Introducing Classification-based Financial Markets Prediction DNN for after-work quants

Takeaway. Dixon–Klabjan–Bang treat next-move prediction as a classification problem: a deep neural net maps lagged market features to up/down (or multi-class) labels—so you can reuse classification metrics instead of chasing noisy return R².

Classification-Based Financial Markets Prediction Using Deep Neural Networks (Matthew Dixon, Diego Klabjan, Jin Hoon Bang, 2017; arXiv:1603.08604) is an early after-work DNN-for-markets paper.

Fit for learners: Open-PDF classic/working paper on ml-finance — useful for after-work Python quant learners building intuition before coding.

This Hub note is a plain-language intro. Not investment advice. Keep factor-zoo scepticism nearby: Harvey–Liu–Zhu.

The problem

Next-period returns are noisy. Regression “accuracy” looks terrible even when direction is slightly better than chance. Learners also mix look-ahead features, unbalanced labels, and accuracy that ignores costs.

The idea / core intuition

• Recast the task as classification of directional (or binned) outcomes rather than point-forecasting returns.
• Feed lagged prices / technical-style inputs into a feed-forward DNN; compare against shallower baselines in the paper.
• Evaluate with classification metrics—and still ask whether a tiny lift survives costs, delay, and regime shifts.
• The research object is the label + architecture + validation recipe, not a live trading bot.

Paper details beat any Hub paraphrase—use this only as a map.

Why after-work Python learners care

This is a clean first DNN homework: sklearn or a tiny PyTorch MLP, train_test_split that respects time, and a confusion matrix. Direction labels are easier to debug than MSE on returns.

Tiny evening exercise: make 200 fake close prices; label each step +1/−1 by next-step sign; fit a tiny MLP on lagged returns with a time-ordered split; print accuracy vs a “always up” dummy—learning only, no live book.

One-evening reading checklist

1. State why they prefer classification over return regression.
2. Name the input window / feature idea at a high level.
3. List two evaluation pitfalls (leakage, class imbalance, costs).
4. Write one simpler baseline you would demand before trusting a DNN chart.

Download PDF

Download PDF

Landing: arXiv abs

Important disclaimer

Not investment advice. This Hub note is educational only. Dixon–Klabjan–Bang is academic ML research—not a recommendation to trade DNNs with real capital.

What this is not

• Not CEF / subsidy framing
• Not Quant or Macro course CTA
• Not an “AI market timer” product pitch

Next step

Single course CTA: SoloQuant Academy’s Introduction to Python Programming—lists, functions, and files first; then time-safe classification notebooks become weekend projects.