................... 326 Patching as inductive bias with PatchTST • 326 Inverting the dimensions with iTransformer • 327 Covariate-rich forecasting with Temporal Fusion Transformers • 327 13.6 Alternative architectures and foundation models .............................................................................. 328 State space models • 329 Time series foundation models • 331 13.7 A practical framework ..................................................................................................................... 334 Step 1: Establish strong baselines • 334 Step 2: Diagnose the problem • 335 Step 3: Apply the selection matrix • 336 Forecasting paths versus predicting horizon labels • 336 Forecasting libraries • 337 13.8 Quantifying prediction uncertainty .................................................................................................. 338 Estimating uncertainty with MC dropout and deep ensembles • 339 Foundation model calibration • 340 13.9 Case study insights .......................................................................................................................... 341 Where the sequence models add ranking content • 341 Which architectures carry the signal • 342 Whether the predictive uncertainty is reliable • 342 Direct prediction versus forecasting • 343 From IC to profitability • 344 13.10 Summary ...................................................................................................................................... 344

Page 20

Table of Contents xix Chapter 14: Latent Factor Models 347 14.1 Making the case for latent factors ..................................................................................................... 349 Three axes of the factor zoo debate • 349 The recurring core • 351 From the zoo to latent factors • 352 14.2 Extracting latent factors with PCA .................................................................................................... 352 The mechanics of eigendecomposition • 352 Assumptions and design choices • 353 Covariance estimation quality • 354 How many PCA factors are real? • 355 14.3 Eigenportfolios for equity strategies ................................................................................................. 355 Constructing and interpreting eigenportfolios • 356 Applications in quantitative finance • 357 Improving interpretability with hierarchical PCA • 357 Fixing eigenvector instability • 358 Practitioner recipe for two-stage production PCA • 359 14.4 Decoding the yield curve ................................................................................................................. 359 Level, slope, and curvature • 359 Why PCA works for yield curves • 360 Practical application for efficient hedging • 361 14.5 Bridging economics and statistics with advanced models .................................................................. 361 The three-stage latent-factor forecasting adapter • 361 Dynamic betas with instrumented PCA • 363 Finding priced factors with risk-premium PCA • 364 Test assets as a design choice • 365 14.6 The conditional autoencoder ........................................................................................................... 367 Preparing the characteristic panel • 368 Specifying the dual-network architecture • 368 Estimating the CAE • 369 Turning latent factors into forecasts • 370 Tuning, validation, and failure modes • 370 Empirical evidence and replication caveats • 370 14.7 The stochastic discount factor and the supervised autoencoder models ............................................. 371 The stochastic discount factor • 372 Building the SDF with adversarial deep learning • 372 The supervised autoencoder • 373 14.8 Case study insights .......................................................................................................................... 374 Where latent factors add ranking content • 374 Which objective extracts the signal • 375 Dimensionality and stability • 377 Regression versus classification • 378 Choosing a method • 378 14.9 Summary ........................................................................................................................................ 379

Machine Learning for Trading A disciplined workflow from research to live execution, with nine case studies and Al agents (Stefan Jansen)（Z-Library）

Machine Learning for Trading A disciplined workflow from research to live execution, with nine case studies and Al agents (Stefan Jansen)（Z-Library）

Registered users can read the full content for free

AI Reading Assistant

Passage locations

Support Author

Recommended for You